Datasets:

bzantium

/

LongPerplexity

like 0

a nucleus is at the heart of every atom , and loosely speaking , is a collection of protons and neutrons that interact pairwise , with much smaller , but significant , three - body interactions . we are fortunate to know that the underlying laws governing the strong interactions result from a quantum field theory called quantum chromodynamics ( qcd ) . it is constructed in terms of quark and gluon fields with interactions determined by a local su(3 ) gauge - symmetry and , along with quantum electrodynamics ( qed ) , underpins all of nuclear physics when the five relevant input parameters , the scale of strong interactions @xmath1 , the three light - quark masses @xmath2 , @xmath3 and @xmath4 , and the electromagnetic coupling @xmath5 , are set to their values in nature . it is remarkable that the complexity of nuclei emerges from `` simple '' gauge theories with just five input parameters . perhaps even more remarkable is that nuclei resemble collections of nucleons and not collections of quarks and gluons . by solving qcd , we are expecting to predict , with arbitrary precision , nuclear processes and the properties of multi - baryon systems . the fine - tunings observed in the structure of nuclei , and in the interactions between nucleons , are peculiar and fascinating aspects of nuclear physics . for the values of the input parameters that we have in our universe , the nucleon - nucleon ( nn ) interactions are fine - tuned to produce unnaturally large scattering lengths in both s - wave channels ( described by non - trivial fixed - points in the low - energy effective field theory ( eft ) ) , and the energy levels in the @xmath6be - system , @xmath7c and @xmath8 are in `` just - so '' locations to produce enough @xmath7c to support life , and the subsequent emergence and evolution of the human species . at a fundamental level it is important for us to determine the sensitivity of the abundance of @xmath7c to the light - quark masses and to ascertain the degree of their fine - tuning . being able to solve qcd for the lightest nuclei , using the numerical technique of lattice qcd ( lqcd ) , would allow for a partial unification of nuclear physics . it would be possible to `` match '' the traditional nuclear physics techniques - the solution of the quantum many - body problem for neutrons and protons using techniques such as no - core shell model ( ncsm ) , greens function monte carlo ( gfmc ) , and others , to make predictions for the structure and interactions of nuclei for larger systems than can be directly calculated with lqcd . by placing these calculations on a fundamental footing , reliable predictions with quantifiable uncertainties can then be made for larger systems . during the 1990 s , the nuclear forces were systematized by the hierarchy emerging from the spontaneously broken chiral symmetries of qcd . the resulting small expansion parameters are powers of the external momenta and powers of the light - quark masses normalized to the scale of chiral symmetry breaking , as pioneered by weinberg , first in the meson sector and then the multi - nucleon sector @xcite . in addition to generating nuclear forces that are consistent with qcd , this construction provides the calculational advantage of parametric estimates of the systematic uncertainty introduced by the truncation of the nuclear interactions at a given order in the expansion . the actual ordering of contributions remains a subject of debate even today , with weinberg s chiral expansion of the potential having its peculiar difficulties , as does the ksw expansion of scattering amplitudes @xcite . calculations are being performed at a sufficiently high order where the size of truncation errors are quite small . weinberg s ordering of operators based upon a chiral expansion of the n - body potentials between nucleons has been carried out to n@xmath9lo , which includes contributions to the three - body ( starting at n@xmath10lo ) and the leading four - body interactions ( starting at n@xmath9lo ) ( for a recent review see ref . @xcite ) . during the last several years , nuclear structure calculations have been performed with the chiral nuclear forces , leading to both postdictions and predictions for nuclei to a given order in the expansion , and compared with experiment , e.g. see fig . [ fig : ncsma ] . li and @xmath6b using chiral nuclear forces @xcite . [ image is reproduced with the permission of p. maris.],title="fig:",scaledwidth=50.0%]li and @xmath6b using chiral nuclear forces @xcite . [ image is reproduced with the permission of p. maris.],title="fig:",scaledwidth=50.0% ] the nuclear forces that are presently used in such calculations are constrained by experimental measurements of nn scattering and light nuclei . as the desired precision increases , which requires working to higher orders in the expansion , the number of required experimental constraints increases . eventually , there are too few experimental constraints to practically reduce the systematic uncertainty below some level in any given calculation . however , lqcd calculations are expected to provide a way to constrain the nuclear forces beyond what is possible with experiment , and hence to further reduce the systematic uncertainties in nuclear structure calculations . beyond providing direct calculations of important quantities , lqcd calculations of the light nuclei and nuclear forces can 1 . verify experimental constraints and/or reduce the uncertainties in the constraints imposed by experiment , 2 . constrain components of the nuclear forces that are inaccessible to experiment , for instance the light - quark mass dependences which dictates some of the multi - pion vertices , and multi - neutron forces , 3 . constrain counterterms at higher orders in the expansion to further reduce the systematic uncertainties lqcd is a technique in which space - time is discretized into a four - dimensional grid and the qcd path integral over the quark and gluon fields at each point in the grid is performed in euclidean space - time using monte carlo methods . a lqcd calculation of a given quantity will deviate from its value in nature because of the finite volume of the space - time ( with @xmath11 lattice points ) over which the fields exist , and the finite separation between space - time points ( the lattice spacing , @xmath12 ) . however , such deviations can be systematically removed by performing calculations in multiple volumes with multiple lattice spacings , and extrapolating using the theoretically known functional dependences on each . supercomputers are needed for such calculations due to the number of space - time points and the monte carlo evaluation of the path integral over the dynamical fields . in order for a controlled continuum extrapolation , the lattice spacing must be small enough to resolve structures induced by the strong dynamics , encapsulated by @xmath13 where @xmath14 is the scale of chiral symmetry breaking . further , in order to have the hadron masses , and also the scattering observables , exponentially close to their infinite - volume values , the lattice volume must be large enough to contain the lightest strongly interacting particle , encapsulated by @xmath15 where @xmath16 is the mass of the pion and @xmath17 is the extent of the spatial dimension of the cubic lattice volume ( this , of course , can be generalized to non - cubic volumes ) . effective field theory ( eft ) descriptions of these observables exist for @xmath18 ( the symanzik action and its translation into chiral perturbation theory ( @xmath19pt ) and other frameworks ) and @xmath15 ( the p - regime of @xmath19pt and other frameworks ) . the low - energy constants in the appropriate eft are fit to the results of the lqcd calculations , which are then used to take the limit @xmath20 and @xmath21 . computational resources devoted to lqcd calculations are becoming sufficient to be able to perform calculations at the physical values of the light quark masses in large enough volumes and at small enough lattice spacings to be relevant , but the majority of present day calculations are performed with pion masses of @xmath22 . therefore , most calculations require the further extrapolation of @xmath23 , but do not yet include strong isospin breaking or electromagnetism . in principle , the gauge - field configurations that are generated in lqcd calculations can be used to calculate an enormous array of observables , spanning the range from particle to nuclear physics . in practice , this is becoming less common , largely due to the different scales relevant to particle physics and to nuclear physics . calculations of quantities involving the pion with a mass of @xmath24 are substantially different from those of , say , the triton with a mass of @xmath25 , and with the typical scale of nuclear excitations being @xmath26 . present day dynamical lqcd calculations of nuclear physics quantities are performed with @xmath27 , lattice spacings of @xmath28 and volumes with spatial extent of @xmath29 . lqcd calculations are approached in the same way that experimental efforts use detectors to measure one or more quantities - the computer is equivalent to the accelerator and the algorithms , software stack , and parameters of the lqcd calculation(s ) are the equivalent of the detector . the parameters , such as lattice spacing , quark masses and volume , are selected based upon available computational resources , and simulations of the precision of the calculation(s ) required to impact the physical quantity of interest , i.e. simulations of the lqcd monte carlos are performed . the size of the computational resources required for cutting edge calculations are such that you only get `` one shot at it '' . a typical work - flow of a lqcd calculation consists of three major components . the first component is the production of an ensemble of gauge - field configurations which contain statistically independent samplings of the gluon fields resulting from the lqcd action . the production of gauge - fields requires the largest partitions on the leadership class computational facilities , typically requiring @xmath30k compute cores . present - day calculations have @xmath31 dynamical light - quark flavors and use the wilson , @xmath32-improved - wilson , staggered ( kogut - susskind ) , domain - wall or overlap discretizations , each of which have their own `` features '' . it is the evaluation of the light - quark determinant ( the determinant of a sparse matrix with dimensions @xmath33 ) that consumes the largest fraction of the resources . roughly speaking , @xmath34 hybrid monte carlo ( hmc ) trajectories are required to produce an ensemble of @xmath35 decorrelated gauge fields , but in many instances this is an under estimate . for observables involving quarks , a second component of production is the determination of the light - quark propagators on each of the configurations . the light - quark propagator from a given source point ( an example of which is shown in fig . [ fig : prop ] ) is determined by an iterative inversion of the quark two - point function , using the conjugate - gradient ( cg ) algorithm or variants thereof such as bicgstab , or the most recently developed multi - grid ( mg ) . [ fig : prop],scaledwidth=60.0% ] during the last couple of years , the propagator production codes have been ported to run on gpu machines in parallel . gpu s can perform propagator calculations faster than standard cpu s by an order of magnitude , and have led to a major reduction in the statistical uncertainties in many calculations . there have been numerous algorithm developments that have also reduced the resources required for propagator production , such as the implementation of deflation techniques and the use of multi - grid methods . the third component of a lqcd calculation is the production of correlation functions from the light - quark propagators . this involves performing all of the wick contractions that contribute to a given quantity . the number of contractions required for computing a single hadron correlation function is small . however , to acquire long plateaus in the effective mass plots ( emps ) that persist to short times , lscher - wolff type methods involve the computation of a large number of correlation functions resulting from different interpolating operators , and the number of contractions can become large . in contrast , the naive number of contractions required for a nucleus quickly becomes astronomically large ( @xmath36 for uranium ) , but symmetries in the contractions , and new algorithms ( e.g. ref . @xcite ) greatly reduce the number of operations that must be performed . a further consequence of the hierarchy of mass scales is that there is an asymptotic signal - to - noise problem in nuclear correlation functions . the ratio of the mean value of the correlation function to the variance of the sample from which the mean is evaluated degrades exponentially at large times . however , this is absent at short and intermediate times and the exponential degradation of the signal - to - noise in the correlation functions can be avoided . capability computing resources provided by leadership class computing facilities are used to produce ensembles of gauge - field configurations , while capacity computing resources , both those operated by usqcd and elsewhere are used to perform observable - dependent calculations of correlation functions using these configurations . thus the capability resources enable a multitude of physics calculations to be accomplished with the capacity resources . in the area of cold nuclear physics there is currently a well - defined set of goals , and a program in place to accomplish these goals , as described in one of the 2013 usqcd whitepapers @xcite : * hadron structure * , * hadron spectroscopy * , * hadronic interactions , nuclear forces and nuclei * , and * fundamental symmetries*. before calculations of nuclei can be sensibly undertaken , the mass and structure of the nucleon must be reproduced in lqcd calculations . the spectrum of the lowest - lying hadrons calculated with lqcd is shown in fig . [ fig : spectrum ] , from which we observe that indeed lqcd is postdicting all of the light - hadron masses within uncertainties . [ fig : spectrum],scaledwidth=70.0% ] beyond its mass , one property of the nucleon that is well known experimentally is the forward - matrix element of the isovector axial current , @xmath37 . significant effort has been put into calculating @xmath37 with lqcd , a summary of which is shown in fig . [ fig : ga ] , but the extrapolated lqcd value has consistently been smaller than the experimental value . with calculations beginning to be performed at the physical pion mass , the community is focused on understanding and quantifying the systematic uncertainties in these calculations . . [ image is reproduced with the permission of h .- w . [ fig : ga ] , scaledwidth=70.0% ] a central element of the physics program at jlab is to determine the excited spectra of mesons and baryons , including searching for exotic states that are beyond the naive nonrelativistic quark model of hadrons , but arise naturally in qcd . a critical component of this program is the lqcd calculations of the spectra . they will play a central role in interpreting and understanding the experimental measurements . the spectra of such states is complicated by the presence of open multi - hadron channels and significant formal developments remain to be put in place before rigorous statements about the spectra can be made . calculations at unphysical pion masses have been performed by the jlab lqcd group , examples of which are shown in fig . [ fig : isoscalars ] , and remarkable progress has been made in the identification of states in these calculations . by the jlab lqcd group @xcite [ image is reproduced with the permission of r. edwards . ] [ fig : isoscalars ] , scaledwidth=85.0% ] the aim is to have lqcd predict the exotic spectra of hadrons before , or at the same time as , the gluex experiment at jlab runs , targeting the 2018 milestone hp15 . multi - hadron lqcd calculations are significantly more challenging than single - hadron calculations for a number of reasons , and systems involving baryons are even more challenging . meson - meson systems are the simplest multi - hadron systems , and impressive progress has been made in the recent past , particularly when the lqcd calculations are combined with @xmath19pt . there is little or no signal - to - noise problem in such calculations and therefore highly accurate lqcd calculations of stretched - isospin states can be performed with modest computational resources . moreover , the efts which describe the low - energy interactions of pions and kaons , including lattice - spacing and finite - volume effects , have been developed to non - trivial orders in the chiral expansion . scattering @xcite . [ image in the left panel is reproduced with the permission of h. leutwyler . ] [ fig : pipi2 ] , title="fig:",scaledwidth=56.0% ] scattering @xcite . [ image in the left panel is reproduced with the permission of h. leutwyler . ] [ fig : pipi2 ] , title="fig:",scaledwidth=34.0% ] the @xmath38 pion - pion ( @xmath39 ) scattering length serves as a benchmark calculation with an accuracy that can only be aspired to in other systems . the scattering lengths for @xmath40 scattering in the s - wave are uniquely predicted at lo in @xmath19pt @xcite : @xmath41 while experiments do not directly provide stringent constraints on the scattering lengths , a determination of s - wave @xmath40 scattering lengths using the roy equations has reached a remarkable level of precision @xcite : @xmath42 the roy equations @xcite use dispersion theory to relate scattering data at high energies to the scattering amplitude near threshold . at present , lqcd can compute @xmath40 scattering only in the @xmath38 channel with precision as the @xmath43 channel contains disconnected diagrams which require large computational resources . it is of great interest to compare the precise roy equation predictions with lqcd calculations , and fig . [ fig : pipi2 ] summarizes theoretical and experimental constraints on the s - wave @xmath40 scattering lengths @xcite . this is clearly a strong - interaction process for which theory has somewhat out - paced the challenging experimental measurements . mixed - action @xmath44 lqcd calculations , employing domain - wall valence quarks on a rooted staggered sea and combined with mixed - action @xmath19pt , have predicted @xcite @xmath45 at the physical pion mass . the agreement between this result and the roy equation determination is a striking confirmation of the lattice methodology , and a powerful demonstration of the constraining power of chiral symmetry in the meson sector . however , lqcd calculations at one or more smaller lattice spacings , and with different discretizations , are required to verify and further refine this calculation . the etm collaboration has performed a @xmath46 calculation of the @xmath38 @xmath40 scattering length @xcite , producing a result extrapolated to the physical pion mass of @xmath47 it is interesting to compare the pion mass dependence of the meson - meson scattering lengths to the current algebra predictions . in fig . [ fig : caplots ] ( left panel ) one sees that the @xmath38 @xmath40 scattering length is consistent with the current algebra result up to pion masses that are expected to be at the edge of the chiral regime in the two - flavor sector . while in the two - flavor theory one expects fairly good convergence of the chiral expansion and , moreover , one expects that the effective expansion parameter is small in the channel with maximal isospin , the lqcd calculations clearly imply a degree of cancellation between chiral logs and counterterms . however , as one sees in fig . [ fig : caplots ] ( right panel ) , the same phenomenon occurs in @xmath48 where the chiral expansion is governed by the strange quark mass and is therefore expected to be much more slowly converging . this remarkable conspiracy between chiral logs and counterterms for the meson - meson scattering lengths remains mysterious . vs. @xmath49 ( left panel ) and @xmath50 vs. @xmath51 ( right panel ) . the solid ( red ) curves are the current algebra predictions . [ fig : caplots ] , title="fig:",scaledwidth=45.0% ] vs. @xmath49 ( left panel ) and @xmath50 vs. @xmath51 ( right panel ) . the solid ( red ) curves are the current algebra predictions . [ fig : caplots ] , title="fig:",scaledwidth=45.0% ] lqcd calculations of the meson - meson scattering phase - shifts are much less advanced than of the scattering length . this is because the calculation of the phase shift , @xmath52 , at a given energy , @xmath53 , requires a lqcd calculation of the two - meson correlation function at the energy @xmath53 . generally speaking , a given calculation can determine the lowest few two - hadron energy eigenvalues for a given momentum of the center - of - mass , and that multiple lattice volumes will allow for additional values of @xmath53 at which to determine @xmath52 . the first serious calculation of the s - wave ( @xmath54 ) @xmath38 @xmath40 phase - shift was done by the cp - pacs collaboration with @xmath46 at a relatively large pion mass @xcite , and more recently two groups have performed calculations at lower pion masses @xcite , the results of which are shown in fig . [ fig : pipidelta ] . scattering phase - shift . the left panel shows the results of the lqcd calculations below the inelastic threshold ( @xmath55 ) at a pion mass of @xmath56 @xcite . the vertical ( blue ) line denotes the start of the t - channel cut . the shaded region in the right panel shows the results of the lqcd calculation extrapolated to the physical pion mass using nlo @xmath19pt , while the points and uncertainties corresponds to the existing experimental data . the vertical ( red ) line corresponds to the inelastic threshold . [ fig : pipidelta ] , title="fig:",scaledwidth=45.0% ] scattering phase - shift . the left panel shows the results of the lqcd calculations below the inelastic threshold ( @xmath55 ) at a pion mass of @xmath56 @xcite . the vertical ( blue ) line denotes the start of the t - channel cut . the shaded region in the right panel shows the results of the lqcd calculation extrapolated to the physical pion mass using nlo @xmath19pt , while the points and uncertainties corresponds to the existing experimental data . the vertical ( red ) line corresponds to the inelastic threshold . [ fig : pipidelta ] , title="fig:",scaledwidth=45.0% ] further , in some nice work by the hadron spectrum collaboration ( hsc ) , the first efforts have been made to extract the d - wave ( @xmath57 ) @xmath38 @xmath40 phase shift @xcite . one of the more exciting recent results is the mapping out of the @xmath58-resonance at @xmath56 from the @xmath59 energy - levels using luscher s method , as shown in fig . [ fig : rho ] @xcite . -resonance at a pion mass of @xmath56 @xcite [ image is reproduced with the permission of r. edwards . ] [ fig : rho ] , scaledwidth=70.0% ] calculations of the nucleon - nucleon scattering lengths have been successfully underway for the last decade @xcite for a range of pion masses . recently , lqcd calculations have been performed at @xmath60 that also provide the effective ranges @xcite , the results of which are shown in fig . [ fig : nnscattnplqcd ] . channel extracted from lqcd calculations at the su(3 ) symmetric point , including the fit to the ere at n@xmath10lo . the right panel shows the ratio of the scattering length to effective range , a quantity that is a measure of the naturalness of the system . [ fig : nnscattnplqcd ] , title="fig:",scaledwidth=50.0% ] channel extracted from lqcd calculations at the su(3 ) symmetric point , including the fit to the ere at n@xmath10lo . the right panel shows the ratio of the scattering length to effective range , a quantity that is a measure of the naturalness of the system . [ fig : nnscattnplqcd ] , title="fig:",scaledwidth=50.0% ] also shown are fits to the effective range expansion ( ere ) , including the shape parameter . the scattering length and effective range in the @xmath61 channel determined from the nlo fit to the ere are @xmath62 the shape parameter obtained from the nnlo fit to the ere expansion is : @xmath63 . an interesting aspect of this result is that the ratio of scattering length to effective range , a measure of the naturalness of the system , is @xmath64 , which is to be compared with @xmath65 at the physical quark masses . this leads one to speculate that the deuteron might be unnatural over a large range of quark masses and not just close to the physical values , indicating that it is not finely tuned . this speculation requires precise calculations at lighter quark masses to determine if this is , in fact , the situation . perhaps some of the most important lqcd calculations of late are those of the ground states of the light nuclei , including the deuteron , @xmath9he , @xmath66he and light hypernuclei . [ fig : ball ] shows the binding energy of the deuteron , @xmath9he and @xmath66he @xcite as a function of the pion mass . he ( middle panel ) and @xmath66he ( right panel ) binding energies from @xmath44 lqcd calculations @xcite . , title="fig:",scaledwidth=32.0% ] he ( middle panel ) and @xmath66he ( right panel ) binding energies from @xmath44 lqcd calculations @xcite . , title="fig:",scaledwidth=32.0% ] he ( middle panel ) and @xmath66he ( right panel ) binding energies from @xmath44 lqcd calculations @xcite . , title="fig:",scaledwidth=32.0% ] not only is it exciting to see nuclei emerge from qcd for a range of the light - quark masses , such calculations are crucial in dissecting and refining the chiral nuclear forces . however , it is clear that calculations at lighter pion masses are required , including at the physical pion mass . ) calculated by nplqcd @xcite at a pion mass of @xmath60 . , scaledwidth=70.0% ] a summary of the energy - levels at the flavor su(3 ) symmetry point found in the s - shell nuclei and hypernuclei @xcite is shown in fig . [ fig : nuclearsummary ] . these energy levels are elements of su(3 ) irreps . which allowed , in some cases , e.g. the h - dibaryon , the hypertriton and @xmath67 , for distinct energy levels with the same spin and parity to be determined . such calculations will become somewhat more complicated at lighter quark masses when the up and down quarks are not degenerate with the strange quark . the calculations of nplqcd and those of yamazaki _ et al _ are already shedding light on how the ground - state energies of the light nuclei approach their values at the physical light - quark masses . they are all bound at the heavier light - quark masses and become less bound as the quarks become lighter . in the case of the dineutron , which is bound at @xmath60 , it becomes unbound at some intermediate value of the pion mass , giving rise to a neutron - neutron system with an infinite scattering length . one of the interesting aspects of the nuclear forces to explore is the tensor interaction , responsible for the mixing between the s - wave and d - wave channels in the deuteron channel . there are a series of lqcd calculations that can be performed that will permit an extraction of the sd mixing parameter , @xmath68 using lscher s method @xcite , see ref . @xcite . it is important to understand how a few of the past roadblocks to progress in this area have been recently overcome . one of the roadblocks of the past was / is the `` signal - to - noise problem '' that afflicts states other than the pion . this problem is seen most simply in the single - nucleon correlation function , generated with a three - quark source and a three - quark sink . the variance of this correlation function is dictated by a 3-quark 3-anti - quark source and a 3-quark 3-anti - quark sink , which overlaps with both the @xmath69 and @xmath70 intermediate states ( and all others with the appropriate quantum numbers ) . at large times , the variance correlation function is dominated by the @xmath70 intermediate state , while the single nucleon correlation function is dominated by the single nucleon , giving rise to an exponentially degrading signal . however , at intermediate times , the behavior of the `` signal - to - noise '' is determined by the overlap of the variance sinks and sources onto the intermediate hadronic states . the momentum projection onto single nucleon blocks , that nplqcd is currently using , provides a volume suppression of the @xmath70 intermediate state compared to the @xmath69 state . thus , there is an intermediate time interval in which the signal - to - noise ratio is not exponentially degrading . it is in this time interval , dubbed the `` golden window '' , that plateaus for the low - lying energy levels in light nuclei can be identified . unfortunately , the window shrinks as the number of nucleons is increased , and so further developments will be required to go to much larger nuclei . a second roadblock that inhibited progress in lqcd calculations of nuclei was the number of wick contractions required to form a correlation function . a system containing @xmath71 up quarks and @xmath72 down quarks requires @xmath73 wick contractions , which is a rapidly growing number as one moves beyond the nucleon . it was recognized that recursion relations relating the wick contractions in systems with n mesons can be related to those with @xmath74 mesons @xcite . further , somewhat more sophisticated algorithms @xcite have been developed for the multi - baryon systems that greatly reduce the computing resources required to perform the contractions . these have led to very efficient calculations of the s - shell nuclei and hypernuclei , moving beyond the s - shell requires extensions of these works , and new ideas are required to calculate heavier nuclei . one of the points of discussion that came up during this presentation was how to optimally couple the results of lqcd calculations to nuclear structure calculations . given the expertise in the nuclear structure community , it makes little sense for lqcd theorists to `` go it alone '' and attempt to calculate the entire periodic table . it makes much more sense for the lqcd theorists to produce sets of quantities that can be handed to the nuclear structure theorists who use them in their machinery to determine the periodic table . the question is what are the optimal quantities to pass along from lqcd . it seems that the minimal set of quantities that could be passed along are the energy eigenvalues for a given system . lqcd calculations of the energy spectrum of an a - nucleon system could be performed in multiple lattice volumes , with multiple lattice spacings and at multiple light - quark masses , and handed to the the nuclear structure theorists who in turn reproduce the energies by tuning the chiral interactions . these tuned interactions are then used to calculate processes in the continuum . this methodology was used to calculate the @xmath75 interactions at the physical pion mass using @xmath19pt @xcite . the chiral interactions were tuned to reproduce the finite - volume energy levels determined in a series of lqcd calculations , and then used to calculate the scattering phase shift at the physical pion mass . progress in this direction is starting to be made , as demonstrated in recent calculations by nir barnea and collaborators @xcite , by using the ground state energies of the deuteron , dineutron and @xmath9he at @xmath60 to reproduce the @xmath66he ground state using the pionless eft . i have summarized the rapid progress that is being made in developing lqcd into a reliable calculational tool for low - energy nuclear physics . it holds the promise to directly connect the structure and properties of nuclei with qcd , and to enable a refinement of the chiral nuclear forces that are used as input into nuclear structure calculations . at present , the ground states of the s - shell nuclei and hypernuclei are being calculated at unphysically heavy light - quark masses , but within the next few years , such calculations at @xmath76 will be performed ( if hardware and software resources increase as expected ) . within the next five years , the spectrum and interactions of the lightest nuclei and hypernuclei will be postdicted or predicted with fully - quantified uncertainties . it is worth emphasizing that the lqcd effort in the us relies heavily on scidac funding to support the scientists who develop and optimize the software to run on the rapidly evolving computational hardware , e.g. gpu - accelerated compute nodes that comprise titan at ornl , or the bg / qs at anl and llnl . further , the effort requires ongoing access to both capability computing resources on leadership - class computing facilities , and capacity computing obtained from nersc , xsede , through usqcd and at local compute clusters . ongoing software ( see fig . [ fig : mgsoftware ] ) and hardware support are critical to progress in this area . ideally , one would start with a lqcd calculation and predict all of the quantities of interest in low - energy nuclear physics . presently , we are not in a position to do this , even if significantly more computing resources were provided to the program . while lscher provided the formalism to relate the two - body s - matrix directly to two - particle energy levels inside a cubic volume with the fields subject to periodic boundary conditions @xcite , which has since been understood and generalized to the two - nucleon systems , e.g. ref . @xcite , such formalism is complicated to apply in coupled - channels systems @xcite . further , the formalism is not in place for the three- and higher - body sectors , but progress is being made in such systems @xcite . _ * happy birthday james * : james vary is one of the first nuclear theorists i met when i arrived in the united states to enter the phd program at caltech in the mid 1980 s . i recall james taking the time to talk physics with me during his stay . his detailed knowledge of , and passion for , important problems of the day left a lasting impression on me . despite having been able to chat with , and even collaborate with , james since that time , when i learned that this conference was in part to celebrate james s 70th birthday , i was taken aback as it seems like yesterday that he was in his early 40 s ( and i was in my early 20 s ) , and he has retained the same passion and energy for science . i should also add that james is responsible for me remembering the value of @xmath77 ! happy 70@xmath78 ! ! _ d. b. kaplan , m. j. savage and m. b. wise , phys . b * 424 * , 390 ( 1998 ) [ nucl - th/9801034 ] . d. b. kaplan , m. j. savage and m. b. wise , nucl . b * 534 * , 329 ( 1998 ) [ nucl - th/9802075 ] . e. epelbaum and u. -g . meissner , ann . nucl . part . * 62 * , 159 ( 2012 ) [ arxiv:1201.2136 [ nucl - th ] ] . p. maris , j. p. vary and p. navratil , phys . c * 87 * , 014327 ( 2013 ) [ arxiv:1205.5686 [ nucl - th ] ] . w. detmold and k. orginos , phys . d * 87 * , 114512 ( 2013 ) [ arxiv:1207.1452 [ hep - lat ] ] . a. s. kronfeld , arxiv:1209.3468 [ physics.hist-ph ] . h. -w . lin , pos lattice * 2012 * , 013 ( 2012 ) [ arxiv:1212.6849 [ hep - lat ] ] . j. j. dudek , r. g. edwards , b. joo , m. j. peardon , d. g. richards and c. e. thomas , phys . d * 83 * , 111502 ( 2011 ) [ arxiv:1102.4299 [ hep - lat ] ] . s. weinberg , phys . * 17 * , 616 ( 1966 ) . g. colangelo , j. gasser and h. leutwyler , nucl . b * 603 * , 125 ( 2001 ) [ arxiv : hep - ph/0103088 ] . h. leutwyler , pos c * confinement8 * , 068 ( 2008 ) [ arxiv:0812.4165 [ hep - ph ] ] . s. m. roy , phys . b * 36 * , 353 ( 1971 ) . s. r. beane , t. c. luu , k. orginos , a. parreno , m. j. savage , a. torok and a. walker - loud , phys . d * 77 * , 014505 ( 2008 ) [ arxiv:0706.3026 [ hep - lat ] ] . x. feng , k. jansen and d. b. renner , phys . b * 684 * , 268 ( 2010 ) [ arxiv:0909.3255 [ hep - lat ] ] . t. yamazaki _ et al . _ [ cp - pacs collaboration ] , phys . d * 70 * , 074513 ( 2004 ) [ arxiv : hep - lat/0402025 ] . j. j. dudek , r. g. edwards , m. j. peardon , d. g. richards and c. e. thomas , phys . d * 83 * , 071504 ( 2011 ) [ arxiv:1011.6352 [ hep - ph ] ] . s. r. beane _ et al . _ [ nplqcd collaboration ] , arxiv:1107.5023 [ hep - lat ] . j. j. dudek , r. g. edwards and c. e. thomas , phys . d * 87 * , no . 3 , 034505 ( 2013 ) [ arxiv:1212.0830 [ hep - ph ] ] . s. r. beane , e. chang , s. d. cohen , w. detmold , p. junnarkar , h. w. lin , t. c. luu and k. orginos _ et al . _ , phys . c * 88 * , 024003 ( 2013 ) [ arxiv:1301.5790 [ hep - lat ] , arxiv:1301.5790 [ hep - lat ] ] . m. fukugita , y. kuramashi , h. mino , m. okawa and a. ukawa , phys . * 73 * , 2176 ( 1994 ) [ arxiv : hep - lat/9407012 ] . m. fukugita , y. kuramashi , m. okawa , h. mino and a. ukawa , phys . d * 52 * , 3003 ( 1995 ) [ arxiv : hep - lat/9501024 ] . s. r. beane , p. f. bedaque , k. orginos and m. j. savage , phys . lett . * 97 * , 012001 ( 2006 ) [ arxiv : hep - lat/0602010 ] . s. r. beane _ et al . _ [ nplqcd collaboration ] , phys . d * 81 * , 054505 ( 2010 ) [ arxiv:0912.4243 [ hep - lat ] ] . n. ishii , s. aoki and t. hatsuda , phys . lett . * 99 * , 022001 ( 2007 ) [ arxiv : nucl - th/0611096 ] . s. aoki , t. hatsuda and n. ishii , comput . * 1 * , 015009 ( 2008 ) [ arxiv:0805.2462 [ hep - ph ] ] . s. aoki , t. hatsuda and n. ishii , prog . theor . phys . * 123 * , 89 ( 2010 ) [ arxiv:0909.5585 [ hep - lat ] ] . t. yamazaki , y. kuramashi , a. ukawa , arxiv:1105.1418 [ hep - lat ] . t. yamazaki , y. kuramashi , a. ukawa , phys . d * 81 * , 111504 ( 2010 ) [ arxiv:0912.1383 [ hep - lat ] ] . de forcrand and m. fromm , phys . lett . * 104 * , 112005 ( 2010 ) [ arxiv:0907.1915 [ hep - lat ] ] . s. r. beane _ et al . _ [ nplqcd collaboration ] , phys . d * 85 * , 054511 ( 2012 ) [ arxiv:1109.2889 [ hep - lat ] ] . t. inoue _ et al . [ hal qcd collaboration ] , nucl . a * 881 * , 28 ( 2012 ) [ arxiv:1112.5926 [ hep - lat ] ] . s. r. beane , e. chang , s. d. cohen , w. detmold , h. w. lin , t. c. luu , k. orginos and a. parreno _ et al . _ , arxiv:1206.5219 [ hep - lat ] . t. yamazaki , k. -i . ishikawa , y. kuramashi and a. ukawa , phys . d * 86 * , 074514 ( 2012 ) [ arxiv:1207.4277 [ hep - lat ] ] . m. luscher , commun . phys . * 105 * , 153 ( 1986 ) . m. luscher , nucl . b * 354 * , 531 ( 1991 ) . r. a. briceno , z. davoudi and t. c. luu , phys . d 88 , * 034502 * ( 2013 ) [ phys . d * 88 * , 034502 ( 2013 ) ] [ arxiv:1305.4903 [ hep - lat ] ] . w. detmold and m. j. savage , phys . d * 82 * , 014511 ( 2010 ) [ arxiv:1001.2768 [ hep - lat ] ] . t. doi and m. g. endres , comput . commun . * 184 * , 117 ( 2013 ) [ arxiv:1205.0585 [ hep - lat ] ] . s. r. beane , e. chang , s. d. cohen , w. detmold , h. -w . lin , t. c. luu , k. orginos and a. parreno _ et al . _ , phys . lett . * 109 * , 172001 ( 2012 ) [ arxiv:1204.3606 [ hep - lat ] ] . r. a. briceno , z. davoudi , t. luu and m. j. savage , arxiv:1309.3556 [ hep - lat ] . n. barnea , _ nuclear physics with heavy pions - eft for lqcd _ , a talk presented at the institute for nuclear theory , july 31 , 2013 . j. c. osborn , r. babich , j. brannick , r. c. brower , m. a. clark , s. d. cohen and c. rebbi , pos lattice * 2010 * , 037 ( 2010 ) [ arxiv:1011.2775 [ hep - lat ] ] . at the computational nuclear physics meeting , washington d.c . , july 23 - 24 , 2012 , by b. joo . m. t. hansen and s. r. sharpe , phys . d * 86 * , 016007 ( 2012 ) [ arxiv:1204.0826 [ hep - lat ] ] . r. a. briceno and z. davoudi , phys . d * 87 * , 094507 ( 2013 ) [ arxiv:1212.3398 [ hep - lat ] ] . p. guo , j. dudek , r. edwards and a. p. szczepaniak , phys . d * 88 * , 014501 ( 2013 ) [ arxiv:1211.0929 [ hep - lat ] ] . r. a. briceno and z. davoudi , arxiv:1204.1110 [ hep - lat ] . k. polejaeva and a. rusetsky , eur . phys . j. a * 48 * , 67 ( 2012 ) [ arxiv:1203.1241 [ hep - lat ] ] .

at the beginning of the 40 s of the past century , the existence of the electric quadrupole moment of the deuteron @xcite was explained by rarita and schwinger by introducing a static tensor term in the nucleon - nucleon ( n - n ) interaction @xcite . since then , tensor terms are unavoidable ingredients of the microscopic n - n interactions , i.e. those interactions constructed to reproduce the properties of two - nucleon systems . despite their relevance in microscopic interactions , the tensor terms are usually neglected when effective interactions and theories are used . in the effective theories some complicated many - body effects are treated , obviously in effective and average manner , by changing the values of the parameters of the interaction . some specific observables of the nucleus are chosen to select these values . for example , in hartree - fock calculations these observables are usually the nuclear binding energies . the effective theory is expected to be able to describe other observables . if this fails , one searches for many - body effects which should be explicitly treated to improve the description of the data . in this manner , we link many - body effects to specific observables . in the case of our interest here , the tensor force , the point is the identification of observables clearly depending on the presence of this term in the effective n - n interaction . in these last years , the interest on the tensor terms of the effective n - n interaction has increased because the inclusion of these terms improves the description of the single particle ( s.p . ) energies of some isotope or isotone chains @xcite when the hartree - fock theory is used @xcite . we see some weak points in using the s.p . energies to define the strength of the tensor terms of the effective interactions . first , s.p . energies are extremely sensitive to the spin - orbit terms of the n - n interaction , and this obscures the possibility of a clear identification of the tensor effects ( see for example the discussion in ref . furthermore , observations are always done on global nuclear properties , therefore the identification of the measured quantities with s.p . properties of the nucleus is done by imposing to the observed quantity the physical interpretation given within a mean field description of the many - body system . the fact that experimental values of spectroscopic factors are usually rather different from the mean field expectations is a clear indication of the limits of this procedure . in this article we propose an alternative approach to select the strength of the effective tensor forces . we have looked at the excitation spectrum to find observables particularly sensitive to the tensor force . we have identified these observables with the energies of the 0@xmath0 charge conserving excitations . the large sensitivity of the 0@xmath0 excitation to the tensor parts of the nucleon - nucleon interaction was pointed out already in ref . @xcite . by using a recursive self - consistent hartree - fock ( hf ) plus random phase approximation ( rpa ) procedure , we have chosen the strength of the tensor term of the effective interaction to reproduce the experimental value of the 0@xmath0 in @xmath1o . with these new interactions we have investigated the ground and excited state properties of various nuclei by doing hf plus rpa calculations . the structure of the tensor terms we have considered and the methodology used to select their strengths are presented in sec . [ sec : int ] . we discuss the results obtained in the description of the ground states of various nuclei in sec . [ sec : gs ] , and in sec . [ sec : rpa ] the results obtained for the excited states . we summarize the main points of our work and draw our conclusions in sec . [ sec : conc ] . the most important tensor component of the microscopic n - n interaction is that related to the tensor - isospin channel @xcite whose long range behaviour is dominated by the exchange of a single pion . since the pion is the lightest meson , the range of the tensor - isospin term is longer than the ranges of the other terms of the n - n interaction . for this reason we have chosen to consider , in our effective interactions , only tensor - isospin terms with finite range . the use of finite range forces requires , in both hf and rpa calculations , the evaluation of direct and exchange interaction matrix elements . the tensor - isospin term of our effective interaction is based on the analogous term of the microscopic argonne v18 interaction @xcite . we have multiplied the radial part of this term by a function which simulates the effect of the short - range correlations @xcite . in our work , the radial part of the tensor - isospin term has the form v_6(r ) = v_6,av18(r ) , [ eq : vtensor ] where we have indicated with @xmath2 the distance between the two interacting nucleons , with @xmath3 the radial function of the argonne v18 tensor - isospin potential @xcite , and with @xmath4 a free parameter . the changes in the tensor - isospin term produced by choosing different values of @xmath4 are shown in panel ( a ) of fig . [ fig : tensq ] , where we present the fourier transformed function @xmath5 , defined by the equation @xmath6 in the above equation @xmath2 and @xmath7 indicate the moduli of @xmath8 and @xmath9 , and we used the definition of the tensor operator s_12(*r * ) = 3 - ( 1)(2 ) , [ eq : tensor ] where @xmath10 are the usual pauli spin matrices . ) , for various parameterizations . in the panel ( a ) we compare the tensor - isospin term of the bare argonne v18 interaction ( full line ) with the interactions obtained from eq . ( [ eq : vtensor ] ) by using different values of the parameter @xmath4 . in panel ( b ) we compare the bare argonne v18 term with the tensor term of the gt2 force of ref . @xcite ( dotted line ) and with that of the m3yp2 force of ref . @xcite ( short - dashed line ) . the tensor forces we have constructed , named d1st and d1mt , are indicated by the dashed - dotted and dashed - doubly - dotted lines . the long - dashed line ( cbf ) shows the tensor term obtained by multiplying the bare argonne v18 interaction with the scalar part of the correlation function obtained in microscopic correlated basis function calculations @xcite . ] the results presented in the panel ( a ) of fig . [ fig : tensq ] show that the correlation effects reduce the strength of the bare tensor force . the smaller is the value of @xmath4 , the more extended in @xmath2 space is this effect , therefore the function to be integrated in eq . ( [ eq : fourier ] ) becomes smaller . the first step of our study consisted in identifying an observable very sensitive to the tensor force with the aim of using it to determine the strength of this part of the effective interaction . we have conducted this study by investigating the excitation spectra of the @xmath11c , @xmath1o , @xmath12ca , @xmath13ca , @xmath14zr and @xmath15pb nuclei within the phenomenological rpa approach developed by the jlich group @xcite . in this approach , based on the landau - migdal theory of finite fermi systems @xcite , the set of s.p . energies and states is obtained by using phenomenological mean field potentials that , in our case , have the shape of woods - saxon wells . the values of the parameters of the potential are chosen for each nucleus in order to reproduce at best the empirical values of the s.p . energies around the fermi surface and the values of the charge root mean square radii . the explicit expression of the potential , and the values of the parameters , can be found in refs . @xcite . following the philosophy of the landau - migdal approach , in the rpa calculations we substituted the woods - saxon s.p . energies with their experimental values , when they are available . all the rpa results presented in this article have been obtained by using a discrete s.p . basis . in the phenomenological calculations discussed in this section we have used a discrete s.p . basis obtained by diagonalizing the woods - saxon well in a harmonic oscillator basis . in analogy with the work of ref . @xcite , we have used configuration spaces large enough that the inclusion of additional s.p . states does not modify the energies of the first excited states , below 20 mev in the lighter nuclei and 15 mev in the heavier ones , within 0.1 mev . in these phenomenological calculations we used as basic n - n effective interaction a density dependent landau - migdal force . the values of the force parameters , given in refs . @xcite , change for each nucleus , and are chosen to reproduce at best the energies of the collective low - lying 3@xmath0 states of each nuclei , and also the energies of the 12@xmath0 excited states in @xmath15pb . we added to this basic interaction a tensor term of the form given by eq . ( [ eq : vtensor ] ) , and we studied the excitation spectra of the nuclei mentioned above . in agreement with previous calculations @xcite we observed that the influence of the tensor term on the natural parity states is negligible . we found few cases of interest in the spectrum of the unnatural parity excitations . the most interesting one was the excitation of the 0@xmath0 states , which show common characteristics in all the nuclei we considered . states for various nuclei as a function of the value of the parameter @xmath4 of eq . ( [ eq : vtensor ] ) ruling the strength of the tensor force . the full lines show the experimental values , the dashed lines the values obtained without tensor , and the dotted lines the values obtained with the full tensor term of the argonne v18 interaction . ] in fig . [ fig : ene0- ] we have summarized the results obtained for the energies of the lowest 0@xmath0 excitations in the nuclei investigated . in this figure , the squares indicate the values of the @xmath16 excitation energies , obtained by adding the tensor term of eq . ( [ eq : vtensor ] ) to the landau - migdal force , as a function of the parameter @xmath4 . the horizontal dashed lines indicate the values obtained without tensor term , while the dotted lines show the values obtained by using the full tensor term of the argonne v18 interaction . the full lines show the experimental values @xcite . we observe that the effect of the tensor term is always attractive , i.e. all the energies obtained with the tensor terms are smaller than those obtained without it . the values of these energies decrease , monotonically and smoothly , from the dashed to the dotted lines when the value of @xmath4 increases , as we have naively expected . discussing the results in more detail , we have obtained variations of the energies from 4 up to 8 mev , and , in relative variations , between 30% and 100% . these are the largest effects of the tensor force on the excitation energies we have found in our investigation . in the @xmath13ca case , the effect of the full tensor term is so strong that we obtained an imaginary solution for the rpa equations . the remarkable sensitivity of the @xmath16 excitation energies to the presence of the tensor force , and their smooth behaviour with respect to the changes of its strength , make these energy values particularly suitable to be chosen as experimental benchmarks to select the strength of the tensor terms in effective interactions . in this work we have performed hf plus rpa calculations . the hf equations were solved with the method used in refs . this method is based on the plane wave expansion technique developed by guardiola and ross @xcite . after the iterative process has reached the convergence , we solved the hf differential equations not only for the s.p . states below the fermi level , the hole states , but also for those states above the fermi level , the particle states . the numerical method automatically produces a set of discrete levels even when the s.p . energies are larger than zero , i.e. in the continuum region . we did not find general criteria for the stability of our results . this problem is not very important in the phenomenological rpa approach , since the effects related to the truncation of the s.p . configuration space are taken into account by changing the parameters of the force . however , in self - consistent calculations this is a more serious problem , since the interaction parameters are chosen to reproduce , in hf calculations , the ground state properties of the various nuclei , and the force remains unchanged in rpa calculations . to keep under control these problems , we restricted our study to the low - lying excited states . for each excitation studied we used the same criterion considered in the phenomenological approach , i.e. we controlled that the energy eigenvalues of the first low - lying states did not vary by more than 0.1 mev against the enlargement of the configuration space . in order to obtain this numerical accuracy , we had to use configuration spaces composed by a few thousands s.p . states , more than 2000 in @xmath15pb . the numerical stability of higher energy excitations , such as giant resonances , requires even larger configuration spaces . in this case we believe it is necessary to abandon the discrete rpa calculations and treat correctly the continuum , as it is done , for example , in refs . @xcite . we have built two new forces by adding to the d1s @xcite and d1 m @xcite parameterizations of the gogny interaction @xcite a tensor - isospin term similar to that given by eq . ( [ eq : vtensor ] ) . we label d1st and d1mt these new interactions . since in gogny - like forces the spin - orbit term is fixed to reproduce the experimental splitting of the s.p . energies of the @xmath17 and @xmath18 neutron states in @xmath1o , we used this nucleus as reference . the new interactions have been fixed by using an iterative procedure . we started with a hf calculation without tensor force to produce a set of s.p . energies and wavefunctions to be used in the rpa calculations . then , we made a rpa calculation with the tensor force and we fixed the value of the parameter @xmath4 of eq . ( [ eq : vtensor ] ) in order to reproduce the energy of the first @xmath16 excitation of @xmath1o at 10.6 mev . with this new interaction we recalculated the hf s.p . energies by changing the spin - orbit interaction to reproduce the splitting of the two @xmath19 states quoted above . these hf and rpa calculations have been repeated until the convergence of the result was obtained . by using this procedure we have found for the parameter @xmath4 the value 0.6 @xmath20 , for the d1st force , and of 0.25 @xmath20 , for the d1mt one . summarizing , we added a tensor term to the d1s and d1 m gogny - like interactions and we modified only the spin - orbit terms from 130 mev , in the original d1s force , to 134 mev , in the d1st interaction , and from 115 mev , in the d1 m force , to 122.5 mev , in the d1mt interaction . no other values of the force parameters have been changed . the tensor terms of the d1st and d1mt interactions are indicated in panel ( b ) of fig . [ fig : tensq ] by the dashed - dotted and dashed - doubly - dotted lines respectively . in this figure they are compared with the tensor - isospin term of the microscopic argonne v18 interaction ( solid curve ) . by construction , the effective tensor terms are smaller than that of the bare n - n interaction . more interesting is the comparison with the long dashed line which has been produced by multiplying the bare interaction with the scalar part of the short - range correlation function obtained in correlated basis function calculations @xcite . the remarkable difference between this line and those of the d1st and d1mt forces indicates that our procedure includes in the effective tensor term not only the effects of the short - range correlations , but also some other many - body effects that the microscopic calculations consider explicitly . in the same panel we make a comparison with other two tensor terms of finite - range interactions used in the literature , the gt2 @xcite and the m3yp2 @xcite forces . the tensor term of the gt2 is constructed to have the same volume integral of the argonne v18 tensor force . the strength of this tensor force is much larger than those of the tensor forces we have built . on the opposite , we observe that the tensor term of the m3yp2 force is much smaller . even though the d1st and d1mt forces reproduce the experimental value of the excitation energy of the @xmath16 state at 10.6 mev in @xmath1o , they produce rather different rpa wave functions , as we have verified by calculating transition densities and inclusive neutrinos cross sections for this excited state . we used the d1st and d1mt interactions , whose construction has been described in the previous section , to make spherical hf calculations for a set of nuclei in different regions of the isotope chart . we have chosen nuclei where the s.p . states below the fermi level are fully occupied to avoid deformation problems and to minimize pairing effects . we have verified these features by controlling the results given by the deformed hartree - fock - bogolioubov calculations of ref . @xcite . ) , between binding energies calculated with the d1st and d1s forces ( open squares ) and with the d1mt and d1 m ( solid circles ) interactions . panels ( b ) and ( c ) : binding energies per nucleon calculated with the various interactions compared with the experimental values ( solid triangles ) taken from ref . the experimental values for the @xmath21o , @xmath22ca , @xmath13ni , @xmath23ni and @xmath24sn nuclei are estimated @xcite . the lines have been drawn to guide the eyes . ] in fig . [ fig : be ] we compare the binding energies obtained for various nuclei we have considered , with the experimental values taken from refs . the experimental energies of the @xmath21o , @xmath13ni , @xmath22ca , @xmath23ni and @xmath24sn nuclei have not been measured but estimated @xcite . the methodology used in our study is already evident . we have investigated the effects of the tensor force by comparing the results obtained by using forces with and without tensor term . to emphasize the effects of the tensor force , we show in the panel ( a ) of fig . [ fig : be ] the quantity e = 100 , [ eq : deltae ] which is the relative percentage differences between the binding energies calculated by using interactions with and without tensor term , @xmath25 and @xmath26 respectively ( @xmath27 ) . in this panel the open squares indicate the results obtained with the d1st and d1s interactions and the solid circles those obtained with the d1mt and d1 m ones . the lines have been drawn to guide the eyes . if we exclude the anomalous values obtained for the @xmath28o nucleus , we observe that all the other results lie in the range @xmath29 . in general , the inclusion of the tensor term in the d1 m force produces more binding , while it has opposite effect in the d1s case . these results show that effect of the tensor on the binding energy of the nuclei we have investigated is rather small , confirming the results of ref . @xcite . as a consequence , and as we can see in panels ( b ) and ( c ) , the inclusion of the tensor term does not modify the agreement with experimental data in a significant manner . ) , between root mean square radii of neutrons ( upper panel ) and protons ( lower panel ) distributions calculated with and without tensor for d1st and d1s ( open squares ) and d1mt and d1 m ( solid circles ) interactions . ] we have investigated the effect of the tensor force on the proton and neutron density distributions obtained from our hf calculations , and , also in this case , we found small effects . we summarize the results of this study in fig . [ fig : radii ] where we have shown the relative percentage difference between root mean square ( rms ) radii obtained by using interactions with and without tensor force , r = 100 . [ eq : deltar ] we show the results for the neutron radii in the upper panel of the figure , and those for the proton radii in the lower panel . we have indicated with the open squares the results obtained with the d1s forces and with the solid circles those obtained with the d1 m ones . the effects of the tensor force are rather small , of the order of few parts on a thousand . we observe again the different sign between d1s and d1 m results . the larger binding produced by the tensor in the d1 m case generates more compact nuclei , i.e. with smaller rms radii . in the case of the d1s forces the effect is just the opposite . this trend is present in both neutron and proton cases . the results we have presented so far indicate that the bulk properties of the nuclear ground states are not greatly affected by the presence of the tensor force . the situation changes when the s.p . energies are considered . in the remaining part of this section we shall discuss results concerning s.p . henceforth we shall distinguish the proton and neutron s.p . levels by using the @xmath30 and @xmath31 labels , respectively . a first quantity we have studied is the difference between the s.p . energies of spin - orbit partner levels s = _ l-1/2 - _ l+1/2 . in particular , we have studied the difference between the values of @xmath32 obtained by using forces with and without tensor terms s = s_d1 - s_d1 . [ eq : deltaep ] in figs . [ fig : lsprot ] and [ fig : lsneut ] we show the values of @xmath33 calculated for the @xmath34 , @xmath35 and @xmath36 proton and neutron levels , respectively , for all the nuclei considered . in these figures the open squares indicate the results obtained with the d1s interactions , and the solid circles those obtained with the d1 m ones . the arrows indicate those nuclei where all the spin - orbit partner levels , for both protons and neutrons , are fully occupied . ) , calculated with and without tensor forces , for the @xmath34 , panel ( a ) , @xmath35 , panel ( b ) , and @xmath36 , panel ( c ) , proton states . the results for d1s and d1st interactions are shown by open squares and solid circles , respectively . the arrows indicate those isotopes where the effect of the tensor is expected to be zero . ] for neutron states . ] let us consider first fig . [ fig : lsprot ] , where , for each isotope chain we show the evolution of @xmath33 values with the increasing number of neutrons . we first observe that , for each s.p . level investigated , the d1s and d1 m results have identical behaviour . minima and maxima are in the same position for both type of calculations . a second observation is that , in general , @xmath33 is negative . this means that the tensor force reduces the energy difference between spin - orbit partner levels . a third observation is that the effects of the tensor force are minimal , almost zero , for those nuclei indicated with the arrows . the first observation indicates that the effects we have pointed out are strictly related to peculiarities of the tensor force and of the nucleus investigated . the small difference between the d1s and d1 m results reflects the difference between the tensor forces in d1st and d1mt , as we have shown in fig . [ fig : tensq ] . the second and third observations are well understood within the scheme proposed by otsuka _ the effect of the tensor interaction between a proton and a neutron occupied s.p . levels is attractive if one of the levels has an angular momentum @xmath37 and the other one @xmath38 . if the angular momenta of the two s.p . levels are of the same type , i.e. both @xmath39 or both @xmath40 , the effect of the tensor force has opposite sign . the effect of an occupied neutron level with angular momentum @xmath39 is to increase the energies of the s.p . proton levels with @xmath39 , and to lower those of the levels with @xmath40 . as a consequence , the splitting between the energies of the proton spin - orbit partner levels is reduced . this effect is reversed when the occupied neutron level has @xmath40 . if both @xmath39 and @xmath40 neutron levels are occupied , the two effects cancel with each other . the results presented in fig . [ fig : lsprot ] are well explained within this picture . in the nuclei marked with the arrows , all the @xmath39 and @xmath40 neutron levels are occupied . in these nuclei we do not expect any tensor effect . in reality the values of @xmath33 are not exactly zero even in these cases because we have changed the strengths of the spin - orbit interactions in the forces with tensor terms ( see sec . [ sec : int ] ) . in all the other nuclei there is , at least , one occupied neutron level with @xmath39 , whose spin - orbit partner level is empty . the effect we have discussed above predicts negative values of @xmath33 , as those shown in the figure . in fig . [ fig : lsneut ] we have shown the values of @xmath33 for the neutron @xmath34 , @xmath35 and @xmath36 levels . since in the oxygen and calcium isotopes all the proton s.p . partner levels below the fermi surface are completely occupied , the effects we observe in these isotopes are due to the tensor interaction acting within neutron s.p . levels only . this is exactly the effect we have discussed above , but acting between s.p . states with the same isospin . this effect is weaker than that between states with opposite isospin , as the results for the @xmath41o , @xmath42o , @xmath13ca and @xmath43ca nuclei indicate . in these nuclei one of the neutron spin - orbit partner levels is unoccupied and the tensor effect is expected to be present . actually , there is also an effect produced by the different values of the spin - orbit forces , but we have verified that , for the cases under consideration , this effect is negligible . the analysis of the @xmath33 results for the ni , zr , sn and pb isotopes is more complicated , since , in these cases , in both the proton and neutron sectors , there are spin - orbit partner s.p . levels not fully occupied . for this reason , we must consider the interaction with both proton and neutron levels . the similarity between the behaviour of the results obtained with the different interactions indicate that we are observing an effect due to the peculiarities of the tensor force . another quantity of interest related to the s.p . energies is the gap , @xmath44 , between the energies of the levels just above and just below the fermi surface . in fig . [ fig : gap ] we have shown the proton ( lower panels ) and neutron ( upper panels ) energy gaps calculated for the nuclei we have investigated with the four forces we are using . our results are compared with the experimental values ( solid triangles ) extracted from the binding energies of nuclei with atomic numbers differing by one unit . the first remark related to the results shown in fig . [ fig : gap ] is that , in general , the effects of the tensor force are relatively small and they are similar for both type of interactions . they are negligible for neutrons , while some noticeable effects are present in the case of protons . the neutron results , obtained with and without tensor , follow reasonably well the behaviour of the experimental energy gaps . the effects of the tensor force are better emphasized in fig . [ fig : deltagap ] where we show the difference g = g_d1 - g_d1 , [ eq : deltag ] between the gap values obtained by using interactions with and without the tensor force . also in this case we have indicated with the arrows the nuclei where all the spin - orbit partner levels are fully occupied , for both protons and neutrons . the results obtained with the d1s interactions are shown by the open squares and those obtained with the d1 m interactions by the solid circles . it is interesting to observe that also in this case the behaviour of the d1s and d1 m results is similar , though the effects produced by the d1st force are slightly larger than those obtained for d1mt . also the results in fig . [ fig : deltagap ] can be well explained within the scheme proposed by otsuka . in the spin unsaturated oxygen isotopes , the unpaired neutron levels are always of @xmath39 type . the tensor force lowers the energy of the @xmath45 occupied proton level ( @xmath40 type ) and increases that of the @xmath46 empty level ( @xmath39 type ) . for this reason , the results for the oxygen isotopes , in the proton case , have positive values . an analogous effect is present also for the @xmath13ca and @xmath43ca isotopes . in this case , the states to be considered are the holes @xmath47 or @xmath48 and the particle @xmath49 . for the heavier isotopes , the situation is more complicated because the unpaired levels can be more than one , and because the levels involved in the gap calculation could be both of the same type ( @xmath39 or @xmath40 ) . this is the case for @xmath50 or @xmath51=28 or 50 , but not for @xmath52 . for this reason , in the case of protons , the values of @xmath53 are negative for ni , sn and pb isotopes and positive for @xmath14zr . for the neutrons , panel ( a ) , the situation is more difficult to discuss since the nuclei we have investigated are not isotones . for each set of isotopes there are different s.p . levels related to the neutron gap , therefore the effect of the tensor term can be different for each nucleus considered . this is the reason of the oscillating behaviour we observe in the ni isotopes . since oxygen and calcium isotopes are spin saturated in protons , the results presented in the panel ( a ) of fig . [ fig : deltagap ] for these nuclei , are produced by the interaction of an unpaired neutron s.p . level of @xmath39 type with the neutron s.p . levels just below and above the fermi surface . the comparison of the results of these nuclei shown in the two panels indicates that the effect of the tensor interaction between like nucleons is smaller than that between neutrons and protons , reflecting the fact that we have used a tensor - isospin term in the interaction . we have seen that the tensor term affects more the s.p . than the bulk properties of the nucleus . the effects of the tensor force on the s.p . levels can modify their relative order . if this occurs for the levels near the fermi surface , the spin of even - odd nuclei neighboring the nuclei investigated , which is determined by the last unpaired nucleon , should be modified . we found some cases where this happens and few of them are presented in fig . [ fig : spetens ] , where we show the evolution of the states near the fermi surface for the different interactions we have used . in the @xmath13ca case we consider the two proton states below the fermi surface . the @xmath47 is a @xmath40 state , therefore its energy is lowered by the tensor , while that of the @xmath48 state remains essentially unchanged . this effect inverts the order of the two states , as we observe in the results of the d1st column . the calculations done with the d1 m interaction give the two proton states in an inverted order with respect to that of the d1s one . in this case the tensor effect enlarges the energy difference between the two states . since @xmath13ca and @xmath13ni are mirror nuclei , we expected an analogous effect for the two neutron hole states in @xmath13ni . this effect is present in our calculations as it is shown in the figure . since all the spin - orbit partner levels are occupied for both protons and neutrons , we do not expect tensor effects in @xmath22ca . this is what we observe in fig . [ fig : spetens ] . we show the results for this nucleus since the order of the proton hole states obtained with the d1s and d1st interactions is inverted with respect to that obtained with the d1 m and d1mt forces . in the @xmath23ni case , the tensor force produces a large lowering of the @xmath54 level , and this , for the d1 m interaction , generates an inversion with the @xmath55 level . the figure shows that the tensor force reduces the energy gap between @xmath54 and @xmath49 levels . for the @xmath24sn nucleus we show both proton and neutron cases , since in both situations we obtain the inversion of the @xmath56 and @xmath57 levels , when the d1st interaction is used . the tensor force lowers the energies of the @xmath56 levels , which are @xmath40 states , and enhances those of the @xmath57 levels , which are @xmath39 states . we obtain a large effect on the energies of the @xmath56 levels while the modification of the @xmath57 energy is minimal . we observe an inversion of the order of the @xmath58 and @xmath59 levels in @xmath60sn , when the d1st interaction is used . the tensor lowers the energy of the @xmath58 state , of @xmath40 type , and has no effect on the @xmath59 state . the effect is present also when the d1mt interaction is used , but it is not large enough to invert the order of the states . in @xmath61sn we obtain an inversion of the order of the @xmath62 and @xmath63 levels when both d1st and d1mt interactions are used . also in this case we observe the tensor effect predicted by the otsuka s scheme . the energy of the @xmath62 state , of @xmath40 type , is lowered , while that of the @xmath63 state , of @xmath39 type , is enhanced . to summarize the results presented in this section , we may say that the tensor effects have remarkable consequences on those observables which we relate to s.p . properties of the nuclear system , such as s.p . energies , gaps and spin of the system . in the previous section we have presented the results regarding the ground state properties of some spherical nuclei . in this section we show the results we have obtained for the excited states of these nuclei by doing rpa calculations . in this study , we have also considered @xmath11c , a well studied nucleus from both experimental and theoretical points of view , despite the fact that its ground state contains deformations @xcite . this feature is less important in the description of the excitation spectrum than for the ground state observables . the input required by any rpa calculations is composed by the s.p . basis and the effective interaction . the results shown in this section have been obtained by using the s.p . basis and the effective interaction provided by hf calculations . in the rpa jargon this procedure is called self - consistent to distinguish it from the phenomenological one described in sec . [ sec : intro ] . to be precise , we do not strictly use the same interaction in hf and rpa , since in the latter case we neglect the coulomb and spin - orbit terms of the interaction . we should keep in mind this fact , even though there are indications that , in rpa calculations , these two terms of the interaction produce very small effects that , in addition , have the tendency of canceling with each other @xcite . the strategy of our investigation is analogous to that used in the previous section , i.e. we compare results obtained by using interactions with and without tensor terms . while for the hf calculations this comparison provides a clear indication of the effects of the tensor force , in the rpa case the situation is more complicated . in effect , the differences in the rpa results can be due to genuine effects of the tensor interaction in the rpa calculations , and also to the different values of the s.p . energies and wave functions provided by the hf results . we have disentangled the effects produced by these two different sources by presenting results where the tensor force is switched on and off in both hf and rpa calculations . to distinguish the results of the different type of calculations we have indicated with @xmath64 the rpa excitation energies , where the first subindex , a , refers to the interaction used in the hf calculation , and the second subindex , b , to that used in the rpa one . the a and b labels can be t if the interaction includes the tensor term , i.e. if the d1st or the d1mt interaction has been used , and n otherwise . for example , @xmath65 indicates the excitation energy obtained with a hf calculation without tensor and a rpa calculation with the tensor force . we present also results obtained by switching off the residual interaction . we label these results as @xmath66 , where the superscript ipm means independent particle model . the energies produced in this type of calculations need only one subindex which indicates the presence , t , or the absence , n , of the tensor force in the hf calculation . in our study we have investigated various types of multipole excitations , and , in agreement with the results of refs . @xcite , we found that the natural parity excitations are practically insensitive to the effects of the tensor force . for this reason , we present here only results obtained for the unnatural parity excitations . we start or discussion by presenting the results related to the excitation of the first @xmath16 state in the various isotopes we are studying . in the phenomenological calculations of sec . [ sec : int ] we have pointed out the large sensitivity of the excitation energy of these states to the tensor force . for this reason we have chosen the energy of this excitation in @xmath1o to select the strengths of our tensor forces . .energies , in mev , of the first @xmath16 excited state for those isotopes we have studied in this work where the energy values have been measured @xcite . the theoretical energies are obtained by doing rpa calculations with different interactions . [ cols=">,^,^,^,^,^",options="header " , ] the detailed analysis of the results shows that the effects of the tensor force on the energy values improve the agreement with the experimental values . the situation regarding the @xmath67 values is more complex . the tensor force lowers the @xmath67 values of the first state , and this improves the agreement with the experiment . the situation is reversed for the second excited state . our results are compatible with those found in literature @xcite . the large difference between the theoretical and experimental @xmath67 values of the second state can be attributed to the limitations of the rpa which is unable to describe the large fragmentation of the second 1@xmath68 state . we show in fig . [ fig : pb1 + ] the inelastic electron scattering responses @xcite , calculated for these two states and we compare them with the data of ref . we observe that there is a larger sensitivity to the tensor force in the case of the first excited state . we have studied the effects of the tensor terms of interactions used in hf and rpa theories . we considered only finite - range terms in the tensor - isospin channel , because they are the most important terms in microscopic interactions . we constructed these tensor terms by multiplying the analogous term of the microscopic argonne v18 interaction by a function which modifies its behaviour at short internucleonic distances . this function contains a single parameter whose value determines the strength of the tensor force . to determine this strength , we searched for a global observable particularly sensitive to the tensor force , and , by using phenomenological rpa calculations , we found it in the excitation of the @xmath16 states . we have also observed that the tensor force strongly influences the values of the s.p . we constructed new effective interactions by adding a tensor - isospin term to finite range interactions , and we selected the strength of the tensor and spin - orbit forces to reproduce , in the @xmath1o nucleus , the experimental energy of the first @xmath16 excited state , and the splitting between the s.p . energies of the neutron @xmath18 and @xmath18 levels . we based our work on the d1s and d1 m parameterizations of the gogny interaction , and we called d1st and d1mt the two new effective interactions constructed with a recursive procedure where hf and rpa calculations have been repeated until both experimental quantities have been reproduced . the study of the effects of these tensor forces has been done by comparing the results obtained in hf and rpa calculations , using interactions with and without tensor terms . we have repeated each calculation with both d1s and d1 m type of effective interactions to extract genuine tensor effects from those related to the peculiarities of a specific choice of the force parameters . we always found great similarities between the results obtained with the two different type of parameterizations . since the strength of the tensor force in d1mt is weaker than that of d1st , we found smaller tensor effects in the results obtained with d1mt than in those for d1st . we have done calculations for a set of spherical nuclei chosen such as all the s.p . levels below the fermi surface are fully occupied . with this choice we avoided the effects of the deformation and we minimized those of the pairing . our hf calculations indicate that tensor forces do not produce sensitive effects on bulk observables such as binding energies , radii and density distributions . the effects of the tensor force on quantities related to the s.p . properties of our theoretical approach are certainly more remarkable . we have calculated energy splitting between spin - orbit partner levels , energy gaps between the s.p . states just above and below the fermi surface , and we found noticeable effects produced by the tensor force . also the ordering of the s.p . levels around the fermi surface , which determines the spin of the even odd nuclei neighboring those we have studied , is strongly influenced by the presence of the tensor force in some cases . we could explain all our results within the picture proposed by otsuka and collaborators @xcite , eventually by extending it to consider the interaction between nucleons of the same type . we have already mentioned in the introduction that the clear identification of the tensor effects on the s.p . properties from the comparison with experiment is rather problematic . in the mf picture the effects of the tensor force are of the same size of those produced by the spin - orbit force . beyond the mf model , it is well known that s.p . properties are strongly affected by correlation effects , the most important ones , for the quantities of our interest , are those related to the coupling of s.p . states with low - lying collective vibrations @xcite . the study of tensor effects in the excitation spectra is more complex , since the effects of the tensor force in the rpa calculations add to those already present in the hf calculation that produces the s.p . bases . in our investigation we have disentangled the effects coming from these two different sources . we have verified the well know fact @xcite that natural parity excitations are essentially unaffected by the tensor force . for this reason we have presented results regarding unnatural parity excitations only . we started our investigation by studying the @xmath16 excitations in various nuclei , and we found that the excitation energies obtained with the tensor forces are always smaller than those obtained without it . we studied the different role played by the tensor force in is and iv type of excitations . to identify clearly these different type of excitation modes , we have considered nuclei with equal number of protons and neutrons , and states dominated by particle - hole transitions with the same angular momentum coupling for both protons and neutrons . the energies of the iv modes are always greater than those of the is modes with or without tensor . these results agree with the experimental observations . we found that the is excitations are more sensitive to the tensor force than the iv ones . the tensor force increases the energy difference between is and iv excitations . we made a systematic study of the effects of the tensor force on the excitation of the 1@xmath68 states in nuclei with different number of protons and neutrons . we have considered the states showing the largest @xmath67-values , and we found that the tensor force consistently lowers the values of their excitation energies . this is essentially obtained as the incoherent sum of the effect generated by the hf calculation and that obtained by the rpa . the study of the excitation of the first two 1@xmath68 excited states in @xmath15pb indicates that the lower energy state is more sensitive to the tensor force than the other one . the presence of the tensor force modifies energies and @xmath67-values , and slightly improves the agreement with the experimental data @xcite , even though the limitations of the rpa theory do not allow a description of the fragmentation of the strength of the higher energy state . we have restricted our study to charge conserving nuclear excitations . there are indications that the effects of the tensor - isospin force are more relevant on the charge - exchange excitations @xcite . the presence of the tensor forces seems to be relevant also in the description of the reactions between heavy nuclei @xcite .

the addition of a @xmath0 gauge singlet supermultiplet to the mssm ( minimal supersymmetric extension of the standard model ) has been motivated by several issues : ( i ) the solution of the so - called @xmath1-problem ; ( ii ) the possibility of spontaneous breaking of the cp symmetry ; ( iii ) the upper limits on the higgs mass that are relatively low in the mssm could be relaxed ; ( iv ) the neutral particle spectrum is enriched by two scalars and one fermion , which could mix with other higgs bosons and neutralinos and modify their physical properties ; ( v ) possible deviations from the mssm could urge to enlarge the experimental searches , which have been centred upon the mssm . we refer here to the supersymmetric model with a gauge singlet in its simplest version , as defined below , as the ( m+1)ssm . however , according to our analysis , the strong constraints from lep experiments , when combined with theoretical and phenomenological requirements , select the ( m+1)ssm parameters in such a way that the gauge singlet scalars and fermions are basically decoupled and the rest of the spectrum effectively reproduces a restricted class of the mssm . because of this similarity , the motivations ( ii)-(v ) fade out as these issues are excluded or diminished after such a detailed investigation . nevertheless , this also means that phenomenological analyses of the mssm apply to a large extent to the ( m+1)ssm as well ! still , if supersymmetry is found , one can go further and look for some peculiarites that could distinguish the two kind of models , specially in the case in which the lightest supersymmetric particle is the singlet fermion , or singlino . the ( m+1)ssm can also be experimentally tested from the fact that the effective parameters in the equivalent mssm are very constrained . the present paper is devoted to a thorough discussion of the model in its version with only dimensionless couplings in its superpotential , and universal boundary conditions for the soft supersymmetric breaking terms at the gut scale . with these assumptions , the dimensionality of the parameter space of the ( m+1)ssm equals that of the mssm . our aim is to understand the pattern of the particle spectrum , correlations among the particle masses , and features which are in common with or different from the mssm . an analysis of the complete parameters space of the model requires numerical methods , as a scanning over the independent parameters at the gut scale , their subsequent evolution using the rge down to the weak scale , the minimization of the higgs potential , the implementation of theoretical and experimental constraints and finally the computation of the particle masses and some mixing angles . many phenomenological analyses of the model have appeared before @xcite . the purpose of the present work is twofold . first , we implement the most recent experimental constraints on sparticles , higgs and top quark masses from the lep and the tevatron @xcite . second , we explain the results of our numerical procedure with the help of analytic approximations and inequalities among the parameters . once the vacua which break colour , electromagnetism or correspond to @xmath2 are forbidden and the experimental constraints are imposed , these formulae allow us to understand the essential features of the bounds on and correlations among some particle masses . quite a lot of attention is paid to the property found in @xcite , namely , that the singlet scalar acquires a v.e.v . which is much larger than the higgs ones . indeed , this has the important consequence that the singlet fermion and scalars quite decouple from the the other multiplets as already mentioned . in section [ sec : properties ] we review the motivations for the model , its essential differences with respect to the mssm due to the enlargement of higgs sector and the status of the spontaneous cp violation . the universality assumption for the soft terms and some of its consequences are explained in section [ sec : universal ] . section [ sec : mass ] contains the minimum conditions for the scalar potential as well as the scalar higgs masses with the radiative corrections included in an approximation that is generally good , though we use the full expressions @xcite in our numerical calculations . in section [ sec : stability ] we derive some approximate constraints from the stability of the vacuum to understand the pattern of the physical predictions . in section [ sec : inos ] we discuss some correlations in the mass spectrum along the same approximations . then we summarize our implementation of the experimental and theoretical constraints on the model , in section [ sec : cuts ] . the main results of the careful scanning of the parameter space are displayed in the figures in section [ sec : spectrum ] , with some emphasis on the sparticles more likely to be discovered at the colliders , with the exception of the higgs sector that has been already presented @xcite . the results are summarized in the last section . some useful solutions of the rge for the soft terms ( with generic boundary conditions , for completeness ) are listed in the appendix . the superpotential of the ( m+1)ssm ( where family mixing is neglected and irrelevant for the analysis in this paper ) is given by @xmath3 where the chiral superfields are denoted as follows : @xmath4 ( singlet ) , @xmath5 and @xmath6 ( higgs doublets ) , @xmath7 ( antitop ) , @xmath8 ( antibottom ) , @xmath9 ( antitau ) , @xmath10 ( top - bottom doublet ) , @xmath11 ( tau - neutrino doublet ) . let us briefly review the present status of these motivations for the ( m+1)ssm . a supersymmetric mass term @xmath12 is added to the mssm superpotential in order to avoid a potentially dangerous peccei - quinn symmetry and to allow for an acceptable minimum of the higgs potential . the analysis of the @xmath13 breaking then requires @xmath1 to be roughly of the same order of magnitude as the soft terms that embody the supersymmetry breaking effects in the two higgs sectors . this interesting hierarchy problem , the so - called @xmath1-puzzle , has motivated many a suggestion to link the supersymmetric mass @xmath1 to supersymmetry breaking @xcite . in any instance , some kind of new physics is requested and in this sense the mssm that looks mostly economical at low energies is not necessarily so at the unification @xmath14 level . a natural solution @xcite is provided by the addition of a singlet @xmath15 soft terms that break supersymmetry can induce a v.e.v . for @xmath4 and an effective supersymmetric mass @xmath16 of the right order magnitude . this mechanism asks for the @xmath17 self - coupling in the superpotential which also breaks the unwanted peccei - quinn symmetry . however , this coupling would be forbidden and the singlet model inoperative if @xmath4 has non - trivial quantum numbers with respect to broken symmetries beyond those of the standard model . this phenomenon is frequently found in superstring compactifications . we take here the pragmatic viewpoint that all the dimension three interactions that preserve the low - energy local and global symmetries are to be included . as we shall discuss later , very low values of the coupling @xmath18 are consistent with the phenomenological constraints . in this case , it could be a relic of non - renormalizable interactions after the decoupling of , @xmath19 extra @xmath20 gauge symmetries . actually , the `` sliding '' singlet v.e.v . compensates for the smallness of the couplings @xmath18 and @xmath21 and this considerably reduces the fine - tuning problem of the @xmath1-term . such a possibility is indeed supported by our analysis . supersymmetric mass terms are forbidden by the @xmath22 symmetry of the purely cubic superpotential assumed in ( [ a ] ) . this discrete symmetry also prevents @xmath4 from getting a large v.e.v . through large radiative corrections to singlet tadpole terms , which invalidate other supersymmetric models with gauge singlets @xcite . on the other hand , the @xmath23 symmetry can give rise to cosmological problems in the form of domain walls which are produced during the electroweak phase transition @xcite . possible solutions to this problem have been discussed in @xcite and include inflation of the weak scale , embedding of the discrete symmetry into a gauge symmetry at the planck scale or re - introducing of the @xmath24-term in the superpotential . ( in the presence of the singlet , @xmath24 can be orders of magnitudes smaller than the weak scale , in which case our subsequent analysis is not affected ; however , the @xmath24-problem was part of the motivations of the model . ) in any case , since we concentrate in this paper on the particle physics aspects of the model and do not discuss cosmological issues , our results will be independent of the form of the solutions of the domain wall problem . the study of spontaneous cp breaking in the ( m+1)ssm also differs from the mssm . indeed , cp violating vacua can be induced only if there are more than one phase - dependent interaction in the scalar potential . in the mssm , the only tree - level term is the soft scalar mass term @xmath25 . analytic terms of higher dimensions generated at the one loop level can conspirate with the tree - level one to induce spontaneous cp violation . however , this has been shown @xcite to entail a very low upper limit on one of the scalar masses which is inconsistent with present experiments at lep . the classical scalar potential processes a richer structure in the ( m+1)ssm , with three additional complex scalar couplings ( one of dimension four , two cubic soft terms , see next section ) . the conditions for cp spontaneous breaking in this type of models have been shown @xcite to imply a negative eigenvalue in the scalar ( mass@xmath26 matrix . however , top quark - squark loop corrections could shift this eigenvalue to positive values . thus spontaneous cp violation would occur for appropriate values of the parameters and one or two light scalars are predicted @xcite . this issue certainly deserves a more detailed comparison with phenomenology ; it lies outside the purview of the present investigation . because the very small region in the ( m+1)ssm parameter space consistent with spontaneous cp violation and those selected by our cp conserving solutions are apart , we disregard _ ab initio _ the question of non - trivial phase vacua . upper bounds have been derived for the mass of the lightest higgs scalar in the mssm that take into account the large one loop corrections @xcite as well as some two - loop effects @xcite . this prediction is of paramount importance in view of the experimental searches of the higgs scalar in the lep 200 and the lhc . the ( m+1)ssm has an additional contribution @xmath27 to the higgs ( mass@xmath28 it has been shown @xcite that requiring the theory to remain perturbative up to the gut scale reduces this additional contribution to @xmath29(30 gev ) . the upper bounds on the lightest higgs boson mass are considerably lowered if one also imposes all phenomenological constraints needed for the model to be realistic . this has been done in the case of universal supergravity couplings @xcite and also for some special non - universal models @xcite . note that a relatively small increase in the higgs mass may affect the experimental signatures . since the predictions of the phenomenologically constrained ( m+1)ssm for the higgs sector has been the subject of a recent publication @xcite , we shall only give a short account of this important matter here . as for the particle spectrum , this model presents several new features as compared to the mssm . the supermultiplet @xmath4 consists of two scalars of opposite cp and one majorana fermion . after the @xmath13 breaking , these states get mixed with the neutral states in the higgs doublet and gauge supermultiplets , leading to a slightly richer spectrum . in particular , since the visibility of physical states depends on their couplings to the gauge bosons and to matter , some particles could be relatively light if they are mainly gauge singlets . thus , some of the limits on neutral particles obtained at lep in the mssm framework do not immediately apply to the alternative ( m+1)ssm . the discussion in this paper as well as in our previous ones @xcite is also aimed to persuade experimentalists to analyze their data on a more general basis . in the @xmath30 limit , the gauge singlet decouples and the remaining theory looks like the mssm @xcite . however , this limit leads to a particular version of the mssm with a strong correlation among the soft parameters , as we shall discuss later on . the singlet scalar takes a very large v.e.v . of @xmath31 , but the fermion singlet and both scalar singlets are rather light . of course , the large dimension of the scalar field manifold allows for a rich structure in the overall scalar potential and leads to the existence of many local extrema @xcite . one has to carefully check that the phenomenologically viable vacuum is favoured with respect to the unwanted ones . among these there are the usual @xmath32 breaking ones , corresponding to vacua with quark and lepton quantum numbers @xcite , @xcite , @xcite . in the ( m+1)ssm there is the additional possibility of a non - vanishing charged higgs field vacuum . however , the necessary condition for this particular charge violating classical solution is a relatively large coupling of the higgses to the singlet @xcite , so that this situation is quite naturally avoided by a bound on that coupling.our theoretical and experimental constraints turn out to automatically selected the right region of the parameter space . actually , as we shall discuss here below , an acceptable minimum of the potential requires some inequalities among the soft supersymmetry breaking terms which would tend to induce @xmath32 breaking vacua as well . the radiative corrections due to the gauge interactions then allow for a compromise while restricting the useful region in the parameter space . since the electroweak gauge symmetry breaking is induced by radiative corrections involving the top yukawa coupling a well - known property in supersymmetric extensions of the standard model the existence of a physically acceptable vacuum in the ( m+1)ssm is a purely quantum effect . the supersymmetric part of the scalar potential is obtained from the superpotential through the well - known expression as a sum of @xmath33 and @xmath34 terms . the supersymmetry breaking part of the lagrangian contains all soft terms consistent with the symmetries of the superpotential : gaugino masses , trilinear analytic scalar couplings and scalar mass terms . in the ( m+1)ssm they are as follows ( with the same notation for the chiral matter supermultiplets and their first component complex scalars ) : @xmath35 where @xmath36 are the gauginos associated to the @xmath37 and @xmath38 gauge symmetries . only the terms involving the top scalars have been retained in ( [ b ] ) , with @xmath39 but similar interactions for all other squarks and sleptons are to be understood . all the parameters are taken to be real ( up to irrelevant phases ) and flavour mixing is neglected . notice that these soft terms contain all the dimensionful parameters of the model , which are all naturally related to the supersymmetry breaking scale @xmath40 . in order to study the particle spectrum and @xmath13 symmetry breaking in the ( m+1)ssm , one should take the parameters in ( [ a ] ) and ( [ b ] ) at the relevant scales , the fermi scale @xmath41 gev or @xmath42 instead , it is natural to fix these parameters at the level of the underlying supergravity theory . we shall consider here the unification scale @xmath43 gev to define them . this is just a few orders of magnitude below the planck mass and one can presumable assume an effective renormalizable theory at @xmath44 once the parameters are given at @xmath45 their corresponding values at @xmath46 or @xmath47 are calculated by the integration of the renormalization group equations ( rge ) for the ( m+1)ssm @xcite . we present in the appendix the analytic solutions of the ( m+1)ssm rge in the approximation @xmath48 which turns out to be good for most of the set of parameters that will be selected on a phenomenological basis . however , for the sake of precision , we numerically integrate the rge in our analysis , the expressions in the appendix being reserved for analytical discussions of the numerical results . in practice , we renormalize the parameters down to the scale @xmath46 through the supersymmetric rge , and then apply the coleman - weinberg corrections to the higgs potential to take into account the decoupling of the stops at @xmath40 . the important hypothesis to be made in this paper is the so - called `` universality '' or flavour - independence for the soft - terms in ( [ b ] ) . for the supersymmetry breaking scalar interactions it amounts to assume a common value for all the parameters @xmath49 in ( [ b ] ) @xmath50 @xmath51 and the same values ( at @xmath52 ) for the scalar masses in ( [ b ] ) , @xmath53 this universality follows if the direction of supersymmetry breaking , characterized by the goldstino components , corresponds to fields with purely gravitational couplings to the relevant chiral superfields ( for instance , if it is along the dilaton - axion direction in superstring inspired theories @xcite ) . of course , it follows in general if the goldstino components have equal couplings to all relevant chiral superfields ; an example is given by the `` large radius limits '' of orbifold compactifications in string theory if supersymmetry breaking occur along the modulus direction @xcite . as for gaugino masses , the universality hypothesis at @xmath52 reads @xmath54 in grand unified gauge theories many of these relations will be required by the enlarged gauge group at the tree - level . at the one loop level one expects some threshold corrections to the universality of the soft terms already at the reference scale @xmath55 the deviations from ( [ c ] ) - ( [ e ] ) in the matter sector can be phenomenologically important because of the existent experimental bounds on fcnc effects but they are less relevant to our discussion in this paper . deviations from the universality assumption are also possible in the framework of superstring theory , since the moduli sector can couple differently to the various chiral fields . since the assumptions @xmath56 and @xmath57 play an important role in our phenomenological analysis , some of our results depend on the universality of the supersymmetry breaking parameters . however a similar analysis of the ( m+1)ssm has been published @xcite for the case of non - universal soft terms in the framework of an orbifold compactification of the superstring , and these results are quite analogous to those presented here . the most important consequences of the flavour independence of soft terms concern the @xmath13 symmetry breaking . it could be induced by a negative value of @xmath58 , but for the potential to be bounded from below and to avoid colour or @xmath59 symmetry breaking the universal parameter @xmath60 has to be positive , so that @xmath61 . for this reason , @xmath62 has to be enforced by renormalization effects . this is a well - known feature of the mssm as well , entailing a lower limit on the top yukawa coupling . in the ( m+1)ssm , the sign of @xmath58 could be made to flip by the coupling of the singlet fields even for a light top quark @xcite . the experimental value of top quark mass is perfectly consistent , however , with a negative @xmath58 at low energies from the effects of top - stop quantum loops . furthermore the cubic soft terms can also induce spontaneous symmetry breaking if the value of the parameter @xmath63 in ( [ c ] ) is large enough . hence large values of @xmath64 are preferred to give the singlet scalar a v.e.v .. however , with the universality assumption ( [ c ] ) , the squark and sleptons could also get non - vanishing v.e.v.s @xcite . interestingly enough , the gauge interactions reduce the @xmath63-parameters for matter fields while preserving the value of @xmath65 as can be seen from the expressions in the appendix . thus , phenomenologically consistent ( m+1)ssm s with the condition ( [ c]),([d]),([e ] ) , can be build at the price of restricting the space of the supersymmetry breaking parameters . this is the most important restriction imposed on the model by the universality hypothesis , as compared to the mssm @xcite . otherwise , the effects of this universality on the @xmath13 spontaneous breaking are similar to those in the mssm . another consequence of this assumption is to reduce the number of free bare parameters in the ( m+1)ssm to five ( before imposing the physical constraints ) just like in the mssm : the dimensionful parameters @xmath8 and @xmath1 of the latter are replaced in the former by the dimensionless parameters @xmath21 and @xmath66 . at low energies , the soft terms as obtained from the rge depend on the parameters @xmath67 on the choice of the gut scale , @xmath68 through @xmath69 @xmath70 ( a typical value being @xmath71 and on the gauge and yukawa couplings . this calculation has to be done numerically in general . however , from the dimensionality of these running parameters , they can be generically expressed at low energy in terms of their universal initial values as @xmath72 where the coefficients depend on the various gauge and yukawa couplings . for the simple case where the top yukawa coupling dominates over the others , the dependence on the latter reduces to simple expressions in terms of the ratio @xmath73 where @xmath74 is the fixed point value of the top yukawa coupling . the results of this approximation are presented in the appendix and are used for the sake of some qualitative understanding of the numerical results . with the recent measurements of the top quark mass , @xmath75 is bounded by @xmath76 ( this is higher than the minimum value required to induce mediative @xmath13 breaking ) . as an example , we list the low energy soft parameters at the fixed point value @xmath77 the coefficients are approximated to a reasonable precision . @xmath78 1truecm let us first restrict the discussion to the physically relevant vacua with ( real ) v.e.v.s for the field @xmath79 the scalar potential restricted to this sector reads @xmath80 @xmath81 where @xmath82 is the quantum correction to the effective scalar potential beyond the rge effects included in the running parameters of ( [ f ] ) . these radiative corrections are important , in particular in the evaluation of the lightest higgs boson mass . the relevant contribution comes from the top - stop sector because of the relatively large value of the top yukawa coupling . hence we only include the top quark and the two stop states , @xmath83 in the usual coleman - weinberg expression , @xmath84 the appropriate mass terms in ( [ g ] ) are then the top quark mass @xmath85 and the stop masses @xmath86 all the parameters in ( [ f ] ) and ( [ g ] ) are to be taken at the scale @xmath87 these radiative corrections are mostly due to the splitting between the quarks and the squarks . however , even if one starts from universal parameters at @xmath45 the rge running introduce some asymmetry in the stop sector as can be seen from the formulae in the appendix . the resulting splitting in @xmath88 produces a sizeable effect in @xmath89 and in the scalar mass spectrum . the full one loop expressions for the mass matrices and the potential minimization are available in the literature . all the numerical calculations in the present paper are performed by using the formulae in @xcite . however , for the sake of the analytic approximations discussed below we use a simple approximation by keeping only the lowest relevant order in the development in terms of the stop splitting , @xmath90 the minimization of the scalar potential amounts to the following conditions @xmath91 $ } & $ = $ & $ 0 $ \\ & & \\ $ h_{2}\left [ m_{2}^{2}+\hat{\mu}^{2}\left ( 1+\frac{h_{1}^{2}}{s^{2}}\right ) -% \frac{1}{2}\overline{g}^{2}\left ( h_{1}^{2}-h_{2}^{2}\right ) \right ] + $ & & \\ & & \\ $ + \hat{b}\hat{\mu}\frac{h_{1}}{h_{2}}+\beta _ { t}\left [ \left ( m_{t}^{2}+m_{q_{3}}^{2}+a_{t}^{2}\right ) \left ( l+\varepsilon \right ) \right . $ & & \\ & & \\ \multicolumn{1}{r}{$\left . -\left ( m_{t}^{2}+m_{q_{3}}^{2}\right ) + 2m_{t}^{2}l\right ] $ } & $ = $ & $ 0 $ \\ & & \\ $ s\left [ m_{s}^{2}+\hat{\mu}\frac{h_{1}^{2}+h_{2}^{2}}{s^{2}}+2\nu ^{2}+a_{\kappa } \nu + \left ( \hat{b}+\nu \right ) \hat{\mu}\frac{h_{1}h_{2}}{% s^{2}}+\right . $ & & \\ & & \\ \multicolumn{1}{r}{$\left . + \beta _ { t}\hat{\mu}^{2}\left ( \frac{h_{2}}{s}% \right ) ^{2}\left ( l+\varepsilon \right ) \right ] $ } & $ = $ & $ 0$% \end{tabular } \label{i}\ ] ] where @xmath92 and the variables @xmath93 have been introduced . an one loop term has been conveniently included in the definition of the effective parameter @xmath94 ( of the mssm ) . the @xmath95 mass matrix for the cp even scalars ( in the same approximation ) has the following matrix elements @xmath96 the parameter l characterizes the top - stop mass splitting and can be relatively large , while @xmath97 depends on the low energy scale chosen to define the running parameters we take @xmath98 in this paper so that @xmath99 the cp odd scalar ( mass)@xmath100 matrix with the would - be goldstone boson projected out is as follows , @xmath101 the minimization of the potential fixes @xmath102 and @xmath103 to be negative , so that the diagonal entries in ( [ m ] ) are positive . the ( mass)@xmath100 of the charged scalar reads @xmath104 the radiative corrections are incorporated into the cp odd neutral and charged scalar masses through the shift of the @xmath105 parameter as defined in ( [ k ] ) @xcite . although all the numerical results in this paper are obtained by numerical minimization of the complete one loop potential ( [ f ] ) , ( [ g ] ) and the numerical determination of the mass eigenvalues , it is interesting to get some insight into the results through simple analytic approximations . indeed , many features of the particle spectrum of the ( m+1)ssm are related to an important property , first noticed in @xcite , namely : the only physically acceptable solutions of the vacuum equations ( [ i ] ) are large singlet solutions : @xmath106 indeed , solutions with @xmath107 lead to one or more light states in the spectrum that are excluded by experiments , specially by lep experiments . examples can be found in @xcite , where the parameters have been chosen to yield light spectrum . the general character of this feature of the ( m+1)ssm has been carefully checked in our numerical analysis . having settled this fact one can go on to establish several approximate consequences . \(a ) the @xmath4 v.e.v is approximately obtained from the third equation in ( [ i ] ) with @xmath108 , @xmath109 the potential at this value of @xmath110 has also to be negative in the allowed region of the parameter space , as discussed in more detail in section [ sec : stability ] . this additional condition requires @xmath111 this approximation turns out to be very accurate . the parameters @xmath112 and @xmath113 are only slightly renormalized if @xmath114 actually , these relatively low values of the couplings are also required by our numerical study of the model , so that the condition ( [ neccon ] ) approximately applies to the bare universal parameters @xmath115 yielding the strong constraint in the parameter space : @xmath116 the @xmath4 v.e.v . in ( [ o ] ) defines the effective parameters @xmath105 and @xmath117 analogous to those often denoted @xmath8 and @xmath1 in the mssm . indeed , up to corrections of @xmath118 the first two equations in ( [ i ] ) are the same as in the mssm with the introduction of these parameters . notice that @xmath119 should be of @xmath120 hence the well known `` fine - tuning '' on the mssm free parameter , @xmath121 is replaced here by the requirement of no strong hierarchy between the @xmath21 and @xmath122 yukawa couplings , as the @xmath13 invariant scale of the singlet field is dynamically fixed by supersymmetry breaking , accordingly to ( [ o ] ) . with @xmath123 one recovers , in this approximation , the mssm relation @xmath124 only the dominant terms of the radiative corrections are retained . because the denominator in ( [ p ] ) has terms of either sign @xmath125 it is relatively sensitive to radiative corrections which tend to decrease @xmath126 with respect to the tree - level approximation . \(b ) in the neutral scalar mass matrices ( [ l],[m ] ) as well as in the neutralino mass matrices , the mixing of the singlet fields to the others is always proportional to @xmath127 and @xmath128 , hence small . this is the reason why , in most of the allowed points of the parameter space , the singlet sector of the theory , after dynamically producing the effective parameters ( [ k ] ) , is almost decoupled from the rest of the theory that resembles to the mssm . \(c ) the mssm is formally obtained in the limit @xmath129 with @xmath130 and @xmath117 fixed @xmath131 in this limit the singlet sector decouples from the doublet higgs sector , and the singlet masses are : .5 truecm @xmath132 .5 truecm the singlet particles become light in the small @xmath133 limit . we argue in section [ sec : stability ] that this limit is always associated to the singlet decoupling limit @xmath134 some of the physical consequences of a light singlino are discussed in section [ sec : inos ] . the neutral and charged higgs ) there is a negative contribution @xmath135 to @xmath136 this is the origin of the condition @xmath137 which is sufficient to avoid @xmath59 charge breaking through @xmath138 which is satisfied anyway by the models that are consistent with the other phenomenological constraints . ] sector , as well as other neutralinos and charginos follow the mssm pattern . in this sense the ( m+1)ssm reduces to the mssm in this limit but with an important restriction in the parameter space needed to dynamically implement the physically acceptable vacuum . the spontaneous breaking of @xmath13 symmetry in the ( m+1)ssm restricts even more the @xmath139 parameter subspace than in the mssm under the same assumption of universality . the minimum discussed in section 3 has to be cosmologically flavourless and stable with respect to other possible minima . the first constraint comes from the fact that the @xmath140 v.e.v . is induced through the `` @xmath63-mechanism '' @xcite with @xmath141 as given in ( [ neccon1 ] ) . it is well known that large @xmath133 values may induce charged and coloured vacua . in order to avoid slepton v.e.v.s one has to impose @xmath142 at a scale @xmath143 where @xmath144 characterizes the scale of the would - be slepton @xmath145 from the rge solutions in the appendix this implies @xmath146 @xmath147 . \label{s}\end{aligned}\ ] ] the most dangerous v.e.v.s comes from the electron sector with the smallest yukawa and the largest scale @xmath148 , and ( [ r ] ) at this scale finally becomes @xmath149 this can be combined with the condition @xmath150 to obtain restrictions on the ratio @xmath151 for this purpose we introduce a parameter to characterize the fine - tuning of @xmath152 with respect to @xmath153 @xmath154 and obtain from ( [ t ] ) the upper bound on @xmath155 @xmath156 for @xmath157 ( [ t ] ) yields the limit @xmath158 for @xmath159 can become much larger than @xmath160 at the price of a fine - tuning of @xmath161 corresponding to @xmath162 of @xmath163 though these conditions are only valid under the universality assumption for soft terms , analogous ones can be derived along the same lines for other models ( see , @xmath19 @xcite ) . ( strictly speaking , ( [ t ] ) can only be analytically derived @xcite if the scalars are relatively degenerate . but recent numerical studies suggest that it is approximatelly valid under more general conditions @xcite ) the next step is to avoid minima with @xmath164 with @xmath165 in this analysis the radiative corrections discussed in section 3 can be relevant . of course , in our numerical analysis they are included , but in order to understand the origin of some of the constraints on the ( m+1)ssm parameter space , they are neglected in some of the expressions here below . in the effective two - step minimization of the scalar potential , one finds for the phenomenologically acceptable solution discussed in the previous section the approximate value at the minimum @xmath166 where the first term comes from the pure singlet sector and the last one from the effective mssm potential defined with the parameters given by ( [ k ] ) . this physical minimum has to be compared with another one , with @xmath167 and @xmath168 the potential at this minimum is @xmath169 in principle one should look for the cosmological formation and stability of the minimum ( [ x ] ) with respect to ( [ y ] ) . for simplicity we replace the rigorous constraints by the simpler condition that @xmath170 _ i.e. , _ that the absolute minimum is the physical one.this stability condition differs from the corresponding one within the mssm in two respects : ( i ) the presence of the pure singlet negative term in ( [ x ] ) favours @xmath171 ( ii ) in the mssm , @xmath172 is replaced by @xmath173 in ( [ y ] ) , so that @xmath174 is more dangerous in the ( m+1)ssm . the comparison of ( [ x ] ) and ( [ y ] ) leads to the condition ( at the scale @xmath46 ) , @xmath175 on the other hand , from the rge solution in the appendix one gets @xmath176 and the measured values of the top mass does not allow for small @xmath177 so that @xmath178 . in addition , in the next section we derive from the experimental limits on the chargino masses a bound on @xmath179 @xmath180 ( at this point , one is introducing experimental information on the spectrum . as already stressed , it is the interplay between the lep bounds on sparticle masses and the vacuum conditions that constrains the parameter space . ) now , from ( [ minm0 ] ) and ( [ a1 ] ) , ( [ z ] ) can be well approximated by @xmath181 notice that the necessary condition @xmath182 , which follows from ( [ minm0 ] ) and ( [ z ] ) , implies @xmath183 namely , the condition ( [ neccon1 ] ) . for @xmath184 one obtains from ( [ v ] ) and ( [ a2 ] ) the bound @xmath185 so that @xmath18 is always relatively small , @xmath186 this is related to the large values taken by the @xmath4 v.e.v .. indeed by replacing ( [ o ] ) and ( [ minm0 ] ) in ( [ a2 ] ) , we obtain @xmath187 of course , the limits ( [ a3 ] ) and ( [ a4 ] ) are relatively loose bounds and most of the solutions of our numerical studies correspond to smaller ( larger ) values of @xmath18 ( @xmath188 respectively ) . also to be noticed is the behaviour of @xmath110 for fine - tuning of the parameter @xmath162 in ( [ u ] ) : @xmath189 as @xmath190 while @xmath18 decreases as @xmath191 for @xmath192 the situation is more involved since one can allow for large values of @xmath133 with some fine - tuning of the @xmath161 ratio . the relations corresponding to ( [ a3 ] ) and ( [ a4 ] ) are respectively @xmath193 for moderate values of @xmath194 ( [ a5 ] ) corroborate our conclusions from ( [ a3 ] ) and ( [ a4 ] ) . in the fine - tuning limit , @xmath195 , the singular behaviour of @xmath18 in the parameter @xmath162 is apparent . however , in this case our approximations are not powerful enough to explain the lower bound @xcite @xmath196 that is found in our numerical analysis . another interesting constraint follows from ( [ p ] ) which implies @xmath197 neglecting the top - stop radiative corrections and using ( [ a1 ] ) and ( [ minm0 ] ) yields , @xmath198 from ( [ o ] ) and ( [ v ] ) , one derives the limits : @xmath199 for moderate values of @xmath162 this implies @xmath200 at the price of some fine - tuning and @xmath201 one can obtain @xmath18 slightly larger than @xmath202 concerning the important parameter @xmath203 some simple qualitative predictions can be obtained from the following expression , @xmath204 which is equivalent to ( [ p ] ) by the minimum conditions ( [ i ] ) . notice that @xmath205 . from the renormalized expression ( [ z6 ] ) in the appendix , one sees that for @xmath206 the sign of @xmath105 is preserved and its magnitude increased by the gauge renormalisation proportional to @xmath207 which increases the numerator in ( [ q ] ) . therefore , in this case , @xmath208 and it can take relatively small values for @xmath209 instead , for @xmath210 where ( [ w ] ) also applies , the @xmath207 term in ( [ z6 ] ) tends to exceed the @xmath211 term so that in most cases @xmath212 and its magnitude is relatively large . ( exceptions are the special cases with small @xmath75 and @xmath213 with a positive large @xmath214 . ) therefore , with these simple analytic approximations for the vacuum conditions , together with the approximated condition from the experimental limits on the gauginos ( to be discussed later on ) , one easily understands the qualitative pattern of the allowed parameter space as obtained from the detailed numerical investigation . for instance , ( [ a4 ] ) and ( [ a5 ] ) are related to the decoupling of the singlet sector , whereas sizeable mixing - which requires @xmath215 - is possible only with the fine - tuning ( and signs ) discussed below ( [ a5 ] ) . let us first consider the chargino masses given by the expression @xmath216 with @xmath217 for the lep experimental limit we take @xmath218 gev . of course , the constraints from this limit on the parameters are the same as in the mssm ( with @xmath214 in the more restricted range determined by the ( m+1)ssm as discussed in section [ sec : spectrum ] ) . on the other hand , this constraint plays an important role in the discussion of the previous section as well as in this section . because the scale of the soft supersymmetry breaking parameters is set by the weak scale , we conveniently express the dimensionful parameters in terms of @xmath219 in this section . one can use the experimental bound , together with ( [ p ] ) and the ( [ a1 ] ) , to deduce the relations @xmath220 let us now define a parameter @xmath221 by , @xmath222 which , for @xmath223 ( which turns out to be always the case in our numerical analysis ) , satisfies the approximate inequalities @xmath224 then , from ( [ e1 ] ) and ( [ e2 ] ) one gets the following expression for the lightest chargino mass , @xmath225 so that the approximate bounds , @xmath226 are obtained by varying the parameters @xmath227 and @xmath203 and by using the relation between the gluino and wino masses @xmath228 this defines a relatively narrow band for @xmath229 as a function of the gluino mass @xmath230 , that is reasonably reproduced in our numerical analysis . it just expresses the fact that the lightest chargino is mostly a wino , with a component of @xmath231 of the higgsino . now , from ( [ e5 ] ) and the experimental lep bound , @xmath232 one gets a limit on the gluino mass , @xmath233 gev ( from the previous lep bound , @xmath234 the corresponding result was @xmath235 gev , but needed a more detailed approximation than ( [ e5 ] ) ) . this corresponds to @xmath236 gev . notice that these are only approximations aimed to explain rather than predict the numerical results that take all the parameters into account . as already stressed , this result is analogous to the limits on the chargino masses in the mssm . for @xmath237 the off - diagonal terms become relatively unimportant and @xmath238 therefore the lightest chargino is mostly a wino with @xmath239 and the other chargino has mass @xmath240 we next consider the neutralino sector , with five majorana fermions . it is particularly important since it includes the lightest odd r - parity state , or lsp , and since its phenomenology can deviate from that predicted in the mssm because of the presence of the singlet fermion ( singlino , @xmath241 the mass terms for the neutralinos are , @xmath242+\hat{\mu}\widetilde{h}_{1}\widetilde{h}_{2}\nonumber + m_{1}\widetilde{b}\widetilde{b}+m_{2}\widetilde{w}_{3}\widetilde{w}_{3}\\ & & + m_{z}\left ( \widetilde{b } \sin \theta _ { \mathrm{w}}+ \widetilde{w}_{3}\cos \theta _ { \mathrm{w}}\right ) \left ( \widetilde{h}_{1}\cos \beta + \widetilde{h}_{2}\sin \beta \right ) \label{e6 } \end{aligned}\ ] ] the point to be stressed is the relatively small mixing of the singlino to the higgsinos . this is because @xmath243 and @xmath244 , @xmath245 hence the singlino remains an almost pure state of mass @xmath246 , and the higgsino - gaugino sector is analogous to the charged one . the experimental constraints on the chargino imply , as just discussed , that the lightest chargino is mostly a gaugino , which will also imply an analogous situation in the non - singlet neutralino sector . one finds indeed that the lightest states are : than a `` bino '' @xmath247 but this can be easily incorporated in the discussion that follows where a @xmath248-state is assumed for simplicity . ] one which is mostly @xmath249 with a mass @xmath250 , and another one , mostly @xmath251 with mass @xmath252 from the bound on the chargino mass one has @xmath253 the most important difference between the ( m+1)ssm and the mssm could be in the lsp character : unless one badly violates the universality assumption for the scalar masses , the lsp is mostly a @xmath248 in the mssm , while the @xmath254 is also a lsp candidate in the ( m+1)ssm . let us consider the condition for a _ singlino _ lsp , @xmath255 which in terms of the bare parameters yields : @xmath256 this corresponds to some constraints in the coupling constants as well.from ( [ a3 ] ) one infers @xmath257 the @xmath4 v.e.v . has to be relatively large because of ( [ a7 ] ) , which gives @xmath258 from ( [ a6 ] ) and ( [ e8 ] ) one derives the constraint , @xmath259 on the other hand , ( [ e8 ] ) also implies @xmath260 @xmath261 . in other words , the light singlino scenario corresponds to gaugino masses being the largest soft terms , a `` gaugino dominated scenario '' . this implies many correlations between sparticle masses . if the singlino is the lsp , the second lightest neutralino being mostly a @xmath262 should be quite long - lived . indeed , its decay into @xmath263 has to go through its mixing to higgsinos . then the decay @xmath264 is proportional to @xmath265 conversely , if ( [ e8 ] ) is not realized , then @xmath254 will decay into the lsp , mostly @xmath266 with a similar coupling ( @xmath254 would be produced in higgs or higgsino decays ) . let us now turn to the slepton sector . the lowest lying states are the @xmath267 sneutrinos @xmath268 and the `` right - handed '' sleptons , @xmath269 . context , the @xmath270 superpartners can become considerably splitted and much lighter than the other sleptons , as a consequence of renormalization between the planck and @xmath271 scales , as noticed in @xcite . however in this paper we are assuming @xmath272 at @xmath273 and we consistently neglect this possibility . ] this follows from the assumption of universality for the soft terms . the approximated expressions for the slepton masses as given in the appendix are @xmath274 searches at lep 1 put a limit of roughly @xmath275 on both of them . for the relatively large values of @xmath126 that tend to prevail in the ( m+1)ssm @xmath276 with the lep 1 results on the chargino masses , as pointed out above , one had a bound @xmath277 and the bounds on @xmath278 gave some useful information . with the lep 1.5 results , the limit on @xmath160 implied by ( [ e6 ] ) is enough to ensure @xmath279 a more interesting issue is provided by the interplay between chargino and slepton searches at lep 200 . in order to compare their masses , we first replace @xmath207 in ( [ e9 ] ) in terms of the lightest chargino mass in ( [ e4 ] ) to obtain @xmath280 which gives a qualitative explanation of figure ( 2 ) in the next section . the difference between @xmath281 and @xmath282 is almost a measure of the parameter @xmath283 in the future searches for these particles at relatively high energies . as a matter of fact , this reverses the situation at lep 1 , where the chargino was expected to be lighter than the selectron . indeed by comparing ( [ e9 ] ) and ( [ e4 ] ) , we find that the chargino is always lighter than the selectron if @xmath284 or , equivalently , as @xmath285 this is quite well reproduced by our numerical results . in particular , in the light singlino scenario , the term @xmath60 can be almost neglected in ( [ e11 ] ) so that the selectron and chargino masses are strongly correlated . in this section we describe the numerical procedure employed to produce the figures ( 1 ) - ( 6 ) . as mentioned above , the essential independent parameters of the model are ( apart from the lepton and light quark yukawa couplings and the known gauge couplings ) three yukawa couplings @xmath286 , @xmath287 and @xmath66 and , under the universality hypothesis at @xmath288 , the three soft parameters @xmath207 , @xmath289 and @xmath211 . since these are the only dimensionful parameters in the theory and the scale is finally set by the experimental value of @xmath219 , the true independent parameters are the five dimensionless quantities @xmath290 , @xmath291 , @xmath292 , @xmath293 and @xmath294 . the numerical results presented here have been obtained by scanning over @xmath295 points in this five dimensional parameter space . ( here we used a logarithmic measure , but we have checked the results with different measures as well . a lot of scanning of particular regions of the parameter space has been carried out in order to verify the boundaries of the parameter space and the mass ranges , but the results of these scannings are not shown in the figures . ) in each case we integrate the one loop renormalization group equations down to the electroweak scale @xmath41 gev . to one loop accuracy it is sufficient to use @xmath46 as the low energy scale , independently from the exact particle masses of @xmath296 . at the one loop level , one can also take advantage of ( [ obvious ] ) and compute the coefficients in these equations , what saves a lot of computer time to be used for a better scanning . actually , also two loop renormalization group equations have been considered in @xcite . in this case , however , the low energy scale has to be defined more precisely : the decoupling of particles with masses of @xmath296 has to be properly taken into account ; these masses , in turn , are only known once the higgs potential ( which depends on these masses via the radiative corrections ) has been minimized . as a consequence , the numerical procedure becomes much more envolved and allows only to study much less points in the parameter space . on the other hand , the corresponding numerical results hardly deviate from ours . individual points in the parameter space may well lead to somewhat different particle masses , once two loop contributions have been taken into account . allowed mass ranges and correlations , however , remain practically unchanged . we found it much more important numerically to minimize the full one loop coleman - weinberg higgs potential including the non - logarithmic contributions than to include two loop logarithms . having obtained the parameters at low energy it is most convenient to check first the absence of slepton v.e.v.s according to eq . ( [ t ] ) at the scale @xmath297 . next we minimize in each remaining case the effective potential ( [ f ] ) including the coleman - weinberg radiative corrections ( [ g]),([h ] ) , numerically . we test , whether the minimum with @xmath298 , @xmath299 and @xmath300 is the lowest one , and dismiss the corresponding set of parameters otherwise . we calculate @xmath126 and discard the initial conditions that lead to @xmath301 ( since we assume @xmath302 in the present work ; larger values turn out to be somewhat disfavoured in the model ) . in the remaining cases we determine the overall scale of the dimensionful parameters by identifying @xmath303 with @xmath304 , and compute the physical masses of all particles . then we impose the following experimental constraints : for the top quark pole mass @xmath305 ( where we take the leading @xmath306 corrections to the pole mass into account @xcite ) we require @xmath307 @xcite . next we impose the lep @xmath308 lower bound on the lightest chargino mass @xmath229 : @xmath309 gev @xcite . as a result , nearly all other experimental bounds on new particles turn out to be satisfied automatically , with some exceptions discussed in section [ sec : spectrum ] . indeed , though the experimental bounds on sleptons , stop , gluinos , as well as on the decays of the @xmath310 into neutralinos and higgs scalars are implemented in our code , these conditions only eliminate a few marginal points in the parameter space . this remarkable property also can be seen from the plots in the next section . let us briefly discuss the range of the bare parameters , which turns out to be consistent with our theoretical and experimental constraints . concerning the yukawa couplings we first remark that , as in the mssm , the known range of the top quark mass together with the solution of the rge for @xmath286 as given in the appendix allows easily to obtain the allowed range for @xmath290 : @xmath311 gev leads to @xmath312 . the other two bare yukawa couplings @xmath291 and @xmath292 turn out to be fairly small , @xmath313 and @xmath314 , and relatively closely related , @xmath315 . as already remarked below eq . ( [ e5 ] ) , the experimental lower limit on @xmath229 implies a lower limit on the soft susy breaking parameter @xmath207 . the numerical analysis is in good agreement with the approximate analytic result and leads to @xmath316 . no strict upper limits ( in the absence of fine tuning constraints ) on the soft susy breaking parameters have been obtained . the scalar mass term @xmath60 turned out to be unconstrained by our analysis ; we assumed , however , @xmath317 . the numerical analysis also confirms the allowed range for the third susy breaking parameter @xmath211 as a function of @xmath207 and @xmath289 , eq . the corresponding lower limit on @xmath211 is a particularity of the ( m+1)ssm , since it is required by the need to destabilize the scalar potential in the singlet direction . the correlations between the particle masses and the bare parameters can finally be obtained from the approximate analytic solutions given in sect ( 6 ) and the appendix ; in table 1 we present the precise numerical parameters and masses for two particular generic cases : one with @xmath211 negative and a non - singlet lsp , and one with @xmath211 positive and a singlet lsp . in order to allow to study the correlations among the particle masses we choose to plot all masses against the lightest chargino mass @xmath229 . each point in these plots corresponds to one of the @xmath318 points in the parameter space , which satisfy all our theoretical and phenomenological constraints . the density of points in the plots is clearly not uniform ; regions of low density ( if not completely empty and hence forbidden ) , typically towards larger particle masses , correspond to regions where more and more fine tuning is required . we will not , however , employ quantitative `` fine tuning constraints '' in this paper in order to constrain the particle masses from above ; a reader , who is interested in `` probable particle masses '' , can obtain them by investigating the relative densities in our plots . let us first consider the gluino mass @xmath230 , fig . 1 . as already noted in eq . ( [ e5 ] ) , given the present experimental bounds , @xmath319 and @xmath230 turn out to be strongly correlated . 1 agrees well with the approximate analytic relation ( [ e5 ] ) , and one sees that the lower experimental bound on @xmath229 turns into a lower bound @xmath320 gev on the gluino mass within our model . for large masses , the ratio between the lightest chargino and the gluino is close to @xmath321 as predited by the chargino being a wino , but slightly smaller as the parameters in the chargino mass matrices are dynamically correlated ( as discussed in section [ sec : inos ] ) . next we turn to the slepton masses . the lightest charged `` right - handed '' slepton mass @xmath322 is plotted against @xmath229 in fig . we see that @xmath309 gev implies @xmath323 gev in our model . in most cases , as noted below ( [ e11 ] ) , the charged slepton mass @xmath324 has the tendency to be smaller than @xmath229 ( except for large values of the bare scalar mass @xmath289 ) , although it always satisfies @xmath325 in agreement with the analytic approximation ( [ e11 ] ) . in fig . 3 we plot the sneutrino mass @xmath326 against @xmath229 . given @xmath309 gev , the lep 1 lower limit on @xmath326 only eliminates a tiny additional corner in the parameter space . the neutral scalar higgs sector has already been discussed in detail in ref . @xcite , in particular with respect to future experiments at lep @xmath327 . although in @xcite only the lep @xmath328 bound on the chargino mass @xmath229 had been taken into account , the results concerning the neutral higgs sector remain essentially unchanged . therefore we briefly repeat only the essential features . in order to take the experimental constraints on neutral higgs bosons properly into account both their masses @xmath329 their couplings to the @xmath310 bosons have to be evaluated . in fact , the negative lep @xmath328 results on neutral cp - even higgs scalars @xcite eliminate only a tiny region of the parameter space of the present model , corresponding to @xmath330 gev and a coupling @xmath331 as large as in the non - supersymmetric standard model . on the other hand , a large region with smaller values of @xmath332 , but small couplings @xmath331 remains unconstrained by lep @xmath328 . in this region the lightest cp - even higgs scalar is essentially a gauge singlet and hence decoupled . in view of this possibility the experimental searches for and the theoretical upper limits on the masses of the lightest cp - even higgs scalar have to be turned into considerations of the `` lightest visible cp - even higgs scalar '' . fortunately it turns out within the present model @xcite , that upper limits exist also on the lightest visible cp - even higgs scalar ( which could be the second lightest ) , varying from @xmath333 to @xmath334 gev for gluino masses ( as a measure of the susy breaking scale ) from @xmath328 to @xmath335 tev . similar complications arise also in the cp - odd higgs sector , which contains a gauge singlet state as well . here no upper limit can be derived , and we just remark that a visible cp - odd higgs scalar below @xmath336 gev is impossible within the present model and the present experimental constraints on @xmath229 . as in the mssm the mass of the visible ( non - singlet ) cp - odd higgs scalar is actually just somewhat below the mass @xmath337 of the charged higgs ; cf . the upper left entry of the cp - odd mass matrix eq . ( 15 ) and eq . ( 16 ) . in fig . 4 we plot @xmath337 against @xmath229 , and we see that the present lower limit on @xmath229 implies @xmath338 ( increasing rapidly with increasing lower limits on @xmath229 ) . the range of @xmath339 is fairly restricted , and depends strongly on the sign of the bare susy breaking parameter @xmath211 : for @xmath340 we have @xmath341 , whereas for @xmath342 we find both signs , but large absolute values for @xmath343 : @xmath344 or @xmath345 . since these inequalities hardly depend on the mass of the chargino , a figure does not provide additional information . next we turn to the squarks . because of renormalization effects between @xmath346 and @xmath219 ( cf . the appendix ) the lightest stop eigenstate is the lightest of all squarks ; subsequently we concentrate on this particle . in fig . 5 we plot its mass @xmath347 against @xmath229 , and we see that the present lower limit on @xmath229 implies @xmath348 gev . a possible `` supersymmetric '' top quark decay mode is thus practically excluded within the present model . finally we consider the neutralino sector . as discussed in chapter ( 6 ) , two different scenarios are possible within the present model , depending on the mass of the nearly pure singlino : in the case of a heavy singlino , the two lightest neutralino states are a practically pure bino with a mass @xmath349 and a practically pure wino with a mass @xmath350 . the numerical procedure confirms this approximate analytic result to a very high accuracy . since furthermore the chargino mass @xmath229 is close to @xmath351 , cf . ( [ e5 ] ) , the two lightest neutralino masses can easily be obtained as a function of @xmath229 and a figure is of no further use . here the lsp is , of course , the bino , and the supersymmetric decays of all sparticles will proceed as in the mssm with a suitable set of parameters . the case of a light singlino is actually more interesting . from the numerical analysis we find , that now the second and third neutralinos correspond to the above mentioned nearly pure bino and wino states . in fig . 6 we plot the mass of the lightest singlino , @xmath352 , against the bino mass @xmath349 ( for those parameters where @xmath353 ) . now supersymmetric decays of all sparticles will start as in the mssm , until the bino state is produced . ( a direct decay into the singlino lsp will be heavily suppressed due to its tiny coupling . ) finally the bino will decay into the singlino lsp , producing an additional cascade , whose consequences on experimental susy searches still have to be worked out in more detail . 1truecm in the present paper we have presented an exhaustive overview over the parameter space and the particle spectra , which are consistent with the correct vacuum structure and present experimental constraints , within the ( m+1)ssm under the universality hypothesis . since the scalar potential and the corresponding minimization conditions differ considerably from the mssm due to the presence of the additional singlet superfield , the outcome of this analysis was far from predictable . we have seen that still a large region in the parameter space consistent with all constraints exists . this region is characterized , however , by a ) small values of the new yukawa couplings @xmath287 and @xmath66 , b ) a large value of the scalar singlet @xmath354 , c ) a strong tendency of the singlet states to decouple from the rest of the spectrum . in spite of the complexity of the model this region allows for fairly accurate analytical approximations to the vacuum minimization conditions and particle masses , which we have worked out and presented in some detail . the near decoupling of the singlet states leads to a particle spectrum , which is close to the mssm for a suitable set of the corresponding parameters . our figures show strong correlations or inequalities among the particle masses , which allow for experimental falsifications of our underlying hypothesises . for instance , as in the mssm with the universality hypothesis , sparticles are not light enough to play an important role for the radiative corrections to @xmath355 @xcite . only the higgs boson , the lightest chargino , sleptons and neutralinos could be light enough to be discovered at lep 2 ; gluinos , squarks and charged or cp - odd higgs scalars are necessarily too heavy in our model . our approximate analytic results allow to translate future possibly negative sparticle searches into constraints on the soft susy breaking parameters . obvious are lower limits on the universal gaugino mass @xmath207 from negative chargino searches ( cf . ( [ e5 ] ) ) and , of course , negative gluino searches . from eqs . ( 50 ) , one can furthermore easily deduce excluded regions in the @xmath356 plane from negative `` right - handed '' slepton and sneutrino searches . these relations are actually not different from the mssm . ( 28 ) , however , allows to turn these constraints on @xmath289 and @xmath207 into constraints on @xmath211 , which is a particular feature of the ( m+1)ssm . if future sparticle searches turn out to be successful , and the mass pattern predicted by our model turns out to be confirmed , the question arises whether the model can be distinguished from the mssm with corresponding parameters . the answer depends on the corresponding parameters of the ( m+1)ssm . first , there still exist small corners in the parameter space , where mixing with the singlet states either in the neutral higgs sector or in the neutralino sector could be possible ; in this case the couplings of the corresponding visible particles could be substantially and hence measurably smaller than in the mssm . a more likely possibility - but by no means certain - is the singlino lsp scenario . as discussed in the previous section , in this case sparticle decays will produce an additional cascade with respect to the mssm . this scenario desserves further phenomenological investigations . in this appendix we list the analytic solutions of the rge for soft susy breaking terms in the approximation where the dependence on all yukawa couplings but the top one , @xmath357 are neglected . also , flavour mixing is neglected in these expressions ( see @xcite ) for solutions that fully take family mixing into account ) . however , for the sake of generality , universality is _ not _ assumed for the soft terms at the unification scale , @xmath358 all quantities are defined at the scale @xmath359 ( unless stated otherwise ) and the boundary conditions at the scale @xmath360 are generally denoted as @xmath361 we define the following quantities : @xmath362 the index @xmath363 refers to the @xmath364 and @xmath365 gauge groups , respectively , and @xmath366 is the sum of the their quadratic casimir eigenvalues for the three fields in the i - th yukawa coupling . we replace the integral @xmath367 by the value @xmath368 of @xmath369 at the scale @xmath359 corresponding to a landau pole at @xmath370 by defining @xmath371 the solutions for the yukawa coupling are then as follows @xmath372 the soft trilinear terms are as follows : @xmath373 soft supersymmetry breaking scalar masses can also be written in terms of @xmath374 and j ( or @xmath375 as follows : @xmath376 _ { t=0 } \nonumber \\ \overline{m}_{0}^{2 } & = & \frac{1}{3}\left ( m_{2}^{2}+m_{t}^{2}+m_{q}^{2}\right ) _ { t=0 } \nonumber \\ k & = & \left ( 1-\rho \right ) \left ( a_{t_{0}}+\xi m_{0}\right ) ^{2}-\xi ^{2}m_{0}^{2}+ \nonumber \\ & & + 2\left ( \frac{16}{3}g_{3}^{2}+3g_{2}^{2}+\frac{13}{9}g_{1}^{2}\right ) \left ( \xi + 1\right ) tm_{0}^{2 } \label{z5}\end{aligned}\ ] ] finally , let us recall that the renormalization effects on the effective variables @xmath105 and @xmath377 are easily obtained by taking the limit in which the singlet fields are decoupled at its classical value . then , @xmath378 99 j. ellis , j. f. gunion , h. e. haber , l. roszkowski , f. zwirner , _ _ phys . rev . _ _ * * d 39 * * ( 1989 ) 844;l . durand and j. l. lopez , _ phys . * b 217 * ( 1989 ) 463;m . drees , _ int . j. mod . phys . _ * a 4 * ( 1989 ) 3635 . u. ellwanger , m. rausch de traubenberg , c. a. savoy , _ phys . * b 315 * ( 1993 ) 331 . t. elliott , s. f. king and p. l. white , _ phys . * b 305 * ( 1993 ) 71;t . elliott , s. f. king and p. l. white , _ phys . * b 314 * ( 1993 ) 56;t . elliott , s. f. king , p. l. white , _ _ phys . _ _ * * d 49 * * ( 1994 ) 2435 . s. f. king and p. l. white , _ _ phys . _ _ * * d 52 * * ( 1995 ) 4183 . u. ellwanger , m. rausch de traubenberg and c. a. savoy , _ _ z . phys.__**c 67 * * ( 1995 ) 665 . r. m. barnett et al . , _ _ phys . _ _ * * d 54 * * ( 1996 ) 1 . aleph collaboration , _ _ phys . _ _ * * b 373 * * ( 1996 ) 246 . u. ellwanger , _ _ phys . lett . _ _ * * b 303 * * ( 1993 ) 271 . g. f. giudice and a. masiero , _ _ phys . _ _ * * b 206 * * ( 1988 ) 480 . nilles , m. srednicki , d. wyler , _ phys . * b 120 * ( 1983 ) 346 . h. nilles , m. srednicki , and d. wyler , _ phys . * b 124 * ( 1983 ) 337 . a. lahanas , _ phys . * b 124 * ( 1983 ) 341 . u. ellwanger , _ phys . * b 133 * ( 1983 ) 187;j . bagger , e. poppitz , _ phys . lett . _ * 71 * ( 1993 ) 2380;j . bagger , e. poppitz , l. randall , _ nucl . * b 455 * ( 1995 ) 59 . s. a. abel , s. sarkar and p. l. white , _ _ nucl . _ _ * * b 454 * * ( 1995 ) 663 , s. a. abel , preprints hep - ph 9603301 and hep - ph 9609323 . a. pomarol , _ phys . _ * b 287 * ( 1992 ) 331 . j. romo , _ phys . _ * b 173 * ( 1986 ) 309 . k. s. babu and s. m. barr , _ phys . _ * d49 * ( 1994 ) 2156 . y. okada , m. yamaguchi and t. yanagida , _ progr . * 85 * ( 1991 ) 1 , _ phys . * b 262 * ( 1991 ) 54 . j. ellis , g. ridolfi and f. zwirner , _ phys . * b 257 * ( 1991 ) 83 , _ phys . _ * b 262 * ( 1991 ) 474 . h. haber and r. hempfling , _ phys . * 66 * ( 1991 ) 1815.p . chankowski , s. pokorski and j. rosiek , _ phys . _ * b 274 * ( 1992 ) 191 . j. a. casas , j. r. espinosa , m. quiros and a. riotto , _ nucl . phys . _ * b 436 * ( 1995 ) 3 . m. carena , j. espinosa , m. quiros , c. wagner , _ phys . * b 335 * ( 1995 ) 209 . u. ellwanger , m. rausch de traubenberg , _ z. phys . _ * c 53 * ( 1992 ) 521 ; p. binetruy and c. a. savoy , _ phys . * b 277 * ( 1992 ) 453.j . r. espinosa and m. quiros , _ phys . * b 279 * ( 1992 ) 92;g . kane , c. kolda and j. wells , _ _ phys . _ _ * * 70 * * ( 1993 ) 2686;j . r. espinosa and m. quiros , _ _ phys . _ _ * * b 302 * * ( 1993 ) 51.u . ellwanger and m. lindner , _ _ phys . lett . _ _ * * b 301 * * ( 1993 ) 365 ; ph . brax , u. ellwanger and c. a. savoy , _ phys . * b 347 * ( 1995 ) 269 . derendinger , c. a. savoy , _ nucl . _ * b 237 * ( 1984 ) 307 . frre , d. r. t. jones , s. raby , _ _ nucl.phys . _ _ * * b 222 * * ( 1983 ) 11 . g. gamberini , g. ridolfi and f. zwirner,__nucl.phys . _ _ * * b 331 * * ( 1990 ) 331 . m. olechowski and s. pokorski , _ _ phys . lett . _ _ * * b 214 * * ( 1988 ) 393 . j. a. casas , a. lleyda and c. muoz , _ _ nucl.phys . _ _ * * b 471 * * ( 1996 ) 3.a . j. borner , hep - ph/9506409 . r. barbieri , s. ferrara and c. a. savoy , _ phys . * b 119 * ( 1982 ) 343.r . arnowitt , a. chamseddine and p. nath , _ _ phys . _ _ * * d 49 * * ( 1982 ) 970.l . hall , j. lykken and s. weinberg , _ _ phys . _ _ * * d 27**(1983 ) 2359 . v. s. kaplunovski and j. louis , _ _ phys . _ _ * * b**306 ( 1993 ) 269.r . barbieri , j. louis and m. moretti , _ _ _ * * b**312 ( 1993 ) 451.a . brignole , l. e. ibez and c. muoz , _ _ nucl . _ _ * * b**422 ( 1994 ) 235 . r. tarrach , _ _ nucl . phys.__**b 183 * * ( 1981 ) 384 . j. kamoshita , y. okada and m. tanaka , _ phys . * b 328 * ( 1994 ) 67 . j. d. wells , c. kolda and g. l. kane , _ phys . * b 338 * ( 1994 ) 219 . g. l. kane , r. g. stuart , j. d. wells _ phys . * b 354 * ( 1995 ) 350 . d. garcia and j. sola , _ _ phys . lett.__**b 354 * * ( 1995 ) 335 . p. h. chankowski and s. pokorski , _ phys . * b 366 * ( 1996 ) 188 . j. ellis , j. l. lopez and d. v. nanopoulos , _ phys . * b 372 * ( 1996 ) 95 . r. barbieri , l. hall and a. strumia , _ phys . _ ph . brax and c. a. savoy , _ nucl . * b 447 * ( 1995 ) 227 . = 25 truecm * figure :* gluino mass @xmath230 versus the chargino mass @xmath229 ( all in gev ) slepton mass @xmath324 versus @xmath229 sneutrino mass @xmath379 versus @xmath229 lightest stop mass @xmath347 versus @xmath229 charged higgs mass @xmath337 versus @xmath229 mass of the second lightest neutralino @xmath380 versus the mass of the lightest neutralino @xmath381 in those cases where @xmath382 is essentially a singlino * table :* parameters and particle masses ( in gev ) for two particular cases ( with @xmath383 not too large ) . the three neutral cp even higgs scalars are denoted by @xmath384 , @xmath385 and @xmath386 ( in the order of increasing masses ) , the two neutral cp odd higgs scalars by @xmath387 and @xmath388 , and the two lightest neutralinos by @xmath382 and @xmath389 , respectively . in the first case @xmath382 is essentially a non - singlet , in the second case it practically a pure singlino .

the charge and magnetic instabilities of one dimensional materials are well understood in terms of fermi surface ( fs ) nesting . here it is shown that for two - dimensional materials , features in maps of the bare susceptibility can be understood in terms of a series of fs ` nesting curves ' , and the dominant instabilities are generally related to _ double nesting _ features . the analysis shows how to locate these nesting features in momentum ( @xmath2 ) space , often providing analytical expressions . as an application , the full evolution with doping of the leading magnetic instabilities for several families of cuprates is presented , both in the conventional hartree - fock ( hf ) plus rpa ( hf+rpa ) and in a gutzwiller approximation ( ga+rpa ) calculation . the analysis provides a pseudogap candidate and makes the surprising prediction that the ` checkerboards ' seen in scanning tunneling microscopy ( stm ) studies are _ not _ the same phase as the ` stripes ' in la@xmath3sr@xmath4cuo@xmath5 ( lsco ) . the present results should find extensive utilization . first , for weakly correlated two - dimensional systems the hf+rpa results provide an essentially complete solution to the nesting problem , as long as the interaction @xmath0 is @xmath2-independent . the ga+rpa extends the results into the intermediate coupling regime . in both cases , the magnetic instabilities are determined by zeroes of the stoner denominator , @xmath6 here for cuprates @xmath7 is the hubbard @xmath0 in hf+rpa , and a more complicated @xmath2-dependent object @xmath8 in the ga+rpa calculation . thus , the leading hf+rpa instability is simply associated with the maximum of the bare susceptibility @xmath9 , while the leading gutzwiller instability can be shifted by the @xmath2-dependence of @xmath8 . it will be clear that the same analysis can be extended to any two - dimensional material . the cuprates appear to be in an intermediate coupling regime where the gutzwiller results can be expected to provide a good approximation to the phase diagram at @xmath10 . thus , in the cuprates , recent quantum monte carlo ( qmc)@xcite and ` quasiparticle - gw ' ( qp - gw)@xcite calculations have been able to reproduce experimental arpes spectra of optimally and overdoped cuprates , starting essentially from lda bands and calculating the self - energy self consistently . in the qp - gw approach , the self energy is calculated as a convolution over a renormalized rpa susceptibility . not only is the low - energy dispersion renormalization reproduced , but also the ` waterfall ' effect@xcite , which represents the dressing of low - energy quasiparticles by ( mainly ) spin fluctuations . extension of these results to the hole - underdoped regime will require identification of the phase or phases responsible for the pseudogap . since these are most likely to be incommensurate density wave or ` stripe ' phases , the qmc calculations have a severe problem of limited @xmath2-resolution , while the qp - gw calculations are ideally suited to handle this . such an approach has had considerable success with electron - doped cuprates , where the leading instability is commensurate @xmath11 . ( we set the lattice constant @xmath12 . ) the paper is organized around the stoner criterion as follows . sections 2 - 3 describe the calculations of the zero frequency bare susceptibility @xmath13 , and determining its maximum @xmath14 . in section 2 we introduce the concept of nesting curve , associated with the nesting criterion @xmath15 , where @xmath16 is the ( anisotropic ) fermi wave vector , and we demonstrate that @xmath14 is generally associated with a double nesting . we apply this concept to the cuprates , taking the hopping parameters from tight - binding one - band fits to the lda dispersions , and show that over most of the doping range there is a unique @xmath14 , with a surprising electron - hole symmetry . in section 3 we extend this analysis to the more complicated hole doping regime for doping @xmath17 between half filled and the doping of the van hove singularity ( vhs ) . here competing magnetic phases are found , and the dominant phase is sensitive to material parameters , being different for different cuprates . section 4 describes the ga+rpa technique and introduces the corresponding @xmath18 . section 5 presents the resulting gutzwiller magnetic phase diagrams . it is found that the phase diagram of la@xmath3sr@xmath4cuo@xmath5 ( lsco ) is distinct from that of most other cuprates , but that for all cuprates , using a bare hubbard @xmath19 , the paramagnetic state is unstable at the ga level for all relevant dopings , including the overdoped regime . a discussion is presented in section 6 , and conclusions in section 7 . the present calculations are based on tight - binding parametrizations of typical dispersions for single - layer cuprates , including models of la@xmath3sr@xmath4cuo@xmath5 ( lsco ) , nd@xmath3ce@xmath4cuo@xmath5 ( ncco ) , and bi@xmath20sr@xmath20cu@xmath21o@xmath22 ( bi2201 ) , table i in appendix [ sec : band - parameters ] . for ncco , bi-2201(2 ) , and lsco(2 ) , the parameters are based on a tight - binding fit to the lda bands@xcite , while sets bi-2201(1 ) and lsco(1 ) are direct fits to the experimental bands . in all cases , @xmath23 dispersion is neglected , approximating the cuprates as two - dimensional . 0.5 cm the real part of the susceptibility @xmath24 is dominated by a series of plateaus , with the largest susceptibility systematically shifting from one plateau to another as a function of doping . figure [ fig:1 ] shows that these and related features dominate @xmath25 over the full doping range , and further reveals a striking quasi - electron - hole symmetry of the evolution . however , instead of being symmetric about @xmath26 , the evolution is symmetric about the doping of the van hove singularity ( vhs ) , @xmath27 , where the fermi energy coincides with the vhs , @xmath28 , and the density of states ( dos ) diverges logarithmically . figure [ fig:1](a ) shows @xmath25 calculated along the high symmetry axes of lsco , for the full electron - doping range @xmath29 from 1 to 2 , with an extension to the hole doping of the vhs ( blue curves ) . near half filling the susceptibility is dominated by the well - known plateau@xcite near @xmath30 ( red and blue curves ) , which peaks at the vhs . as electron - doping increases , the peak shifts to a second plateau near @xmath31 ( green curves ) , then to a third near @xmath32 for a nearly full band ( brown curves ) . the _ same _ sequence is followed for hole doping , fig . [ fig:1](b ) . in addition to these plateaus , there is an additional @xmath32-centered feature , which is prominent near the doping of the vhs , but has largely disappeared by half filling . this feature will be referred to as the antinodal nesting ( ann ) plateau . bi2201 displays the same plateaus , with the same quasi - electron - hole symmetry , fig . [ fig:1](c , d ) . a key difference is that the @xmath30-plateau of bi2201 is convex over a considerably wider doping range than in lsco , before turning concave near the vhs . this has a consequence that the ann peak dominates over a wide doping range in bi2201 , but only in the immediate vicinity of the vhs in lsco . at the vhs , the bare susceptibility diverges logarithmicly both at @xmath32 ( where it is equal to the dos ) and at @xmath30 , but the latter divergence is quite weak and not apparent in the numerical calculations of fig . 1 . despite the apparent complexity of these evolving susceptibility patterns , the plateau evolution can be understood in detail , with analytic formulas for the positions of all dominant peaks . and folded back into the first bz ( dashed ) . ( b ) electron - like fermi surface ( solid ) shifted by the double nesting vector @xmath33 and folded back into the first bz ( dashed ) . the ( double ) nesting points are marked by a solid dot . ( c , d ) construction of nesting curves ( dashed ) for hole - like ( c ) and electron - like ( d ) fs s ( solid lines ) from condition eq . [ eq : g2 ] . the fs is shown in the first bz whereas the nesting curves are folded into the momentum space @xmath34 as defined by the dashed square . the arrows indicate scattering processes which lead to ` double nesting ' as explained in the text . the dotted lines corresponds to the boundary of the magnetic bz in the first and enlarged zone . ( a , c ) parameters for ncco , @xmath35 ; ( b , d ) parameters for lsco(1 ) , @xmath36.,title="fig:",width=340 ] and folded back into the first bz ( dashed ) . ( b ) electron - like fermi surface ( solid ) shifted by the double nesting vector @xmath33 and folded back into the first bz ( dashed ) . the ( double ) nesting points are marked by a solid dot . ( c , d ) construction of nesting curves ( dashed ) for hole - like ( c ) and electron - like ( d ) fs s ( solid lines ) from condition eq . [ eq : g2 ] . the fs is shown in the first bz whereas the nesting curves are folded into the momentum space @xmath34 as defined by the dashed square . the arrows indicate scattering processes which lead to ` double nesting ' as explained in the text . the dotted lines corresponds to the boundary of the magnetic bz in the first and enlarged zone . ( a , c ) parameters for ncco , @xmath35 ; ( b , d ) parameters for lsco(1 ) , @xmath36.,title="fig:",width=340 ] 0.5 cm the generic evolution of the plateaus from @xmath37 can be understood with reference to the ` nesting curves ' , fig . [ fig : new ] . for the generic case of two fermi surface segments , a nesting curve can be defined as the locus of all points @xmath38 , where @xmath39 is a point on the @xmath40th fs , fs@xmath41 , with the restriction that when fs@xmath21 is shifted by @xmath2 it is tangent to fs@xmath20 . for the cuprates , there is usually just a single fs section [ an exception is given below in section iii , fig . 6 ] , and the nesting curves simplify to plots of @xmath42 for any fermi momentum @xmath16 , fig . [ fig : new ] . since the point @xmath30 lies on horizontal and vertical planes of reflection symmetry for the susceptibility , the nesting curves must be supplemented by their reflections about the lines @xmath43 , @xmath44 . [ fig:2 ] shows three sets of nesting curves corresponding to three hole dopings for lsco ( dispersion 1 ) , @xmath17 = 0.41 ( red curves ) , 0.62 ( blue ) , and 0.79 ( green ) . frames ( b - d ) show the corresponding susceptibility maps @xmath45 , and it can be seen that the ridges in @xmath25 are exactly given by the nesting curves . [ for convenience , the @xmath36 data is replotted in fig . note that the susceptibility maps are plotted over the range of @xmath46 between 0 and @xmath47 , whereas the nesting curves are plotted over the wider range 0 to @xmath48 . ] furthermore , the dominant peaks in @xmath25 correspond to the intersection of two nesting curves . by drawing the original and @xmath2-shifted fermi surfaces ( fss ) , figs . [ fig : new](a ) , ( b ) , it can be seen that this overlap corresponds to the simultaneous nesting of two different sections of fs . hence the term ` double nesting ' . as shown in figs . [ fig : new](c ) , ( d ) this kind of ` double nesting ' can originate from either the scattering between points on opposite segments ( cf . example in fig . [ fig : new](c ) ) or adjacent segments ( cf . example in fig . [ fig : new](d ) ) of the fs . in the example shown in fig . [ fig : new](c ) we denote the scattered states on the fs as @xmath49 and @xmath50 which yield the points on the nesting curves @xmath51 and @xmath52 . thus in this case ` double nesting ' ( @xmath53 ) occurs when @xmath54 which generally can only be fulfilled for points near the antinodes of hole - like fs s . these are thus referred to as antinodal nesting ( ann ) features . more common is the situation sketched in fig . [ fig : new](d ) where we consider the scattered states @xmath49 and @xmath55 . in this case the ` double nesting ' condition @xmath53 can only be fulfilled for @xmath56 ( which is trivial since initial and final states of the two scattering processes are identical ) or @xmath57 . the latter condition implies that this kind of ` double nesting ' generally affects the nodal states and leads to scattering vectors @xmath2 close to @xmath58 , generating the @xmath30-plateau . a slightly different point of view might help clarify the role of double nesting . tangency of two fermi surface segments [ ` nesting ' ] means a stability of fs overlap , in that the nesting persists if one displaces the nesting vector in a particular direction . then double nesting means tangency along two surface segments , so nesting persists if one displaces the nesting vector in two ( rather than one ) directions . 0.5 cm the origin of the plateau transitions is now apparent : as hole doping increases , the fs shrinks to a small pocket near @xmath32 before disappearing . the nesting curves shrink in a similar fashion , but with a doubled radius , since @xmath15 . thus the dominant overlap shifts from near @xmath59 at the vhs ( fig . 2(b ) ) towards @xmath31 ( fig . 2(c ) ) , and finally towards @xmath32 , in fig . 2(d ) , thereby explaining the plateau evolution . for electron doping , the fermi surface ultimately closes at @xmath30 , leading to the same sequence of plateau transitions , as illustrated in fig . [ fig:3 ] , where the nesting maps are directly superposed on the susceptibility curves . in the doping range relevant to the cuprates , the physics of ncco is dominated by the @xmath30 plateau , fig . [ fig:3]a and red curves in fig . 1(a ) , which shrinks to a point at the end of the ` hot - spot ' regime . 0.5 cm in all cases , the dominant peak lies along a high symmetry axis , and the doping dependence of its position can readily be found from the dispersion @xmath60 - 4t'c_x(a)c_y(a ) \nonumber \\ -2t''[c_x(2a)+c_y(2a ) ] \nonumber \\ -4t'''[c_x(2a)c_y(a)+c_y(2a)c_x(a ) ] \ > , \label{eq:2}\end{aligned}\ ] ] where @xmath61 and @xmath62 is an integer ( or half - integer ) . for this dispersion , the vhs is generally at @xmath31 , or @xmath63 . the positions of the peaks in fig . [ fig:2](b , c ) and fig . [ fig:3](b , c ) lie along the zone boundary ( @xmath64 ) with @xmath65 , \label{eq:1x}\ ] ] with @xmath66 $ ] and @xmath67/[4t'']$ ] . for electron [ hole ] doping the peak is exactly at @xmath31 when @xmath682(t-2t''')$ ] . beyond this point , the peak is at @xmath69 , @xmath70 given by eq . [ eq:1x ] with @xmath712t'+2t''')/[4(t''-[+]2t''')]$ ] and @xmath722(t-2t''')]/[4(t''-[+]2t''')]$ ] . since a peak at @xmath32 corresponds to ferromagnetism , the above reproduces the common finding that a nearly empty or nearly full band tends to be ferromagnetic . the ann peak @xmath53 in fig . [ fig : new]a satisfies @xmath73/[2(t'-2t'')]$ ] . thus the peak susceptibility is generally associated with double nesting . the only exception we have found to this is a tendency to remain commensurate for a finite doping range about high symmetry points such as @xmath11 or @xmath74,@xcite a form of van hove nesting@xcite . 0.5 cm while the above sequence is completely generic , holding for all cuprates and being electron - hole symmetric , the additional features associated with the vhs are considerably more variable . [ fig:4 ] illustrates the low - hole doping regime for three dispersions proposed for the cuprates . at half filling , all three have a peak susceptibility on the @xmath30 plateau , associated with conventional staggered antiferromagnetism ( saf ) . at low doping the peak is either commensurate or @xmath75 incommensurate , with @xmath76 increasing with doping . for finite hole doping a new feature emerges , a roughly @xmath77-shaped peak along the zone diagonal , at @xmath78 . from the nesting curves , it can be seen that the peak is associated with nesting of the flat sections of the bands near @xmath31 hence the name antinodal nesting ( ann ) but the largest susceptibility lies along the zone diagonal , where both @xmath31 and @xmath79 nesting occur simultaneously , fig . [ fig : new]a . this ann feature has an interesting relation with ` hot spot ' physics@xcite . a ` hot spot ' is a point of the fermi surface which is simultaneously on the antiferromagnetic ( af ) zone boundary [ diagonal of the paramagnetic brillouin zone ] . from fig . [ fig : new](c ) it becomes apparent ( cf . large dot ) that the image of the af zone boundary in the extended @xmath80-map intersects the ` nesting curves ' exactly at the points of ` double nesting ' . but the image of the af zone boundary gets folded in the @xmath2-map onto the zone diagonal , @xmath81 , thereby generating the usual ann feature . thus all the diagonal ann peaks in @xmath2 arise from hot spots in @xmath82 . as the doping of the vhs is approached the hot spots move toward @xmath83 and the nesting curve moves to @xmath32 . since the susceptibility at @xmath32 is equal to the dos , it diverges at the vhs , thereby controlling the magnetic instability . the appearence of the ann peak leads to a competition between two different kinds of magnetic order , and the doping evolution of the susceptibility maps diverges . for most dispersions studied , including the left and central columns of fig . [ fig:4 ] , the ann intensity grows and the dominant peak changes discontinuously from the saf plateau to ann nesting . for the bi-2201 dispersion ( left column ) this ann peak becomes dominant at about @xmath17=0.2 , and evolves smoothly to @xmath32 at @xmath17=0.4 . if one unfolds the nesting curve , one sees that this happens exactly at the doping of the vhs , when the fs passes through @xmath31 . the central column , corresponding to an extreme dispersion proposed for lsco to enhance antinodal nesting , displays a much more complicated nesting map over a limited doping range close to the vhs , but also displays a dominant peak at @xmath32 exactly at the vhs . this map will be explained below . finally , for the right column , corresponding to a more conventional dispersion for lsco , the ann peak is weaker than the saf peak except in the immediate vicinity of the vhs . in this case , the dominant susceptibility peak remains commensurate at @xmath30 until the doping of the vhs , jumps to @xmath32 at the vhs , then jumps back and smoothly evolves to @xmath84 incommensurate [ this sequence can also be followed in fig . 1(a),(b ) ] . note further , in fig . 5(i , j ) , that there is a wide doping range where the susceptibility peak remains commensurate , at @xmath30 . [ since @xmath30 lies along several mirror planes , the corresponding susceptibility is in general a local maximum or minimum . ] 0.5 cm the complicated nesting map of the middle column near the vhs is explained in fig . [ fig:5 ] . the dispersion is such as to produce an ` extended vhs',@xcite which first intersects the @xmath85 axis at a point @xmath86 . for larger hole doping , the fs has two sheets , one a squarish barrel centered at @xmath32 , the other a pocket centered at @xmath31 , fig . [ fig:5](b ) . the intrasheet @xmath15 nesting maps are shown by the red curves in fig . [ fig:5](a ) . in addition to the one associated with the barrel fs , there are two overlapping segments associated with the pockets , but translated by the @xmath2 folding to the vicinity of @xmath32 . as in the other frames of figs . 2 - 4 , these nesting curves exactly match some of the ridges seen in the susceptibility . however , there are additional ridges , associated with nesting between two different fs segments . thus , the blue curves in fig . [ fig:5](a ) represent barrel - pocket nesting that is , the locus of the @xmath2-vectors needed to shift the pocket until it is tangent to some point on the barrel fs . for instance , the points labelled 1 , 2 , 3 represent translations equivalent to those shown in fig . [ fig:5](b ) . the new nesting curves can easily be found numerically , by requiring that the two fs sections have a common tangent at the point of osculation . similarly , the green curve near @xmath30 represents inter - pocket nesting , with point 4 illustrating one particular nesting vector . it can be seen that the full collection of nesting curves explains all of the ridge - like features seen in the susceptibility map , and in particular allow the determination of the points of maximal susceptibility , except for the above - noted commensurability effects . note however , that despite these complications , the susceptibility peak moves to @xmath32 in a finite doping range about the point where the antinodal electron pocket shrinks to zero width . the above discussion can be summarized : in the doping between half filling and the vhs , a new susceptibility peak arises , associated with nesting of the antinodal parts of the large fs . two kinds of behavior are found : when the ann is dominant , the peak evolves to @xmath32 at the vhs . on the other hand , for lsco(2 ) with small @xmath87 , the ann peak is inherently weaker than the @xmath30 peak , in which case the susceptibility remains commensurate at @xmath30 from half filling almost to the vhs , then smoothly develops a @xmath88 incommensurability . yet even here , in the immediate vicinity of the vhs , the ann peak at @xmath32 becomes dominant in a very limited doping range . hence , for most dispersions the susceptibility will have strong fm fluctuations near the vhs , while for other dispersions the fluctuations remain mostly afm . in the ga+rpa calculation@xcite , the electronic paramagnetic ground state energy is calculated in the ga@xcite and then expanded to second order in the on - site and intersite magnetic fluctuations in the spirit of vollhardt s fermi liquid approach@xcite . response functions are computed using linear response in the presence of small external field.@xcite one obtains rpa like susceptibilities but with strong vertex corrections . as shown in appendix [ sec : spin - rotat - invar ] longitudinal and transverse susceptibilities are trivially related and lead to the same stoner criteria as required by spin rotational invariance . in terms of a tensor bare transverse susceptibility ( cf . appendix [ sec : spin - rotat - invar ] and appendix c ) @xmath89 with @xmath90 , the ga+rpa dressed susceptibility @xmath91 is given by @xmath92 here the ratio of dressed to bare dispersion is given by @xmath93 , with the gutzwiller renormalization factor @xmath94 ^ 2}{1-x^2}\ ] ] which depends on the ga double occupancy variational parameter @xmath95 and the doping @xmath17 . the interaction matrix is@xcite @xmath96 which is defined in appendix [ sec : spin - rotat - invar ] . while eq . [ eq : g4 ] is a tensor equation , it can be expanded into the form of eq . [ eq : g1 ] with @xmath97 and @xmath98 , @xmath99 with @xmath100 , @xmath101 . the @xmath102 factors appear now in @xmath103 since we want to use the bare @xmath104 in eq . [ eq : g1 ] . all correlation effects are incorporated in the definition of @xmath103 which therefore can be viewed as a vertex corrected interaction term in the magnetic @xmath105 channel . note that @xmath103 is not strictly a pure interaction term , but is weighted by kinetic terms which enhance its @xmath2-dependence . 0.5 cm figure [ fig:6 ] compares @xmath106 , @xmath107 , @xmath108 , and the derived @xmath109 , @xmath110 the results are fairly insensitive to doping or band parameters . the sharp structures associated with nesting show up only in @xmath106 , while the other @xmath111 s are smooth , and the @xmath112 s contain only a weak structure imposed from @xmath106 . hence @xmath103 remains a smooth , weakly varying function of @xmath2 , and the instabilities are controlled by the peaks in @xmath104 . figure [ fig:7](a ) shows how @xmath103 varies as the bare @xmath0 is increased . at small @xmath0 , @xmath113 and the conventional hf+rpa is recovered . again the results do not depend strongly on doping ( fig . 8(b ) ) or dispersion . the general trend is that at large @xmath0 @xmath103 tends to saturate , and the large-@xmath2 components of @xmath103 are reduced most strongly , tending to favor instabilities nearer @xmath114 , suggestive of ferromagnetic domains . [ fig:7 ] should be compared with fig . 2 of vilk and tremblay@xcite , who find a similar saturation . since these authors employ a very different perscription for including vertex corrections , the similarity of our results gives considerable further confidence to the trends we find . in addition since the gutzwiller approximation has a simple interpretation in terms of kinetic energy suppression due to correlation our results shed further light on the physical meaning of the vertex corrections . 0.5 cm 0.5 cm figure [ fig:8 ] shows our main result , the magnetic phase diagrams of lsco and bi2201 calculated in the gutzwiller ga+rpa approach , based on the stoner criterion , eq . 1 . the phase diagram for bi2201 is fairly generic , and we have found similar results for ncco , bi2212 [ neglecting bilayer splitting ] and scoc , while the lsco phase diagram is limited to the dispersion parameters of lsco(1 ) . thus , for electron doping the phase diagram is very simple , dominated by simple af order with @xmath2 very close to @xmath11 , while for hole doping there is a competition between the saf and ann order , except in lsco , where saf order dominates . it is possible that ann order is relevant to the ` checkerboard ' phase seen in stm studies of several cuprates , as discussed briefly below . results for hf+rpa are similar , but since @xmath7 in eq . 1 is not bounded ( @xmath115 ) , there would be magnetic instabilities for all dopings , with transitions to the @xmath31 and @xmath32 plateaus , as in fig . [ fig:1 ] . in contrast , in the ga+rpa calculation @xmath103 saturates , and magnetic order exists only in a limited doping range in the cuprates , primarily in the range where the @xmath30 plateau exists . however , this doping range encompasses the full doping range of relevance to cuprate physics including the overdoped regime if a doping independent @xmath116 is assumed . thus , to have a magnetic quantum critical point near optimal doping , one must go beyond the one - band ga+rpa approach . the limited doping range for the magnetic phases has also been seen in a @xmath117 study.@xcite analogously to hf+rpa , ga+rpa determines the transition line at which the mean field state becomes unstable , but the energies are evaluated at mean field level . fluctuation contributions to the energy will most likely lower the energy of the paramagnetic ( disordered ) phase further reducing the stability range of the magnetic phases . there have by now been countless calculations concerning nanoscale phase separation or stripe phases in cuprates , and this is becoming a paradigm in many other correlated electronic systems as well . we believe , however , that the present results are unique in providing a systematic phase diagram , covering the full doping dependence@xcite and all possible @xmath1-vectors for realistic band dispersions , and indeed finding that the dominant instabilities may be different in different cuprates . while the above has provided a thorough analysis of possible magnetic instabilities , nothing has been said about competing instabilities in the charge channel charge density waves ( cdws ) or ` stripes ' . of course , one possibility is that charge order arises as a secondary effect following the spin order , with @xmath118 . the case where the charge instability is primary will be somewhat harder to analyze . that is because the corresponding stoner factor analysis is likely to require an extension beyond the hubbard model , and hence involve greater uncertainty in the choice of @xmath119 . for charge instabilities , the corresponding stoner denominator is nothing but the ( zero frequency ) dielectric constant @xmath120 , where @xmath121 is a background dielectric constant and @xmath122 is a coulomb potential . when @xmath122 is taken as either the hubbard @xmath0 or a long - range coulomb interaction,@xcite @xmath123 is always positive and there is no instability . moreover , from the theory of dielectric stability , it is known that @xmath123 can not fall in the range between 0 and 1 [ equivalently @xmath124 , and an instability @xmath125 must be approached through negative values of @xmath123.@xcite while a purely electronic instability could still arise via inclusion of local field effects , the most natural situation arises when electron - phonon coupling is included . a simple @xmath126 wave instability will be suppressed by the on - site coulomb repulsion as shown in ref . but one can still have instabilities in other channels . to the extent that the matrix elements which determine the channel have a smooth behavior one will have a stoner denominator similar to eq . ( [ eq : g1 ] ) with @xmath127 , where @xmath128 is the electron - phonon coupling parameter , @xmath129 the corresponding bare phonon propagator , @xmath130 a bare phonon frequency , and @xmath131 an electron - electron dielectric constant.@xcite thus , since the instability is also controlled by a stoner factor , the results of the present paper will also apply for charge instabilities . this is consistent with the common expectation that cdws and spin density waves ( sdws ) are controlled by the same nesting instabilities . stated differently , the stoner criterion is a formal expression of the idea that @xmath132 for an instability , where @xmath133 , the interaction energy , is represented by @xmath7 while the kinetic energy @xmath134 is measured via @xmath135 . the kinetic energy is closely tied to the band structure , and carries important material - specific information . in contrast , the potential energy is fairly featureless , leading to smoothly varying @xmath119 . thus , the locations of the instabilities are controlled by peaks in @xmath104 , and a @xmath2-dependent @xmath0 can only shift the dominant instability between two peaks of comparable height . in the remainder of this section , when we discuss comparison to experiment , it will be seen that the present model has correctly determined the dominant @xmath2 values , but that in several cases experiment points more toward cdws than sdws . there have been a number of recent hints that the ` stripe ' order in underdoped la@xmath3a@xmath4cuo@xmath136 , a = sr ( lsco ) or ba ( lbco ) is not the same phenomenon as the ` checkerboard ' order found in sr@xmath20cuo@xmath20cl@xmath20 ( scoc ) and bi@xmath20sr@xmath20cacu@xmath20o@xmath137 ( bi-2212 ) . thus , resonant soft x - ray scattering experiments find evidence of charge order in lbco ( stripes)@xcite but not in ca@xmath20cuo@xmath20cl@xmath20 ( checkerboards)@xcite , while evidence for time reversal symmetry breaking has been found in yba@xmath20cu@xmath138o@xmath139 ( ybco ) , but not in lsco@xcite . the present results suggest a connection between the ann and checkerboard phases , in that the periodicity of the latter also scales with the antinodal nesting vector@xcite , and a further connection between the @xmath30 plateau and stripes . thus it is quite interesting to observe that the ann phase is virtually absent for the lsco(2 ) dispersion , where evidence for conventional stripes is strongest . in 1d systems the susceptibility diverges at @xmath15 at all dopings , due to perfect ( flat band ) nesting . the present results generalize this to 2d systems : the dominant instability is generally associated with double nesting , but since @xmath104 in general does not diverge , a finite coupling @xmath0 is required to drive an instability . an exception is vhs nesting@xcite , for which @xmath104 has a logarithmic divergence . at the vhs , there are two competing instabilities , af at @xmath30 and fm at @xmath32 , corresponding to the magnetic branch of the @xmath140 phase diagram of the cuprates , and these are responsible for the commensurate pinning near @xmath30 and @xmath32 , respectively.@xcite doped away from the vhs , these instabilities evolve into the two dominant peaks of the susceptibility , and can be considered as ` generalized vhs nesting ' . as such , they dominate the magnetic physics of the cuprates over the full doping range , coextensive with the limits of the @xmath141-plateau in the susceptibility . recently , storey _ et al._@xcite proposed that the generic behavior of high - t@xmath142 cuprates could be understood if the pairing interaction ( or pairing energy cutoff ) falls off rapidly near the vhs . a recent experiment@xcite does indeed confirm that @xmath143 scales with @xmath144 , but @xmath145 . the present results suggest that most cuprates will have strong fm fluctuations near the vhs , which are incompatible with simple d - wave superconductivity . empirical evidence for fm fluctuations has been noted previously@xcite . while some previous calculations@xcite have found evidence of ferromagnetism near a vhs , others@xcite have suggested that fm instabilities were unlikely in competition with incommensurate susceptibility . the present calculations confirm that a dominant fm susceptibility should be present near the vhs . the advantage of two - dimensional materials is that it is straightforward to display the susceptibility maps and nesting curves . in 3d , the curves become _ nesting surfaces _ and the susceptibility maps are 4-dimensional . one can speculate that the dominant susceptibility peaks correspond to triple - nesting points . for quasi-2d materials , it should be possible to analyze a series of cuts perpendicular to the ( weakly - dispersing ) @xmath146-axis . in principle , the present results could provide information on nanoscale phase separation ( nps ) as well . one model of nps is that there are different instabilities associated with particular dopings [ e.g. , half filling and the vhs ] , and that these two phases are more stable than uniform phases at intermediate dopings . in this case , the two end phases could still be described by nesting maps , only at particular dopings . for instance , in the hf+rpa analysis of lsco(2 ) , the fm phase is stable only very close to the vhs , and could lead to nps with a second phase at the undoped insulator . the stoner criterion determines which @xmath2 value is most unstable , and the minimum @xmath0 needed to drive that instability . however , as @xmath0 increases above threshold , the @xmath2 of the ordered phase may shift . thus , at half filling , when a full gap can be formed , @xmath2 will no longer be determined by best nesting , but by the largest gap . this tends to favor more commensurate @xmath2 values , leading to a pinning of @xmath2 at these commensurate values over a wide range of parameters . we find that as @xmath0 increases , there is a first order transition from an incommensurate phase with fermi surface pockets to a commensurate [ @xmath30 or @xmath31 ] phase that is fully gapped.@xcite away from half filling , luttinger s theorem ensures the persistence of a fermi surface , so nesting instabilities should persist over a wider range of @xmath0 s . furthermore , as we have seen above , the @xmath2-value corresponding to the largest @xmath148 can shift with @xmath0 . typically , for @xmath103 the shift is to a smaller @xmath2-value , associated with an instability towards fm order ( in the magnetic channel ) or nps ( in the charge channel ) . the present results provide a constructive scheme for identifying the dominant nesting instabilities for any two dimensional material . clearly , for fermi surfaces with multiple sections , a large number of nesting curves are possible , leading to extremely complicated susceptibility maps . nevertheless , the present scheme will automatically sort out the possible double nesting peaks and follow their evolution with doping . this should allow a much more detailed understanding of 2d phase diagrams , particularly for magnetic phases , where the interaction @xmath0 has negligible @xmath2-dependence . the good agreement of the ga+rpa calculations with more detailed variational calculations at half filling@xcite and with exact and numerical results in @xmath117,@xcite and of the af+sc model in electron doped cuprates with experiments gives us confidence in the model . accordingly , we note the following three points . first , within ga+rpa , the paramagnetic state is unstable over the full hole - doping range in the cuprates , including overdoped . to avoid this conclusion , and restore a fermi liquid phase in the overdoped regime , it may be necessary to include non gaussian fluctuations@xcite or an additional doping dependence for @xmath0 . the origin of any such doping dependence lies outside the ga treatment of the hubbard model . second , the @xmath11 phase , or its incommensurate extension , is unstable against a competing ann order in [ most ] hole - doped cuprates . third , there is a material dependence to the phase diagram , and lsco may have a very different doping dependence than other cuprates . the results have been applied to a number of model dispersions for the cuprates . a possible pseudogap candidate has been identified and a distinction between stripes and checkerboards proposed . these findings will be discussed in greater detail in ensuing publications . this work is supported by the us department of energy , office of science , basic energy sciences contract de - fg02 - 07er46352 , and benefited from the allocation of supercomputer time at nersc , northeastern university s advanced scientific computation center ( ascc ) . rsm s work has been partially funded by the marie curie grant piif - ga-2008 - 220790 soqcs , while gs work is supported by the vigoni program 2007 - 2008 of the ateneo italo - tedesco deutsch - italienisches hochschulzentrum . .i . band parameter sets [ cols="^,^,^,^,^,^",options="header " , ] in table i , all hopping parameters are given in mev . the @xmath149 term for lsco(1 ) is the coefficient of a term @xmath150 in eq . ncco , lsco(2 ) , and bi2201(2 ) data sets are fit to lda band dispersions of the near - fermi level antibonding cuo@xmath20 band , and are appropriate for the gutzwiller analysis . the data sets lsco(1 ) and bi2201(1 ) are taken from fits to experimental arpes data . to the extent that the bands are renormalized by a @xmath2-independent factor @xmath151 , the effective bare susceptibilities @xmath152 can be used in the gutzwiller analysis , while @xmath14 occurs at the same @xmath2 . recently it has been suggested that the enhanced nesting , such as observed in lsco(1 ) , may be associated with stripe formation@xcite . here we sketch the ga+rpa formalism and define the relevant quantities . for a review see ref . . the susceptibility can be computed in the longitudinal@xcite or in the transverse channel . @xcite as shown bellow both results are equivalent as dictated by spin rotational invariance on a paramagnetic ( singlet ) ground state . for later use we define the density matrix associated with the unprojected slater determinant , @xmath153 , as @xmath154 and as a shorthand we define @xmath155 . we start from the spin - rotational invariant gutzwiller energy functional for the hubbard model defined in terms of @xmath156 and the double occupancy in the gutzwiller variational state @xmath157 . as derived e.g. in ref . the functional reads , @xmath158 where we have defined the spinor operators @xmath159 and the @xmath160-matrix @xmath161\cos\varphi_i \\ \frac{s_i^+}{2s_i^z}[z_{i\uparrow}-z_{i\downarrow}]\cos\varphi_i & z_{i\uparrow}\sin^2\frac{\varphi_i}{2}+z_{i\downarrow}\cos^2\frac{\varphi_i}{2 } \end{array } \right ) \\ \tan^2\varphi_i&=&\frac{s_i^+s_i^-}{(s_i^z)^2}.\end{aligned}\ ] ] with the @xmath146 factors given by @xmath162 and for clarity spin expectation values are denoted by @xmath163 , @xmath164 , @xmath165 , and @xmath166 . in the limit of a vanishing rotation angle @xmath167 the @xmath160-matrix becomes diagonal and the renormalization factors reduce to those of the standard gutzwiller approximation . for a homogeneous , paramagnetic system ( @xmath168 , @xmath169 ) the expansion of the gutzwiller energy functional up to second order in the particle - hole excitations reads , @xmath170 the first contribution is the free fermion part , @xmath171 @xmath172 denotes the dispersion of the gutzwiller quasiparticles whereas @xmath173 corresponds to the unrenormalized dispersions , i.e. @xmath174 . the remaining part in eq . ( [ eq : ega2 ] ) is due to quasiparticle interactions and separates into contributions from the charge ( @xmath175 ) , the longitudinal ( @xmath176 ) and transverse ( @xmath177 ) spin channel . @xmath178 and @xmath179 have been derived in detail in ref . . for our purposes we report in the following the expansion in the longitudinal channel : @xmath180 and in the transverse magnetic channel which has been derived in ref . @xmath181 the relevant fluctuations have been defined as @xmath182 the 11 element of the interaction kernel is given by @xmath183 with @xmath184 here @xmath185 is the brinkman - rice energy . @xmath186 is similar , but requires separate averages for different components of the energy : @xmath187\bigl<{c_{kx}+c_{ky}}\bigr > -4t'c_{qx}c_{qy}\bigl < c_{kx}c_{ky}\bigr > \nonumber \\ & -&t''[c_{2qx}+c_{2qy}]\bigl<{c_{2kx}+c_{2ky}}\bigr > \nonumber \\ & -&2t'''[c_{qx}c_{2qy}+c_{qy}c_{2qx}]\bigl<{c_{kx}c_{2ky}+c_{ky } c_{2kx}}\bigr>,\nonumber \label{eq : g17}\end{aligned}\ ] ] in an obvious notation . furtheron we define the susceptibility matrices for the bare time - ordered correlation functions both in the longitudinal @xmath188 and in the transverse channel @xmath189 where a hat has been added to distinguish fluctuations ( @xmath190 ) form operators ( @xmath191 ) . the longitudinal susceptibility describes the @xmath192 , singlet to triplet excitations of the paramagnetic state while the transverse describes @xmath193 spin excitations . spin rotational invariance dictates that these excitations should be degenerate . one obtains for these correlation functions @xmath194 and @xmath195 for the non - interacting gutzwiller quasiparticles spin - rotational invariance is thus guaranteed from the relation @xmath196 this identity is preserved within the ga+rpa when we compute the interacting susceptibilities from the rpa series @xmath197 since from eqs . ( [ eq : vzz ] , [ eq : vpm ] ) the interaction kernels are related by @xmath198 . clearly the energies of particle hole excitations described by these equations are degenerate . in particular they lead to the same stoner criteria in both channels . the expressions involve the following derivatives of @xmath199 in the longitudinal channel @xmath200 @xmath201 @xmath202 @xmath203 @xmath204 the derivatives in the transverse channel can be related to those in the longitudinal channel as follows , @xmath205 on the right we have given explicit expressions in terms of @xmath76 , the doping measured with respect to half - filling and the double occupancy @xmath95 in the paramagnetic state . 99 a. macridin , m. jarrell , t. maier , and d.j . scalapino , phys . lett . * 99 * , 237001 ( 2007 ) . markiewicz , s. sahrakorpi , and a. bansil , phys . rev . b*76 * , 174514 ( 2007 ) . f. ronning , k.m . shen , n.p . armitage , a. damascelli , d.h . lu , z .- x . shen , l.l . miller , and c. kim , phys . b*71 * , 094518 ( 2005 ) ; j. graf , g .- h . gweon , k. mcelroy , s.y . zhou , c. jozwiak , e. rotenberg , a. bill , t. sasagawa , h. eisaki , s. uchida , h. takagi , d .- h . lee , and a. lanzara , phys . lett . * 98 * , 067004 ( 2007 ) . markiewicz , s. sahrakorpi , m. lindroos , hsin lin , and a. bansil phys . rev . b*72 * , 054519 ( 2005 ) . aligia , phys . b*39 * , 6700 ( 1989 ) ; q. si , y. zha , k. levin , and j.p . lu , phys . rev . b*47 * , 9055 ( 1993 ) ; p. bnard , l. chen , and a .- m.s . tremblay , phys . rev . b*47 * , 15217 ( 1993 ) ; m. lavagna and g. stemmann , phys . rev . b*49 * , 4235 ( 1994 ) ; f. onufrieva and p. pfeuty , phys . b*61 * , 799 ( 2000 ) ; m.r . norman , phys . * , 14751 ( 2000 ) . for an example of this extended commensurate nesting , see frames j , k of fig . rice and g.k . scott , phys . lett . * 35 * , 120 ( 1975 ) . r. hlubina and t.m . rice , phys . rev . b*51 * , 9253 ( 1995 ) . andersen , o. jepsen , a.i . liechtenstein , and i.i . mazin , phys . rev . b*49 * , 4145 ( 1994 ) . a. diciolo , j. lorenzana , m. grilli , and g. seibold , phys . b * 79 * , 085101 ( 2009 ) . f. gnther , g. seibold , and j. lorenzana , phys . 98 * , 176404 ( 2007 ) . some previous determinations of stripe incommensurability vs doping , but for a limited range of @xmath1-values , are : j. lorenzana and g. seibold , phys lett . * 89 * , 136401 ( 2002 ) ; g. seibold and j. lorenzana , phys . rev . b*69 * , 134513 ( 2004 ) . f. gebhard , phys . b*41 * , 9452 ( 1990 ) . g. kotliar and a.e . ruckenstein , phys . lett . * 57 * , 1362 ( 1986 ) . d. vollhardt , rev . 56 * , 99 ( 1984 ) . y.m . vilk and a .- m.s . tremblay , j. phys . i * 7 * , 1309 ( 1997 ) . markiewicz and a. bansil , phys . rev . b*75 * , 020508(r ) ( 2007 ) . dolgov , d.a . kirzhnits , and e.g. maksimov , rev . * 53 * , 81 ( 1981 ) . p. abbamonte , a. rusydi , s. smadici , g.d . gu , g.a . sawatzky , and d.l . feng , nature phys . * 1 * , 155 ( 2005 ) . s. smadici , p. abbamonte , m. taguchi , y. kohsaka , t. sasagawa , m. azuma , m. takano , and h. takagi , phys . b*75 * , 075104 ( 2007 ) . macdougall , a.a . aczel , j.p . carlo , t. ito , j. rodriguez , p.l . russo , y.j . uemura , s. wakimoto , g.m . luke , phys . * 101 * , 017001 ( 2008 ) . shen , f. ronning , d.h . lu , f. baumberger , n.j.c . ingle , w.s . lee , w. meevasana , y.kohsaka , m. azuma , m. takano , h. takagi , and z .- x . shen , science * 307 * , 901 ( 2005 ) ; w.d . wise , m.c . boyer , k. chatterjee , t. kondo , t. takeuchi , h. ikuta , y. wang , and e.w . hudson , nature physics * 4 * , 696 ( 2008 ) . markiewicz and m.t . vaughn , phys . rev . b*57 * , r14052 ( 1998 ) . storey , j.l . tallon , and g.v.m . williams , phys . b * 76 * , 174522 ( 2007 ) . palczewski , t. kondo , r. khasanov , n.n . kolesnikov , a.v . timonina , e. rotenberg , t. ohta , a. bendounan , y. sassa , a.v . fedorov , s. pailhs , a.f . santander - syro , j. chang , m. shi , j. mesot , h.m . fretwell , and a. kaminski , phys . b*78 * , 054523 ( 2008 ) . a. kopp , a. ghosal , and s. chakravarty . usa * 104 * , 6123 ( 2007 ) . r. hlubina , s. sorella , and f. guinea , phys . lett . * 78 * , 1343 ( 1997 ) ; r. hlubina , phys . rev . b*59 * , 9600 ( 1999 ) ; d. vollhardt , n. blmer , k. held , m. kollar , j. schlipf , and m. ulmke , z. phys . b*103 * , 283 ( 1997 ) ; b. valenzuela , m.a.h . vozmediano , and f. guinea , phys . b*62 * , 11312 ( 2000 ) . igoshev , a.a . katanin , and v.yu . irkhin , j.e.t.p . , to be published , ( extended version : arxiv:0709.3219 ) . m. hashimoto , t. yoshida , h. yagi , m. takizawa , a. fujimori , m. kubota , k. ono , k. tanaka , d.h . lu , z .- x . shen , s. ono , and yoichi ando , phys . b*77 * , 094516 ( 2008 ) . we thank tanmoy das for providing us with these values , based on lda . norman , phys . rev . b*75 * , 184514 ( 2007 ) . zabolotnyy , a.a . kordyuk , d.s . inosov , d.v . evtushinsky , r. schuster , b. buechner , n. wizent , g. behr , sunseng pyon , h. takagi , r. follath , s.v . borisenko , europhys . lett . * 86 * , 47005 ( 2009 ) . markiewicz , phys . * , 174518 ( 2004 ) . markiewicz , j. lorenzana , and g. seibold , unpublished . s. andergassen , s. caprara , c. di castro , and m. grilli , phys . lett . * 87 * , 056401 ( 2001 ) . j. lorenzana and g. seibold , low . physics * 32 * , 320 ( 2006 ) . g. seibold , f. becca , p. rubin , and j. lorenzana , phys . rev . b * 69 * , 155113 ( 2004 ) .

wz sge - type dwarf novae are an unusual subtype of su uma - type dwarf novae ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . wz sge - type dwarf have properties common to su uma - type dwarf novae ( @xcite ; @xcite ; @xcite ) in that they show superoutbursts and superhumps . the most unusual properties of wz sge - type dwarf novae are characterized by the extremely low frequency of outbursts ( usually once in years to decades ) , the extremely large outburst amplitude ( exceeding 6 mag ) and long ( in some cases reaching @xmath0100 d before the quiescence is reached ) duration , and the lack or extreme deficiency of normal ( short ) outbursts , which , in ordinary su uma - type dwarf novae , occur more frequently than superoutbursts ( cf . @xcite ) . the origin of such unique characters has been long sought , both from the observational and theoretical sides . two basic ideas have been historically proposed , just like a descendant from historical discussions on the mechanism of dwarf nova outbursts . one is the mass - transfer instability model , which assumes the outer atmosphere of the extreme low - mass secondary in wz sge - type dwarf novae become unstable to the irradiation from the outbursting accretion disk , thus giving rise to an enhanced mass - transfer from the secondary , which leads to a long - lasting superoutburst ( the best historical example of this application being @xcite ) . the other is an application of the disk - instability model , which , with an assumption of very low quiescent viscosity of the accretion disk , can produce the main characteristics of wz sge - type dwarf novae , namely the long outburst intervals and the absence of normal outbursts ( @xcite ; @xcite ) . the required quiescent viscosity ( @xmath1 ) to reproduce such long outburst intervals has been shown to be extremely small ( @xmath2 : @xcite ; @xmath3 : @xcite ) . these values of the required quiescent viscosity has usually been considered too small for a disk with a developed magneto - rotational instability @xcite , although this problem is recently becoming more positively resolved by considering a suppression of this instability in a cold disk with a low electric conductivity @xcite . the apparent difficulty with the low quiescent viscosity has historically invoked modifications of the disk - instability model exemplified by the truncated inner disk @xcite . this model supposed a cold steady state - solution during quiescence , and a slight variation in mass - transfer rate was supposed to be the cause of infrequent outbursts . @xcite proposed another possibility of the magnetically truncated disk . although these truncated disks were successful in reproducing one of the characteristics of wz sge - type dwarf novae ( the long recurrence time ) , these models with a high @xmath1 ( @xmath0 0.01 ) have been shown by @xcite to meet difficulties in reproducing long - lasting superoutbursts . although there have been attempts ( e.g. @xcite ) to reproduce long outbursts by considering an enhanced mass - transfer during superoutbursts , the resultant light curve is very different from observation . furthermore , the supposed observational evidence for an enhanced mass - transfer is excluded by later careful analysis @xcite . in addition to these outburst properties , wz sge - type dwarf novae are shown to have other properties common to all members , but are absent in ordinary su uma - type dwarf novae . one is the presence of _ early superhumps _ during the earliest stage of superoutbursts of wz sge - type dwarf novae ( cf . the early superhumps have a stable period almost identical with the orbital period [ during the best observed 2001 superoutburst of wz sge , the period of the early superhumps was reported to be very slightly shorter than the orbital period @xcite ] . this feature was historically first detected and described in wz sge itself ( @xcite ; @xcite ) , although the origin of this variation was not properly discussed or identified at this time . the common existence of early superhumps among wz sge - type dwarf novae has been progressively confirmed : al com ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) , hv vir ( @xcite ; @xcite ) , rz leo ( @xcite ) , and the main focus of this paper , eg cnc ( @xcite ; @xcite ) . the other is occasional dips " ( transient fading episodes lasting for one to several days ) during the later stage of superoutburst , which was first clearly described in the 1978 outburst of wz sge ( e.g. @xcite ) , and was repeated in the 1995 outburst of al com ( @xcite ; @xcite ; @xcite ; @xcite ) . @xcite repoeted that the stage after the dip phenomenon shares the characteristics common to superoutbursts of ordinary su uma - type dwarf novae : the presence of a likely precursor and the subsequent growth of usual superhumps . from these findings , the 1977 - 1978 outburst of wz sge and the 1995 outburst of al com were interpreted as double superoutbursts " @xcite.http://www.kusastro.kyoto - u.ac.jp / vsnet / mail / alert1000/msg00195.html@xmath4 . ] an argument naturally rises why such a sequence of superoutburst can be only realized in wz sge - type dwarf novae . a naive explanation may be a consequence of enhanced - mass transfer , but is rather difficult to reconcile considering the low proper angular momentum of the fresh " transferred matter , accumulation of which will more easily produce normal outbursts than a second superoutburst . @xcite , @xcite , and @xcite instead suggested a possibility of the cooling and heating waves originating before the termination of the entire superoutburst . @xcite presented observational evidence in al com that the second superoutburst was triggered by a short outburst , which , from the absence of superhumps , was attributed to a normal outburst as seen in ordinary su uma - type dwarf novae . independent observations during the second ( super)outburst are critically needed to confirm the universality of the claimed observational features , and to explain the unique outburst pattern of wz sge - type dwarf novae . on 1996 november 30 , an alert ( vsnet - alert 599 ) through the vsnet alert system @xcite notifying a bright outburst of the suspected wz sge - type star , http://www.kusastro.kyoto - u.ac.jp / vsnet / mail / vsnet - alert/ + msg00599.html@xmath4 and @xmath5http://www.kusastro.kyoto - u.ac.jp / vsnet / dne / egcnc.html@xmath4 . ] eg cnc @xcite announced the opening of a new story . eg cnc was originally discovered by @xcite from his photographic archive on the occasion of an outburst in 1977 . subsequent searches for other photographic materials between 1977 and 1983 , and searches for further outbursts by huruhata himself and a number of amateur astronomers have yielded a null result until 1995 . @xcite correctly identified the quiescent counterpart on palomar sky survey prints , and suggested from its marked blue color and the apparent low frequency of outbursts that this object is similar to the dwarf nova wz sge . upon the alert by schmeer , we started time - resolved ccd photometry using telescopes at osaka kyoiku university ( oku ) and ouda station @xcite , kyoto university . the initial results proving the existence of early superhumps ( @xmath6 d ) and the subsequent development of usual superhumps ( @xmath7 d ) were already discussed in our earlier paper ( @xcite ) . we discuss mainly on its peculiar late - stage phenomenon , particularly focusing on repetitive rebrightenings . the methods of observation and reduction were the same as in @xcite . the observations at osaka kyoiku university were carried out using the 50-cm telescope at the cassegrain focus ( focal length @xmath8 ) equipped with an astromed eev 88200 ccd camera ( @xmath9 pixels ) and a @xmath10-band filter . the observations at ouda station were carried out using the 60-cm telescope at the cassegrain focus ( focal length @xmath11 ) equipped with a thomson th 7882 ccd camera ( @xmath12 pixels ) . we used an interference filter designed to reproduce the johnson @xmath10 band . the @xmath13 on - chip binning mode was used . the frames obtained at osaka kyoiku university were reduced using the iraf . the frames obtained at ouda station were reduced with a microcomputer - based automatic aperture and psf photometry packages developed by one of the authors ( tk ) . we used gsc 1948.1452 ( @xmath10 = 12.82 ) as the local comparison star , whose constancy during the observation was confirmed to 0.01 mag using the check star gsc 1948.1193 ( @xmath10 = 13.86 ) . heliocentric correction were applied before the following analysis . table [ tab : log ] lists the log of ccd photometry . ccccccc & start end ( ut ) & @xmath14 & exp(s ) & mean @xmath10 mag & obs + 1996 & december & 2.625 2.791 & 195 & 30 - 90 & 12.38 & oku + & & 3.625 3.708 & 66 & 60 - 80 & 12.51 & oku + & & 7.679 7.811 & 208 & 45 & 12.42 & ouda + & & 8.715 8.870 & 176 & 60 & 12.48 & ouda + & & 9.629 9.755 & 195 & 45 & 12.54 & ouda + & & 11.635 11.870 & 345 & 45 & 12.76 & ouda + & & 13.746 13.877 & 166 & 60 & 13.18 & ouda + & & 14.683 14.882 & 322 & 45 & 14.32 & ouda + & & 20.666 20.860 & 112 & 120 & 16.13 & ouda + & & 21.550 21.871 & 478 & 45 - 60 & 13.29 & ouda + & & 22.770 22.824 & 49 & 80 - 90 & 14.10 & ouda + & & 23.555 23.856 & 264 & 90 & 15.39 & ouda + & & 24.555 24.870 & 277 & 90 & 15.92 & ouda + & & 25.586 25.697 & 102 & 60 & 15.59 & ouda + & & 26.700 26.883 & 140 & 45 - 60 & 16.23 & ouda + & & 27.619 27.883 & 322 & 60 & 16.22 & ouda + & & 30.600 30.871 & 320 & 45 - 90 & 15.84 & ouda + & & 31.519 31.745 & 195 & 90 & 16.30 & ouda + 1997 & january & 2.543 2.753 & 144 & 90 & 16.25 & ouda + & & 3.862 3.863 & 2 & 120 & 16.35 & ouda + & & 4.473 4.888 & 721 & 40 - 50 & 13.37 & ouda + & & 6.616 6.730 & 72 & 90 & 15.98 & ouda + & & 10.725 10.795 & 82 & 60 & 13.92 & ouda + & & 11.640 11.804 & 97 & 90 & 15.29 & ouda + & & 12.614 12.755 & 78 & 90 & 16.32 & ouda + & & 13.605 13.751 & 122 & 90 & 16.23 & ouda + & & 15.647 15.811 & 137 & 90 & 16.27 & ouda + & & 19.756 19.745 & 75 & 90 & 16.35 & ouda + & & 20.501 20.843 & 284 & 90 & 16.22 & ouda + & february & 1.664 1.669 & 5 & 90 & 17.17 & ouda + & & 3.433 3.439 & 5 & 120 & 17.38 & ouda + & & 5.443 5.454 & 6 & 180 & 17.29 & ouda + & & 27.512 27.697 & 87 & 90 - 120 & 17.82 & ouda + & & 28.490 28.617 & 61 & 150 & 17.78 & ouda + & march & 1.619 1.727 & 57 & 150 & 17.83 & ouda + & & 3.589 3.726 & 63 & 150 & 17.95 & ouda + & & 4.634 4.727 & 49 & 150 & 17.94 & ouda + & & 5.522 5.680 & 82 & 150 & 18.02 & ouda + & & 7.631 7.639 & 4 & 150 & 17.92 & ouda + & & 8.493 8.675 & 96 & 150 & 18.01 & ouda + & & 11.486 11.616 & 58 & 180 & 18.09 & ouda + & & 28.431 28.516 & 39 & 150 & 17.94 & ouda + & & 30.503 30.600 & 50 & 150 & 18.25 & ouda + & & 31.469 31.604 & 61 & 180 & 18.26 & ouda + & april & 1.435 1.602 & 58 & 240 & 18.15 & ouda + & & 13.455 13.463 & 5 & 150 & 17.97 & ouda + & & 14.467 14.472 & 4 & 150 & 17.73 & ouda + & & 24.442 24.450 & 5 & 170 & 18.45 & ouda + & & 25.459 25.463 & 3 & 150 & 18.55 & ouda + + ( 160mm,80mm)fig1.eps figure [ fig : vis ] represents the overall course of the outburst from ccd and visual observations reported to the vsnet , including the available @xmath10-band ccd observations , averaged to one to a few points per night . the object was first detected in outburst on 1996 november 30.917 ut at @xmath1512 @xcite . the last available negative observation was done on november 22.685 ut , giving an upper limit of 13.3 ( t. watanabe , private communication ) . the early rise was unfortunately not observed due to the interference by the moon . the outburst thus started some time between november 22 and 30 . the existence of early superhumps @xcite , which are known to rapidly decay within several days of the maximum light of a wz sge - type superoutburst , and the recorded maximum brightness ( @xmath16 = 11.8 ) close to that of the 1977 outburst , suggest that the detection by schmeer was made within a few days of the start of the outburst . the main superoutburst lasted until december 15 ( hjd 2450433 ) , followed by a precipitous decline . the object then faded close to or below @xmath10 = 16 . on december 20 ( hjd 2450438 ) , the first hint of rebrightening was recorded in the ouda data . on the next night , the object reached the peak brightness of @xmath10 = 13.29 , and soon started fading rapidly . this rebrightening was first considered to be of the rather common phenomenon among short orbital - period su uma - type dwarf novae ( e.g. v1028 cyg : @xcite , see also @xcite ) , not necessarily confined to wz sge - type dwarf novae . the unique property of eg cnc , however , emerged after this . the star remained about 3 mag brighter than its quiescence , and again showed a rebrightening 7 days after its first one . similar rebrightenings repetitively occurred six times in total ( counting those exceeding @xmath10 = 14 at maximum ) , separated by five to eleven days . table [ tab : reblist ] summarizes the observed post - superoutburst rebrightenings . although there was a suggestion of two rebrightenings after the 1994 superoutburst of another wz sge - type candidate , uz boo @xcite , , such a large number of repetitive brightenings had been never observed in any dwarf novae until this outburst of eg cnc . ccc date of maximum & magnitude & @xmath17 ( d ) + 50439 & 13.3 & @xmath18 + 50446 & 13.4 & 7 + 50453 & 13.3 & 7 + 50458 & 13.2 & 5 + 50465 & 13.3 & 7 + 50475 & 13.2 & 10 + + + + after its sixth rebrightening on 1997 january 26 , eg cnc then entered a stage of slow , steady decline , keeping the rate of slow fading between rebrightenings . this stage lasted at least until our last observation on 1997 april 25 ( 146 d after schmeer s detection of the outburst ) , when eg cnc had a magnitude of @xmath10 = 18.6 , which was still 0.6 mag above the quiescence ( cf . eg cnc thus reproduced the long - fading tail of wz sge observed in 1978 ( cf . @xcite ) and al com in 1995 ( @xcite ) , accompanied by numerous short rebrightenings instead of a superoutburst - looking second outburst in wz sge and al com . figure [ fig : wzcomp ] shows the comparison of outburst light curves of three best - observed wz sge - type dwarf novae in the recent years . as already described in @xcite , doubly humped early superhumps were observed during the earliest two nights of our observations . since the raw light curve was not presented in @xcite , we present the light curve in figure [ fig : earlylc ] . then textbook " superhumps ( often referred to as _ common superhumps _ or _ ordinary superhump _ when discussing on wz sge - type stars ) emerged ( e.g. figure [ fig : shlc ] ) . ( 88mm,120mm)fig3.eps ( 88mm,60mm)fig4.eps ( 88mm,120mm)fig5.eps ( 88mm,120mm)fig6.eps figures [ fig : earlysh ] and [ fig : sh ] show the comparison of the profiles of these two kinds of superhumps between the recent three well - observed wz sge - type dwarf novae ( the data of al com from @xcite , hv vir from @xcite ) . the periods ( 0.05877 d and 0.06038 d ) used in folding the data of eg cnc are our revised estimates , using the pdm ( phase dispersion minimization : @xcite ) technique to the datasets for 1996 december 23 ( the stage showing early superhumps ) and 1996 december 713 ( common superhumps ) , after subtracting the trend of linear decline in each dataset . the new best periods were determined using a minimum determination algorithm developed by one of the authors ( tk ) . the 1-@xmath19 errors for these periods were determined to be 0.00017 and 0.00001 d , respectively , using fernie s application to lafler - kinman class of statistics @xcite . statistical f - tests yielded significance levels of the signals 87% ( @xmath14 = 261 ) and @xmath2099.9% ( @xmath14 = 805 ) , respectively . as is evident from these figures , all of these three wz sge - type dwarf novae showed doubly peaked early superhumps , sometimes with a rather complex structure ( cf . @xcite ) . the smallness of the amplitude in eg cnc probably suggests that this early modulation had already started to decay at the time of observations , or that this system has a low orbital inclination ( cf . @xcite ) . the existence of these early superhumps , whose periods are considered to be very close to the orbital periods ( wz sge : @xcite ; @xcite ; @xcite ; @xcite ) , al com : @xcite ) , is now regarded as an independent criterion for discriminating wz sge - type dwarf novae @xcite . the common superhumps appearing later are very similar between different objects , and have almost identical profiles to those of ordinary su uma - type dwarf novae ( cf . @xcite ; @xcite ) . we then analyzed short - term variation during the post - superoutburst rebrightening stage . we first divided the observations during these stages into the following three representative stages : ( 1 ) _ between rebrightenings _ ( faint state between repetitive rebrightenings ; data for 1996 december 26 - 31 and 1997 january 2 were used ) , ( 2 ) _ rebrightening maximum _ ( around peak brightness of rebrightenings ; the 1997 january 4 data were used ; note that the unique period identification is improbable because of the short duration of the data segment ) , and ( 3 ) _ fading tail _ ( 1997 february 27 and after ) . we then removed the long - term trend of each dataset by subtracting a linear fit , and applied the pdm technique to the datasets . the resultant theta diagrams , together with those of the main superoutburst , are shown in figure [ fig : pdm ] ( the weaker signals arounds 0.0569 and 0.0643 d are one - day aliases , which are not true signals ) . ( 88mm,120mm)fig7.eps although the result of the second set is less conclusive in terms of period , the common presence of signals ( minima in theta ) around 0.0603 d suggest that the period of usual ( later ) superhumps during the main superoutburst persisted throughout the rebrightening phase and the fading tail . statistical f - tests yielded significance levels of the signals 59% ( @xmath14 = 1041 ) , 87% ( @xmath14 = 716 ) , and 84% ( @xmath14 = 494 ) for the above three stages , respectively . ( 88mm,120mm)fig8.eps figure [ fig : fold ] shows the phase - averaged light curves of individual data sets . the hump profiles , though amplitudes are very different , resemble each other and that of usual superhumps . this observation of strong superhump modulation after the main superoutburst safely preclude the possibility of the cooling white dwarf as a main source of the fading tail of wz sge - type dwarf novae , as suggested by @xcite . ( 88mm,60mm)fig9.eps we further examined the evolution of these persistent superhumps with respect to the rebrightening stages . we used the data between 1996 december 20 and 23 , which best covered the rebrightening ( the second rebrightening ) . the resultant phase - averaged light curves are shown in figure [ fig : minifold ] , which again show the stability of the 0.0603-d persistent superhumps , despite that these persistent superhumps become weaker when the system brightens . figure [ fig : miniamp ] shows the time - variation of the pulsed ( superhump ) flux and the total flux of the object . both fluxes are shown in magnitude scale corresponding to apparent @xmath10 magnitudes ( i.e. 10% flux modulations at mean magnitude @xmath10 = 14.0 corresponds to a pulsed flux corresponding to @xmath10 = 16.5 ) . despite the @xmath0 3 mag magnitude brightening of the system during the rebrightening , the pulsed ( superhump ) flux remained rather constant within a factor of two . ( 88mm,60mm)fig10.eps cccccccc @xmath21 & hjd@xmath222450000 & source & @xmath23 & @xmath21 & hjd@xmath222450000 & source & @xmath23 + -39 & 422.882 & 2 & 0.007 & 128 & 432.959 & 2 & -0.008 + -38 & 422.943 & 2 & 0.008 & 129 & 433.020 & 2 & -0.007 + -23 & 423.853 & 2 & 0.011 & 159.5 & 434.862 & 2 & -0.008 + -7 & 424.809 & 2 & 0.000 & 209.5 & 437.890 & 2 & -0.002 + -6 & 424.872 & 2 & 0.003 & 214.5 & 438.188 & 1 & -0.006 + -5 & 424.935 & 2 & 0.005 & 215.5 & 438.252 & 1 & -0.003 + -4 & 424.993 & 2 & 0.003 & 216.5 & 438.311 & 1 & -0.004 + -3 & 425.055 & 2 & 0.004 & 223.5 & 438.728 & 2 & -0.010 + 0 & 425.236 & 1 & 0.004 & 230.5 & 439.158 & 1 & -0.003 + 1 & 425.299 & 1 & 0.007 & 258.5 & 440.856 & 2 & 0.003 + 10 & 425.839 & 2 & 0.003 & 259.5 & 440.913 & 2 & -0.000 + 11 & 425.900 & 2 & 0.003 & 260.5 & 440.974 & 2 & 0.000 + 12 & 425.961 & 2 & 0.004 & 261.5 & 441.039 & 2 & 0.005 + 13 & 426.022 & 2 & 0.005 & 262.5 & 441.091 & 1 & -0.004 + 17 & 426.263 & 1 & 0.004 & 263.5 & 441.163 & 1 & 0.008 + 18 & 426.324 & 1 & 0.004 & 264.5 & 441.214 & 1 & -0.002 + 25 & 426.743 & 2 & 0.000 & 265.5 & 441.288 & 1 & 0.012 + 26 & 426.805 & 2 & 0.002 & 291.5 & 442.855 & 2 & 0.008 + 27 & 426.861 & 2 & -0.002 & 307.5 & 443.820 & 2 & 0.006 + 28 & 426.923 & 2 & -0.001 & 387.5 & 448.654 & 2 & 0.005 + 32 & 427.165 & 1 & -0.001 & 487.5 & 454.700 & 2 & 0.008 + 33 & 427.225 & 1 & -0.001 & 503.5 & 455.675 & 2 & 0.017 + 41 & 427.707 & 2 & -0.002 & 521.5 & 456.760 & 2 & 0.014 + 42 & 427.768 & 2 & -0.002 & 538.5 & 457.787 & 2 & 0.014 + 43 & 427.828 & 2 & -0.002 & 586.5 & 460.690 & 2 & 0.016 + 44 & 427.892 & 2 & 0.001 & 668.5 & 465.638 & 2 & 0.009 + 45 & 427.950 & 2 & -0.001 & 684.5 & 466.599 & 2 & 0.003 + 65 & 429.162 & 1 & 0.002 & 720.5 & 468.783 & 2 & 0.011 + 67 & 429.282 & 1 & 0.001 & 738.5 & 469.858 & 2 & -0.002 + 68 & 429.343 & 1 & 0.002 & 752.5 & 470.704 & 2 & -0.002 + 93 & 430.846 & 2 & -0.006 & 853.5 & 476.812 & 2 & 0.003 + 94 & 430.909 & 2 & -0.003 & 868.5 & 477.716 & 2 & 0.001 + 95 & 430.970 & 2 & -0.003 & 885.5 & 478.740 & 2 & -0.003 + 96 & 431.031 & 2 & -0.002 & 934.5 & 481.700 & 2 & -0.004 + 100 & 431.272 & 1 & -0.003 & 1064.5 & 489.555 & 2 & -0.005 + 101 & 431.330 & 1 & -0.005 & 1083.5 & 490.699 & 2 & -0.009 + 104 & 431.510 & 2 & -0.007 & 1148.5 & 494.627 & 2 & -0.009 + 106 & 431.630 & 2 & -0.007 & 1513.5 & 516.678 & 2 & -0.015 + 107 & 431.691 & 2 & -0.007 & 1544.5 & 518.561 & 2 & -0.005 + 108 & 431.753 & 2 & -0.005 & 1613.5 & 522.717 & 2 & -0.019 + 110 & 431.870 & 2 & -0.009 & 1762.5 & 531.733 & 2 & -0.007 + 111 & 431.932 & 2 & -0.008 & 1794.5 & 533.669 & 2 & -0.005 + 112 & 431.992 & 2 & -0.008 & 1811.5 & 534.705 & 2 & 0.004 + 116 & 432.233 & 1 & -0.009 & 1845.5 & 536.755 & 2 & -0.001 + 117 & 432.295 & 1 & -0.007 & 1853.5 & 537.252 & 2 & 0.013 + 118 & 432.355 & 1 & -0.008 & 1861.5 & 537.727 & 2 & 0.004 + 125 & 432.780 & 2 & -0.006 & 1877.5 & 538.701 & 2 & 0.011 + 126 & 432.842 & 2 & -0.004 & 1886.5 & 539.240 & 2 & 0.007 + 127 & 432.900 & 2 & -0.006 & & & & + + + + we then estimated the times of superhump maxima by eye examination of the light curve ( except at the peak of rebrightening ) and from the phase - averaged light curve ( around the rebrightening maximum ) . the values , together with the literature values , are listed in table [ tab : shmax ] , with cycle counts ( @xmath21 ) starting with @xmath21 = 0 at hjd 2450425.236 . the accuracy of maximum determination is typically 0.001 d during the main superoutburst , and 0.002 d during the rebrightening stage . in order to see whether the observed maximum times can be expressed by a continuous period change of a coherent variation , or we need to introduce a discontinuous phase jump ( cf . @xcite ) , we calculated residuals to linear fits to the observed times in both cases assuming the superhumps having no phase jump and a 0.5 phase jump . the latter ( 0.5 phase jump , or phase reversal ) corresponds to the phenomenon what is called late superhumps ( @xcite ; @xcite ; @xcite ; @xcite ) . ( 88mm,60mm)fig11.eps ( 88mm,120mm)fig12.eps in the former case , the superhump period was confirmed to show a discontinuous change and larger @xmath23 values ( figure [ fig : oc1 ] ) , while the latter case showed a more reasonable , smooth period change when we assume @xmath24159 maxima are late superhumps . this identification looks reasonable that @xmath21 = 159 corresponds to the termination of the main superoutburst , at which a 0.5 phase jump is known to be most frequently observed ( e.g. @xcite ; @xcite ) . in table [ tab : shmax ] , we provide @xmath21 numbers assuming the 0.5 phase jump for @xmath24159 . a linear regression to the observed superhump times ( table [ tab : shmax ] ) yielded the following ephemeris : @xmath25 the @xmath23 s given in table [ tab : shmax ] were calculated against this linear regression ( see figure [ fig : oc2 ] ) . the superhump period initially increased during the main superoutburst and the first rebrightening , and the period started to decrease following the main superoutburst . a similar sequence of the superhump period change is frequently observed in other wz sge - type dwarf novae or infrequently outbursting su uma - type dwarf novae ( e.g. @xcite ; @xcite ; @xcite ) . a quadratic fit to the observed @xmath23 s for @xmath26210 yielded a definitely positive period derivative of @xmath27 . to summarize , the 0.0603-d superhump signal persisted throughout the rebrightening stage and the fading tail rebrightenings do not seem to affect the profile , superhump phase and pulsed flux . these findings make a clear contrast to what are observed during the initial stages of an ordinary su uma - type superoutburst @xcite . the repetitive brightenings less likely represent the start of another superoutburst in spite of accompanying superhumps . instead , they bear more characteristics of normal outbursts . the persistent superhumps should be rather considered to be the consequence of a persistent eccentricity in the accretion disk , and rebrightening must have occurred independent of these superhumps . as briefly discussed in section [ sec : intro ] , no confirmed outburst was observed since the 1977 ( discovery ) outburst until 1995 . in order to estimate the possibility of missed outbursts ( owing to the observational gaps ) we applied monte - carlo simulations on actual observations by the vsolj members . the total number of observations was 238 , distributing between jd 2445344 and 2450933 , spanning @xmath015 yr . the results were that 52% of simulated superoutbursts ( the light curve was assumed to identical with the present one ) and 86% of simulated normal outbursts ( maximum magnitude of 13.3 and decay rate of 1 mag d@xmath28 assumed ) were missed . these probabilities can be considered as upper limits , since the vsolj database contains no observations by huruhata himself , and the object had been equally intensively monitored by a number of skilled visual observers , including p. schmeer . though it is still premature to conclude on the previous occurrence of outbursts , the 1-@xmath19 lower limit of recurrence times of superoutbursts ( @xmath29 ) and normal outbursts ( @xmath30 ) can be set 4 and 1 yr , respectively , assuming the poisson statistic of the outburst occurrence . the low number of outbursts has been also confirmed by the absence of a confirmed outburst in the vsnet observations after the 19961997 one . these lower limits of recurrence times are located at the longest end of the distribution of @xmath29 among all su uma - type dwarf novae ( e.g. @xcite ; @xcite ) . according to @xcite , the 1977 outburst started between 1977 november 9.8 and 12.8 ut . the brightest measured magnitude was 11.9 on 1977 november 12.8 . since the photometric sequence of the comparison stars by huruhata closely matches the present scale ( m. huruhata , private communication ) , we can safely compare huruhata s values with our modern @xmath10-magnitudes . the object declined to 12.4 mag on november 22.8 ut , 10.9 d after the initial detection . the duration of the outburst indicates that the 1977 outburst was also a superoutburst . after an observational gap of 11 d , @xcite further positively detected the variable on three films taken in two separate occasions in early december , giving estimates @xmath014.0 mag 2122 d and 25 d after the initial detection . considering that the duration of the main superoutburst in 1996 did not exceed 23 d , even taking the observational gap into account , some of these december positive detections may be interpreted as a hint of rebrightenings , or a part of a second ( super)outburst . the main difference from the 19961997 outburst is the faintness ( 14.0 mag ) during this stage . it is , however , clear no evidence was found regarding repeated rebrightenings as in 19961997 , and the 1977 outburst seems to more resemble the 1913 and 1946 outbursts of wz sge . it would be important to note the same star at times exhibit different patterns of activity . based on our own analysis , we propose the period of early superhumps ( @xmath6 d ) to be the candidate orbital period . @xcite alternatively proposed a different ( 0.05997 d ) period , which is extremely close to the superhump period . we suspect that this period identification suffered from ambiguity from two reasons . the spectroscopic period by @xcite was _ not _ directly from radial velocity variations but from a periodic variation of the asymmetry of the profile . we must note that this method of period determination could have suffered from the remaining eccentricity in the disk , as would be naturally expected from the long persistence of the superhumps . the photometric method was even more ambiguous ; our own analysis of the early - stage variations gave a different result @xcite , which was also favored by quiescent photometry @xcite . we absolutely need an independent determination from _ true _ radial velocity observation before concluding on the true identification of the orbital period . in this paper , we adopt our own identification ( @xmath6 d ) in the following discussion . lcccc & eg cnc & al com & hv vir & wz sge + year of outbursts & 1977 , 1996 & 1892 , 1941 , 1961 & 1929 , 1939 , 1970 , & 1913 , 1946 , 1978 , + & & 1965 , 1974 , 1975 , & 1981 , 1992 & 2001 + & & 1995 , 2001 & & + maximum magnitude & 11.9 & 12.0 & 11.5 & 7.0 + minimum magnitude & 19.1 & 20.5 & 19.1 & 15.5 + outburst amplitude ( mag ) & 7.2 & 8.5 & 7.6 & 8.5 + supercycle ( yr ) & 19 ? & 1020 & 10 : & 2333 + normal outbursts ? & no ? & yes & yes ? & no + superoutburst duration ( d ) & @xmath20100 & @xmath2070 & @xmath2050 & 130 + long fading tail & yes & probably yes & yes & yes + type(s ) of rebrightening & repeated short & superoutburst - like & probably yes & long ( similar + & outbursts & & ( short ? ) & to al com ) + early superhumps & yes & yes & yes & yes + period of early superhumps ( d ) & 0.05877 & 0.05666 & 0.05708 & 0.05667 + period of common superhumps ( d ) & 0.06038 & 0.05722 & 0.05820 & 0.05721 + superhump excess ( % ) & 2.7 & 1.0 & 2.0 & 1.0 + quiescent humps & ? & double & ? & double+eclipses + + + + + as @xcite pointed out , the optical behavior of the present outburst bears all the characteristics common to those of known wz sge - type dwarf novae . table [ tab : wzcomp ] summarizes the comparison of the observed properties between these objects . the late - stage behavior of superoutbursts has a wide range of diversity between objects , and even between different outbursts of the same object , in spite of the marked similarity of the general light curve and the time evolution of superhumps during the early stage of outbursts . phenomenologically , the superoutbursts of wz sge - type dwarf novae seem to make a continuous sequence ranging from superoutburst without noticeable rebrightenings ( wz sge in 1913 and 1946 ) , ones accompanied by a single rebrightening ( eg cnc in 1977 ; possibly hv vir in 1992 , cf . @xcite ) , double to multiple rebrightenings ( eg cnc in 19961997 ; uz boo in 1994 : @xcite ) , and to two distinct outbursts separated by a dip " ( wz sge in 19781979 , 2001 ; al com in 1961 , 1995 and 2001 ) , some of which may be double superoutbursts . a similar idea of a continuum of types of outbursts of large - amplitude , infrequently outbursting dwarf novae ( tremendous outburst amplitude dwaef novae or toads : @xcite ) was discussed in @xcite , who mainly treated the diversity of duration and brightness of superoutbursts in these systems . an ideal model of wz sge - type dwarf novae should reproduce this wide and continuous variety of outburst activity by changing some input parameters . in observationally interpreting the rebrightening phenomenon , discrimination of _ inside - out_-type ( heating wave originating from the inner accretion disk ) , or _ outside - in_-type outbursts ( @xcite ; the terminology corresponding to type b and type a outbursts in @xcite ) , is potentially important : different parameters of the disk instability model predict different origins of the instability ( e.g. @xcite ; @xcite ) , which may provide a potential observational diagnostic to the underlying mechanisms . ( 88mm,120mm)fig13.eps a close examination of the light curve of each rebrightening revealed the presence of five rebrightenings which showed a rapid rise and one which showed a much slower rise . figure [ fig : minicomp ] demonstrates enlarged light curves of each rebrightening . the first five rebrightenings showed rather common outburst behavior : an abrupt rise and a 3-mag fade in 3 d. the time needed for the rise is well confined to less than 0.4 d from the observations of the fourth and fifth rebrightenings . the sixth rebrightening is different , accompanied with a slow rise lasting more than a day before reaching maximum . ( 88mm,60mm)fig14.eps ( 88mm,60mm)fig15.eps a similar phenomenon , but with a lesser amplitude , was also observed in time - resolved ccd photometry of the first rebrightening ( figures [ fig : reb1 ] and [ fig : miniprof ] ) . a fast - rising type rebrightening occurred at around hjd 2450439 , while a more slowly rising brightening followed at around hjd 2450443 ( not counted as one of six major rebrightenings because of its faintness ) shows much a slower rise and faint peak brightness ( the faintness of the peak was also confirmed from the subsequent ccd observations reported to the vsnet ) . 1 d ) of the phenomenon , as well as the continued slow rising for 7 hr , was much longer than the superhump period . ] although whether or not this small brightening can be treated in the same context of other ( major ) rebrightenings is an open question , the slow rising stage of this small brightening was more analogous to the sixth rebrightening , possibly implying the common underlying mechanism . the existence of two types ( fast - rising and slow - rising ) of rebrightenings may represent the coexistence of _ inside - out_-type and _ outside - in_-type outbursts , more usual ones being the faster _ outside - in_-type . the sixth rebrightening may be related to the former , slow " rebrightening . since the disk instability theory predicts the cycle length of normal outbursts , for a certain range of parameters , is roughly inversely proportional to the mass - transfer rate ( e.g. @xcite ) , it would be natural to attribute the increased outburst activity after the main superoutburst to an enhanced mass - transfer somehow caused by the superoutburst . since the measured short outburst intervals ( 510 d ) are one of the shortest even among frequently outbursting su uma - type dwarf novae ( cf . @xcite ; @xcite ) , this interpretation would require ( assuming no change in other parameters before and after the superoutburst ) a mass - transfer rate something between ordinary su uma - type dwarf novae and er uma stars ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . the er uma stars are peculiar su uma - type stars which are proposed to have mass - transfer rates several times higher than in ordinary su uma - type dwarf novae ( @xcite ; @xcite ) . although the initial pattern and amplitudes of rebrightenings of eg cnc closely resembled those of er uma stars , the enhanced mass - transfer and the enhanced hot spot brightness would not solely by themselves explain the later behavior , since the brightness of the system was shown to decrease gradually ( see figure [ fig : vis ] ) , even after the sixth rebrightening . this smooth decrease in brightness should require _ smoothly decreasing _ frequency of rebrightenings , which is contrary to the observation if the system brightness actually reflects the brightness of the hot spot or the mass - transfer rate . @xcite instead proposed a model to reproduce these rebrightenings , by assuming increased quiescent viscosity after the main superoutburst . this model , by adjusting the time dependence of the quiescent viscosity and its radial dependence through the disk , could at least reproduce the observed rebrightenings without an assumption of an enhanced mass - transfer . @xcite originally suggested that such increased quiescent viscosity is somehow related to the turbulent motion in the disk resulting from the tidal instability . they also proposed the extremely low mass - transfer rate in wz sge - type dwarf novae would be responsible for the slow circularization of the eccentric disk , leading to the slow decrease in the quiescent viscosity after the outburst and multiple rebrightenings never seen in usual dwarf novae . the underlying physical mechanism for the decay of the disk viscosity has been recently proposed by @xcite by considering a decay of mhd turbulence under the condition of the low magnetic reynolds numbers in the cold accretion disk @xcite . a potential difficulty of this model @xcite is that the disk after the main superoutburst would not expand enough to trigger a second superoutburst as reported in al com @xcite . [ note that the outburst of al com after the dip " was not composed of recurring small rebrightenings , as observed in the 2001 superoutburst of wz sge . ] @xcite and @xcite proposed the idea that some material can be piled up just behind the cooling wave , causing it to be reflected as a heating wave , which they considered to cause an additional brightening . regarding the reason why such a condition is predominantly met in wz sge - type dwarf novae , or toads , @xcite suggested that the stored disk material survives being depleted by accretion , owing to the low occurrence of normal outbursts in these systems . however , the degree of depletion of the disk mass during one supercycle is rather expected to affect the initial condition of superoutburst , but not necessarily that of the late - stage condition of superoutburst . as @xcite has shown , the late stage of wz sge - type superoutbursts is expected to follow the same time - evolution as those of usual dwarf novae . the initial excessive matter is effectively accreted on a viscous time - scale ; it is not still convincingly clear how the low occurrence of normal outbursts ( or its precondition ) naturally explains the unique late - stage phenomenon in wz sge - type superoutbursts . though the discussion was more limited to sxts , @xcite could reproduce a similar pattern of rebrightenings by modifying the radial dependence of the @xmath31 viscosity . this model , however , would suffer , as in the model proposed by @xcite , from the depletion of high - angular momentum matter necessary to invoke an immediate second superoutburst . we alternately consider a possibility that the matter in the outer accretion disk may be left from being accreted during the main superoutburst , by way of the premature quenching of the hot state in the outer disk . we propose the weaker turbulence and resultant heating caused by the weaker tidal instability in the outer disk than in ordinary su uma - type dwarf novae can be responsible for this possibility . this weaker tidal torque can be reasonably achieved under conditions of extreme mass ratios in wz sge - type dwarf novae , which can hold the 3:1 resonance radius well inside the roche lobe of the primary , or the maximum dimension of the expanded disk . this possibility was originally proposed in 1997 by @xcite and @xcite , which was further explored by @xcite and @xcite . the unusual presence of na d absorption line @xcite could be attributed to the cool reservoir in the disk required by this interpretation . superoutbursts in ordinary su uma - type dwarf novae are believed to be triggered when the disk radius , upon the ignition of the outburst , first reaches the radius of the 3:1 resonance during the normally outbursting cycle @xcite . the eccentricity wave @xcite caused by the tidal instability should naturally start in the outermost region of the disk , and should propagate inward , accounting for the general decrease in the superhump period during superoutbursts of ordinary su uma - type dwarf novae @xcite . in wz sge - type dwarf novae , however , the absence of normal outbursts may lead to an accumulation of a large amount of matter , leading to more violent expansion of the disk during superoutburst enabling the disk expanding beyond the 3:1 resonance . the above picture then predicts the eccentricity wave can propagate outward . the presence of positive period derivatives of superhump periods in some wz sge - type dwarf novae and infrequently outbursting , large - amplitude su uma - type dwarf novae ( e.g. al com : @xcite ; v1028 cyg : @xcite ; sw uma : @xcite ; @xcite ; hv vir : @xcite ; see also @xcite for a recent summary of period derivatives in su uma - type dwarf novae ) would strengthen the idea of the relative difference in the location of generation and propagation of tidal instability between wz sge - type and ordinary su uma - type dwarf novae . the extremely low mass - ratios in wz sge - type dwarf novae could play the key role in generating tidal instability well inside the entire disk radius . we obtained extensive ccd photometry during the 19961997 superoutburst of eg cnc . this outburst showed an unprecedented series of six major rebrightenings spaced by 510 d. during this rebrightening stage and the following slow fading tail , photometric modulations with a period of 0.0603 d persisted , which is equal to the superhump period observed in the latter course of the main superoutburst . we suggest that the superhumps observed during the rebrightening stage and the fading tail are a remnant " of usual superhumps , and are not a result of rebrightenings . by comparison with the 1977 outburst of this object and outbursts of other wz sge - type dwarf novae , we propose an activity sequence of wz sge - type superoutbursts , in which the current outburst of eg cnc is placed between a single - rebrightening event and distinct outbursts separated by a dip . we propose a new potential explanation of post - superoutburst behavior of the wz sge - type dwarf novae , by considering the effect of extreme mass ratios in these binaries on the generation and propagation of the tidal instability . we are grateful to many amateur observers for supplying their vital visual and ccd estimates via the vsnet alert network , and especially to patrick schmeer for his detection and early notification of the long - awaited outburst . we are grateful to the vsolj staffs and members providing earlier observations . we are also grateful to jonathan kemp for providing up - to - date information , and people involved in extensive on - line discussions in _ vsnet - chat _ ; part of this log can be electronically accessed at http://www.kusastro.kyoto-u.ac.jp/vsnet/dne/egcnc.html and its related links . we express our sincere gratitude to yoji osaki and shin mineshige for valuable discussions made at the annual meeting of the astronomical society of japan . part of this work was supported by a research fellowship of the japan society for the promotion of science for young scientists ( dn ) .

in this article we consider self - similar solutions to smoluchowski s mean - field model for coagulation @xcite that is given by the equation @xmath12\ , , \label{s1e1}\end{split}\ ] ] where @xmath13 denotes the number density of clusters of size @xmath14 at time @xmath15 . the kernel @xmath16 describes the rate of coalescence of clusters of size @xmath14 and @xmath17 and subsumes all the microscopic aspects of the coagulation process . it is well - known , cf . for example the review article @xcite and references therein , that if the kernel grows at most linearly , then the initial value problem for ( [ s1e1 ] ) is well - posed for initial data with finite mass and the mass is conserved for all times . for homogeneous kernels it is furthermore expected that solutions converge to a self - similar form for large times . in fact , for the kernels @xmath18 and @xmath19 , the constant and additive kernel respectively , this issue has by now been completely solved . it has been established @xcite that besides explicitly known exponentially decaying self - similar solutions also solutions with algebraic decay exist . in @xcite their domains of attraction could also be completely characterized . however , not much is known about self - similar solutions for other kernels than these solvable ones . existence of fast - decaying self - similar solutions has been established for a large class of kernels @xcite , and local properties of such solutions have been investigated in @xcite , but it is not known whether they are unique in the class of solutions with finite mass . it is also not clear , not even on the formal level , whether self - similar solutions with algebraic decay exist . in this article we will consider mass - conserving self - similar solutions to coagulation equations with the following product kernel of homogeneity @xmath20 : @xmath21 existence of a solution for this kernel has been established in @xcite . the purpose of this paper is to give an asymptotic description of such a solution in the regime @xmath22 . let us describe briefly what is known for self - similar solutions for kernels as in ( [ s1e2 ] ) and compare it to results for the so - called sum kernels . to fix ideas , we restrict ourselves to kernels of the form @xmath23 even though most of the results that we mention also apply to more general kernels with the same growth behavior as the ones in ( [ s1e2b ] ) . we will need in particular to distinguish between the case @xmath24 , the sum kernel , and the case @xmath25 , the product kernel . it has long been predicted @xcite that self - similar solutions for kernels as in ( [ s1e2b ] ) exhibit a singular power - law behavior of the form @xmath26 with @xmath27 in the case @xmath24 , and @xmath28 for the case @xmath25 . this has been rigorously proved for the case @xmath24 in @xcite and for the case @xmath25 in @xcite . as has been pointed out in @xcite , the next order behavior for small clusters in the case @xmath24 can then easily be established and is as predicted by the physicists . in the case @xmath25 , however , the next order behavior has not been known @xcite , and only numerical simulations suggested that it is oscillatory @xcite . our goal in this paper is to describe formally how to construct mass - conserving self - similar solutions for kernels as in ( [ s1e2 ] ) in the limit @xmath22 and by this also describe their asymptotic behavior on the whole positive real line . while we believe that such self - similar solutions are unique , our approach only gives the construction of one such solution . in forthcoming work @xcite we will also show how to make this construction rigorous . we will see that in the limit @xmath22 the oscillatory character of the solutions becomes very explicit and can be interpreted in terms of simple odes coupled with explicitly solvable equations . in the limit @xmath29 the product kernel becomes close to the sum kernel @xmath30 . as described above the power law for the sum kernel is different from the one for the product kernel . this is due to the fact that , thinking in terms of coagulating particles , the physical behaviour of these particles is different in these two cases . indeed , the power law obtained for the sum case shows that particles with small @xmath8 coalesce mostly with particles that are much bigger than themselves . on the contrary , in the case of the product kernel small particles interact mostly with the ones having a comparable size . the point @xmath31 can be thought as a bifurcation point where both kind of behaviours take place simultaneously . we are now going to derive the equations that are solved by self - similar solutions . it is known that the coagulation equation ( [ s1e1 ] ) can be written in the following conservative form that makes the conservation of the number of monomers @xmath32 transparent . @xmath33 using the self - similar variables @xmath34 as well as the form of the kernel ( [ s1e2 ] ) , equation ( [ s2e1 ] ) becomes @xmath35 integrating this equation , assuming decay of the solutions as @xmath36 and imposing absence of particle fluxes , we obtain @xmath37 the balance of the terms on both sides of ( [ s2e2 ] ) suggests the following power law behaviour for @xmath38 near the origin , @xmath39 where @xmath40 is the classical beta function @xcite . for further reference we notice that @xmath41 in order to remove the power law behaviour we introduce the new function @xmath42 then @xmath9 solves @xmath43 in the rest of the paper we will study solutions of ( [ s2e5 ] ) in the limit @xmath22 that satisfy @xmath44 as @xmath45 . notice that ( [ s2e5 ] ) has the explicit constant solution @xmath46 that corresponds to the well - known solution with infinite mass @xmath47 in the formulation ( [ s2e2 ] ) . a main idea in our approach is to reformulate the problem of finding solutions of ( [ s2e5 ] ) as a shooting problem for a volterra integro - differential equation . indeed , we can rewrite ( [ s2e5 ] ) as @xmath48 differentiating ( [ s3e5 ] ) we obtain the equations @xmath49 thus the solution of ( [ s2e5 ] ) solves ( [ s3e7 ] ) with @xmath9 as in ( [ s3e4 ] ) , that has the advantage of being a volterra integro - differential equation . however , the problem ( [ s3e4 ] ) , ( [ s3e7 ] ) admits solutions that in general do not decay as @xmath50 and therefore they do not provide solutions of ( [ s2e5 ] ) . therefore , among all the solutions of ( [ s3e4 ] ) , ( [ s3e7 ] ) we must select those satisfying @xmath51 we now sketch the asymptotics of the solution that we are going to construct . it passes , roughly speaking , through three stages that are described in detail in sections [ s.zero ] , [ combregion ] and [ s.asymptotics ] respectively . section [ s.zero ] discusses the behavior of the solution near @xmath52 . as pointed out before , near the origin the solution is oscillatory and behaves as @xmath53 where @xmath54 and @xmath55 as @xmath22 . our analysis consists of constructing a solution by starting with @xmath9 as in ( [ oscillations ] ) and using @xmath56 as a shooting parameter . however , since equation ( [ s2e5 ] ) is invariant under the rescaling @xmath57 for @xmath58 , we can restrict the range of @xmath59 to @xmath60 ( see also the comment in section [ ss.variables ] ) . in this regime near the origin we can approximate ( [ s3e4 ] ) and ( [ s3e5 ] ) by a nonlinear ode ( see ( [ s5e6 ] ) ) , that is a perturbation of a simple ode system . for the perturbed system we can use an adiabatic approximation to compute the increase of an associated energy @xmath61 along trajectories . as we discuss in section [ ss.validity ] this first ode approximation is valid as long as @xmath62 . when @xmath63 we enter a new regime that is described in section [ combregion ] . in this regime @xmath9 develops peaks of height @xmath10 and width of order one that are separated by wide regions in which @xmath9 is small . the regime where @xmath9 is small is again described by an ode ( cf . section [ ss.ode ] ) , while the peaks are described by an integro - differential equation ( see section [ ss.integro ] ) . the analysis of these respective regimes is done in sections [ ss.intanalysis ] and [ ss.odeanalysis ] , while their coupling is described in section [ asymptpeaks ] . in section [ s.shooting ] we will then show by a continuity argument that there exists a shooting parameter such that the corresponding solution @xmath9 converges to zero as @xmath45 . in the third regime , that is discussed in section [ s.asymptotics ] , @xmath9 then decays exponentially fast to zero . it has been observed in @xcite that the self - similar solutions associated with ( [ s1e1 ] ) exhibit oscillations for small values of @xmath65 this oscillatory behaviour can be seen as follows . if we make the ansatz @xmath66 with some @xmath67 , plug this into ( [ s2e5 ] ) and keep only the leading order terms , we obtain that @xmath68 must satisfy @xmath69 after some computations , we find that this is equivalent to @xmath70 one can show that @xmath71 has two complex conjugate roots @xmath72 with positive real part and nonzero imaginary part . indeed , in the limit @xmath6 this can be seen easily as the dominating terms in ( [ s3e3 ] ) are , assuming @xmath73 , @xmath74 and thus the roots of @xmath71 satisfy @xmath75 consequently we expect the following asymptotics for the solutions of ( [ s2e5]):@xmath76 for suitable real constants @xmath77 and @xmath78 notice that the asymptotics ( [ s2e6 ] ) indicate that two consecutive local maxima of @xmath9 , called @xmath79 and @xmath80 , satisfy @xmath81 for small @xmath82 in comparison , to double the amplitude of ( [ s2e6 ] ) we must multiply @xmath8 by a number of order @xmath83 . thus , the behaviour ( [ s2e6 ] ) is essentially oscillatory , with a growth in the amplitude that takes place on a much larger scale . for our forthcoming analysis it will be more convenient to use the following variables : @xmath84 then ( [ s3e4 ] ) , ( [ s3e7 ] ) becomes @xmath85 \ , , \label{s3e10}\\ \frac{du}{dx } & = -\left ( 1{-}\lambda\right ) u + \left ( 1{-}\lambda\right ) h\ , , \qquad \frac{dv}{dx } = \lambda v -\lambda h\ , , \label{s3e12}\ ] ] where @xmath86 \left ( x\right ) = h_{\lambda}\int_{-\infty}^{x}dye^{\left ( 1-\lambda\right ) \left ( y - x\right ) } h\left ( y\right ) \int_{\log\left ( e^{x}-e^{y}\right ) } ^{x}dze^{-\lambda\left ( z - x\right ) } h\left ( z\right)\ , . \label{s3e13}\ ] ] the system ( [ s3e10])-([s3e13 ] ) has two explicit solutions , namely @xmath87 @xmath88 and @xmath89 . our goal is to construct a solution of ( [ s3e10])-([s3e13 ] ) for small values of @xmath90 with the property @xmath91 we will see that a solution that satisfies ( [ s4eb ] ) decays exponentially fast to zero . this in particular implies that the corresponding self - similar solution @xmath38 has finite mass . equations ( [ s3e8 ] ) , ( [ s3e9 ] ) and ( [ s2e6 ] ) yield the following asymptotics of the solutions as @xmath92 : @xmath93 where @xmath94 is as in ( [ d1e2 ] ) and @xmath95 the complex number @xmath59 determines the solution of ( [ s3e10])-([s3e13 ] ) uniquely . however , we notice that ( [ s3e10])-([s3e13 ] ) is invariant under translations @xmath96 and thus all the complex numbers in the spiral @xmath97 yield the same solution up to translations . thus we can identify the solutions of ( [ s3e10])-([s3e13 ] ) satisfying ( [ s4ea ] ) , up to translations , with the set of positive real numbers contained between two consecutive intersections of @xmath98 with the real axis . in particular , given @xmath99 we have , since @xmath100 , that the next consecutive point in @xmath101 is @xmath102 therefore , there is a one - to - one correspondence between the points in the interval @xmath103 and the solutions of ( [ s3e10])-([s3e13 ] ) satisfying ( [ s4ea ] ) . the main result of this paper is that we show , using asymptotic arguments , that for small @xmath90 a value of @xmath104 exists such that ( [ s4eb ] ) holds . the key idea in computing the asymptotics of the solutions of ( [ s3e10])-([s3e13 ] ) is to obtain suitable approximations of the operator @xmath105 \left ( x\right ) $ ] in ( [ s3e13 ] ) for small @xmath90 . the formula for @xmath106 ( cf . ( [ s2e3 ] ) ) , combined with ( [ s2e3b ] ) and ( [ s3e13 ] ) , suggests the approximation @xmath86 \left ( x\right ) = \lambda\int_{-\infty}^{x}e^{\left ( y - x\right ) } h\left ( y\right ) dy\int_{\log\left ( e^{x}-e^{y}\right ) } ^{x}h\left ( z\right ) dz\ , . \label{e1}\ ] ] we will discuss the consistency of this assumption in section [ s.consistency ] . the operator @xmath107 \left ( x\right ) $ ] can be further approximated if @xmath108 changes significantly faster than @xmath109 , as we saw is the case for @xmath110 . then it is natural to approximate @xmath111 by @xmath112 and we obtain @xmath86 \left ( x\right ) \sim\lambda h\left ( x\right ) \int_{-\infty}^{x}e^{\left ( y - x\right ) } dy\int_{\log\left ( e^{x}-e^{y}\right ) } ^{x}h\left ( z\right ) dz\ \ ] ] and changing the order of the integrals @xmath86 \left ( x\right ) \sim\lambda h\left ( x\right ) \int_{-\infty}^{x}h\left ( z\right ) e^{z - x}dz\ , . \label{s5e1}\ ] ] from ( [ s3e12 ] ) we have @xmath113 which together with ( [ s5e1 ] ) leads to @xmath86 \sim\lambda h u \ , . \label{s5e2}\ ] ] using ( [ s5e2 ] ) as well as ( [ s2e3 ] ) and ( [ s2e3b ] ) we obtain the following approximation of ( [ s3e10 ] ) up to order @xmath90 : @xmath114 and plugging this approximation into ( [ s3e12 ] ) , ( [ s3e13 ] ) we obtain@xmath115 as ( [ d1e2 ] ) and ( [ s4e0 ] ) imply that the changes of @xmath116 compared to those of @xmath117 are of order @xmath118 , we make the change of variables @xmath119 that transform ( [ s5e3 ] ) , ( [ s5e4 ] ) up to order @xmath118 into @xmath120 let us denote by @xmath121 any solution of ( [ s5e6 ] ) . we first notice that the asymptotics of the solutions of ( [ s5e6 ] ) as @xmath122 agree with those obtained in ( [ s4e0 ] ) . indeed , in the limit @xmath123 we can compute the asymptotics of @xmath124 as @xmath125 using ( [ s4e0 ] ) , ( [ d1e2 ] ) and ( [ s5e5 ] ) : @xmath126 linearizing ( [ s5e6 ] ) near @xmath127 we obtain the same asymptotics for the functions @xmath128 as @xmath129 this gives the desired matching between the solutions of the linearization of ( [ s3e10])-([s3e13 ] ) around @xmath130 and the solutions of the approximated problem ( [ s5e6 ] ) . the approximation ( [ s6e2 ] ) is valid as long as @xmath131 is small . however , a detailed analysis of the nonlinear problem ( [ s5e6 ] ) is needed if @xmath132 becomes of order one . in order to describe the solutions of ( [ s5e6 ] ) in this regime , we notice that this equation is a perturbation of the equation @xmath133 this equation can be explicitly integrated , since the following quantity is conserved along trajectories : @xmath134 the solutions of ( [ s6e3 ] ) are periodic , but this behavior is not compatible with the exponential growth obtained in ( [ s6e2 ] ) . this growth is due to the increase in @xmath61 produced by the terms of order @xmath135 in ( [ s5e6 ] ) . we now compute this change of energy for values of @xmath124 of order one . using ( [ s5e6 ] ) and ( [ s6e4 ] ) we obtain @xmath136 \label{s6e5}\,.\ ] ] since the solutions of ( [ s5e6 ] ) are close to those of ( [ s6e3 ] ) during finite time intervals , we can adiabatically compute the change in @xmath137 more precisely , the solutions of ( [ s6e3 ] ) satisfying ( [ s6e4 ] ) are periodic with a period @xmath138 given by @xmath139 where the functions @xmath140 are defined as the roots of the equation @xmath141 for any given value of @xmath142 and @xmath143 integrating ( [ s6e5 ] ) and using the adiabatic approximation , we can then approximate the change of @xmath61 during an interval of length @xmath144 as @xmath145 d\xi=\phi\left ( e\right ) \label{s6e7}\ ] ] where @xmath146 is a solution of ( [ s6e3 ] ) satisfying ( [ s6e4 ] ) . it turns out that the function @xmath147 is positive for any @xmath148 this follows , using the second equation in ( [ s6e3 ] ) , via @xmath149 d\omega -\int_{-\sqrt{2e}}^{\sqrt{2e}}\left [ \left ( 1-u_{+}\left ( e,\omega\right ) \right ) + \omega^{2}\right ] d\omega\,,\ ] ] whence @xmath150 we can compute the asymptotics of ( [ s6e8 ] ) as @xmath151 using ( [ s6e6 ] ) . notice that the leading contribution to the integral is given by @xmath152 since @xmath153 then @xmath154 we have obtained that , as long as the approximation ( [ s5e6 ] ) is valid , we can approximate the functions @xmath155 by @xmath156 while @xmath61 is computed via the iterative recursion ( [ s6e7 ] ) . due to ( [ s6e8a ] ) these values of @xmath61 increase an amount of order @xmath157 in each period of size @xmath158 we now proceed to discuss the range of validity of the different approximations that have been used to derive ( [ s6e7 ] ) and ( [ s5e6 ] ) . to examine the range of values of @xmath61 for which the adiabatic approximation is valid , we need to show that the value of @xmath159 remains close to @xmath160 for @xmath161 due to ( [ s6e6 ] ) and ( [ s6e8 ] ) the adiabatic approximation is valid as long as @xmath162 let us remark that ( [ s6e9 ] ) does not follow immediately from the inequality @xmath163 because the right - hand side of ( [ s6e5 ] ) does not have a sign and therefore it is not clear that @xmath164 for some @xmath165 a priori it is not possible to rule out the possibility of big changes of @xmath166 along each cycle balanced in such a way that the overall change of @xmath61 along the cycle is just @xmath167 however , as shown in appendix [ a.energy ] , it turns out that ( [ s6e10 ] ) holds true . using ( [ s6e7 ] ) , ( [ s6e8a ] ) and ( [ s6e10 ] ) it follows that the adiabatic approximation is valid as long as @xmath168 and this is satisfied as long as @xmath169 we now study the range of validity of the approximation ( [ s5e2 ] ) that is the main ingredient in deriving the approximate problem ( [ s5e6 ] ) . the main assumption used in the derivation of ( [ s5e2 ] ) is that the characteristic length scale for which @xmath170 has changes comparable to itself is much larger than one . due to ( [ s5e3a ] ) it follows that as long as @xmath171 remains close to @xmath172 the characteristic length scale for @xmath173 is the same as for @xmath174 notice that @xmath175 is small if @xmath176 since we have @xmath177 , the condition @xmath178 holds if ( [ s6e11 ] ) is true , and so the characteristic length scales for @xmath173 and @xmath179 are the same under the assumption ( [ s6e11 ] ) . for @xmath180 of order one , the evolution of ( [ s6e3 ] ) takes place in the length scale @xmath181 or equivalently for changes in @xmath182 of order @xmath183 therefore , the condition in ( [ s6e11 ] ) that allows us to obtain ( [ s5e2 ] ) is immediately satisfied and we can restrict our analysis to the case @xmath184 the first equation in ( [ s6e3 ] ) combined with ( [ s5e5 ] ) implies that @xmath185 the left - hand side measures the relative variations in @xmath186 with respect to changes in @xmath187 the characteristic length scale that describes the changes in this quantity is large as long as @xmath188 which is again true if ( [ s6e11 ] ) holds . therefore , if the condition ( [ s6e11 ] ) is satisfied , we can apply simultaneously all the approximations that lead to ( [ s5e6 ] ) , the adiabatic approximation yielding ( [ s6e7 ] ) as well as the condition @xmath189 however , the three assumptions break down simultaneously if @xmath61 becomes of order @xmath190 . since ( [ s6e7 ] ) and ( [ s6e8a ] ) imply that @xmath61 increases to arbitrarily large values , the failure of ( [ s6e11 ] ) happens for every solution of ( [ s3e10])-([s3e13 ] ) satisfying ( [ s4e0 ] ) . in order to describe the solutions of ( [ s3e10])-([s3e13 ] ) when @xmath191 we introduce a new group of variables . it turns out to be convenient to split the regime into the one where @xmath186 ( or @xmath192 ) is of order one or smaller , and the one where @xmath186 ( or @xmath192 ) is large . notice that ( [ s6e6 ] ) implies that for a given value of @xmath193 the minimum of @xmath186 scales as @xmath194 , and the maximum scales as @xmath10 . we remark that the forthcoming analysis will show that as long as @xmath171 remains of order one and is not small the values of @xmath192 and @xmath186 have the same order of magnitude . this assumption will be made implicitly in all the remaining computations and will be justified by the self - consistency of the derived asymptotics . we begin by studying the case in which @xmath186 ( and @xmath192 ) are large . since we are interested in the case @xmath195 , equation ( [ s6e6 ] ) suggests the rescaling @xmath196 then ( [ s6e12 ] ) yields a characteristic length scale for @xmath182 of order one . this suggests introducing the new set of variables @xmath197 plugging ( [ s7e1 ] ) into ( [ s3e10])-([s3e13 ] ) we obtain @xmath198\ , , \label{s7e4}\\ \frac{d\mathcal{u}}{dx } & = -\left ( 1{-}\lambda\right ) \mathcal{u } + \left ( 1{-}\lambda\right ) \mathcal{h}\ , , \ , \qquad \frac{d\mathcal{v}}{dx } = \lambda\mathcal{v } -\mathcal{h}\ , , \label{s7e6}\ ] ] where the operator @xmath199 $ ] is as in ( [ s3e13 ] ) . using the approximation ( [ e1 ] ) that will be seen to be still valid in this region we obtain @xmath200 \left ( x\right ) = \left ( 1+o\left ( \lambda\right ) \right ) \int_{-\infty}^{x}e^{\left ( y - x\right ) } \mathcal{h}\left ( y\right ) dy\int_{\log\left ( e^{x}-e^{y}\right ) } ^{x}\mathcal{h}\left ( z\right ) dz\,.\ ] ] taking the limit @xmath201 in ( [ s7e4])-([s7e6 ] ) we obtain @xmath202 if @xmath191 but @xmath186 ( and @xmath192 ) remains bounded we can approximate ( [ s3e10])-([s3e13 ] ) as follows . keeping just the leading terms in ( [ s3e10 ] ) , ( [ s3e13 ] ) we arrive at the approximation @xmath203 . plugging this approximation into ( [ s3e12 ] ) we obtain @xmath204 we are going to compute the solutions of ( [ s7e7 ] ) , ( [ s7e9 ] ) explicitly using laplace transforms . in order to bring ( [ s7e7 ] ) , ( [ s7e9 ] ) into the form of a convolution equation we introduce the set of variables @xmath205 that transforms ( [ s7e7 ] ) , ( [ s7e9 ] ) into @xmath206 notice that this system of equations can be obtained from ( [ s3e4 ] ) , ( [ s3e7 ] ) by means of the rescaling @xmath207 and taking the limit @xmath208 we can compute explicitly a family of solutions of ( [ s8e1 ] ) , ( [ s8e2 ] ) that will be used to describe the solutions of ( [ s3e10])-([s3e13 ] ) in the limit @xmath209 [ solint ] for any given @xmath210 and @xmath211 there exists a solution to ( [ s8e1 ] ) , ( [ s8e2 ] ) that satifies @xmath212 it is given by @xmath213 \frac{e^{\zeta}}{\zeta}d\zeta\label{f7e5}\ ] ] where @xmath214 is a contour in the complex plane contained in the domain @xmath215 satisfying @xmath216 as @xmath217 and @xmath218 as @xmath219 with @xmath220 if @xmath221 we have @xmath222 and @xmath223 for any @xmath224 the following asymptotics hold : @xmath225@xmath226@xmath227 if @xmath228@xmath229 the solutions with different values of @xmath230 are essentially the same up to rescaling . notice also that ( [ f7e7 ] ) implies that @xmath231 becomes negative for @xmath8 sufficiently large if @xmath232 some comments about the choice of the range of values of @xmath233 are in order . we need the restriction @xmath234 to avoid the singular points @xmath235 crossing the imaginary axis . it would also be possible to include the value @xmath236 in the second formula of ( [ g1e1 ] ) and ( [ g1e2 ] ) since the right - hand side vanishes . however this would result in asymptotic formulas like @xmath237 that do not have a precise meaning . finally the constraint @xmath238 in the second formula of ( [ g1e3 ] ) is strictly needed , since the asymptotics of @xmath239 for @xmath240 is proportional to @xmath241 it is interesting to note that the solutions described in theorem [ solint ] for @xmath242 are just the self - similar solutions with algebraic decay that have been obtained for the coagulation equation with constant kernel in @xcite . this correspondence can be seen as follows . the solutions we obtained in theorem [ solint ] satisfy @xmath243 with @xmath244 using this formula to eliminate @xmath245 from ( [ s8e1 ] ) we obtain @xmath246 differentiating ( [ a1 ] ) we obtain the equation that is satisfied by the self - similar solutions of @xmath247 -\frac{\left ( 1-a\right ) } { \tau}f\ ] ] having the form @xmath248 with @xmath249 $ ] as in ( [ s1e1])@xmath250 equation ( [ a2 ] ) can be transformed into ( [ s1e1 ] ) using the change of variables @xmath251 and @xmath252 and correspondingly the solutions obtained in theorem [ solint ] are transformed into those obtained in @xcite if @xmath253 integrating the first equation in ( [ s8e2 ] ) we arrive at @xmath254 using the second equation in ( [ s8e2 ] ) we obtain @xmath255 assuming that all the integrals involved are convergent . therefore , using ( [ s8e2a ] ) and ( [ s8e4 ] ) we find @xmath256 eliminating @xmath257 and @xmath258 from ( [ s8e1 ] ) , the second equation in ( [ s8e2 ] ) , as well as ( [ s8e2a ] ) , ( [ s8e3 ] ) , ( [ s8e4 ] ) , implies @xmath259 integrating by parts in the last equation we obtain @xmath260 equation ( [ s8e6 ] ) is a convolution equation . in order to solve it we introduce the laplace transform of @xmath261 @xmath262 then ( [ s8e6 ] ) becomes @xmath263 . this equation can be integrated explicitly . its only solution that does not have singularities and takes real values along the line @xmath264 is @xmath265 \ , . \label{s8e8}\ ] ] inverting the laplace transform we obtain@xmath266 \frac{e^{\zeta x}}{\zeta}d\zeta\ , . \label{s8e9}\ ] ] the change of variables @xmath267 transforms this formula into ( [ f7e5 ] ) . the choice of the contour of integration ensures the exponential convergence of the integral as well as the absence of singularities of the integrand along the contour of integration . taking the limits @xmath268 and @xmath50 we obtain ( [ f7e7 ] ) . the negativity of @xmath269 for @xmath221 can be proved by differentiating ( [ s8e9 ] ) and deforming the contour of integration to a double half - line following @xmath270 in a positive and negative direction with @xmath271 and @xmath272 respectively . then @xmath273 the asymptotics in ( [ g1e1 ] ) can be obtained rewriting ( [ s8e9 ] ) as @xmath274 and@xmath275 differentiating these formulas we obtain @xmath276 and using the change of variables @xmath267 in both integrals it follows that @xmath277 taking the limit @xmath278 in the first formula and @xmath36 in the second and using the fact that @xmath279 we obtain ( [ g1e1 ] ) . using then ( [ s8e2 ] ) we obtain ( [ g1e2 ] ) , ( [ g1e3 ] ) . formula ( [ g1e4 ] ) follows by integrating by residues . we can reformulate the results in theorem [ solint ] in terms of the original functions @xmath280 in the following we choose @xmath281 as a suitable normalization . [ thintas]for any @xmath234 there exists a solution of ( [ s7e7 ] ) , ( [ s7e9 ] ) satisfying @xmath282 . it is given by @xmath283@xmath284@xmath285 if @xmath221 we have @xmath286 for any @xmath287 . we have the asymptotics @xmath288@xmath289@xmath290 if @xmath228@xmath291 equation ( [ s7e10 ] ) that approximates the behaviour of the solutions of ( [ s3e10])-([s3e13 ] ) if @xmath191 and @xmath186 is of order one , can be solved explicitly since @xmath292 is preserved along trajectories . the solutions of ( [ s7e10 ] ) will be used to match the solutions of ( [ s7e7 ] ) , ( [ s7e9 ] ) obtained in the previous section for @xmath293 for this range of values @xmath171 is almost constant if @xmath294 moreover , since we are away from the ode regime described in section [ ode ] we can assume that @xmath295 in ( [ s6e6 ] ) is of order one and therefore @xmath296 is of order one , whence also @xmath297 is of order one . suppose that we consider a matching region ( to be made precise later ) where @xmath296 is of order one , @xmath171 is almost constant and satisfies @xmath298 . then the first equation in ( [ s7e10 ] ) implies that @xmath186 decreases exponentially . more preciesely , let us assume that @xmath299 for @xmath300 then , the asymptotics of @xmath301 are @xmath302 for some @xmath303 . the asymptotics ( [ s9e2 ] ) will be shown to match with suitable solutions among the ones described in theorem [ solint ] . moreover , notice that ( [ s7e10 ] ) implies that these asymptotics are valid as long as @xmath186 remains larger than some small number @xmath304 that we can assume to be of order one , although small . the exponential decay of @xmath186 in ( [ s9e2 ] ) implies that the transition between @xmath299 and @xmath305 takes place on a length scale @xmath306 of order one . we now describe the solution when @xmath186 is less than @xmath307 . for this purpose let us define by @xmath308 and @xmath309 the two roots of the equation @xmath310 . then , in the limit @xmath29 , the curves given by ( [ s9e1 ] ) in the plane @xmath311 are approximately the two horizontal lines @xmath312 and @xmath313 connected by a curve that is approximately vertical . along this connecting curve @xmath171 changes an amount of order one and @xmath314 continues to be of order one since @xmath315 is preserved . when @xmath186 is less than @xmath304 we can approximate the second equation in ( [ s7e10 ] ) as @xmath316 then @xmath317 increases exponentially . the first equation in ( [ s7e10 ] ) indicates that for the solutions of ( [ s7e10 ] ) under consideration , @xmath186 remains small while @xmath171 changes from @xmath318 to@xmath319 then we can use the approximation @xmath320 where @xmath321 is the range of @xmath182 that it takes for @xmath186 to decrease from @xmath299 to @xmath322 we recall that @xmath321 is of order one . notice that the structure of the curve ( [ s9e1 ] ) implies that eventually @xmath186 becomes again of order @xmath304 with @xmath171 close to @xmath323 it then follows from ( [ s7e10 ] ) that for small @xmath324 @xmath186 reaches again the value @xmath172 at @xmath325 after an additional length @xmath326 of order one . notice that , using ( [ s9e3 ] ) , we have @xmath327 and @xmath186 has the behaviour , for @xmath328 , @xmath329 we need to estimate the relation between @xmath308 and @xmath330 from ( [ s9e1 ] ) we deduce that to leading order @xmath331 on the other hand ( [ s9e4 ] ) allows us to obtain an approximation for the transition length @xmath332 indeed , we have @xmath333 whence , since @xmath334 are of order one and @xmath304 can be made arbitrarily small @xmath335 as @xmath209 we have described the transition of the solution @xmath336 of ( [ s3e10])-([s3e13 ] ) for small @xmath90 and @xmath337 the following result will play a crucial role in describing the solutions , because it will show that the amplitude of the oscillations increases with @xmath187 [ l1]suppose that @xmath338 @xmath339 satisfy ( [ s9e6 ] ) . then @xmath340 . this follows from the inequality @xmath341 that is just a consequence of @xmath342 . we now describe the behaviour of the solutions of ( [ s3e10])-([s3e13 ] ) for @xmath201 and @xmath343 with @xmath61 as in ( [ s6e6 ] ) . it turns out that these solutions can be described for this range of values by a sequence of intervals where the solutions can be approximated alternately by solutions of ( [ s7e7 ] ) , ( [ s7e9 ] ) or by solutions of ( [ s7e10 ] ) with both types of regions connected with suitable matching regions . more precisely , suppose that @xmath344 is a point where @xmath345 . notice that at such a point we can expect to be able to use the approximation ( [ s7e10 ] ) . this will be seen matching the ode - integrodifferential equation regime with the ode regime described in section [ odelolkavolterra ] . we will assume for the moment that this approximation is valid . then @xmath192 is also of order one and we can approximate it to the leading order as @xmath203 ( cf . section [ ss.ode ] ) . we can also use the approximation ( [ s9e5 ] ) with @xmath346 and approximate @xmath171 by a constant . then we obtain for @xmath347 of order one that @xmath348 we can now match the asymptotics ( [ t2e1 ] ) with those obtained for the solutions of ( [ s7e7 ] ) , ( [ s7e9 ] ) . we use ( [ s7e1 ] ) to obtain the matching condition @xmath349 this behaviour must be matched with the one for the solutions of ( [ s7e7 ] ) , ( [ s7e9 ] ) obtained in theorem [ thintas ] . more precisely we will match these behaviours with the ones of @xmath350 for a suitable choice of @xmath351 and @xmath352 the asymptotics of @xmath353 combined with the one in ( [ t1e3 ] ) yield @xmath354 since in the matching region we expect to have @xmath355 on the other hand ( [ t1e5 ] ) , ( [ t1e6 ] ) give the choice @xmath356 with these choices of @xmath357 and @xmath358 we obtain , using ( [ t1e3 ] ) , ( [ t1e5 ] ) , ( [ t1e6 ] ) and ( [ t2e2 ] ) , a matching to the first order between ( [ t2e1 ] ) and @xmath359 in the common region of validity where @xmath360 and @xmath361 . we can then use the approximation @xmath359 to describe the solutions of ( [ s3e10])-([s3e13 ] ) for small @xmath90 and @xmath362 of order one . this approximation breaks down for @xmath363 sufficiently large . indeed , ( [ t1e5 ] ) and ( [ t1e6 ] ) imply that @xmath364 and @xmath365 converge exponentially to zero as @xmath366 however , the approximation ( [ s7e7 ] ) , ( [ s7e9 ] ) is only valid if @xmath367 in the region where @xmath368 becomes of order @xmath90 we must use again the approximation ( [ s7e10 ] ) . in particular this region can be described using the analysis in section [ ss.odeanalysis ] . the asymptotics of the solutions for @xmath186 of order one is as in ( [ s9e2 ] ) for suitable choices of @xmath369 more precisely , since this matching is made for @xmath370 we can use the asymptotics ( [ t1e3 ] ) , ( [ t1e5 ] ) , ( [ t1e6 ] ) for @xmath371 we then obtain , using also the rescaling ( [ s7e1 ] ) , the matching conditions @xmath372 we now match these formulas with ( [ s9e2 ] ) ( combined with the approximation @xmath203 ) . to this end we must choose @xmath373 we remark that all this analysis will be meaningful only if @xmath374 therefore @xmath375 the region where @xmath376 can then be described using the analysis in section [ ss.odeanalysis ] . the conclusion of this analysis is that @xmath377 moves close to the point @xmath378 for some suitable @xmath379 in a characteristic length given by ( [ s9e7 ] ) . more precisely if we define ( cf . ( [ s9e6 ] ) ) @xmath380 @xmath381 we can approximate @xmath301 by means of ( [ t2e1 ] ) with @xmath344 replaced by @xmath382 combining ( [ t2e3 ] ) , ( [ t2e4 ] ) and ( [ t2e6 ] ) we obtain to the leading order@xmath383 we first remark that equation ( [ s5e6 ] ) agrees with the approximating equations ( [ s7e7 ] ) , ( [ s7e9 ] ) and ( [ s7e10 ] ) if @xmath384 in their respective regimes of validity . indeed , we can rewrite ( [ s5e6 ] ) using ( [ s5e5 ] ) as @xmath385 and these equations yield the same behaviour as ( [ s7e10 ] ) if @xmath186 is bounded . on the other hand , in order to approximate the solutions of ( [ s7e7 ] ) , ( [ s7e9 ] ) for @xmath384 we use the fact that for small @xmath386 the solutions of ( [ s7e7 ] ) , ( [ s7e9 ] ) in theorem [ thintas ] have the characteristic length scale @xmath387 then , for @xmath233 small , which corresponds to the matching region indicated above , we have a characteristic length very large compared with one . then , as in ( [ s5e2 ] ) we can approximate @xmath388 \sim \lambda \mathcal{h } \mathcal{u}$ ] . since @xmath389 we have @xmath390 and we obtain from ( [ s7e7 ] ) the approximation @xmath391 and plugging this formula into ( [ s7e9 ] ) and using the rescaling ( [ s7e1 ] ) we obtain @xmath392 that agrees with ( [ t2e7 ] ) in the range of values @xmath393 the previous computations show that the equations used in both transition regimes agree in the intermediate matching regime . actually it is possible to check that the main characteristics of the computed solution also agree , as could be expected . notice that the length scale @xmath394 can be computed using the function @xmath138 in ( [ s6e5a ] ) for @xmath384 : @xmath395 in the computation of this integral , we have split the region of integration into the two subsets @xmath396 and @xmath397 where @xmath398 is a small but fixed number . it turns out that the contribution of the second integral is much smaller than the one due to the first integrand , exactly in the same manner that the contributions in @xmath399 computed in subsection [ iteration ] are due mostly to the region with small values of @xmath400 using the rescaling ( [ s5e5 ] ) we then obtain the approximation @xmath401 on the other hand , we can compute the same length using ( [ t2e6a ] ) . taking into account that in the matching region @xmath402 are small , we obtain from ( [ t2e5 ] ) that @xmath403 using ( [ t2e6a ] ) we then obtain to leading order that @xmath404 next , equation ( [ s6e6 ] ) and the definition of @xmath357 by means of @xmath405 imply @xmath406 and plugging ( [ t3e4a ] ) into ( [ t3e4 ] ) we obtain the sought - for matching with ( [ t3e2 ] ) . finally we obtain the matching for the formula of the change of energy in each iteration . we have obtained using the ode approximation that the change of energy in each cycle is given by ( [ s6e7 ] ) where in the matching region @xmath407 we can use the approximation ( [ s6e8a ] ) . then @xmath408 on the other hand , using again @xmath409 and plugging it in ( [ t3e3 ] ) we obtain again ( [ t3e5 ] ) and this yields the desired matching . the structure of the solutions of ( [ s3e10])-([s3e13 ] ) described in section [ asymptpeaks ] by means of alternate approximate solutions of ( [ s7e7])-([s7e9 ] ) and ( [ s7e10 ] ) yields several sequences of numbers @xmath410 that measure the amplitude of the oscillations , the positions of the points where @xmath411 and the position of the peaks where @xmath186 and @xmath192 are of order @xmath190 respectively . it is relevant to notice that the sequence @xmath412 is increasing due to lemma [ l1 ] . the values of @xmath357 approach zero in the matching region ( see the approximation ( [ t3e4a ] ) for @xmath407 ) , but they become of order one for @xmath413 with increases in each iteration of order one . eventually , this sequence of points becomes larger than or equal to one . a detailed study of the solutions of ( [ s3e10])-([s3e13 ] ) shows that there are several possibilities . if @xmath357 becomes strictly larger than one , theorem [ thintas ] shows that during the peak regime @xmath171 becomes strictly negative . in particular @xmath171 reaches the value @xmath414 at some finite @xmath415 if @xmath171 remains positive for larger values of @xmath182 , something that could happen if @xmath416 it would be possible to approximate ( [ s3e10])-([s3e13 ] ) by means of ( [ s7e10 ] ) . therefore @xmath171 would increase , and the value of @xmath357 would be increased to a new value @xmath417 at the next iteration . the only possibility of not having such a behaviour would be with @xmath418 as @xmath419 such approximation to zero can not be described by the approximate equation ( [ s7e10 ] ) , because in such a regime the integral terms in ( [ s3e10])-([s3e13 ] ) must be taken into account in full detail . we now recall that for any @xmath420 there exists a unique solution of ( [ s3e10])-([s3e13 ] ) satisfying ( [ s4e0 ] ) . the discussion in section [ ss.variables ] shows that all the solutions of ( [ s3e10])-([s3e13 ] ) with such a behaviour can be obtained , up to rescaling , by choosing @xmath59 in the interval @xmath421 a continuity argument taking into account the behaviour of the trajectories shows that there exists a value of @xmath422 in this interval for which the corresponding solution of ( [ s3e10])-([s3e13 ] ) satisfies @xmath423 this also implies @xmath424 and @xmath425 using ( [ s3e10 ] ) , ( [ s3e12 ] ) . more precisely , the inversion of ( [ s3e10 ] ) allows us to write @xmath426 in terms of @xmath427 then @xmath428 and ( [ s3e12 ] ) yields @xmath429 we finally describe the asymptotics of the solution of ( [ s3e10])-([s3e13 ] ) satisfying ( [ s4e0 ] ) with @xmath422 described in the previous section . we have seen that @xmath431 to leading order the behaviour of this trajectory is described by the solution of ( [ s7e7 ] ) , ( [ s7e9 ] ) given in theorem [ thintas ] with @xmath432 the asymptotics @xmath433 is valid as long as @xmath434 , where , by assumption , to the leading order @xmath435 it is convenient to reformulate ( [ t4e1 ] ) in the original variables ( cf . ( [ s3e8 ] ) , ( [ s3e9 ] ) ) @xmath436 that is valid for @xmath437 we can match ( [ t4e2 ] ) with the following exponential asymptotics for the solutions of ( [ s2e5 ] ) as @xmath50 : @xmath438 the asymptotics ( [ t4e2 ] ) , ( [ t4e3 ] ) match for small @xmath90 if @xmath439 this gives the desired exponential decay for the obtained solution . we now show that the solution we obtained is self - consistent with the approximations made for the integral operator @xmath107 \left ( x\right ) $ ] defined in ( [ s3e13 ] ) . we have made two main approximations . the first one is ( [ e1 ] ) , that approximates @xmath107 $ ] by an operator with a simple dependence on @xmath82 the second approximation is ( [ s5e2 ] ) and it allows us to replace the integral operator by a much simpler local operator . concerning the validity of ( [ e1 ] ) we remark that its precise meaning is that the error made in the approximation is smaller than the right - hand side . in order to check its validity we distinguish between the regions where @xmath192 is of order @xmath190 and the regions where @xmath192 is smaller than that quantity . notice that for any region we can expect the errors made in the approximation to be of order @xmath440 and we can expect these two terms to be very small compared with the right - hand side of ( [ e1 ] ) . indeed , in the case of @xmath441 , the contribution due to the region @xmath442 is small compared with the term in ( [ e1 ] ) if @xmath304 is small . on the other hand , if @xmath443 we can use the smallness of the exponential factor @xmath444 to obtain estimates for the corresponding terms in @xmath441 as @xmath445 and since @xmath446 it then follows that this contribution is exponentially small . concerning @xmath447 we can argue similarly for @xmath448 smaller than @xmath304 . if @xmath449 , since @xmath450 it follows that the size of the region of integration can be bounded by @xmath451 and therefore , it gives also a very small contribution . concerning the approximations of @xmath107 $ ] by means of local terms that have been made in the derivation of the ode approximations ( [ s5e6 ] ) or ( [ s7e10 ] ) two main ingredients are required . in the derivation of ( [ s5e6 ] ) we have used just the fact that @xmath452 has significant changes over distances much longer than one . since the order of magnitude of @xmath192 is roughly the same for the range of values described by means of ( [ s5e6 ] ) no special care is required to estimate the values of @xmath452 with @xmath453 on the other hand , in the case of the approximation ( [ s7e10 ] ) we use the fact that @xmath107 $ ] is a corrective term that is completely ignored in ( [ s7e10 ] ) . notice that @xmath452 takes values for @xmath454 much larger than the ones of @xmath426 in the region where the approximation ( [ s7e10 ] ) is used . however , due to the exponential factor @xmath444 as well as the exponential decay of @xmath452 in the region described by the integro - differential equation ( cf . subsection [ iteration ] ) the corresponding contribution of such values of @xmath452 is negligible . during the part of the dynamics dominated by a perturbation of a conservative ode it has been assumed that the energy @xmath61 does not change significantly during each cycle of length @xmath158 this is the justification for the adiabatic approximation and we will check now that , indeed , the adiabatic approximation yields such smallness for the variation of the energy during the trajectory and thus the assumption is self - consistent . let us denote by @xmath455 a value where @xmath456 our goal is to estimate this quantity for @xmath457where @xmath458 is the next value of @xmath14 where @xmath459 and @xmath460 again . due to the invariance of ( [ s5e6 ] ) we can assume without loss of generality that @xmath461 due to ( [ s6e5 ] ) and assuming the adiabatic approximation we obtain the following approximation for the change of the energy along a trajectory : @xmath462 ds=\sigma\left ( \xi\right)\ , . \label{a1e1}\ ] ] since @xmath463\ ] ] and given the form of the curves @xmath464 satisfying ( [ s6e4 ] ) we obtain the existence of numbers @xmath465 satisfying @xmath466 as well as @xmath467 moreover @xmath468 , @xmath469 and @xmath470 the value of the total change of the energy during each cycle , namely @xmath471 has been computed in ( [ s6e7 ] ) , ( [ s6e8 ] ) and approximated asymptotically as @xmath151 ( cf . ( [ s6e8a ] ) ) . notice that @xmath472 , assuming that @xmath473 due to ( [ a1e2 ] ) , ( [ a1e3 ] ) we need to compute @xmath474 in order to estimate the range of variation of @xmath475 . we then have @xmath476 where we use that @xmath477 in order to estimate these quantities we define an auxiliary quantity @xmath478 via @xmath479 that measures the maximum value of @xmath480 for a given value of @xmath61 . it will be more convenient to compute @xmath481 using @xmath480 as independent variable . to this end we define @xmath482 by means of the roots of the equation @xmath483 taking into account ( [ s6e3 ] ) we can use @xmath480 as variable of integration instead of @xmath484 in order to compute @xmath485 for @xmath486 then , with @xmath487 we obtain @xmath488 d\omega_{0}\ \ , \ \ k=1,2,3\ , , \label{a1e4}\ ] ] where the integration is made along the curve defined by ( [ u2e3 ] ) . notice that we use @xmath489 in order to approximate @xmath490 for small @xmath478 we use taylor in ( [ u2e3 ] ) to obtain the approximation @xmath491 that combined with ( [ a1e4 ] ) yields @xmath492 d\omega_{0}\ \mbox { as } \omega_{0}\rightarrow0 \,.\ ] ] on the other hand , we can approximate @xmath493 by @xmath494 for @xmath495 whence @xmath496 as @xmath497 for @xmath495 and @xmath498 since the region where @xmath179 is of order one gives a small contribution to the integrals , we obtain the approximations @xmath505 combining this with ( [ s6e8a ] ) we obtain that @xmath506 for some @xmath507 independent of @xmath502 it then follows that the change of energy during each cycle can be estimated by the final change . therefore , this justifies the adiabatic approximation . * acknowledgment : * bn and jjlv gratefully acknowledge the warm atmosphere at the isaac newton institute for mathematical sciences where part of this work was done during the program on pde in kinetic theories . this work was also supported by the epsrc science and innovation award to the oxford centre for nonlinear pde ( ep / e035027/1 ) and through the dges grant mtm2007 - 61755 . r. l. drake , _ a general mathematical survey of the coagulation equation _ , in : topics in current aerosol research ( part 2 ) . international reviews in aerosol physics and chemistry , oxford . pergamon press ( 1972 ) , 203 - 376 . p. laurenot and s. mischler , _ on coalescence equations and related models _ , modeling and computational methods for kinetic equations , eds . p. degond , l. pareschi , g. russo , series modeling and simulation in science , engineering and technology , birkhuser , ( 2004 ) , 321 - 356 .

in the high - energy regime of qcd , an interesting new phenomenon is expected to show up : parton saturation @xcite . saturation changes qualitatively the usual equation for the evolution of scattering cross sections with the energy of the reactions , namely the so - called balitsky - fadin - kuraev - lipatov ( bfkl ) equation @xcite , by introducing nonlinearities . in turn , due to these nonlinearities , the intrinsic stochasticity of partonic evolution may start to have a sizable effect on observables . the basic nonlinear evolution equation beyond the linear bfkl equation is the balitsky - kovchegov ( bk ) equation @xcite ( or , alternatively , the balitsky - jalilian marian - iancu - mclerran - weigert - leonidov - kovner ( b - jimwlk ) equations @xcite ) which however neglects the stochastic effects alluded to before , and is therefore a kind of mean - field approximation . while a formulation which would include all stochastic effects has not been fully achieved yet ( the most recent advances may be found in ref . @xcite ) , it is believed that the complete evolution has a lot in common with some reaction - diffusion processes described by equations of the fisher - kolmogorov - petrovsky - piscounov ( fkpp ) type @xcite . at the mean - field level , this analogy is a formal identity between the fkpp equation and the bk equation in the so - called diffusive limit and assuming uniformity in the transverse space ( i.e. the gluon distribution is assumed to evolve in the same way at all points of the impact - parameter space ) @xcite . beyond the mean - field approximation , the conjecture made so far for the full problem is that at any fixed position in transverse space , the rapidity evolution of say the gluon content of a hadronic object is like the time evolution of a one - dimensional reaction - diffusion process , whose space variable would be the logarithm of the transverse size ( or momentum ) of the gluons . generally speaking , the dynamics of such systems is described by equations equivalent to stochastic extensions to the fkpp equation . ( the first ideas on how to go beyond the bk equation were presented in ref . @xcite ; the equivalence with reaction - diffusion processes was conjectured in ref . @xcite ; deeper insight can be found in e.g. ref . @xcite , and a review in ref . @xcite . ) while this analogy is useful to find asymptotic properties of observables that involve one unique impact parameter , so far little is known about the correlations and the fluctuations of the gluon distribution between different points in transverse space , which would show up in observables that probe several points in impact - parameter space simultaneously . the correlations of the gluon number densities at different points in transverse space were computed exactly in ref . @xcite in the context of the systematic approximation to full qcd provided by the color dipole model @xcite , which is an accurate representation of the physics described by the bfkl equation but which does not take into account saturation effects . in ref . @xcite , a calculation was done in a theory with full saturation based on a similarity with the liouville gravity . but that calculation was valid only up to distances of the order of the inverse saturation momentum . on the numerical side , on one hand , the bk equation was solved taking into account the full impact - parameter dependence @xcite , and on the other hand , the relevance of the one - dimensional stochastic fkpp equation at each fixed impact parameter was tested in toy models for qcd evolution beyond the mean - field approximation @xcite . in this paper , we would like to investigate how the saturation scale varies in impact - parameter space in the presence of both saturation and fluctuations . our method will consist in proposing a simple toy model which contains the main physical features of qcd , which may be implemented as a monte - carlo event generator and for which analytical calculations will be possible . in these respects , our approach follows the one developed in ref . @xcite , but while the latter work was purely numerical , our main results will consist in analytical expressions of the correlation of the saturation scale between two points in impact - parameter space , as a function of the distance between the points and as a function of the rapidity . the model is introduced in the next section . we then provide the derivation of the analytical expression for the correlations . finally , we check our calculations against numerical simulations of different versions of the model . the model that we introduce here is a simplified version of the model proposed in ref . @xcite . the starting point is the qcd color dipole model @xcite , supplemented with some ad hoc saturation mechanism which limits the number of dipoles in any given phase space cell . the dipole model accurately represents the qcd evolution in the high - energy regime and in the limit of a large number of colors . it provides an equation for the change of the density of gluons ( represented by a set of color dipoles of different sizes and positions in the two - dimensional plane transverse to the flight axis of the hadron ) inside a hadron , when rapidity is increased . the basic process from which the evolution is built is the splitting of a dipole represented by its two endpoints @xmath2 into two dipoles @xmath3 and @xmath4 with the rate @xcite = d^2x_2 . [ eq : dpdy ] as usually , @xmath5 , where @xmath6 is the strong coupling constant , @xmath7 the number of colors and @xmath8 is the rapidity . the splitting of dipoles is a linear process , which generates the bfkl equation when averages over dipole configurations ( `` events '' in an experimental language ) are taken . the rapidity @xmath8 is an effective evolution time . when the rapidity becomes very high , then gluons and thus dipoles may start to interact among themselves , which induces nonlinear terms in the evolution equations . the effect of these interactions is to tame the growth of the phase - space number density of dipoles as soon as it reaches @xmath9 , which would otherwise be exponential with the rapidity . the precise mechanism for these effects is still not known in qcd , but the main observables should be quite independent of these details . with respect to qcd , we assume the following simplifications : _ ( i ) _ dipoles evolve by giving birth to one dipole of half size ( the left or the right half of the parent dipole ) , or to one dipole of double size ( in such a way that the parent be the left or right half of its offspring ) at some fixed rates , _ ( ii ) _ dipoles do not disappear in the evolution , that is to say , the parent dipoles are not removed , _ ( iii ) _ the positions and dipole sizes are discrete , and _ ( iv ) _ the configuration space of the dipoles is a line instead of the full two - dimensional space . we thus give up two main properties of the qcd dipole model : the collinear singularities , which cause the dipole endpoints to emit an arbitrary number of dipoles of arbitrarily small sizes , and the continuous and two - dimensional nature of the dipole sizes and positions . the first simplification is the diffusion approximation , which has been studied in the context of bfkl physics ( see e.g. ref . @xcite ) , but which was not assumed in ref . @xcite . the second simplification was instead already assumed in @xcite . these model simplifications may introduce some artefacts , but that we believe are under control , and many results which we will obtain within such simple models are likely to apply to qcd since they will not depend on the details . let us now specify completely the model . according to the evolution rules given above , starting from a dipole of size 1 , the sizes of all dipoles present in the system after evolution are powers of 2 . in practice , we shall only consider fractions of 1 , i.e. the sizes may be written as @xmath10 , where @xmath11 . for each value of @xmath12 , there are @xmath13 possible values of the position @xmath14 of the center of the dipoles : @xmath15 . let us number these bins by the index @xmath16 running from the negative to the positive positions . the model may be represented as a hierarchy of bins that contain a discrete number of dipoles , see fig . [ fig : sketch0 ] . note that to any given impact parameter @xmath14 between @xmath17 and @xmath18 corresponds one unique bin at each level of size . for example , at position @xmath19 , one sees the bins @xmath20 , @xmath21 , @xmath22 etc ... at position @xmath23 , one sees the bins @xmath20 , @xmath21 , @xmath24 etc ... more generally , at position @xmath25 , one sees @xmath26)$ ] , where the square brackets represent the integer part . during the rapidity ( or time ) interval @xmath27 , a dipole in the bin @xmath28 has a probability @xmath29 to give birth to a dipole in the bin @xmath30 , @xmath31 to give a dipole in the bin @xmath32 , and @xmath33 to give a dipole in the bin @xmath34 if @xmath35 is even and @xmath36 if @xmath35 is odd . note that @xmath27 may be infinitesimal ( which is generally speaking convenient for analytical calculations ) , but also finite ( which is convenient for numerical simulations ) . as for the saturation mechanism , we assume the simplest one : we veto splittings to bins which already host the number @xmath37 of dipoles . we can consider that the number density of `` gluons '' of a given size seen at one impact parameter is proportional to the number of dipoles in the corresponding bin @xmath28 . as rapidity is increased , the occupation of the bins with low values of @xmath12 gets higher until the number of objects they contain reaches @xmath37 . the subsequent filling of the bins indexed by larger values of @xmath12 ( smaller dipole sizes ) can be seen as the propagation of traveling wave fronts at each impact parameter , with possibly complicated relationships between them . the ( logarithm of the ) saturation scale @xmath38 at impact parameter @xmath14 is related to the position of the front seen there at time @xmath39 . there are several equivalent ways to define the position of the front . it could be , for example , the largest value of @xmath12 for which the number of objects becomes some given fraction of @xmath37 . ( later , we will use a slightly different definition ) . let us denote by @xmath40 the number of dipoles present in bin @xmath28 at time @xmath39 . then , according to the rules given above , we can write the following stochastic evolution equation : @xmath41)}(t)\\ + \delta^{\beta/2}_{(k+1,2j)}(t)+\delta^{\beta/2}_{(k+1,2j+1)}(t ) \bigg ] , \label{eq : defmodel1}\end{gathered}\ ] ] where the @xmath42 are drawn according to the binomial distribution = n_(k , j)(t ) + l ( x dt)^l(1-x dt)^n_(k , j)(t)-l . [ eq : defmodel2 ] this is a rather complicated equation which we do not know how to solve except numerically . this model does not a priori look like a stochastic fkpp model . we may assume uniformity in impact parameter : this would amount to imposing the same @xmath43 and @xmath44 respectively for all @xmath35 at any given @xmath12 . in this case , the model would be projected to the fkpp class , but by definition , this would wash out the fluctuations between the different impact parameters . this simplified model , that we call `` fip '' ( for `` fixed impact parameter '' , since effectively , the model is completely defined by a single impact parameter ) in the terminology of ref . @xcite , is nevertheless useful since it provides a benchmark to evaluate how the fluctuations between different impact parameters may alter the fkpp picture . in this paper , we will rely on ( and check again in the case of our model ) the conclusion reached in ref . @xcite that thanks to saturation , locally at each impact parameter , the full model is still well - described by a one - dimensional fkpp equation , and the fluctuations between different positions in impact - parameter space do not qualitatively change the picture . let us first apply the well - known treatment of fkpp equations to the fip case . we know that the large - rapidity realizations of the model are stochastic traveling waves , whose main features can be determined from a simple analysis of the linear part of the evolution equation . in this model , only the number of dipoles @xmath45 in the bins say @xmath46 ( i.e. at impact parameter @xmath17 ) is relevant . the evolution equation reads instead of the last term in eq . ( [ eq : evolutionfip ] ) . ( this may even be a more literal implementation of the fip approximation ) . but this would not make a large difference , which anyway , we would be unable to capture analytically . ] n_k(t+dt)=. [ eq : evolutionfip ] the mean - field ( or balitsky - kovchegov ) approximation to the evolution leads to the equation n_k(t+dt)= , where the @xmath45 are now real functions of @xmath12 . the linearized equation ( equivalent to the bfkl equation ) is simply obtained by discarding the `` @xmath47 '' in the previous equation : n_k(t+dt)= n_k(t)+dt n_k-1(t)+dt n_k+1(t ) . [ eq : bfkl ] from standard arguments , we know that for asymptotically large @xmath39 and @xmath37 , the velocity of the wave front , that is the time derivative of the position @xmath48 of the front , is given by @xcite v_0==^(_0 ) , [ eq : v0 ] where @xmath49 is the eigenvalue of the kernel of the linearized evolution equation ( [ eq : bfkl ] ) corresponding to the eigenfunction @xmath50 , namely ( ) = ( 1 + dte^+dte^- ) , [ eq : eigenvalue ] and @xmath51 minimizes @xmath52 . we recall that @xmath27 may be finite or infinitesimal , in which case eq . ( [ eq : eigenvalue ] ) is to be understood as the derivative of @xmath53 . the shape of the front is a decreasing exponential to the right of the saturation region , n_k(t)~n e^-_0 ( k - x(t ) ) , [ eq : exposhape ] extending to @xmath54 for large times . the corrections due to the discreteness of @xmath45 are known . taking the latter into account , the front has now a finite extension , of the order of l_0=. [ eq : l0 ] indeed , typically , it can not extend further than the point where @xmath55 , and eq . ( [ eq : l0 ] ) then follows from eq . ( [ eq : exposhape ] ) . its mean velocity reads @xcite v_=^(_0 ) - . [ eq : vbd ] this velocity was obtained in a still deterministic ( mean - field ) approximation with appropriate cutoffs to mimic saturation and discreteness . taking furthermore fluctuations into account , the velocity becomes @xcite v=^(_0 ) - , [ eq : vbdmm ] and realization - to - realization fluctuations are characterized by the following cumulants of the position of the front : = , [ eq : cumbdmm ] where a priori the model leads to @xmath56 , but empirically , a better fit to the results of the numerical simulations is obtained by adding a subleading correction of the form @xmath57 . in particular , the diffusion constant of the front , that is to say the slope of the @xmath39-dependence of the second - order cumulant reads d=. [ eq : d ] these expressions are valid in the limit of large @xmath58 , i.e. for exponentially large values of @xmath37 . there exists a subclass of these models that may be reduced exactly to a collection of fkpp models . let us set @xmath59 , that is to say , authorize only splittings to smaller - size dipoles . consequently , there may not be any influence of the bins at any size level @xmath12 on the content of the bins of level less than @xmath12 ( i.e. of larger sizes ) . then , at each impact parameter , one has a fip model , i.e. a model of the one - dimensional fkpp type . however , the relationship between the different copies of fip models is not trivial , since part of the evolution is common between different impact parameters . even for @xmath60 , we believe that this minimal model , consisting in considering two one - dimensional systems appropriately correlated , is a good approximation to the full model . in our investigations , we will have in mind the latter class of models , and we will check numerically that for more general models for which @xmath60 , the results that we shall obtain are not significantly altered . our aim is to study the correlations between the point at position @xmath19 in transverse space ( left edge of the system , see fig . [ fig : sketch0 ] ) and the one at position @xmath61 with @xmath62 . we calculate the average of the squared difference of the positions of the front between these points , which is formally related to the two - point correlation function of ( the logarithm of ) the saturation scales , and which we deem a good estimator of the spatial fluctuations of the saturation scale . in the hierarchical model , all bins with index @xmath12 less than or equal to @xmath63 $ ] ( the notation `` @xmath64 $ ] '' stands for the integer part ) and @xmath65 overlap both impact parameters , and thus the dipoles of size larger than @xmath66 seen at these points are exactly the same . for @xmath67 instead , the bins seen at the two points are distinct and nonoverlaping . so in particular , in our model with @xmath59 , as soon as the position of the front at one point or at the other is larger than @xmath68 , that is to say , as soon as there are of the order of @xmath37 dipoles in the bin @xmath69 , then the evolutions are completely uncorrelated at the two points in the corresponding bins . ( we expect that for finite @xmath70 of order 1 , the discussion would not be qualitatively changed . ) this matches to the picture that we may infer for the qcd dipole model : the dipoles at two positions in impact - parameter space separated by a distance larger than the typical saturation scales in that region evolve ( almost ) independently towards larger rapidities . note that choosing pairs of points around impact parameter 0 , one with positive impact parameter and another one with negative impact parameter , would not satisfy this property , due to the rigidity of the sizes and positions of the dipoles . indeed , these two points would decorrelate very soon in the evolution since their common ancestors necessarily sit in the bin @xmath20 , see fig . [ fig : sketch0 ] . as a consequence of these features of qcd reproduced in the toy model , studying two - point correlations between points in impact - parameter space as a function of their distance @xmath71 and of the time ( = rapidity ) @xmath39 is equivalent to studying the time dependence of the correlations of the saturation scales of two realizations of the model whose evolutions are identical until the tip of the front reaches @xmath68 . on the average , it takes a time @xmath72 , @xmath73 being the mean velocity of the individual fronts , for the front whose tip is at @xmath74 at the beginning of the evolution to have its tip at @xmath68 . then the bins such that @xmath67 evolve independently between the two realizations over the remaining time interval t = t - t_b , t_b=. [ eq : ttb ] note that this is very close to assuming that the realizations are identical for @xmath75 and completely uncorrelated for @xmath76 . from this discussion , we see that the basic input of our calculation will be the mechanism for the propagation of a fkpp front . we will review it in the next subsection , then we will proceed to the formulation of the calculation of the correlations . in this section , we review the recent progress in the understanding of the mecanism for front propagation @xcite , which has eventually led to the formulation of a model , and from which eqs . ( [ eq : vbdmm ] ) , ( [ eq : cumbdmm ] ) follow @xcite . instead of solving the nonlinear problem , it was proposed @xcite to replace the nonlinearities by absorptive boundaries , and to treat the evolution equation as a linear ( branching diffusion ) equation between these boundaries . there are two types of nonlinearities : _ ( i ) _ the saturation condition that keeps the number of particles in each bin less than or equal to @xmath37 and _ ( ii ) _ the very discreteness of this number of particles , whose effect mainly shows up in the region where @xmath77 . the position of the boundaries is adjusted in such a way that the distance between them be the size of the front ( [ eq : l0 ] ) , namely @xmath78 , and that the large - time solution of the diffusion equation be stationary . this procedure leads to the expression of the front velocity for large @xmath37 given in eq . ( [ eq : vbd ] ) , but it does not predict the fluctuations of the position of the front . in order to incorporate the latter , we add to this picture the possibility that there be one ( or a few ) particles randomly sent ahead of the tip of the front @xcite , at a distance @xmath79 . we attribute to this event a probability per unit time p()d = c_1 e^-_0()d , [ eq : p ] in the continuation of the shape of the front ( [ eq : exposhape ] ) solution in the large-@xmath39 and large-@xmath37 limit . ( @xmath80 is some constant , undetermined at this stage ) . in ref . @xcite , we treated such fluctuations as localized extra weights of appropriate `` mass '' a unit or so to the left of the right boundary . we then computed the effect of this weight on the position of the front at large time , and found a forward shift of the position equal to r()= ( 1+c_2 ) . [ eq : r0 ] @xmath81 is another constant . at large times after a fluctuation has occurred , the front relaxes to its mean - field shape . we assumed that the relevant fluctuations are rare enough in such a way that the front has time to completely relax between two fluctuations , which turns out to be true for @xmath82 . from the two quantitative elements ( [ eq : p ] ) and ( [ eq : r0 ] ) , together with the solution of the stationary diffusion problem between the boundaries ( [ eq : vbd ] ) , we may write down an effective theory for stochastic front propagation , with however two unknown parameters , namely @xmath80 and @xmath81 . while we were not able to determine @xmath80 and @xmath81 separately , it is the product @xmath83 that appears in all cumulants of the position of the front , and in particular in the correction to the velocity induced by fluctuations . assuming that the velocity of the front taking into account the fluctuations is the velocity of a mean - field front whose size is extended by @xmath84 with respect to @xmath85 ( i.e. with the substitution @xmath86 ) , we got a determination of the product @xmath83 : c_1c_2=^2^(_0 ) . [ eq : c1c2 ] the effective theory for front propagation then leads to eqs . ( [ eq : vbdmm ] ) and ( [ eq : cumbdmm ] ) . as we will see below , the only new ingredient that will be needed is the time dependence of the shift of the front @xmath87 , whose large - time asymptotics is eq . ( [ eq : r0 ] ) . in line with the above discussion , we wish to compute the correlations of the position of two fronts whose evolutions are identical for @xmath75 and uncorrelated for @xmath76 . note that strictly speaking , we would need to keep the content of all bins @xmath88 identical between the two realizations at all times , even after time @xmath89 . but these two formulations give quantitatively similar results . let us introduce @xmath90 the position of the front at time @xmath39 in the frame in which @xmath91 . we focus on what happens slightly before the initial time @xmath92 . on one hand , @xmath93 if no fluctuation has occurred between times @xmath94 and @xmath92 , on the other hand , @xmath95 if a fluctuation has occurred at a position @xmath79 ahead of the front ( which happens with probability @xmath96 ) . it is straightforward to write an equation for the generating function of the cumulants of @xmath97 : -e^x(t_0,t)=v_+dp ( ) ( e^r(t - t_0,)-1 ) . one now considers two such independent fronts and add up the generating functions . one gets @xmath98 expanding for @xmath99 close to 0 , the coefficients of the second power of @xmath99 obey the equation ( x_1-x_2)^2=2dp ( ) r^2(t - t_0 , ) , [ eq : corr ] where we have used the fact that @xmath100 and @xmath101 are independent random variables for @xmath102 , and we have traded @xmath92 for @xmath39 in the derivative , taking advantage of the fact that both @xmath103 and @xmath87 only depend on @xmath104 . in practice , @xmath92 will be equal to @xmath89 , the time at which the tip of the single front reaches @xmath68 . from eq . ( [ eq : ttb ] ) , this time is @xmath105/v$ ] . we see that the basic ingredient is the time evolution of the shift of the front due to a forward fluctuation . this shift was given in eq . ( [ eq : r0 ] ) in the limit of large times . we have to repeat the steps that led to eq . ( [ eq : r0 ] ) keeping however track of the full time dependence . the problem amounts to solving a diffusion equation between two fixed absorptive boundaries , with various initial conditions . we shall discuss the scaled dipole number @xmath106 , and for the simplicity of the formalism , consider that it is a function of a real variable @xmath12 . let us write the general branching diffusion equation : _ t u(t , k)=(-_k)u(t , k ) , [ eq : equ ] where @xmath107 is an appropriate kernel that encodes the linear evolution of the dipoles . ( the operators that appear here are denoted as differential operators , but they could also be finite differences as in eq . ( [ eq : bfkl ] ) . analytical calculations are usually easier with differential operators ) . in the case of qcd , @xmath108 would be the bfkl kernel ( that can easily be deduced from eq . ( [ eq : dpdy ] ) ) , @xmath109 , @xmath110 ( where @xmath111 is the size of the dipoles ) , and @xmath112 would be the scattering amplitude . further , we define @xmath51 to be the eigenvalue of @xmath107 that satisfies @xmath113 . following ref . @xcite , we write the ansatz u(t , k)=e^-_0(k - x(t ) ) l ( , ) . [ eq : ansatz ] @xmath114 is the size of the front , which is essentially @xmath78 for large @xmath37 . when @xmath107 is expanded to second order around the eigenvalue @xmath51 , then @xmath115 obeys the partial differential equation _ y=_^2 + ( ^(_0)-x^(t ) ) , [ eq : diffusion0 ] where we have defined y= = . [ eq : changevar ] we have only kept the dominant terms for large @xmath114 . we see that @xmath116 has to scale like @xmath117 for all terms of this equation to be relevant . the coefficient must be chosen in such a way that in the large-@xmath118 limit , there is a nontrivial stationary solution . we shall use the already known result ( [ eq : vbd ] ) to write x^(t)=^(_0)-+o(1/l^2 ) [ eq : xprime ] and check a posteriori that it is the correct expression . equation ( [ eq : diffusion0 ] ) then becomes _ y=_^2 + , [ eq : diffusion ] up to higher - order terms when @xmath114 is large . we shall admit that the saturation of the number of dipoles and the stochasticity may be appropriately implemented , within well - controlled approximations , by respectively an absorptive boundary at @xmath119 and another one at @xmath120 ( which corresponds to a distance @xmath114 between the boundaries in @xmath12-coordinates , i.e. to the natural width of the stationary front which travels at the velocity @xmath121 ) . the boundary conditions formally read ( y,=0)=0 ( y,=1)=0 . [ eq : bc ] let us now discuss the initial condition . the fluctuations that will generate the front - shifts @xmath87 responsible for the corrections to @xmath121 are dipoles sent at a distance @xmath79 ahead of the tip of the steady front . formally , in order to describe such a fluctuation occurring at time @xmath122 , we would write @xmath123 , i.e. from eq . ( [ eq : ansatz ] ) , @xmath124 . but since there is an absorptive boundary at @xmath120 , we had better move the fluctuation to the left of the right boundary . this should not qualitatively change the problem as long as @xmath125 . thus we write ( y=0,)=(-1+|a ) , [ eq : ic ] where @xmath126 is a constant of order @xmath127 , and therefore @xmath128 . the solution of eq . ( [ eq : diffusion ] ) with the conditions ( [ eq : bc ] ) and ( [ eq : ic ] ) reads _ ( y,)= _ n=1^(-1)^n+1 n |a n e^- . [ eq : psi ] the subscript @xmath79 recalls that this solution was obtained starting with an initial condition consisting in one dipole at a distance @xmath79 ahead of the deterministic front . we first discuss the stationary solution . we see that for large @xmath118 , the higher harmonics are suppressed exponentially with respect to the fundamental mode @xmath129 , which gives the following contribution : _ _ 0(y,)= |a . [ eq : psi0 ] thanks to the choice ( [ eq : xprime ] ) for @xmath130 , this solution has no @xmath118 dependence , and leads to a stationary @xmath112 in the frame of the front . the expression ( [ eq : psi0 ] ) is independent of the initial condition except for the overall normalization . the value of @xmath131 , which characterizes the initial condition , will be adjusted later . undoing the changes of variables which trade @xmath112 for @xmath115 , @xmath12 for @xmath132 and @xmath39 for @xmath118 ( eq . ( [ eq : ansatz ] ) ) , the stationary solution @xmath133 reads u__0(t , k)= e^-_0(k - x(t ) ) . [ eq : u0 ] we require that @xmath134 for @xmath135 , where @xmath136 is a constant of order 1 . this condition is satisfied if we set @xmath137 . indeed , with this choice , u__0(t , x(t)+al)2 ^ 2|a a l^2 e^-_0 a l. [ eq : pointnorm ] since @xmath138 and @xmath139 are constants , the right - hand side is just a number of order 1 . we now add a fluctuation to the stationary front , that is to say , an extra particle at a position @xmath79 ahead of the tip of the steady front . the solution of the diffusion equation is the superposition of the large - time stationary solution @xmath133 given by eq . ( [ eq : u0 ] ) with @xmath140 , and of the solution @xmath141 of the diffusion equation with the generic initial condition characterized by @xmath79 ( see eq . ( [ eq : psi ] ) ) , up to a multiplicative constant @xmath81 of order 1 that we do not control in this calculation , since it certainly depends on the detailed shape of the fluctuations . we write u(t , k)&=u__0(t , k)+c_2 u_(t , k ) + & = e^-_0(k - x(t ) ) l [ eq : u ] up to the replacement of the variables by their expressions ( [ eq : changevar ] ) . the presence of the second term alters the shape of the front ( the front eventually relaxes back to the sine shape in eq . ( [ eq : u0 ] ) ) , see fig . [ fig : psi ] . but of course , we want to keep the normalization condition for @xmath112 , namely for some appropriate value of @xmath12 , @xmath112 is required to equate eq . ( [ eq : pointnorm ] ) at all @xmath39 . this is possible by shifting the value of @xmath12 at which we enforce the normalization condition from @xmath135 to say @xmath142 . this is equivalent to shifting the position of the front @xmath143 . equation ( [ eq : u ] ) then leads to u(t , x(t)+al+r(r , ) ) = 2 ^ 2 |a a l^2 e^-_0 a l-_0 r(t , ) . [ eq : ufin ] equating the right - hand sides of eq . ( [ eq : ufin ] ) and eq . ( [ eq : pointnorm ] ) , we get r(t,)= , [ eq : r ] where only the lowest orders in @xmath126 , @xmath144 in the expansion of @xmath145 must be kept . with the help of eq . ( [ eq : psi ] ) , it is then straightforward to arrive at an explicit expression of @xmath87 . note that by performing the shift @xmath146 , we have actually added a @xmath39-dependent term to @xmath147 in eq . ( [ eq : xprime ] ) , which was initially assumed to be a constant . however , since the time scale for the variations of @xmath87 is @xmath148 , @xmath149 is suppressed for large @xmath114 , and thus eq . ( [ eq : diffusion ] ) is not modified within our approximations . it is interesting to note that @xmath145 is related to some jacobi @xmath150 function @xcite . since _ 4(z|q)=1 + 2_n=1^(-1)^n(2nz)q^n^2 , we may rewrite eq . ( [ eq : psi ] ) as _ ( y,)= . the notation qe^- = e^- [ eq : q ] has been introduced . using eq . ( [ eq : r ] ) and performing the appropriate expansion for small @xmath126 and @xmath144 , we arrive at an expression for @xmath151 in terms of the @xmath152-function which is particularly compact : [ eq : rtheta ] with -_q_4(0|q)=2_n=1^+ ( -1)^n+1n^2 q^n^2 - 1 . [ eq : theta4def1 ] it is actually quite natural that the jacobi theta functions appear , since the latter are defined as solutions of the one - dimensional heat equation with periodic boundary conditions . we turn to the analysis of the obtained result . first , for large @xmath118 , only the fundamental mode contributes to @xmath145 , and it is clear that eq . ( [ eq : r ] ) reduces to eq . ( [ eq : r0 ] ) . looking back at eq . ( [ eq : psi ] ) , we see that higher harmonics would give a series of exponentially decreasing corrections . but at a finite time , a large number of modes have to be taken into account , typically all modes such that @xmath153 . a few low - lying modes are not enough to describe the small - time behavior . instead , it is a saddle point ( in an appropriate integral reformulation ) that dominates the sum ( [ eq : psi ] ) . in this regime , it would be useful to find a way to write the series of harmonics such that at asymptotically large @xmath118 , only the first term contributes instead of the whole series . this is actually possible using the poisson summation formula _ n=-^+f(n)=_k=-^+ dxf(x)e^-2ik x. [ eq : poisson ] in order to get @xmath151 , we need the value of @xmath145 at @xmath154 . hence we choose f(x)=- x |a x a q^x^2 - 1 e^ix . we then perform the integral over @xmath155 in the r.h.s . of eq . ( [ eq : poisson ] ) . introducing @xmath156 and @xmath157 , we get the following expression for @xmath145 : @xmath158 since we eventually want to apply eq . ( [ eq : r ] ) in order to get an expression of the shift of the front , we expand the latter formula for @xmath159 . the leading order reads _ ( y , a)= _ k=1^+ e^. the shift of the front due to a fluctuation is obtained from @xmath145 with the help of eq . ( [ eq : r ] ) : @xmath160 e^{-\frac{(2k-1)^2l^2}{2\chi^{\prime\prime}(\gamma_0)t } } \bigg\}. \label{eq : rcomplet}\end{gathered}\ ] ] @xmath161 is the function of @xmath39 given by eq . ( [ eq : q ] ) . this formula is extremely useful , since the series indexed by @xmath12 converges fast . even for moderately large values of @xmath39 , a few terms accurately describe the whole function . this is actually the best formula for numerical evaluations of @xmath87 . we shall now examine the limit of small @xmath39 ( @xmath162 ) . then only the term @xmath163 has to be kept . the expression for @xmath87 boils down to r(t,)= ( 1+c_2 ) . as a final remark , let us note that the poisson summation ( [ eq : poisson ] ) that we have used to rewrite the series of harmonics corresponds to a jacobi identity for the @xmath150 functions @xcite . equation ( [ eq : rcomplet ] ) results from eq . ( [ eq : rtheta ] ) with the replacement -_q_4(0|q)= _ k=1^+(^2(2k-1)^2 + 2q ) e^. [ eq : theta4def2 ] with the elements presented in the previous sections , we can write the expression for @xmath164 . it is enough to insert the expression for the probability of fluctuations ( eq . ( [ eq : p ] ) ) and for the time - dependent shift ( eq . ( [ eq : rtheta ] ) ) into eq . ( [ eq : corr ] ) : = _ 0^+de^-_0 ^2 , [ eq : calsigma2 ] where for @xmath165 we use either one of the equivalent expressions ( [ eq : theta4def1 ] ) , ( [ eq : theta4def2 ] ) according to the limit that we want to investigate . we now have to fix the value of @xmath114 . in ref . @xcite , @xmath114 was taken to be a constant . ( the phenomenological model predicted @xmath166 , but empirically , we saw that it was better to add a subdominant correction , namely @xmath167 . ) in this case , a change of variable can be made in the integrand . all the parameters may be factored out , leaving us with a simple numerical integral to perform : _ 0^+^2(1+x)=2(2)=. thus = . replacing the product of the unknown constants by eq . ( [ eq : c1c2 ] ) and @xmath161 by eq . ( [ eq : q ] ) and integrating over the time variable between 0 and @xmath168 , we arrive at a parameter - free expression for @xmath169 as a function of @xmath170 , namely [ eq : sigma122f ] we now investigate the two interesting limits , i.e. @xmath171 and @xmath172 . for large @xmath170 , the integral is dominated by the region @xmath173 , thus @xmath174 may be replaced by its value at @xmath175 ( @xmath176 ) . performing the remaining integration , we get _ 12 ^ 2 t , which is twice the second - order cumulant of the position of the fip front , see eq . ( [ eq : d ] ) . for small @xmath170 instead , say @xmath177 , we use the expansion of @xmath165 for @xmath178 , i.e. the first term in eq . ( [ eq : theta4def2 ] ) , which reads _ q_4(0|q)= - e^. equation ( [ eq : sigma122f ] ) boils down to the following expression : _ 12 ^ 2 ( - ) . [ eq : smalltasymptotics ] so far , we have chosen the size of the front @xmath114 constant , of the order of @xmath85 . another possible model for @xmath114 would be to promote it to a function of @xmath79 at the level of eq . ( [ eq : calsigma2 ] ) , namely l = l_0++ , where the constant has to be determined empirically . this choice takes maybe into account more accurately the extension of the front by @xmath79 that generates the fluctuations . the @xmath79-integral can not be performed analytically in eq . ( [ eq : calsigma2 ] ) except in some limits , so a priori , there is no simpler expression than eq . ( [ eq : calsigma2 ] ) . thus we need to know the values of @xmath80 and @xmath81 individually . we can consider that @xmath179 is the natural normalization of the probability distribution @xmath180 . then , we must set @xmath181 in order to satisfy eq . ( [ eq : c1c2 ] ) . the above - mentioned two models , in which @xmath114 is either constant or @xmath79-dependent , differ by subleading terms in the large-@xmath114 limit . since the values of @xmath79 which dominate the @xmath79-integral in eq . ( [ eq : calsigma2 ] ) are of order @xmath182 , like the first correction to @xmath85 in the case of constant @xmath114 , the models are not expected to differ significantly . we will check this statement numerically . looking back at eq . ( [ eq : sigma122f ] ) , we see that @xmath169 has a nice scaling property . indeed , we may rewrite the latter equation as _ 12 ^ 2= _ e^-_0(v_0-v)t^1 in terms of the properties of a single front ( its velocity @xmath73 and the diffusion constant @xmath183 whose analytical expressions were given in eq . ( [ eq : vbdmm ] ) and ( [ eq : d ] ) ) , where @xmath184 can be read in eq . ( [ eq : v0 ] ) . in particular , we have the following scaling : [ eq : scaling ] from eq . ( [ eq : smalltasymptotics ] ) , we see that the function in the right - hand side is exponentially damped when its argument is smaller than 1 , i.e. parametrically for @xmath172 . once one knows the characteristics of the traveling waves in the fip model ( i.e. @xmath73 and @xmath183 ) , this scaling of the correlations is a pure prediction . thus it will be interesting to check it in the numerical calculations . let us try and evaluate the limits on the validity of our calculations . the latter were essentially based on the assumption that the eigenvalue @xmath185 of the kernel @xmath107 dominates . while this statement is clearly true at large times , when the traveling - wave front is well formed , ( see e.g. ref . @xcite ) , it must break down at early times right after a fluctuation has occurred : indeed , a fluctuation has an initial shape that is far from the one of the asymptotic front , see fig . [ fig : psi ] . we wish to estimate the order of magnitude of the dispersion of the relevant eigenvalues about @xmath51 . to this aim , neglecting for the moment the boundary conditions and the prefactors , we write the solution of eq . ( [ eq : equ ] ) as u(t , k)~d e^-k+()t . the interesting values of @xmath12 are the ones around the position of the wave front , therefore we write @xmath186 , where @xmath187 is of the order of the size @xmath114 of the front . expanding @xmath49 about @xmath51 , we write u(t , v_0t+k)~ e^-_0k d ( ) e^-k + 12 ^(_0 ) ( ) ^2t+ , [ eq : uapprox ] where @xmath188 . it is clear from this equation that the relevant values of @xmath189 are of the order of @xmath190 . since the order of magnitude of @xmath187 is the size @xmath114 of the front , we would a priori conclude that the dispersion of @xmath191 around @xmath51 is small and hence that the calculation is valid as soon as @xmath192 . however , we have also expanded @xmath49 to second order . this means that for a generic kernel @xmath107 , we have neglected terms of the form @xmath193 ( which would fit in the dots in eq . ( [ eq : uapprox ] ) ) . the expansion is a good approximation if the latter term is small , i.e. if tl^3/2 . so far , we have been working with the minimal model , consisting in two realizations of the fip model which evolve in the same way until their common tip reaches @xmath68 , and which decorrelate for @xmath67 . the only relevant parameter which determined the decorrelation of the positions of the fronts of the realizations was the time @xmath168 after the tip had reached @xmath68 . we now wish to discuss the transcription of the obtained results to impact - parameter space , which was our initial problem . to this aim , we will of course make use of eq . ( [ eq : ttb ] ) to express @xmath89 with the help of the mean front velocity @xmath73 . but we also need a length scale to which the distance in impact - parameter space @xmath71 may be compared . the natural length is the dipole size at the position of the front , namely l_s(t)=2 ^ -x(t)=l_s(t_b)2 ^ -vt . [ eq : ls ] on the other hand , according to eq . ( [ eq : ttb ] ) and disregarding the integer part operator , @xmath194 and the tip of the front @xmath68 is ahead of the bulk @xmath195 by @xmath114 : @xmath196 . using the previous equation , we may now express @xmath170 as a function of @xmath71 and of the length scale @xmath197 : t=. the scaling ( [ eq : scaling ] ) reads _ 12 ^ 2~ . this formula , together with the behavior of the scaling function ( see eq . ( [ eq : smalltasymptotics ] ) ) , shows that there is little @xmath14-dependence until @xmath198 , that is to say , until @xmath199 . in other terms , the size @xmath71 of the domain around impact parameter @xmath14 in which the fluctuations in the position of the fronts are negligible is , in notations more familiar to qcd experts , where @xmath200 is the usual saturation momentum at impact parameter @xmath14 . note that since the fronts are statistically independent as soon as @xmath201 , this result may seem a bit surprising : it says that the effective correlation length between different points in impact - parameter space is much larger than @xmath202 in the parametrical limit of small @xmath6 . this is the main qualitative result of this paper . in this section , we confront our analytical calculations to numerical simulations of the toy model . first , we consider the full model and test the validity of the assumption that the minimal model is a good approximation to the full model also for @xmath203 , i.e. when splittings to larger - size dipoles are authorized . second , we compare the minimal model to the analytical results for the fluctuations between different positions in impact - parameter space ( given essentially by eqs . ( [ eq : calsigma2]),([eq : sigma122f ] ) ) . the model defined by eqs . ( [ eq : defmodel1 ] ) and ( [ eq : defmodel2 ] ) is straightforward to implement numerically in the form of a monte - carlo event generator . the simplest is to store the number of dipoles in each bin in an array whose index @xmath204 is related to @xmath12 and @xmath35 through @xmath205 . the splitting dynamics relates bin @xmath204 to @xmath206 ( down left ) , @xmath207 ( down right ) and @xmath208 $ ] ( up ; the square brackets stand once again for the integer part ) . we have to deal with an array whose size grows exponentially with time . it is thus very difficult to pick large values of @xmath39 , and thus also large values of @xmath37 . indeed , the relevant time scale grows with @xmath37 like @xmath209 , and consequently the minimum number of entries in the array one wants to consider grows like @xmath210 . in practice , we limit ourselves to @xmath211 and @xmath212 . as for the time step @xmath27 , the most convenient is to take it small but finite . we set @xmath213 . we start with one particle and evolve it for a few hundred units of time using the fip version of the model . we obtain a traveling wave front , whose tip we eventually label @xmath163 . ( the complete front sits in the bins @xmath214 ) . from the initial condition built in this way , we evolve all bins for which @xmath215 using the fip model , and all bins for @xmath11 using the full model . one event in shown in fig . [ fig : sketch1 ] . although @xmath37 and @xmath39 are small in this calculation , we see that the regions in impact - parameter space which have similar numbers of dipoles are larger than the local length scale @xmath197 ( see eq . ( [ eq : ls ] ) ) . after the evolution times @xmath216 and @xmath217 respectively , we measure the position of the front at various impact parameters on a uniform tight grid ranging from @xmath17 to @xmath218 . we use the following definition of the position of the front : x(b , t)=k_0 + _ k = k_0 + 1^+ where @xmath219 is the largest @xmath12 for which @xmath220)}=n$ ] . note that in principle , we could have chosen @xmath221 . in practice however , because of the discreteness of @xmath12 in our model , this choice would introduce artefacts which we do not expect in real qcd . we compute the squared difference of the front positions between the impact parameters @xmath17 and @xmath25 , and average over events . we plot the result as a function of @xmath222 , where @xmath73 is the average front velocity measured at impact parameter @xmath17 . we compare the results to the correlations obtained in the minimal model , i.e. when we consider two independent realizations of an initial front . we do not attempt to compare to our analytical formulas since the values of @xmath37 that we are able to reach are too small for the approximations that we had to assume to be relevant . the corresponding plot is displayed in fig . [ fig : plot_100_0 ] for @xmath223 , @xmath224 , @xmath59 , and in fig . [ fig : plot_100_2 ] with the same parameters except @xmath225 . first , we see that in the full model , the graph of @xmath169 exhibits steps , i.e. @xmath169 is constant by parts . this is related to the hierarchical structure of the model : the correlations between @xmath19 and any of the points at @xmath226 are identical ; the same is true for @xmath227 , @xmath228 etc ... the logarithmic @xmath14-scale on the @xmath39-axis makes the widths of the steps all equal . next , we see that for small @xmath229 ( i.e. impact parameters close to @xmath17 ) there are very little fluctuations in the front positions . finally , we see that for @xmath59 , as anticipated , the full model and the minimal ones coincide almost perfectly ( fig . [ fig : plot_100_0 ] ) . for @xmath225 , i.e. when splittings towards larger dipole sizes are switched on and therefore new correlations appear beyond the ones taken into account in the minimal model , there are some quantitative differences for large @xmath39 ( fig . [ fig : plot_100_2 ] ) . but we see that using the minimal model instead of the full model that keeps all impact parameters is a good approximation . this corroborates the conclusions of the work in ref . @xcite . we now set @xmath59 , in which case , as discussed earlier and as checked numerically , the model exactly reduces to a collection of one - dimensional fkpp - like models . hence , in order to compute two - point correlation functions , it is enough to evolve two realizations of the corresponding fip model with the constraint that all bins with @xmath88 be identical between the two realizations , and the bins @xmath67 be completely independent . alternatively , we could also generate one single realization and evolve it for @xmath89 time steps , replicate it at time @xmath89 , and then evolve the two replicas completely independently of each other . the difference between these two possible implementations of the minimal model can not be accounted for in our analytical calculations , thus the differences that we shall find numerically will give an indication of the model uncertainty . this time , our aim is essentially to check our analytical formulas , thus we will pick very large values of @xmath37 , even if they appear to be unphysical in the qcd context since they would correspond to exponentially small values of the strong coupling constant @xmath6 . the parameters of the model are obtained from eq . ( [ eq : eigenvalue ] ) with @xmath224 , @xmath59 and @xmath213 : _ 0=1.0136 , v_0=2.6817 , ^(_0)=2.6098these values are close to @xmath230 , @xmath231 and @xmath231 respectively , which would be the correct parameters if @xmath27 were infinitesimal , in which case @xmath232 ( see eq . ( [ eq : eigenvalue ] ) ) . the numerical results are shown in fig . [ fig : plot10 ] for @xmath233 with the two versions of the model ( we generated about @xmath234 realizations ) , and compared with the analytical predictions . we test the two possible choices for the size @xmath114 of the front : either @xmath114 is a constant , which from our previous experience with fkpp traveling waves @xcite , we set to l = n+n- or it is @xmath79-dependent , namely l=+-1.4 . the numerical constants , which are not determined in our theory , were chosen empirically so that they properly describe all numerical data for @xmath235 . in the first case , eq . ( [ eq : sigma122f ] ) is used . in the second case , eq . ( [ eq : calsigma2 ] ) is integrated numerically over @xmath39 and @xmath79 . we see that the agreement between the numerical calculation and the analytical predictions is good , except maybe for very small values of @xmath170 where the calculations are not expected to be accurate . indeed , for the same values of @xmath170 , we also see in fig . [ fig : plot10 ] a sizable discreapancy between the two versions of the minimal model . the calculations for @xmath236 are shown in fig . [ fig : plot50 ] . the numerical results and the theoretical expectations ( eq . ( [ eq : sigma122f ] ) ) coincide almost perfectly . finally , we check that the scaling in eq . ( [ eq : scaling ] ) is well reproduced by the numerical data . the monte - carlo simulations are shown in fig . [ fig : scaling ] , plotted in the appropriate scaling variables . the diffusion constant of a single wave front @xmath183 as well as the velocity @xmath73 are measured from the same data . we see that all curves nicely superimpose for @xmath235 ( we show data for values of @xmath37 as large as @xmath237 ) , while there are clear deviations for smaller @xmath37 ( see the curve for @xmath223 ) , as expected . in this paper , we have built a model that possesses the main features of the qcd dipole model including the dynamics in impact - parameter space , and which is furthermore very easy to implement numerically . we have obtained analytical expressions for the fluctuations of the ( logarithm of the ) saturation scale from one position in impact - parameter space to another one nearby , which gives an indication on the homogeneity of the gluon number density . since our analytical calculations are only based on some rather general properties of the model , they should go over to full qcd after appropriate replacement of the parameters , hopefully giving the correct small-@xmath6 ( large @xmath37 ) asymptotics . we have found that the saturation scale varies quite slowly . in the usual notations of qcd , if at position @xmath14 in impact - parameter space the local saturation scale is @xmath200 , then the saturation scale is uniform over a region of size @xmath238 around that position . our calculations suffer the usual limitations in this kind of models : analytical expressions are derived for exponentially large @xmath37 , that is to say , exponentially small values of the strong coupling constant @xmath6 . although they often may be successfully extrapolated down to @xmath239 ( @xmath240 ) , this is at the cost of tuning constants , and so far , we have not found a systematic procedure to compute finite-@xmath37 corrections . the next step would probably be a numerical calculation in the case of full qcd , using , as proposed earlier @xcite , a combination of a monte carlo implementation of the color - dipole model in the low - density regime ( such an implementation is already available ; see ref . @xcite and ref . @xcite for more recent work ) and of a numerical solution of the bk equation at the transition to saturation @xcite . this looks very challenging , but maybe the nice smoothness of the saturation scale in impact - parameter space that we have found in this paper will help . finally , it would be interesting to find an observable which would directly be sensitive to the gluon density at two points in impact - parameter space simultaneously . this work was supported in part by the department of energy ( usa ) , and in part by the agence nationale pour la recherche ( france ) , contract anr-06-jcjc-0084 - 02 . l. v. gribov , e. m. levin and m. g. ryskin , phys . rept . * 100 * , 1 ( 1983 ) . a. h. mueller and j. w. qiu , nucl . b * 268 * , 427 ( 1986 ) . l. n. lipatov , sov . . phys . * 23 * , 338 ( 1976 ) ; e. a. kuraev , l. n. lipatov , and v. s. fadin , sov . jetp * 45 * , 199 ( 1977 ) ; i. i. balitsky and l. n. lipatov , sov . j. nucl * 28 * , 822 ( 1978 ) . i. balitsky , nucl . b * 463 * , 99 ( 1996 ) . y. v. kovchegov , phys . d * 60 * , 034008 ( 1999 ) ; phys . rev . d * 61 * , 074018 ( 2000 ) . j. jalilian - marian , a. kovner , a. leonidov and h. weigert , nucl . b * 504 * , 415 ( 1997 ) ; phys . d * 59 * , 014014 ( 1999 ) ; e. iancu , a. leonidov and l. d. mclerran , phys . lett . b * 510 * , 133 ( 2001 ) ; nucl . a * 692 * , 583 ( 2001 ) ; h. weigert , nucl . phys . a * 703 * ( 2002 ) 823 . t. altinoluk , a. kovner , m. lublinsky and j. peressutti , jhep * 0903 * , 109 ( 2009 ) . for reviews on the fkpp equation and its stochastic extensions , see w. van saarloos , phys . rep . * 386 * , 29 ( 2003 ) ; d. panja , phys . rep . * 393 * , 87 ( 2004 ) . a. h. mueller and a. i. shoshi , nucl . b * 692 * , 175 ( 2004 ) . e. iancu , a. h. mueller and s. munier , phys . b * 606 * , 342 ( 2005 ) . e. iancu and d. n. triantafyllopoulos , nucl . a * 756 * , 419 ( 2005 ) , phys . b * 610 * ( 2005 ) 253 . s. munier , phys . rept . * 473 * , 1 ( 2009 ) . k. j. golec - biernat and a. m. stasto , nucl . b * 668 * , 345 ( 2003 ) . s. munier , g. p. salam and g. soyez , phys . d * 78 * , 054009 ( 2008 ) . g. camici and m. ciafaloni , phys . b * 395 * , 118 ( 1997 ) . k. j. golec - biernat , l. motyka and a. m. stato , phys . rev . d * 65 * , 074037 ( 2002 ) . a. h. mueller and d. n. triantafyllopoulos , nucl . b * 640 * , 331 ( 2002 ) . e. brunet , b. derrida , phys . e * 56 * , 2597 ( 1997 ) ; comp . 121 - 122 ( 1999 ) 376 ; j. stat . 103 ( 2001 ) 269 . e. brunet , b. derrida , a. h. mueller and s. munier , phys . e * 73 * , 056126 ( 2006 ) . m. abramowitz and i. a. stegun , _ handbook of mathematical functions _ , ( 1964 ) dover publications , new york . isbn 0 - 486 - 61272 - 4 . g. p. salam , nucl . b * 461 * , 512 ( 1996 ) . g. p. salam , comput . commun . * 105 * , 62 ( 1997 ) . a. h. mueller and g. p. salam , nucl . b * 475 * , 293 ( 1996 ) . e. avsar , g. gustafson and l. lonnblad , jhep * 0507 * ( 2005 ) 062 ; jhep * 0701 * ( 2007 ) 012 .

very massive stars were long ago predicted to get hot enough that the ambient photons in the interior are sufficiently energetic to create electon / positron pairs ( rakavy & shaviv 1967 ; barkat , rakavy & sack 1967 ; rakavy & shaviv 1968 ; rakavy , shaviv & zinamon 1968 ; fraley 1968 ; see also wheeler 1977 ; el eid & hilf 1977 ; el eid , fricke & ober 1983 ; carr , bond & arnett 1984 ; stringfellow & woosley 1988 ) . the conversion of energy to rest mass rather than thermal energy alters the equation of state so that the pressure does not increase sufficiently with density upon compression to maintain hydrostatic equilibrium . if a sufficiently large , mass - averaged region of the star is in the pair - formation regime , such that @xmath1 @xmath2 4/3 , the structure will be dynamically unstable . models predict that this happens after stars have undergone their central helium burning and have formed massive cores composed primarily of oxygen ( with a small mass fraction of carbon ) . in models , the instability occurs in an off - center shell . the rapid contraction of this shell drives it inward , leading to the rapid compression and heating of the inner core of oxygen . unlike the case of iron - core collapse , the oxygen in these stars is subject to strong energy release by rapid thermonuclear burning . the result is the prediction that the star is totally disrupted , leaving no remnant , but with the production of a very large mass of radioactive @xmath3ni , the decay of which could power the light output . the original calculations , cited above , found that the pair - instability regime was encountered for stars with massive oxygen cores , greater than about 60 @xmath0 . at the time , it was not clear how massive a star needed to be to develop a sufficiently massive oxygen core , but estimates were in the range of 100 @xmath0 . it was also not clear that any stars sufficiently massive to reach this condition of instability existed . the latter point was resolved theoretically with the understanding that massive , radiation - pressure dominated stars subject to dynamical instability would be stabilized in the non - linear regime ( appenzeller 1970 ; ziebarth 1970 ) and observationally with the discovery that young clusters such as r 136 in 30 doradus contained very massive stars ( panagia et al . 1983 ; hunter et al . 1995 ; crowther et al . a subsequent important development was the prediction that in the context of @xmath4cdm models of the universe , the first stars forming at zero metallicity after the dark ages might preferentially form especially massive stars that would , in turn , be subject to pair instability ( abel et al . 1998 ; abel et al . 2000 ; bromm et al . 2002 ; bromm & larson 2004 ) . considerable effort has gone into the computation of the formation of the first stars , the evolution of stars that will reach pair - instability conditions , and the predicted observational properties of the resulting explosions ( heger & woosley 2002 , ohkubo et al . 2003 ; scannapieco et al . 2005 ; blinnikov & heger 2007 ; waldman 2008 ; ohkubo et al . 2009 ; heger & woosley 2010 ; kasen et al . 2010 ; whalen & fryer 2010 ; kasen et al . 2011 ; joggerst & whalen 2011 ) . pair - instability supernovae ( pisn ) models have a characteristic nucleosynthetic yield ( heger & woosley 2002 ; ohkubo et al . 2003 ) , but searches for evidence of such a distribution in the lowest metallicity stars has not revealed the expected pattern ( christlieb et al . 2002 ; frebel et al . 2005 ) . in general , lower metallicity will suppress mass loss and allow relatively lower mass main sequence stars to encounter the pair - instability regime . for a given main sequence mass , higher metallicity will tend to lead to lower mass oxygen cores , thus avoiding this regime . two subsequent developments have altered the context in which pair instability is considered . one is the recent understanding that conditions of the first stars may be more susceptible to fragmentation so that the first stars may have been of smaller mass than once thought ( stacy et al . 2010 ; greif et al . another important development was the discovery in the contemporary universe of a category of super - luminous supernovae ( slsn ) that are relatively rare , but brighter by a factor of 10 to 100 than most contemporary supernovae ( smith et al . 2007 ; quimby et al . 2007 ; miller et al . 2009 ; gal - yam et al . 2010 ; chatzopoulos , wheeler & vinko 2011 ) . some of these slsn show evidence of high mass , but can not be pair - instability supernovae ( smith et al . 2007 ; miller et al . 2009 ; chatzopoulos , wheeler & vinko 2011 ; chatzopoulos , wheeler & vinko 2012 ; leloudas et al . 2012 ) . if their brightness derived from the radioactive decay of @xmath3ni , as demanded by the pair - instability model , they would require a greater mass of @xmath3ni than is allowed for the total mass of the ejecta . the ejecta mass is constrained by the width of the light curve that is a measure of the diffusion time . the great luminosity of super - luminous supernovae like sn 2006gy probably derives from the collision of the ejecta with a shell of matter previously ejected by the progenitor star ( smith et al . 2007 ; chevalier & irwin 2011 ; chatzopoulos , wheeler & vinko , 2011 ) . several of these super - luminous events do , however , have all the characteristics expected of a pisn ( sn 2007bi , gal - yam et al . 2010 ; ptf 10nmn , yaron et al . in preparation ) . these events have occured in low , but not zero , metallicity environments . the possibility that the first stars are not of especially high mass and the discovery of contemporary super - luminous supernovae that have the characteristics of pisn brings a new focus to the issue of just which stars can undergo pair instability and under what conditions . maeder ( 1987 ; see also maeder & meynet 2011 for a review ) discussed the effects of rotationally - induced mixing on the evolution of massive stars . he concluded that there would be substantial mixing produced by the small - scale three - dimensional tail of the turbulent spectrum of the baroclinic instability and that the diffusion coefficient could be sufficiently large to mix most massive stars during their main sequence lifetimes . in particular , above a critical rotation , maeder predicted that the evolutionary tracks go upwards and bluewards , very close to those of fully homogeneous evolution . the immediate implication was that stars in the mass range predicted to reach pair instability ( already radiation - pressure dominated and hence close to neutral dynamical instability ) were susceptible to quasi - homogeneous evolution such that nearly all the main sequence mass could be burned to heavier elements . quasi - homogeneous evolution would significantly decrease the main sequence mass that leads to pair instability . the notion that quasi - homogeneous evolution might lead to larger core masses has been applied in the context of the progenitors of grbs ( heger , woosley & spruit 2005 ; yoon & langer 2005 ) and explored to a certain extent in recent work ( langer et al . 2007 , eskstrom et al . 2008 , 2011 ; brott et al . 2011a , 2011b ) , but has not been pursued in detail in the context of pair instability . in this paper , we conduct a thorough parameter study to explore the minimum main sequence mass that encounters the pair instability regime . in his unpublished phd thesis , barkat ( 1967 ) noted that some models that skirted the pair - instability regime ejected shells of matter , but survived the instability and continued to evolve . somewhat higher mass models became sufficiently unstable that a dynamical explosion , a pisn , was produced directly . woosley , blinnikov & heger ( 2007 ) invoked the former effect , that they called a pulsational pair instability supernova ( ppisn ) to address the nature of slsn 2006gy . their simulations showed a repeated ejection of shells . the second , faster shell collided with the first and produced a luminous display reminiscent of the light curve of slsn 2006gy . here we will delineate models that encounter the ppisn and those that encounter the full pisn explosion . section 2 describes our assumptions and models , 3 gives the results and presents a model that invokes a different mixing criterion and another with solar metallicity and mass loss . section 4 discusses our conclusions . as discussed in the introduction , the masses of stars that evolve from the main sequence to either ppisn or pisn will be a function of metallicity , mass loss , and rotationally - induced mixing . the outcome will also depend on the treatment of convective instability , semi - convection , and overshoot . dissipation of shear by magnetic effects may also attend the rotationally - induced mixing . rather than explore this whole parameter space where the physics is , in any case , uncertain , we have focused on the portion of parameter space that is expected to lead to the minimum mass to encounter pair formation . in particular , we have explored conditions of zero metallicity and have neglected mass loss for the majority of our models in order to establish the proof of principle . we have run a few models at solar metallicity in order to put our results in context . we have adopted rates of rotation on the main sequence that run from non - rotating to rotating at 80 % of equatorial keplerian velocity . we have focused on schwarzschild convection for two reasons . one is that this prescription will tend to enhance the mass of the oxygen core for a given main sequence mass , in keeping with the philosophy of this exploratory work . the second reason is based on a suspicion that neither the schwarzschild nor the ledoux criterion really captures the three - dimensional , plume - driven convection in real stars . as we will describe in a future work ( chatzopoulos , dearborn & wheeler , in preparation ) , we have grounds to believe that in the late stages , when the oxygen core forms , the schwarzschild criterion is the more appropriate . multi - dimensional effects tend to swamp the stabilizing effect of composition gradients . our simulations have also adopted the effects of magnetic viscosity as parametrized by heger , woosley & spruit ( 2005 ) based on the prescriptions of spruit ( 1999 , 2002 ) . this is not because we believe that this particular parametrization captures all the relevant multi - dimensional mhd instabilities and related phenomona in rotating , shearing , stars , but because it is a widely used and recognized algorithm so that our results can be readily compared to others using the same prescription . we have used the modules for stellar experiments in astrophysics ( mesa ; paxton et al . 2011 ) code to calculate the evolution of a grid of @xmath5 0 massive stars ranging from zero age main sequence ( zams ) mass of 35 @xmath0 to 200 @xmath0 for four different degrees of zams rotation . we assume initially rigid body rotation on the zams with surface rotation corresponding to 0 , 30% , 50% and 80% of the critical keplerian rotation @xmath6 @xmath7 where @xmath8 @xmath9 is the gravitational acceleration at the surface " of the star , @xmath10 the gravitational constant , @xmath11 the mass , @xmath12 the radius of the star and @xmath13 the eddington factor where @xmath14 and @xmath15 is the total radiated luminosity and the eddington luminosity respectively . the 35 @xmath0 model was only run for maximum rotation ( 80% ) in order to establish the lower mass limit for ppisn and the models above 85 @xmath0 for lower ( 30% ) and zero degrees of rotation to establish the corresponding minimum zams mass for ppisn . the models with zams masses 40 - 85 @xmath0 were run in bins of 5 @xmath0 for all the selected degrees of rotation in order to better resolve the limits for different final fates of the stars . in order to benchmark against the results of heger & woosley ( 2002 ) and woosley , blinnikov & heger ( 2007 ) for non - rotating stars that will produce a pure pisn and a ppisn , respectively , we have run a 110 @xmath0 and a 200 @xmath0 model ( zams masses ) without rotation . the reason that we chose this range of zams masses is the fact that heger & woosley ( 2002 ) and heger et al . ( 2003 ) predict that oxygen core masses ( @xmath16 ) ranging from 40 - 64 @xmath0 will undergo ppisn while 64 @xmath17 133 @xmath0 will explode as direct pisne for non - rotating progenitors , but with significantly rapid rotation a lower zams mass can produce oxygen core masses in this range . mesa was run with the schwarzschild criterion for convection implemented for reasons discussed above and for zero mass loss . it should be mentioned that rotationally induced mass loss may reduce the final oxygen core masses . ekstrom et al . ( 2008 ) found that rapidly rotating zero metallicity stars with zams masses above @xmath18 50 @xmath0 may lose up to @xmath18 11 @xmath0 ; however due to the lack of understanding of mass loss mechanisms in massive stars this remains uncertain . nevertheless , we also conducted calculations with mass loss included in models with initially zero metallicity , as described in 3.2 , to estimate the impact of mass loss on the final fate of the models . mesa employs a combination of prescriptions for the equation of state ( eos ) , but for high density and temperature plasma the helm eos ( timmes & swesty 2000 ) is used . the helm eos accounts for pressure induced by radiation , ions , electrons , positrons and corrections for coulomb effects and therefore for the effects of electron - positron pair formation , which drives the adiabatic index @xmath19 below 4/3 . for the treatment of nuclear processes with mesa we employ the approx21 " network ( timmes 1999 ) , which covers all major stellar nuclear reaction rates . the effects of angular momentum transport via rotation and magnetic fields are treated based on the one - dimensional approximations of spruit ( 1999 , 2002 ) and heger , woosley & spruit ( 2005 ) . mesa is capable of running stellar models up to the core collapse ( cc ) and pre - supernova ( pre - sn ) stage ; however for high mass stellar models that encounter a degree of instability induced by electron - positron pair production the effects become very dynamic and a challenge for a stellar evolution code to handle . future expansions of mesa will be able to handle those dynamical effects and to follow the supernova ( sn ) explosions of stars with 1-d hydrodynamics implemented ( paxton ; private communication ) . in the current work , stellar models for which a signficant fraction of their core approaches the @xmath20 4/3 regime due to pair formation are stopped before core oxygen ignition and within the carbon burning phase , shortly before becoming dynamic . those models are then mapped into the multi - dimensional hydrodynamics code flash ( fryxell et al . 2000 ) , and their evolution is followed in 1-d . the newest version of flash ( flash4-alpha release ) is used for these simulations . flash is very suitable to follow the dynamical transition of the models from mesa because it uses the same eos ( helm ; timmes & swesty 2000 ) and similar nuclear reaction network . ppisne are characterized by a violent contraction and pulsation that heats the core up to a temperature such that some of the oxygen is burned to produce primarily @xmath21si and @xmath22s ( woosley , blinnikov & heger 2007 ) . pure thermonuclear pisne are heated significantly enough from the dynamical collapse induced by the softening of the eos due to electron - positron pair formation that they burn oxygen explosively and large amounts of @xmath3ni are produced . flash is capable of reproducing those basic features of the events and is therefore used to establish the final fate of the models . we note that in the 1-d flash hydrodynamic simulations the effects of rotation are not considered . in a future work we plan to investigate the effects of rotation in the hydrodynamical stage of pisne with multi - dimensional flash simulations . table 1 summarizes the characteristics of all the models considered in this work ( for zero metallicity and mass loss turned off ) . the first column lists the zams mass ( @xmath23 , in solar masses ) , the second is the critical rotational ratio @xmath24 , the third and fourth the maximum central temperature ( in units of @xmath25 k and density ( in units of @xmath26 g @xmath27 ) that were encountered due to the pair formation dynamical instability , the fifth the mass of the oxygen core , @xmath16 ( in units of solar masses ) , that each model produced , the sixth the surface abundance of @xmath28n and the seventh the final fate of the model as observed in the flash hydrodynamics simulations ( cc for core collapse , ppisn for pulsational pair instability supernova and pisn for pair instability supernova ) . in all figures , the models that ended as cc will be represented by black curves , the models that encountered ppisn by green curves and the models that underwent pisn with red curves . to compare with previous results for ppisn presented by woosley , blinnikov & heger ( 2007 ) in the case of a 110 @xmath0 zams star and for direct pisn presented by heger & woosley ( 2002 ) for a variety of stars with @xmath29 140 @xmath0 we run our own non - rotating 110 @xmath0 and 200 @xmath0 models . the model with 110 @xmath0 formed a 56 @xmath0 oxygen core and then encountered a violent pulsation that heated the center of the star up to @xmath30 k , in a similar manner to that suggested by woosley , blinnikov & heger ( 2007 ) . the 200 @xmath0 model produced a direct pisn that synthesized a massive amount of @xmath3ni ( @xmath18 21 @xmath0 ) and totally disrupted the star , behaving exactly as predicted by woosley , blinnikov & heger ( 2007 ) . additionally , in order to establish a lower mass end for the production of ppisne , we have run a 35 @xmath0 model with rotation at 80% of the critical velocity that converted essentially all of its mass to oxygen . this model was able to evolve up to cc in mesa . for benchmarking it was also mapped to flash at the time of core oxygen ignition where it also kept slowly evolving toward higher densities and temperatures always avoiding the pair formation regime . for all the models , increased levels of rotation led to a more chemically homogeneous evolution and produced higher oxygen core masses . zams rotation at 80% of the critical keplerian velocity was able to convert all stellar mass into oxygen for all the models . the evolution of some models rotating at 80% yielded brief stages when rotation became mildly super - critical ( 110 - 120% ) at the surface which means that results from these models may not be completely accurate . for all other degrees of rotation the star remained at sub - critical velocities throughout all of its evolutionary track . as a representative example of this effect , we show the evolution of @xmath24 for the rotating 70 @xmath0 models during the main - sequence ( ms ) in figure 1 . the solid green curve , the dashed green curve and the solid red curve show the evolution of @xmath24 for zams rotation at 30 , 50 and 80% , respectively . in order to illustrate the effects of rotation on a zams star with a specific mass we pick the 70 @xmath0 models because , as can be seen in table 1 , all possible final fates are encountered for this model ( cc for zero rotation , ppisn for 30% and 50% critical rotation and pisn for 80% critical rotation ) . the evolution of the central density ( @xmath31 ) and temperature ( @xmath32 ) for these models is presented in the left panel of figure 2 . as can be seen , all rotating models encounter a collapse that heats the core up to higher densities and temperatures , but the one rotating at 80% critical is heated significantly enough to burn oxygen explosively and subsequently ejects of all its mass . higher degrees of rotation lead the @xmath31-@xmath32 track of the star closer to the @xmath20 4/3 pair formation region . the right panel of figure 2 shows the tracks of the models in the hertzprung - russell diagram . it can be seen that the more rapidly rotating models remain bluer and are more luminous that the less rapidly rotating ones . figure 3 illustrates the chemical composition of the 70 @xmath0 models for all degrees of rotation at the time before core oxygen ignition ( upper left panel : no rotation ; upper right panel : 30% critical rotation ; lower left panel : 50% critical rotation ; lower right panel : 80% critical rotation ) . the effects of increased homogeneity and higher @xmath16 , with increasing rotation are clearly illustrated . figure 4 presents the distribution of the cno ( solid black curve ) and 3-@xmath33 process specific nuclear energy inputs are shown . higher degrees of rotation also produced a significantly higher @xmath28n surface abundance than their non - rotating counterparts , a trend also noted by ekstrom et al . this trend , however , is not monotonic . for the extreme level of 80% critical rotation the surface @xmath28n mass fraction is generally found to be reduced while the most significant @xmath28n enrichment is observed in models rotating at 50% of the critical value . increased surface @xmath28n is attributed to the onset of the cno cycle in the outer shell due to the strong rotationally induced mixing . for completely homogeneous evolution that makes an oxygen star , which takes place for models that rotate at the extreme level of 80% critical rotation , the cno contribution in outer layers of the star is confined to a very thin shell . figure 5 illustrates our final results for the fate of rotating massive primordial stars . as can be seen , stars initially rotating with speeds 80% the critical velocity with 40 @xmath0 @xmath34 60 @xmath0 will produce ppisne associated with episodic mass - loss and for @xmath35 65@xmath36 they will explode as pisne . as mentioned before , at this rapid rotation the whole mass of those stars will turn into oxygen , therefore this result is consistent with the findings of heger et al . ( 2003 ) , but for initial zams masses that are only 40 - 50% those of the ones given by their non - rotating calculations . consequently , for this fiducial " degree of rotation at 50% the critical value , the zams mass limits become 45 @xmath0 @xmath34 70@xmath36 for ppisn and @xmath35 75 @xmath0 for pisn progenitors , thus at the level of 50 - 60% of those in the case of no rotation . these results suggest that episodic mass - loss events resulting from ppisne can be encountered for less massive stars and may account for some of the observed lbv - type events . we wish to add that a recent paper by yoon , dierks & langer ( 2012 ) independently calculated the mass limits of ppisn and pisn primordial progenitors and found good agreement with our results . convective mixing and overshoot are some of the factors that control the final mass of the oxygen core . as mentioned above , in this work we have adopted the schwarzschild criterion for convection based on the lack of true knowledge about the nature of convection in realistic three - dimensional situations . one dimensional convective mixing will be supressed if composition gradients are considered , as suggested by the ledoux criterion for convection . this will result in smaller final oxygen core masses and therefore tracks in the @xmath31-@xmath32 plane that are shifted further away from the pair formation @xmath20 4/3 region , something that may alter the results for the minimum zams masses that encounter ppisn and pisn . we ran mesa for the 70 @xmath0 non - rotating and 50% critically rotating models with the ledoux criterion implemented in order to examine the sign of this effect . figure 6 presents the results in the case of zero rotation . the left panel shows the @xmath31-@xmath32 tracks for the model that uses the schwarzschild ( solid black curve ) and the model that uses the ledoux ( dashed black curve ) criterion . the right upper and lower panels show the chemical composition of the two models at the time before core oxygen ignition . it can be seen that when mixing is supressed the non - rotating 70 @xmath0 model evolves towards slightly lower @xmath31 and @xmath32 values and it ends up making an oxygen core that is 86% the one produced in the case where composition gradients are ignored . the result of the same experiment but in the case of rotation at the 50% of the critical value is shown in figure 7 . in this case , we see that the fast rotationally - induced mixing counters the effects of suppression due to the inclusion of composition gradients and the final oxygen core masses are almost equal . the characteristics of the models run with the ledoux criterion implemented are also given in table 1 . these results indicate that selecting a different one - dimensional prescription for convective mixing is unlikely to alter our results for the rotating models by any significant factor . although a thorough study of massive models with @xmath37 0 is beyond the scope of this project , which focuses on the primordial progenitors of pisne , we considered the effects that the presence of metals may have . as is well known , metallicities with @xmath370 can induce significant line - driven mass loss that can cause massive stars to lose a significant amount of mass . this mass loss will drive evolution towards the formation of oxygen cores with smaller mass than in the case of zero mass loss . this will have an effect on the final fate of a very massive star , such that it might miss the pair formation region in the @xmath31-@xmath32 plane and end its life as a cc sn explosion . the high mass loss may lead to the formation of massive circumstellar material ( csm ) environments around the progenitors of these sn so that the sn ejecta will violently interact with the csm producing shock energy that can power their light curve ( lc ) , as manifested by luminous type iin sne . to account for the mass loss with mesa , the prescriptions of de jager , nieuwenhuijzen & van der hucht ( 1988 ) were used , as appropriate for hot o - type stars ( for a recent discussion on mass loss rates for wolf - rayet stars , see yoon , woosley & langer 2010 ) . figure 5 also shows the final fate of the same rotating massive zero metallicity stars for which mass loss is considered in the calculations . as mentioned above , rotationally - induced mixing will dredge up metals to the outer and less gravitationally bound regions of the star inducing line - driven mass loss that will ultimately result in smaller oxygen core mass . we find that when mass loss is considered , the minimum zams mass that produce ppisn events is @xmath18 50 @xmath0 and the minimum mass that produce pisn events is @xmath18 80 @xmath0 at maximum zams rotation ( 80% of the critical value ) . the results for the fate of the 70 @xmath0 model rotating at 50% the critical value with solar metallicity and mass loss are presented in figure 8 . the left panel of figure 8 shows a comparison between the @xmath31-@xmath32 tracks of the two models with ( solid black curve ) and without ( solid green curve ) mass loss , and the upper and lower right panels show the chemical composition of the models at the time before core oxygen ignition . the strong line - driven mass loss ejected the hydrogen and a fraction of the helium shell leaving a 33 @xmath0 star with @xmath38 16 @xmath0 that ended its life as a cc sn . the strong continuous mass loss led the @xmath31-@xmath32 track of the model further from the electron - positron pair @xmath20 4/3 region . furthermore , due to strong rotationally - induced mixing the pre - sn model had an enhanced @xmath28n surface abundance as was the case with most rotating @xmath5 0 models . the extreme amount of mass lost by the star that we compute here is consistent with results presented by ekstrom et al . ( 2011 ) who calculated the evolution of a grid of solar metallicity models from 0.8 to 120 @xmath0 rotating at 40% the critical velocity . ekstrom et al . ( 2011 ) found that even the most massive rotating 120 @xmath0 model ends its life with a mass of 19 @xmath0 , way below the limit for ppisn . this result indicates that ppisn and , even more so , pisn events that may result from local ( @xmath39 @xmath40 ) massive progenitors must be very rare compared to those that result from metal - poor primordial progenitors . recently , stacy et al . ( 2011 ) determined the rotational speed of massive primordial stars to be close to 50% of the critical value using multi - dimensional smooth particle hydrodynamics ( sph ) simulations . in addition , observational evidence presented by dufton et al . ( 2011 ) on the 20 - 30@xmath0 late o - type star vfts102 in 30 doradus indicates a rotational speed @xmath41 50% of the critical value for this star . vfts102 has @xmath37 0 and is not massive enough to encounter pair formation in any circumstances , but it establishes the existence of fast rotation in some high mass stars . using mesa to follow the evolution and flash to compute the late time hydrodynamics for a grid of massive primordial ( @xmath5 0 ) stars with various degrees of rotation and including the effects of magnetic fields as currently parametrized , we established the zams mass ranges required to produce ppisn or pisn as presented in figure 5 . for significant rotational velocities , we find that stars with zams masses as low as 40 - 45 @xmath0 can produce ppisn and stars with zams masses @xmath41 65 - 75 @xmath0 can produce direct pisn explosions due to a more chemically homogeneous evolution that leads to increased oxygen core masses . we also investigated the effects of convective mixing , metallicity and mass loss . we found that suppressed mixing due to composition gradients is unlikely to significantly alter the results for the rotating models since rotationally - induced mixing is the dominant factor . we also found that higher metallicities ( and specifically solar metalicity ) can induce extreme line - driven mass loss so that even some of the most massive zams stars end their lives as cc sne avoiding the pair - formation instability . current results from numerical simulations ( greif et al . 2011 ) suggest a nearly flat primordial initial mass function ( imf ) with typical mass of 100 @xmath0 . if so , pisn and especially ppisn events may be more frequent than estimates suggest based on non - rotating models ( scannapieco et al . this raises the possibility of detecting a larger number of those spectacular explosions with future missions such the james webb space telescope ( jwst ) . as emphasized by smith & owocki ( 2006 ) and smith et al . ( 2007 ) , stars under a wide variety of conditions may undergo the mass ejection process associated with luminous blue variables ( lbv ) . the physical mechanism of the lbv phenomenon is not well understood . ppisn is not likely to be the only mechanism involved in the lbv process , but our result that stars with initial mass as low as 40 @xmath0 may undergo ppisn means that ppisn should be considered as a possible candidate mechanism for the lbv mass loss phenomenon in some circumstances . it would be worthwhile to explore the ppisn process in stars with a variety of envelope compositions and structures , including those that are nearly pure , bare , oxygen cores to understand how ppisn may lead to single or multiple shell - ejection phases . we thank the mesa team for making this valuable tool readily available and especially thank bill paxton for his ready advice and council in running the code . we wish to thank the anonymous referee for useful suggestions and comments . we also wish to thank volker bromm and athena stacy for discussions of the formation and evolution of pop iii stars . ec would like to thank tuguldur sukhbold for help with mesa and selma de mink for useful suggestions . ec would also like to thank sean couch and christopher lindner for help with flash . jcw especially thanks zalman barkat for first introducing him to this topic and for sharing portions of his phd thesis . this research is supported in part by nsf ast-1109801 . abel , t. , anninos , p. , norman , m. l. , & zhang , y. 1998 , , 508 , 518 abel , t. , bryan , g. l. , & norman , m. l. 2000 , , 540 , 39 appenzeller , i. 1970 , , 5 , 355 barkat , z. , rakavy , g. , & sack , n. 1967 , physical review letters , 18 , 379 bond , j. r. , arnett , w. d. , & carr , b. j. 1984 , , 280 , 825 bromm , v. , coppi , p. s. , & larson , r. b. 2002 , , 564 , 23 bromm , v. , & larson , r. b. 2004 , , 42 , 79 brott , i. , de mink , s. e. , cantiello , m. , et al . 2011 , , 530 , a115 brott , i. , evans , c. j. , hunter , i. , et al . 2011 , , 530 , a116 carr , b. j. , bond , j. r. , & arnett , w. d. 1984 , , 277 , 445 chatzopoulos , e. , et al . 2011 , , 729 , 143 chatzopoulos , e. , wheeler , j. c. , & vinko , j. 2012 , arxiv:1111.5237 chevalier , r. a. , & irwin , c. m. 2011 , , 729 , l6 christlieb , n. , bessell , m. s. , beers , t. c. , et al . 2002 , , 419 , 904 crowther , p. a. , schnurr , o. , hirschi , r. , et al . 2010 , , 408 , 731 de jager , c. , nieuwenhuijzen , h. , & van der hucht , k. a. 1988 , , 72 , 259 dufton , p. l. , dunstall , p. r. , evans , c. j. , et al . 2011 , , 743 , l22 ekstrm , s. , meynet , g. , chiappini , c. , hirschi , r. , & maeder , a. 2008 , , 489 , 685 ekstrm , s. , georgy , c. , eggenberger , p. , et al . 2011 , arxiv:1110.5049 el eid , m. f. , & hilf , e. r. 1977 , , 57 , 243 el eid , m. f. , fricke , k. j. , & ober , w. w. 1983 , , 119 , 54 fraley , g. s. 1968 , , 2 , 96 frebel , a. , aoki , w. , christlieb , n. , et al . 2005 , , 434 , 871 fryxell , b. , olson , k. , ricker , p. , et al . 2000 , , 131 , 273 gal - yam , a. , mazzali , p. , ofek , e. o. , et al . 2009 , , 462 , 624 greif , t. h. , springel , v. , white , s. d. m. , et al . 2011 , , 737 , 75 heger , a. , langer , n. , & woosley , s. e. 2000 , , 528 , 368 heger , a. , & woosley , s. e. 2002 , , 567 , 532 heger , a. , fryer , c. l. , woosley , s. e. , langer , n. , & hartmann , d. h. 2003 , , 591 , 288 heger , a. , woosley , s. e. , & spruit , h. c. 2005 , , 626 , 350 heger , a. , & woosley , s. e. 2010 , , 724 , 341 hunter , d. a. , shaya , e. j. , holtzman , j. a. , et al . 1995 , , 448 , 179 joggerst , c. c. , & whalen , d. j. 2011 , , 728 , 129 kasen , d. , woosley , s. , & heger , a. 2010 , american institute of physics conference series , 1294 , 96 kasen , d. , woosley , s. e. , & heger , a. 2011 , , 734 , 102 langer , n. , norman , c. a. , de koter , a. , et al . 2007 , , 475 , l19 leloudas , g. , et al . 2012 , , update , update maeder , a. 1987 , , 178 , 159 maeder , a. , & meynet , g. 2011 , arxiv:1109.6171 meynet , g. , & maeder , a. 1987 , nuclear astrophysics , 287 , 195 miller , a. a. , et al . 2009 , , 690 , 1303 ober , w. w. , el eid , m. f. , & fricke , k. j. 1983 , , 119 , 61 ohkubo , t. , umeda , h. , & nomoto , k. 2003 , nuclear physics a , 718 , 632 ohkubo , t. , nomoto , k. , umeda , h. , yoshida , n. , & tsuruta , s. 2009 , , 706 , 1184 panagia , n. , tanzi , e. g. , & tarenghi , m. 1983 , , 272 , 123 paxton , b. , bildsten , l. , dotter , a. , et al . 2011 , , 192 , 3 quimby , r. m. , aldering , g. , wheeler , j. c. , hflich , p. , akerlof , c. w. , & rykoff , e. s. 2007a , , 668 , l99 rakavy , g. , shaviv , g. , & zinamon , z. 1967 , , 150 , 131 rakavy , g. , & shaviv , g. 1967 , , 148 , 803 rakavy , g. , & shaviv , g. 1968 , , 1 , 429 scannapieco , e. , madau , p. , woosley , s. , heger , a. , & ferrara , a. 2005 , , 633 , 1031 smith , n. , & owocki , s. p. 2006 , , 645 , l45 smith , n. , et al . 2007 , , 666 , 1116 spruit , h. c. 1999 , , 349 , 189 spruit , h. c. 2002 , , 381 , 923 stacy , a. , greif , t. h. , & bromm , v. 2010 , , 403 , 45 stacy , a. , bromm , v. , & loeb , a. 2011 , , 413 , 543 stringfellow , g. s. , & woosley , s. e. 1988 , origin and distribution of the elements , 467 timmes , f. x. 1999 , , 124 , 241 timmes , f. x. , & swesty , f. d. 2000 , , 126 , 501 waldman , r. 2008 , , 685 , 1103 whalen , d. j. , & fryer , c. 2010 , american institute of physics conference series , 1279 , 116 wheeler , j. c. 1977 , , 50 , 125 woosley , s. e. , blinnikov , s. , & heger , a. 2007 , , 450 , 390 yoon , s .- c . , & langer , n. 2005 , , 443 , 643 yoon , s .- c . , woosley , s. e. , & langer , n. 2010 , , 725 , 940 yoon , s .- c . , dierks , a. , & langer , n. 2012 , arxiv:1201.2364 ziebarth , k. 1970 , , 162 , 947 lllllllllllllllllllllllllllllllllllllllllllllllccccccc 35 & 0.8 & - & - & 35 & 4.70 & cc + 40 & 0.0 & - & - & 14 & 0.00 & cc + 40 & 0.3 & - & - & 25 & 0.43 & cc + 40 & 0.5 & - & - & 38 & 22.64 & cc + 40 & 0.8 & 1.16 & 1.03 & 40 & 1.48 & ppisn + 45 & 0.0 & - & - & 17 & 0.00 & cc + 45 & 0.3 & - & - & 30 & 1.31 & cc + 45 & 0.5 & 1.21 & 0.63 & 44 & 15.54 & ppisn + 45 & 0.8 & 0.90 & 0.27 & 45 & 5.49 & ppisn + 50 & 0.0 & - & - & 19 & 0.00 & cc + 50 & 0.3 & - & - & 39 & 0.18 & cc + 50 & 0.5 & 3.19 & 9.25 & 48 & 20.01 & ppisn + 50 & 0.8 & 1.17 & 1.85 & 50 & 1.32 & ppisn + 55 & 0.0 & - & - & 22 & 0.00 & cc + 55 & 0.3 & 2.04 & 3.53 & 41 & 0.15 & ppisn + 55 & 0.5 & 3.22 & 9.57 & 53 & 14.08 & ppisn + 55 & 0.8 & 1.08 & 0.88 & 55 & 1.66 & ppisn + 60 & 0.0 & - & - & 30 & 0.00 & cc + 60 & 0.3 & 2.09 & 3.31 & 45 & 0.04 & ppisn + 60 & 0.5 & 2.91 & 8.09 & 57 & 26.90 & ppisn + 60 & 0.8 & 1.05 & 0.53 & 60 & 15.28 & ppisn + 65 & 0.0 & - & - & 32 & 0.00 & cc + 65 & 0.3 & 2.37 & 5.37 & 46 & 0.00 & ppisn + 65 & 0.5 & 3.49 & 12.16 & 60 & 26.51 & ppisn + 65 & 0.8 & 4.45 & 35.6 & 65 & 0.00 & pisn + 70 & 0.0 & - & - & 35 & 0.00 & cc + 70@xmath42 & 0.0 & - & - & 30 & 0.00 & cc + 70 & 0.3 & 2.58 & 6.90 & 48 & 0.00 & ppisn + 70 & 0.5 & 3.03 & 9.22 & 66 & 20.58 & ppisn + 70@xmath42 & 0.5 & 3.02 & 9.20 & 65 & 18.12 & ppisn + 70@xmath43 & 0.5 & - & - & 16 & 10.51 & cc + 70 & 0.8 & 5.08 & 43.18 & 70 & 5.30 & pisn + 75 & 0.0 & - & - & 36 & 0.00 & cc + 75 & 0.3 & 3.21 & 10.40 & 54 & 0.02 & ppisn + 75 & 0.5 & 3.61 & 20.00 & 67 & 16.68 & pisn + 75 & 0.8 & 5.28 & 61.16 & 75 & 0.00 & pisn + 80 & 0.0 & 1.87 & 2.44 & 40 & 0.00 & ppisn + 80 & 0.3 & 1.99 & 2.30 & 59 & 0.00 & ppisn + 80 & 0.5 & 4.31 & 20.36 & 77 & 26.41 & pisn + 80 & 0.8 & 2.97 & 8.99 & 80 & 0.00 & pisn + 85 & 0.0 & 2.09 & 3.61 & 40 & 0.00 & ppisn + 85 & 0.3 & 1.53 & 0.81 & 65 & 0.39 & pisn + 85 & 0.5 & 3.46 & 10.41 & 79 & 24.93 & pisn + 85 & 0.8 & 3.25 & 9.84 & 85 & 8.64 & pisn + 90 & 0.0 & 1.86 & 1.76 & 45 & 0.00 & ppisn + 90 & 0.3 & 3.98 & 17.44 & 85 & 14.24 & pisn + 95 & 0.0 & 2.75 & 6.59 & 50 & 0.00 & ppisn + 95 & 0.3 & 4.62 & 26.97 & 90 & 33.25 & pisn + 110 & 0.0 & 2.46 & 5.85 & 56 & 0.00 & ppisn + 200 & 0.0 & 5.02 & 34.40 & 120 & 0.00 & pisn +

the global evolution of the observable universe can be separated on four epochs : inflation , radiation domination era , matter domination era and dark energy domination era . all these epochs can be described by general relativity models with minimally coupled scalar fields , because the assumption that the hubble parameter is a monotonically decreasing function does not contradict with the observation data . in such type of models the initial period of the universe evolution with energies above the planck scale should be described by quantum gravity because the classical evolution includes the initial singularity . important question of theoretical cosmology is whether the entire universe evolution can remain classical and has no singularity . it is possible to avoid this singularity considering modified gravity . bouncing universe smoothly transits from a period of contraction to a period of expansion and its evolution remains classical . models of bouncing universes attract a lot of attention @xcite . we can mention , in particular , bouncing models with galileon fields @xcite , @xmath0 gravity models @xcite , non - local gravity models @xcite and models with non - minimal coupling @xcite . the goal of this paper is to explore new mathematical features of cosmological models with non - minimally coupled scalar fields that admit bounce solutions . we consider models with the spatially flat friedmann lematre robertson walker ( flrw ) metric . at the bounce point the period of universe contraction changes to a period of universe expansion . thereby , a bounce point is characterized by two condition : at this point the hubble parameter @xmath1 is equal to zero and its cosmic time derivative is positive . note that the general relativity models with minimally coupled standard scalar fields does not have such a type of solutions , because in these models the hubble parameter is a monotonically decreasing function . the simplest way to get a bounce solution is to consider a model with a phantom scalar field . note that similar models is used to describe dark energy with the state parameter less then minus one @xcite . to get non - monotonic behaviour of the hubble parameter quintom models have been developed @xcite . in phantom field models the null energy condition is violate and instability problems arise @xcite . at the same time it is easy to get increasing or non - monotonic behavior of the hubble parameter in models with a standard scalar field non - minimally coupled to gravity , for example , in induced gravity models @xcite . to obtain bounce solutions in cosmological models with standard scalar fields one can use non - minimal coupling between the scalar fields and gravity and consider the following action : @xmath2 where @xmath3 is the determinant consisting of metric tensor components @xmath4 , @xmath5 is the ricci scalar , @xmath6 and @xmath7 are differentiable functions of a scalar field @xmath8 . from the condition that at the bounce point the hubble parameter @xmath9 , the potential should be negative @xmath10 . the characteristic property of models with non - positive definite potential is the existence of the unreachable domain on the phase plane , which corresponds to non - real hubble parameter @xcite . in this paper we consider models with bounce solutions that generalize the integrable model proposed in @xcite with the ricci scalar @xmath5 as an integral of motion . this model has the potential in the form @xmath11 , with @xmath12 and @xmath13 , and the standard quadratic coupling @xmath14 with @xmath15 . as known in the spatially flat flrw metric the ricci scalar @xmath5 is a function of the hubble parameter @xmath1 and its derivative . so , the condition that @xmath5 is a constant defines the hubble parameter as a solution of the first order differential equation . thereby , we get only monotonic behavior of the hubble parameter ( formulae are presented in section 3 ) . in particular , this integrable model contains bouncing cosmological solutions with smooth future behavior tending to de sitter solution ( while past behavior shows emerging of a universe at a point with zero effective newtonian constant ) . it has been shown in @xcite that the integrable cosmological model @xcite with a constant @xmath5 belongs to one - parameter set of integrable models with one and the same function @xmath16 and different potentials @xmath17 . however , the ricci scalar @xmath5 is not an integral to motion for these integrable models . in this paper we show that the model , proposed in @xcite , is unique in this sense that the property of having @xmath5 to be a constant disappears , when one changes coupling function or scalar field potential ( or both ) . to construct more realistic models with both a bounce and non - monotonic behavior of the hubble parameter one can consider non - integrable models that are close to the above - mention integrable model with a constant @xmath5 . in @xcite the authors got such types of the hubble parameter considering non - integrable models with the same potential and other values of @xmath18 . in the present paper we mostly vary the potential in searching for conditions of bouncing behavior similar to studied in @xcite to exist in these more general models . in section 4 we consider the model with @xmath15 and the higgs - type potential multiplying by a negative constant . in section 5 we consider models with different positive values of @xmath18 , a cosmological constant and different even degrees of monomial potentials . using numerical calculations we demonstrate the possibility of different evolutions of the hubble parameters in the cases of potentials of the forth and sixth degrees . we analyze the reasons of such behavior and find the corresponding conditions of the model parameters . the paper is organized as follows . in sect . 2 we remind a reader the general bounce conditions in non - minimally coupled scalar field theory as well as the conception of the effective potential . in sect . 3 we show that the choice of potential in the paper @xcite leading to the property that the curvature @xmath5 is the integral of motion is the unique one . on the contrary , in sect . 4 and 5 we show that bounce behavior admits much larger class of potentials . in sect . 4 bounce behavior is studied for potentials with quadratic term in addition to quartic and constant terms and @xmath15 . the models with @xmath19 are considered in sect . 6 contains a summary of the result obtained . in the spatially flat flrw metric with the interval : @xmath20 we obtain from action ( [ s1 ] ) the following equations : @xmath21 @xmath22 + 2u'\left[\ddot{\varphi}+2h\dot{\varphi}\right ] = v-\left[2u''+\frac12\right ] \dot{\varphi}^2,\ ] ] @xmath23+v ' = 0 , \label{kgoc}\ ] ] where a `` dot '' means a derivative with respect to the cosmic time @xmath24 and a `` prime '' means a derivative with respect to the scalar field @xmath8 . the function @xmath25 is the scale factor , its logarithmic derivative @xmath26 is the hubble parameter . we note that an effective gravitation constant in the model considered is @xmath27 the dynamics of flrw universe can be prolonged smoothly into the region of @xmath28 ( see , for example @xcite ) , however , since any anisotropic corrections are expected to diverge while @xmath29 tends to infinity @xcite . in this paper , we analyze such bounce solutions that @xmath30 for any @xmath31 and conditions of their existence . if a solution of eqs . ( [ frequoc00])([kgoc ] ) has such a point @xmath32 that @xmath33 and @xmath34 , then it is a bounce solution . let us find conditions that are necessary for the existence of a bounce in models with action ( [ s1 ] ) . using eq . ( [ frequoc00 ] ) we get that from @xmath33 it follows @xmath35 . subtracting equation ( [ frequoc00 ] ) from equation ( [ frequocii ] ) , we obtain @xmath36 therefore , at the bounce point we also get @xmath37v,\ ] ] where functions @xmath16 and @xmath17 and their derivatives are taken at the point @xmath38 . the condition @xmath34 gives the restriction on functions @xmath16 and @xmath17 at the bounce point . to analyze the dynamic it is suitable to introduce a new variable and present equations ( [ frequoc00 ] ) and ( [ frequ21 ] ) in the form similar to the friedmann equations in the einstein frame . following @xcite , we introduce the effective potential @xmath39 if the constant @xmath40 , then @xmath41 coincides with the potential of the corresponding model in the einstein frame . note that we do not transform the metric and consider the model in the jordan frame only . for our purposes any positive value of @xmath42 is suitable ( in @xcite it was chosen @xmath43 ) . also in @xcite functions @xmath44 have been introduced . if @xmath45 , then @xmath46 as well . in terms of these functions eqs . ( [ frequoc00 ] ) and ( [ frequ21 ] ) take the following form : @xmath47 @xmath48 as known the hubble parameter is a monotonically decreasing function in models with a standard scalar field minimally coupled to gravity ( @xmath6 is a constant ) . for a model with an arbitrary positive @xmath6 function @xmath49 has the same property . equation ( [ frequoc00 ] ) is a quadratic equation in @xmath50 that has the following solutions : @xmath51 the values of the function @xmath49 that correspond to @xmath52 are @xmath53 } = \pm\sqrt{\frac{a}{3}{\dot\varphi}^2+\frac{2}{3}k^2v_{eff } } \ , .\ ] ] if the potential @xmath17 is negative at some values of @xmath8 , whereas the function @xmath16 is positive at these points , then one should restrict the domain of absolute values of @xmath54 from below to get a real hubble parameter . in other words , there exists the unreachable domain on the phase plane @xcite . the boundary of this domain is defined by the condition @xmath55 . if @xmath56 then the sign of the function @xmath57 can not change . if in some domain @xmath58 is negative , then the sign of @xmath57 can be changed from plus to minus , but not vice verse . so , the sign of @xmath57 can not be changed twice . equations ( [ frequoc00])([kgoc ] ) can be transformed into the following system of the first order equations which is useful for numerical calculations and analysis of stability : @xmath59u'+2uv'}{2\left(3 { u'}^2 + u\right)},\\ \dot h&={}-\frac{2u''+1}{4(3{u'}^2+u)}\psi^2+\frac{2u'}{{3{u'}^2+u}}h\psi -\frac{6{u'}^2}{3{u'}^2+u}h^2+\frac{u'v'}{2(3{u'}^2+u)}\ , . \end{split } \right . \label{fosequ}\ ] ] if eq . ( [ frequoc00 ] ) is satisfied in the initial moment of time , then from system ( [ fosequ ] ) it follows that eq . ( [ frequoc00 ] ) is satisfied at any moment of time . so , one can use eq . ( [ frequoc00 ] ) to fix initial conditions of system ( [ fosequ ] ) . from eqs . ( [ equp ] ) and ( [ fr21qm ] ) it is easy to get the following system @xcite : @xmath60 system of equations ( [ fr21qm ] ) and ( [ system2 ] ) is equivalent to system ( [ fosequ ] ) . the de sitter solutions corresponds to @xmath61 , and hence , @xmath62 . the corresponding hubble parameter is @xmath63 we analyze the stability of de sitter solutions with @xmath64 and @xmath65 only . from ( [ hf ] ) it follows that @xmath66 . in this case the hubble parameter in the neighborhood of @xmath67 is uniquely defined by eq . ( [ frequoc00 ] ) , so it is enough to consider system ( [ system2 ] ) to analyze the stability of de sitter solution . substituting the de sitter solution with the first order perturbations : @xmath68 into ( [ system2 ] ) , we get the following linear system on @xmath69 and @xmath70 : @xmath71 we find eigenvalues for the system ( [ linsystem ] ) : @xmath72 the real part of @xmath73 always , the condition that the real part of @xmath74 is equivalent to @xmath75 . therefore , for arbitrary differentiable functions @xmath17 and @xmath76 , the model has a stable de sitter solution with @xmath64 only if the potential @xmath41 has a minimum @xcite and @xmath77 . using the obtained values of @xmath78 , it is easy to find @xcite that the de sitter point is a stable node ( the scalar field decreases monotonically ) at @xmath79 and a stable focus ( the scalar field oscillations exist ) at @xmath80 in the paper @xcite the cosmological model with a constant ricci scalar @xmath5 has been considered . in this section we demonstrate how this form of function @xmath16 and the corresponding potential can be found from the requirement that the ricci scalar is an integral of motion . in the spatially flat flrw metric @xmath81 . from eqs . ( [ frequoc00])([kgoc ] ) we get @xmath82 @xmath83 combining eqs . ( [ equtrace ] ) and ( [ kgequ ] ) we obtain @xmath84 from the structure of eq . ( [ equr ] ) it is easy to see that the simplest way to get a constant @xmath5 is to choose such @xmath6 that @xmath85 where @xmath86 is a constant . the first two equations of system ( [ equ ] ) has the following solution @xmath87 where @xmath88 is an integration constant . without loss of generality we put @xmath89 . for such a choice of @xmath6 we get eq . ( [ equr ] ) as follows : @xmath90 considering eq . ( [ equr2 ] ) as a differential equation for @xmath7 , for a constant @xmath5 we get the following solution : @xmath91 where @xmath92 is an integration constant , for other constants we choose @xmath93 thus , requiring that the ricci scalar is a constant one can define both functions @xmath6 and @xmath7 . to get a positive @xmath29 for some values of @xmath8 we choose @xmath94 . substituting @xmath95 and @xmath96 into eq . ( [ bouncecond2 ] ) , we get that the condition @xmath97 is equivalent to @xmath98 , hence , from @xmath99 it follows @xmath13 . this integrable cosmological model has been considered in @xcite , where the behavior of bounce solutions has been studied in detail . equation @xmath100 is a differential equation for the hubble parameter : @xmath101 multiplying this equation by @xmath102 , we get @xmath103= 0.\ ] ] therefore , there exists the following integral of motion : @xmath104 equation ( [ equhr ] ) with a positive @xmath105 has two possible real solutions in dependence of the initial conditions : @xmath106 where @xmath107 is an integration constant . for convenience we locate the bounce at @xmath108 , so hereafter @xmath109 . note that the behavior of the hubble parameter does not depend on the specific dynamics of the scalar field @xmath8 , because two - parametric set of functions @xmath110 corresponds to one - parametric set of @xmath50 . the monotonically increasing hubble parameter @xmath111 is not suitable for construction of a realistic cosmological scenario , whereas @xmath112 is not a bounce solution . in @xcite the authors slightly modify the function @xmath16 to get a bounce solution with non - monotonic behaviour of the hubble parameters . in this section we modify the potential instead of the function @xmath16 and analyze the obtained bounce solutions . in other words , we choose the function @xmath113 , given by ( [ uc ] ) , with @xmath114 . as a minimal generalization of the potential @xmath96 we consider the following potential @xmath115 where @xmath116 are constants . we consider the case @xmath13 only , because the integrable model with the potential @xmath96 has a bounce solution and a stable de sitter solution only if @xmath98 and @xmath13 . we plan to consider the evolution of bounce solutions fixing the initial conditions at bounce point . without loss of generality we can consider as initial value @xmath117 only . the value of @xmath118 is defined by the condition @xmath119 , hence , @xmath120 . our first goal is to find conditions on the coefficients of the potential @xmath121 that corresponds to the existence of bounce solutions . for @xmath113 , a bounce solution exists only if there exists such a point @xmath122 that the following conditions are satisfied @xcite : @xmath123 so , @xmath124 , and we obtain the following conditions on parameters @xmath116 : @xmath125 thus , at least one of the constants @xmath126 or @xmath127 should be positive . a bounce solution has the physical sense only if @xmath128 . the condition @xmath129 means @xmath130 . the effective potential ( [ veff ] ) is @xmath131 the even potential @xmath41 has an extremum at @xmath132 and at points @xmath133 we consider such values of parameter that points @xmath134 are real and @xmath135 . note that for integrable model ( @xmath13 , @xmath136 , @xmath12 ) @xmath137 are non - zero real numbers and the condition @xmath138 is equivalent to @xmath139 . from ( [ bouncecond ] ) , @xmath13 and @xmath130 we get @xmath140 so , the model with a bounce solution has real @xmath134 only at @xmath141 using these conditions , we get @xmath142 so , the potential @xmath41 has a minimum at @xmath132 and maxima at @xmath143 . a few examples of the effective potential @xmath41 with two maxima at nonzero points @xmath134 and a minimum at @xmath132 are presented in fig . [ veffc0 ] . at different values of parameters . in all pictures we choose @xmath144 . the values of parameters are @xmath145 , @xmath146 ( left picture ) , @xmath147 , @xmath146 ( middle picture ) , @xmath147 , @xmath148 ( right picture ) . the parameter @xmath149 ( black curve ) , @xmath150 ( red curve ) , @xmath151 ( blue curve ) , @xmath152 ( green curve ) , and @xmath153 ( cyan curve ) . the yellow curve in the middle picture corresponds to @xmath154.,title="fig:",width=185 ] at different values of parameters . in all pictures we choose @xmath144 . the values of parameters are @xmath145 , @xmath146 ( left picture ) , @xmath147 , @xmath146 ( middle picture ) , @xmath147 , @xmath148 ( right picture ) . the parameter @xmath149 ( black curve ) , @xmath150 ( red curve ) , @xmath151 ( blue curve ) , @xmath152 ( green curve ) , and @xmath153 ( cyan curve ) . the yellow curve in the middle picture corresponds to @xmath154.,title="fig:",width=185 ] at different values of parameters . in all pictures we choose @xmath144 . the values of parameters are @xmath145 , @xmath146 ( left picture ) , @xmath147 , @xmath146 ( middle picture ) , @xmath147 , @xmath148 ( right picture ) . the parameter @xmath149 ( black curve ) , @xmath150 ( red curve ) , @xmath151 ( blue curve ) , @xmath152 ( green curve ) , and @xmath153 ( cyan curve ) . the yellow curve in the middle picture corresponds to @xmath154.,title="fig:",width=185 ] stable de sitter solution at @xmath155 corresponds to @xmath156 . therefore , such a solution exists for @xmath157 only . it means that the stable de sitter solution exists only if @xmath158 . in this case @xmath159 . the potential @xmath41 has the following zeros : @xmath160 at @xmath13 , @xmath12 one gets @xmath161 . therefore , @xmath162 and there exist only two real roots : @xmath163 the bounce point @xmath164 belongs to the interval @xmath165 . we get the condition @xmath166 . for @xmath113 and an arbitrary potential system ( [ fosequ ] ) has the following form @xcite : @xmath167 . \end{split } \label{fosequc}\ ] ] we integrate this system with @xmath168 numerically . we consider a positive @xmath122 such that @xmath169 . the evolution of the scalar field starts at the bounce point with a negative velocity , defined by relation @xmath170 the field @xmath8 can come to zero passing the maximum of the potential . so , we keep in mind that the following subsequence of inequalities : @xmath171 in the case @xmath12 there are three possible evolutions of the bounce solutions , depending on whether the solution passes the points of the maximum of @xmath41 or not . in the left and middle pictures of fig . [ threesolutions ] we present an example of three possible behaviors of the bounce solutions . the solution with less initial value of @xmath8 tends to infinity ( blue curve ) , whereas the bounce solutions with greater initial values tends to zero ( cyan curve ) or to minus infinity as a monotonically decreasing function ( green curve ) . all trajectories start at bounce points . the corresponding behaviors of the hubble parameter are presented in the right picture of fig . [ threesolutions ] ( colors the hubble parameter evolutions on this picture coincide to the colors of the corresponds phase trajectories ) . and the values of constants are @xmath144 , @xmath147 , @xmath172 , and @xmath173 . the initial conditions are @xmath174 , @xmath175 ( blue line ) , @xmath176 , @xmath177 ( cyan line ) , and @xmath178 , @xmath179 ( green line ) . the black curves are the lines of the points that correspond to @xmath9 . the red dash lines are the boundary of unreachable domain defined by @xmath55 . the blue point lines correspond to @xmath180 . a zoom of the central part of phase plane is presented in the middle picture . the corresponding hubble parameters as functions of cosmic time are presented on the right picture.,title="fig:",width=178 ] and @xmath121 . the values of constants are @xmath144 , @xmath147 , @xmath172 , and @xmath173 . the initial conditions are @xmath174 , @xmath175 ( blue line ) , @xmath176 , @xmath177 ( cyan line ) , and @xmath178 , @xmath179 ( green line ) . the black curves are the lines of the points that correspond to @xmath9 . the red dash lines are the boundary of unreachable domain defined by @xmath55 . the blue point lines correspond to @xmath180 . a zoom of the central part of phase plane is presented in the middle picture . the corresponding hubble parameters as functions of cosmic time are presented on the right picture.,title="fig:",width=178 ] and @xmath121 . the values of constants are @xmath144 , @xmath147 , @xmath172 , and @xmath173 . the initial conditions are @xmath174 , @xmath175 ( blue line ) , @xmath176 , @xmath177 ( cyan line ) , and @xmath178 , @xmath179 ( green line ) . the black curves are the lines of the points that correspond to @xmath9 . the red dash lines are the boundary of unreachable domain defined by @xmath55 . the blue point lines correspond to @xmath180 . a zoom of the central part of phase plane is presented in the middle picture . the corresponding hubble parameters as functions of cosmic time are presented on the right picture.,title="fig:",width=178 ] let us compare two bounce solutions with initial conditions : @xmath181 and @xmath182 and negative initial values of @xmath118 . the corresponding hubble parameter that is equal to zero at a bounce point is given by ( [ hpm ] ) : @xmath183 . the initial value of @xmath118 for the first solution we denote as @xmath184 . the potential @xmath10 for all @xmath185 $ ] , therefore , @xmath186 for any solution that pass @xmath187 . thus , at some moment of time @xmath188 the solution @xmath189 comes at the point @xmath187 : @xmath190 . the value of its velocity at the bounce point @xmath191 has a minimal absolute value by comparison with any solutions that pass via @xmath187 with a negative velocity . it follows from ( [ frequoc00 ] ) , because both @xmath192 , and @xmath193 are positive . it proves that if a solution with the bounce point @xmath187 passes through the maximum of the potential and tends to zero , then any solution with the bounce point @xmath194 tends to zero as well . all bounce solutions with negative initial value of @xmath118 come in the domain where @xmath195 , so , @xmath196 . if a solution does not pass the maximum of the effective potential , then after some moment @xmath8 starts to grow and this solution comes to antigravity domain with @xmath197 ( the blue curve in fig . [ threesolutions ] denotes an example of such a solution ) . solutions passing the maximum of the potential can be different and need considering in detail . if @xmath10 for all @xmath8 , then @xmath110 is a monotonic function , because at any point @xmath198 . similar dynamics is possible even if @xmath195 ( see blue curve in fig . [ threesolutions ] ) . for @xmath12 there exists the stable de sitter solution @xmath155 and @xmath199 . it is a stable node at @xmath200 and a stable focus in the opposite case @xmath201 . in fig . [ threesolutions ] ( cyan curves ) , in the left picture of fig . [ phaseportnode ] , and fig . [ phaseportv4c0 ] solutions in the case of a stable focus are presented . an example with a stable node at @xmath132 is given in the middle picture of fig . [ phaseportnode ] . one can see that in the case of a stable node the presented hubble parameter ( the right picture of fig . [ phaseportnode ] ) is close to the hubble parameter obtained in the paper @xcite that has a maximum . note that in the integrable case @xmath136 and there exists a stable node at @xmath132 . , @xmath147 , @xmath202 and @xmath203 . initial values are @xmath204 and @xmath205 ( cyan curve ) , @xmath206 and @xmath207 ( green curve ) . the cyan curve is an example of a stable focus . on the middle and right pictures the example of a stable node at @xmath132 is presented . the field @xmath8 ( blue line ) and the hubble parameter ( red line ) as functions of the cosmic time are presented in the right picture . the values of parameters are @xmath208 , @xmath209 , @xmath172 and @xmath210 . the initial conditions of the bounce solution are @xmath211 , @xmath212 , and @xmath119.,title="fig:",width=178 ] , @xmath147 , @xmath202 and @xmath203 . initial values are @xmath204 and @xmath205 ( cyan curve ) , @xmath206 and @xmath207 ( green curve ) . the cyan curve is an example of a stable focus . on the middle and right pictures the example of a stable node at @xmath132 is presented . the field @xmath8 ( blue line ) and the hubble parameter ( red line ) as functions of the cosmic time are presented in the right picture . the values of parameters are @xmath208 , @xmath209 , @xmath172 and @xmath210 . the initial conditions of the bounce solution are @xmath211 , @xmath212 , and @xmath119.,title="fig:",width=178 ] , @xmath147 , @xmath202 and @xmath203 . initial values are @xmath204 and @xmath205 ( cyan curve ) , @xmath206 and @xmath207 ( green curve ) . the cyan curve is an example of a stable focus . on the middle and right pictures the example of a stable node at @xmath132 is presented . the field @xmath8 ( blue line ) and the hubble parameter ( red line ) as functions of the cosmic time are presented in the right picture . the values of parameters are @xmath208 , @xmath209 , @xmath172 and @xmath210 . the initial conditions of the bounce solution are @xmath211 , @xmath212 , and @xmath119.,title="fig:",width=178 ] and @xmath121 is presented in the left picture . the values of constants are @xmath144 , @xmath147 , @xmath202 , @xmath173 . the initial conditions are @xmath206 and @xmath213 . a zoom of the central part of phase plane is presented in the middle picture . the hubble parameter ( red ) and the scalar field ( blue ) of functions of cosmic time are presented in the right picture.,title="fig:",width=178 ] and @xmath121 is presented in the left picture . the values of constants are @xmath144 , @xmath147 , @xmath202 , @xmath173 . the initial conditions are @xmath206 and @xmath213 . a zoom of the central part of phase plane is presented in the middle picture . the hubble parameter ( red ) and the scalar field ( blue ) of functions of cosmic time are presented in the right picture.,title="fig:",width=178 ] and @xmath121 is presented in the left picture . the values of constants are @xmath144 , @xmath147 , @xmath202 , @xmath173 . the initial conditions are @xmath206 and @xmath213 . a zoom of the central part of phase plane is presented in the middle picture . the hubble parameter ( red ) and the scalar field ( blue ) of functions of cosmic time are presented in the right picture.,title="fig:",width=178 ] in the left picture of fig . [ phaseportnode ] the phase trajectories have been constructed for @xmath144 , @xmath147 , @xmath202 and @xmath203 . let us now change the value of @xmath127 only and consider the model with @xmath173 . in fig . [ phaseportv4c0 ] the corresponding phase trajectory is presented . we see that now the bounce solution that starts at @xmath206 tends to zero and finishes at the point @xmath132 . we can see that the trajectories that revolve around @xmath214 points looks similar at @xmath203 and at @xmath173 . let us consider now the phase trajectory at @xmath215 that is presented in fig . [ phaseportv4cnega ] . we see that trajectories are similar at the beginning only . the scalar field tends to infinity and the system comes to antigravity domain with @xmath216 . the form of the effective potential does not depend essentially from the sign value of @xmath127 ( see fig . [ veffc0 ] ) . but the sign of @xmath217 are different . at the same time the behavior of bounce solutions are essentially different . in the right pictures of fig . [ phaseportv4c0 ] and fig . [ phaseportv4cnega ] we show that the behavior of the hubble parameter also essentially depends on the sign of @xmath127 . and @xmath121 is presented in the left picture . the values of constants are @xmath144 , @xmath147 , @xmath202 , @xmath215 . the initial conditions are @xmath206 and @xmath218 . a zoom of the central part of phase plane is presented in the middle picture . the hubble parameter ( red ) and the scalar field ( blue ) of functions of cosmic time are presented in the right picture.,title="fig:",width=178 ] and @xmath121 is presented in the left picture . the values of constants are @xmath144 , @xmath147 , @xmath202 , @xmath215 . the initial conditions are @xmath206 and @xmath218 . a zoom of the central part of phase plane is presented in the middle picture . the hubble parameter ( red ) and the scalar field ( blue ) of functions of cosmic time are presented in the right picture.,title="fig:",width=178 ] and @xmath121 is presented in the left picture . the values of constants are @xmath144 , @xmath147 , @xmath202 , @xmath215 . the initial conditions are @xmath206 and @xmath218 . a zoom of the central part of phase plane is presented in the middle picture . the hubble parameter ( red ) and the scalar field ( blue ) of functions of cosmic time are presented in the right picture.,title="fig:",width=178 ] the difference between the solutions of system ( [ fosequc ] ) with a positive and a negative @xmath127 is demonstrated in fig . [ c0pm ] as well . the cyan curves correspond to @xmath203 , whereas the red curves correspond to @xmath149 . we see that the phase trajectories of the field @xmath8 and the hubble parameter are similar in the beginning , but stand essentially different in the future . we see that the behavior of solutions essentially depends on the sign of @xmath127 . let us understand the reason of this dependence . , @xmath147 , @xmath219 . the parameter @xmath203 for cyan curves and @xmath149 for red curves . the initial conditions of bounce solution are @xmath204 , @xmath220 ( cyan curve ) and @xmath221 ( red curve).,title="fig:",width=272 ] , @xmath147 , @xmath219 . the parameter @xmath203 for cyan curves and @xmath149 for red curves . the initial conditions of bounce solution are @xmath204 , @xmath220 ( cyan curve ) and @xmath221 ( red curve).,title="fig:",width=272 ] let us consider the domain @xmath222 . from ( [ hpm ] ) it follows that @xmath223 if @xmath224 for all @xmath225 , then this condition is always satisfied and @xmath8 tends to a minimum of @xmath41 . we see such evolution in the right picture of fig . [ phaseportv4c0 ] . the constants @xmath116 are such that @xmath195 for all @xmath226 $ ] . for suitable initial @xmath54 the evolution is finished at @xmath132 . note that in this case if @xmath227 in the moment when potential stands positive and the function @xmath8 tends to zero , then @xmath227 at any moment in future . if the constants @xmath116 are such that the potential change the sign and @xmath228 then the evolution is different ( see the right picture in fig . [ phaseportv4cnega ] ) . the hubble parameter becomes negative and positive again , so , there are two bounce points . after that the hubble parameter tends to infinity . let us consider this case in detail . if @xmath229 , then there is a restricted domain in the neighborhood of @xmath214 point on phase plane such that the values of the scalar field @xmath8 and its derivative correspond to non - real value of the hubble parameter . the boundary of this domain is defines by equation @xmath55 . on fig . [ phaseportv4cnega ] we see that the phase trajectory rotates around this domain . the trajectory can not cross the boundary , but can touch it . let us show that all such trajectories touch the boundary at some finite moment of time . let for some moments of time @xmath230 and @xmath231 we have @xmath232 , then , using @xmath233 and formula ( [ fr21qm ] ) , we get @xmath234 where @xmath235 is a negative number . therefore , this integral has a finite negative value . for any circle value of @xmath57 decreases on some positive value , which does nt tend to zero during evolution , when number of circles increase . we come to conclusion that only a finite number of circles is necessary to get the value @xmath55 . at this point @xmath236 as well , so the function @xmath57 changes the sign . when @xmath55 two possible values of the hubble parameters : @xmath237 and @xmath238 coincide . at this moment the value of the hubble parameter changes from @xmath237 to @xmath238 . the value of the function @xmath57 continues to decrease , so , the distance between the trajectory and the boundary of unreachable domain increases . we do not say that its increase monotonically but @xmath57 that is a characteristic of this distance increases on a finite quantity after any circle . so , after some finite number of circles the absolute value of @xmath8 becomes more then @xmath134 . after this moment @xmath8 monotonically tends to infinity and at some finite moment we get @xmath239 . thus , the final of trajectory will be in antigravity domain always . note that in the domain with @xmath195 the hubble parameter is uniquely defined as a function @xmath8 and @xmath118 by ( [ hpm ] ) . if @xmath240 , then whole evolution of bounce solutions is evolution a solution of the second order system , whereas for @xmath241 the third order system ( [ fosequc ] ) with the additional condition ( [ frequoc00 ] ) is not equivalent to any second order system . in this section we consider the scalar field potential of the form @xmath242 where @xmath243 is an even natural number , and following non - minimally coupled function @xmath244 where @xmath18 is positive . we are interested in cosmological scenarios in the physical region @xmath245 , where the bounce occurs at first and after that stable de sitter solution @xmath246 is realized . so , the potential should change the sign . the potential @xmath41 is an even function , hence , it has an extremum at @xmath132 . if we additionally suppose that @xmath155 , then we get the condition @xmath12 . therefore , the bounce condition @xmath247 for the chosen potential @xmath17 and the function @xmath16 gives ( see ): @xmath248 from the condition @xmath249 it follows @xmath250 taking into account the inequalities @xmath251 and @xmath12 we obtain from conditions and that the bounce exists for * \(i ) @xmath252 * * ( 1 ) . if @xmath253 , then @xmath254 . * * ( 2 ) . if @xmath255 , then @xmath256 . * \(ii ) @xmath257 + @xmath256 . here @xmath122 is the value of the scalar field at the bounce . we see that there is an additional restriction on the location of points of bounce in comparison with the cases studied earlier in @xcite , appearing when @xmath252 ( the case ( i1 ) ) . this happens due to @xmath258 condition of the bounce , which can be violated in the case of small enough @xmath18 , causing a recollapse instead of bounce . the bounce point should corresponds to @xmath245 that gives the following condition @xmath259 let us study properties of the effective potential @xmath41 to analyse the lyapunov stability of de sitter solutions . the effective potential @xmath58 have zeros for @xmath260 they are located in the region @xmath245 only if @xmath261 thus , we have received a restriction from above for @xmath18 . the fist derivative @xmath41 with respect to @xmath8 for the chosen functions @xmath17 and @xmath16 is given by @xmath262 therefore , the effective potential has the extremum at the point @xmath132 for any even @xmath243 . non - zero extrema @xmath134 for @xmath263 are @xmath264 for @xmath265 values @xmath266 are roots of the following cubic equation @xmath267 which has three real roots for @xmath268 . we consider only @xmath269 ( see ) , hence , there are three real roots in this case . only positive ones have the physical sense ( @xmath270 ) . we calculate @xmath271 for de sitter solution @xmath155 , @xmath272 . then it is stable for @xmath273 , namely , @xmath274 from this it follows that there is the restriction from below for the coupling constant @xmath275 in the case of @xmath263 . on the other hand , for @xmath276 the interval of possible @xmath18 allowing evolution towards de sitter solution after the bounce is not restricted from below ( we remind that @xmath277 and we consider only positive values of @xmath18 in this paper ) . when de sitter solution is stable we have ( see and ) a monotonic decreasing of the scalar field for @xmath278 or scalar field oscillations for @xmath279 this result agrees with the one received in @xcite for @xmath280 . let us summarize our requirements to the effective potential @xmath281 and the parameter @xmath18 : * @xmath12 , @xmath251 , * @xmath282 for @xmath276 , * @xmath283 for @xmath263 . the case of @xmath280 and @xmath15 is an integrable model with monotonic behavior of the hubble parameter only . the change of parameter @xmath18 can not only give a hubble parameter with one maximum ( see @xcite ) , but also oscillating hubble parameter that tends to a constant . such solution has been found at @xmath284 and are presented in fig . [ n4xi20 ] . , @xmath285 . the parameters are @xmath284 , @xmath286 , @xmath287 , @xmath288 . the initial conditions are @xmath289 and @xmath290 ( green line ) , @xmath291 and @xmath292 ( blue line ) . the black curves are the lines of the points that correspond to @xmath9 . the blue point lines correspond to @xmath180 . the hubble parameter @xmath50 is presented in the right picture . , title="fig:",width=181 ] , @xmath285 . the parameters are @xmath284 , @xmath286 , @xmath287 , @xmath288 . the initial conditions are @xmath289 and @xmath290 ( green line ) , @xmath291 and @xmath292 ( blue line ) . the black curves are the lines of the points that correspond to @xmath9 . the blue point lines correspond to @xmath180 . the hubble parameter @xmath50 is presented in the right picture . , title="fig:",width=181 ] , @xmath285 . the parameters are @xmath284 , @xmath286 , @xmath287 , @xmath288 . the initial conditions are @xmath289 and @xmath290 ( green line ) , @xmath291 and @xmath292 ( blue line ) . the black curves are the lines of the points that correspond to @xmath9 . the blue point lines correspond to @xmath180 . the hubble parameter @xmath50 is presented in the right picture . , title="fig:",width=181 ] we explore numerically cases with @xmath280 and @xmath265 applying the first order system of differential equations . initial data are chosen @xmath293 , @xmath294 and @xmath295 is real and calculated using @xmath237 ( see ) . values of the scalar field @xmath122 and its time derivative @xmath296 are at the bounce line @xmath297 ( we denote @xmath298 ) . in figs . [ n4xi01 ] , [ n4xi01767 ] and [ v6ksi086 ] parameters @xmath286 , @xmath299 are chosen for all these plots and , therefore , stable de sitter solution is @xmath155 , @xmath300 . what is more interesting , numerical integration reveal another possible fate of a trajectory with initial conditions in the region suitable for bounce . in fig . [ n4xi01767 ] the outcome of the evolution with given initial value have been marked by different colors . black and green zones correspond to a bounce followed by a smooth evolution towards de sitter attractor ( black ) or @xmath8-turn then growing @xmath8 , leading to crossing @xmath180 boundary ( green ) . however , the blue zone exists starting from which a trajectory does not experience any bounce and fall into a singularity . this happens because a trajectory can touch the unreachable part of phase space and jump to another branch of two possible solutions for @xmath1 ( see [ hpm ] ) . we have described such a phenomenon in the previous section , where this happens near @xmath132 due to negative cosmological constant @xmath127 . here it happens in the bounce zone , where we also have negative potential and unreachable part of the phase space . after branch changing , the trajectory can not go through a bounce because bounce region for other branch is located in different phase space region . one can see in fig . [ n4xi01 ] that de sitter attractors do not exist for sufficiently small positive @xmath18 . , @xmath285 : singularities with @xmath301 , @xmath302 ( green points ) and with @xmath303 , @xmath304 ( blue points ) . the parameters are @xmath305 , @xmath286 , @xmath306 , @xmath299 . if the bounce occurs then we mark initial data with circles and in otherwise points . the red domain corresponds to unreachable domain.,width=453 ] , @xmath285 in the left picture : de sitter solution ( black circles ) , the analog of ruzmaikin one ( green circles ) and the singularity with @xmath303 , @xmath304 ( blue points ) . if the bounce occurs , then we mark initial data with circles and in otherwise points . the parameters are @xmath307 , @xmath286 , @xmath306 , @xmath299 . three phase trajectories ( middle picture ) and dependencies @xmath50 ( right picture ) corresponding these attractors are plotted for initial data @xmath308 and @xmath309 ( blue curve ) , @xmath310 ( green curve ) , @xmath311 ( black curve ) . the gold trajectory starts at the bounce point @xmath312 . the black dash line corresponds to @xmath9 . the blue point line corresponds to @xmath180 . the red color in the left picture indicates the unreachable domain . the red dash line in the middle picture is the boundary of unreachable domain defined by @xmath55 . , title="fig:",width=185 ] , @xmath285 in the left picture : de sitter solution ( black circles ) , the analog of ruzmaikin one ( green circles ) and the singularity with @xmath303 , @xmath304 ( blue points ) . if the bounce occurs , then we mark initial data with circles and in otherwise points . the parameters are @xmath307 , @xmath286 , @xmath306 , @xmath299 . three phase trajectories ( middle picture ) and dependencies @xmath50 ( right picture ) corresponding these attractors are plotted for initial data @xmath308 and @xmath309 ( blue curve ) , @xmath310 ( green curve ) , @xmath311 ( black curve ) . the gold trajectory starts at the bounce point @xmath312 . the black dash line corresponds to @xmath9 . the blue point line corresponds to @xmath180 . the red color in the left picture indicates the unreachable domain . the red dash line in the middle picture is the boundary of unreachable domain defined by @xmath55 . , title="fig:",width=181 ] , @xmath285 in the left picture : de sitter solution ( black circles ) , the analog of ruzmaikin one ( green circles ) and the singularity with @xmath303 , @xmath304 ( blue points ) . if the bounce occurs , then we mark initial data with circles and in otherwise points . the parameters are @xmath307 , @xmath286 , @xmath306 , @xmath299 . three phase trajectories ( middle picture ) and dependencies @xmath50 ( right picture ) corresponding these attractors are plotted for initial data @xmath308 and @xmath309 ( blue curve ) , @xmath310 ( green curve ) , @xmath311 ( black curve ) . the gold trajectory starts at the bounce point @xmath312 . the black dash line corresponds to @xmath9 . the blue point line corresponds to @xmath180 . the red color in the left picture indicates the unreachable domain . the red dash line in the middle picture is the boundary of unreachable domain defined by @xmath55 . , title="fig:",width=181 ] in the case of the sixth degree potential and @xmath313 . dependencies @xmath110 and @xmath50 for initial data : @xmath314 and @xmath315 ( gold curve ) , @xmath314 and @xmath316 ( blue curve ) , @xmath317 and @xmath316 ( green curve ) . parameters are chosen @xmath265 , @xmath313 , @xmath286 , @xmath318 , @xmath299 . the critical value of @xmath319 . , title="fig:",width=181 ] in the case of the sixth degree potential and @xmath313 . dependencies @xmath110 and @xmath50 for initial data : @xmath314 and @xmath315 ( gold curve ) , @xmath314 and @xmath316 ( blue curve ) , @xmath317 and @xmath316 ( green curve ) . parameters are chosen @xmath265 , @xmath313 , @xmath286 , @xmath318 , @xmath299 . the critical value of @xmath319 . , title="fig:",width=181 ] in the case of the sixth degree potential and @xmath313 . dependencies @xmath110 and @xmath50 for initial data : @xmath314 and @xmath315 ( gold curve ) , @xmath314 and @xmath316 ( blue curve ) , @xmath317 and @xmath316 ( green curve ) . parameters are chosen @xmath265 , @xmath313 , @xmath286 , @xmath318 , @xmath299 . the critical value of @xmath319 . , title="fig:",width=181 ] finally let us mention the case of sixth degree potentials . we consider @xmath320 . in this case @xmath41 has a minimum at @xmath321 and two maxima in the antigravity domain ( see the left picture of fig . [ v6ksi086 ] ) . both the minimum of @xmath41 at @xmath76 and a maxima of @xmath41 at @xmath322 corresponds to a stable de sitter solutions that are presented in fig . [ v6ksi086 ] . in fig . [ v6ksi086 ] we see oscillations in behavior the hubble parameter . solutions that do not leave the domain @xmath76 are similar to the case of the fourth order potential . there is the difference between solutions that cross the line @xmath180 . namely , such a solution tends to a de sitter solution instead of the infinity point . in this paper we have considered a class of spatially flat flrw models with bounce solutions , generalizing those found in @xcite . we show that despite the model with the potential @xmath323 , studied in @xcite is the unique model with the curvature scalar @xmath5 being the integral of motion , a much wider class of potentials admits bounce solutions followed by smooth future asymptotic behavior . one class of models studied in the present paper generalizes the above mentioned potential by adding a term quadratic in @xmath8 . this generalization allows us to find bounce solutions with non - monotonic behavior of the hubble parameter , in particular we found solutions with essential decreasing hubble parameter , including the solutions with the hubble parameter tending to zero . moreover , we present a list of possible types of the bounce solutions and show that bounce behavior is still present even in the case when @xmath324 being negative , though in this case the nice future asymptotic is evidently lost . another generalization replaces the quartic term in the potential by other even index . we studied the quartic and sixth - power cases , changing also the coupling constant @xmath18 . we show that despite the model becomes non - integrable , the qualitative features of solution found in @xcite are still valid for a not so restricted conditions on the potential . on the other hand , some new interesting features like possible branch changing during the cosmological evolution appears , which can decrease the measure of initial conditions good for bounce , though does not eliminate the possibility for the bounce completely . on the other hand , for sufficient small positive @xmath18 the bounce solution that tends to de sitter point does not exist . note that inflationary models with the ricci scalar multiplied by a function of the scalar field are very popular @xcite . they not only do not contradict the recent observational data @xcite , but also connect cosmology and particle physics . note that models with non - minimally coupled scalar fields are quite natural because quantum corrections to the effective action with minimal coupling include induced gravity terms @xcite . also , the einstein frame model that corresponds to the bouncing model @xcite describes inflation @xcite ( see also @xcite ) . note that non - local model @xcite and its generalization @xcite not only has bounce solution , but also ( at other values of parameters ) can describe inflation @xcite . the observation data @xcite are consistent with ekpyrotic cyclic models in which the universe is smoothed and flattened during a period of slow contraction followed by a bounce ( see @xcite and references therein ) . it would be very interesting to construct cosmological model with a non - minimal coupling standard scalar field , a bounce solution of which is suitable for inflationary scenario . the present paper can be considered as a step in this direction . we plan to consider a possibility to construct such a solution in future publications . research of e.p . is supported in part by grant mk-7835.2016.2 of the president of russian federation . is supported by rfbr grant 14 - 02 - 00894 and by the russian government program of competitive growth of kazan federal university . research of s.v . is supported in part by grant nsh-7989.2016.2 of the president of russian federation . researches of e.p . and s.v . are supported in part by the rfbr grant 14 - 01 - 00707 . m. novello and s.e.p . bergliaffa , _ bouncing cosmologies _ , phys . rept . * 463 * ( 2008 ) 127 ( arxiv:0802.1634 ) d. battefeld and p. peter , _ a critical review of classical bouncing cosmologies _ , * 571 * ( 2015 ) 1 ( arxiv:1406.2790 ) r. brandenberger and p. peter , _ bouncing cosmologies : progress and problems _ , arxiv:1603.05834 [ hep - th ] . m. koehn , j. l. lehners and b. a. ovrut , _ cosmological super - bounce , _ phys . d * 90 * ( 2014 ) 025005 ( arxiv:1310.7577 ) m. libanov , s. mironov , and v. rubakov , _ generalized galileons : instabilities of bouncing and genesis cosmologies and modified genesis _ , arxiv:1605.05992 [ hep - th ] . a. ijjas and p.j . steinhardt , _ classically stable non - singular cosmological bounces _ , arxiv:1606.08880 [ gr - qc ] . k. bamba , a.n . makarenko , a.n . myagky , s. nojiri , and s.d . odintsov , _ bounce cosmology from @xmath325 gravity and @xmath325 bigravity _ , jcap * 1401 * ( 2014 ) 008 ( arxiv:1309.3748 ) s. nojiri , s.d . odintsov , and v.k . oikonomou , _ bounce universe history from unimodular @xmath325 gravity _ , phys . d * 93 * ( 2016 ) 084050 ( arxiv:1601.04112 ) . tretyakova , a.a . shatskiy , i.d . novikov and s. alexeyev , _ non - singular brans - dicke cosmology with cosmological constant _ , phys . d * 85 * ( 2012 ) 124059 ( arxiv:1112.3770 ) b. boisseau , h. giacomini , d. polarski , and a.a . starobinsky , _ bouncing universes in scalar - tensor gravity models admitting negative potentials _ , j. cosmol . * 1507 * ( 2015 ) 002 ( arxiv:1504.07927 ) a.yu . kamenshchik , e.o . pozdeeva , a. tronconi , g. venturi and s.yu . vernov , _ interdependence between integrable cosmological models with minimal and non - minimal coupling _ , class . quantum grav . * 33 * ( 2016 ) 015004 ( arxiv:1509.00590 ) m.p . dabrowski , t. stachowiak and m. szydlowski , _ phantom cosmologies _ , d * 68 * ( 2003 ) 103519 ( hep - th/0307128 ) i.ya . arefeva , _ nonlocal string tachyon as a model for cosmological dark energy _ , aip conf . * 826 * ( 2006 ) 301 ( astro - ph/0410443 ) ; + i.ya . arefeva , a.s . koshelev , and s.yu . vernov , _ exact solution in a string cosmological model _ , * 148 * ( 2006 ) 895 ( arxiv : astro - ph/0412619 ) ; + m.l . pucheu , f.a.p . alves - junior , a.b . barreto and c. romero , _ cosmological models in weyl geometrical scalar - tensor theory _ , arxiv:1602.06966 [ gr - qc ] . yi - fu cai , e.n . saridakis , m.r . setare , and jun - qing xia , _ quintom cosmology : theoretical implications and observations _ , phys . rept . * 493 * ( 2010 ) 160 , arxiv:0909.2776 ; + zong - kuan guo , yun - song piao , xinmin zhang , yuan - zhong zhang , _ cosmological evolution of a quintom model of dark energy _ , b * 608 * ( 2005 ) 177182 ( astro - ph/0410654 ) ; + r. lazkoz , g. len , and i. quiros , _ quintom cosmologies with arbitrary potentials _ , phys . lett . b * 649 * ( 2007 ) 103110 , astro - ph/0701353 ; + m.r . setare , e.n . saridakis , _ quintom cosmology with general potentials _ , int.j.mod.phys . d * 18 * ( 2009 ) 549557 ( arxiv:0807.3807 ) ; + yi - fu cai , taotao qiu , yun - song piao , mingzhe li , xinmin zhang , _ bouncing universe with quintom matter _ , jhep * 0710 * ( 2007 ) 071 ( arxiv:0704.1090 ) i.ya . arefeva , a.s . koshelev , and s.yu . vernov , _ crossing the @xmath326 barrier in the d3-brane dark energy model _ , phys . d * 72 * ( 2005 ) 064017 ( arxiv : astro - ph/0507067 ) ; + s.yu . vernov , _ construction of exact solutions in two - field cosmological models _ , theor . * 155 * ( 2008 ) 544 ( arxiv : astro - ph/0612487 ) ; + i.ya . arefeva , n.v . bulatov , s.yu . vernov , _ stable exact solutions in cosmological models with two scalar fields _ , theor . * 163 * ( 2010 ) 788 ( arxiv:0911.5105 ) a.yu . kamenshchik , a. tronconi , g. venturi , and s.yu . vernov , _ reconstruction of scalar potentials in modified gravity models _ , phys . rev . d * 87 * ( 2013 ) 063503 ( arxiv:1211.6272 ) ; + e.o . pozdeeva and s.yu . vernov , _ stable exact cosmological solutions in induced gravity models _ // aip conf * 1606 * ( 2014 ) 4858 ( arxiv:1401.7550 ) i.ya . arefeva , n.v . bulatov , r.v . gorbachev , s.yu . vernov , _ non - minimally coupled cosmological models with the higgs - like potentials and negative cosmological constant _ , class . quantum grav . * 31 * ( 2014 ) 065007 ( arxiv:1206.2801 ) m. sami , m , shahalam , m. skugoreva and a. toporensky , _ cosmological dynamics of a nonminimally coupled scalar field system and its late time cosmic relevance _ d * 86 * ( 2012 ) 103532 ( arxiv:1207.6691 ) b.l . spokoiny , _ inflation and generation of perturbations in broken symmetric theory of gravity _ , phys . b * 147 * ( 1984 ) 3943 ; + t. futamase , k .- chaotic inflationary scenario in models having nonminimal coupling with curvature _ , phys . d * 39 * ( 1989 ) 399404 ; + r. fakir , w.g . unruh , _ improvement on cosmological chaotic inflation through nonminimal coupling , _ phys . d * 41 * ( 1990 ) 17831791 ; + m.v . libanov , v.a . rubakov , p.g . tinyakov , _ cosmology with nonminimal scalar field : graceful entrance into inflation _ , b * 442 * ( 1998 ) 63 ( arxiv : hep - ph/9807553 ) ; + a. cerioni , f. finelli , a. tronconi , g. venturi , _ inflation and reheating in induced gravity _ , phys . b * 681 * ( 2009 ) 383386 ( arxiv:0906.1902 ) ; r. kallosh , a. linde , d. roest , _ the double attractor behavior of induced inflation _ , j. high energy phys . * 1409 * ( 2014 ) 062 ( arxiv:1407.4471 ) ; + m. rinaldi , l. vanzo , s. zerbini , g. venturi , _ inflationary quasi - scale invariant attractors _ , phys . rev . d * 93 * ( 2016 ) 024040 ( arxiv:1505.03386 ) f.l . bezrukov , m. shaposhnikov , _ the standard model higgs boson as the inflaton , _ phys . b * 659 * ( 2008 ) 703 ( arxiv:0710.3755 ) ; + a.o . barvinsky , a.y . kamenshchik , a.a . starobinsky , _ inflation scenario via the standard model higgs boson and lhc _ , j. cosmol . . phys . * 0811 * ( 2008 ) 021 ( arxiv:0809.2104 ) ; + f. bezrukov , d. gorbunov , m. shaposhnikov , _ on initial conditions for the hot big bang _ , j. cosmol . * 0906 * ( 2009 ) 029 ( arxiv:0812.3622 ) ; + a.o . barvinsky , a.y . kamenshchik , c. kiefer , a.a . starobinsky , c.f . steinwachs , _ asymptotic freedom in inflationary cosmology with a nonminimally coupled higgs field _ , j. cosmol . phys . * 0912 * ( 2009 ) 003 ( arxiv:0904.1698 ) ; + a. de simone , m.p . hertzberg and f. wilczek , _ running inflation in the standard model _ , phys . b * 678 * ( 2009 ) 1 ( arxiv:0812.4946 ) ; + f.l . bezrukov , a. magnin , m. shaposhnikov , s. sibiryakov , _ higgs inflation : consistency and generalisations _ , j. high energy phys . * 1101 * ( 2011 ) 016 ( arxiv:1008.5157 ) ; + a.o . barvinsky , a.yu . kamenshchik , c. kiefer , a.a . starobinsky , c.f . steinwachs , _ higgs boson , renormalization group , and cosmology _ , eur . j. c * 72 * ( 2012 ) 2219 ( arxiv:0910.1041 ) ; + f. bezrukov , _ the higgs field as an inflaton _ , class . quant . grav . * 30 * ( 2013 ) 214001 ( arxiv:1307.0708 ) ; + j. ren , z .- z . xianyu , h.j . higgs gravitational interaction , weak boson scattering , and higgs inflation in jordan and einstein frames _ , j. cosmol . astropart * 1406 * ( 2014 ) 032 ( arxiv:1404.4627 ) d.i . kaiser , _ induced - gravity inflation and the density perturbation spectrum _ , phys . b * 340 * ( 1994 ) 2328 ( arxiv : astro - ph/9405029 ) ; + d.i . kaiser , _ primordial spectral indices from generalized einstein theories _ , phys . d * 52 * ( 1995 ) 4295 ( arxiv : astro - ph/9408044 ) r. kallosh and a. linde , _ universality class in conformal inflation _ , j. cosmol . * 1307 * ( 2013 ) 002 ( arxiv:1306.5220 ) . k. bamba , sh . nojiri and s.d . odintsov , _ reconstruction of scalar field theories realizing inflation consistent with the planck and bicep2 results _ , phys b * 737 * ( 2014 ) 374378 ( arxiv:1406.2417 ) e. elizalde , s.d . odintsov , e.o . pozdeeva , and s.yu . vernov , _ renormalization - group inflationary scalar electrodynamics and @xmath327 scenarios confronted with planck2013 and bicep2 results _ , phys . d * 90 * ( 2014 ) 084001 ( arxiv:1408.1285 ) ; + e. elizalde , s.d . odintsov , e.o . pozdeeva , s.yu . vernov , _ cosmological attractor inflation from the rg - improved higgs sector of finite gauge theory _ , j. cosmol . * 1602 * ( 2016 ) 025 ( arxiv:1509.08817 ) n.a . chernikov , e.a . tagirov , _ quantum theory of scalar fields in de sitter space - time _ , annales poincare phys . theor . a * 9 * ( 1968 ) 109 ; e.a . tagirov , _ consequences of field quantization in de sitter type cosmological models _ , annals phys . * 76 * ( 1973 ) 561 c.g . callan , s.r . coleman , r. jackiw , _ a new improved energy momentum tensor _ , annals phys . * 59 * ( 1970 ) 42

there is evidence that there is a correspondence@xcite between gravitational physics in anti de sitter space time and particular types of conformal field theory in one dimension fewer . this duality is a form of `` holography '' @xcite and a part of the correspondence operates by identifying the field theory as living on the boundary of anti de sitter ( ads ) space time . to be more precise , ads@xmath0@xmath1 is the space time of interest , and there is some @xmath2dimensional theory of gravity compactified on it . the manifold @xmath3 can be an @xmath4sphere , @xmath5 . the corresponding field theory is an @xmath6dimensional conformal field theory living on a space with the topology of the boundary of ads@xmath0 . the isometries of the manifold @xmath3 appear as global symmetries of the field theory : r symmetries if the theory is supersymmetric . this particular form of duality between gravity and field theory is certainly intriguing . the large @xmath7 limit ( where @xmath7 is the rank of the @xmath8 gauge group for the four dimensional yang mills field theory , with appropriate generalizations for other dimensions ) of the field theory at strong t hooft coupling corresponds to classical supergravity . as pointed out in ref . @xcite , following the observations in ref . @xcite , the old program of semi classical quantum gravity finds a new lease on life in this setting , as computations such as those performed with gravitational instantons ( at least in ads ) should have natural field theory interpretations . in this paper , we study the thermal properties of einstein maxwell ads ( emads ) charged black holes , and find behaviour consistent with field theory interpretations . we do this for arbitrary dimensions ( greater than 3 see section [ conclude ] for comments on @xmath9 ) and determine the thermal phase structure of the corresponding field theories . the cases of ads@xmath10 , ads@xmath11 and ads@xmath12 are particularly interesting of course , as they correspond to the theories found on the world volumes of m2 , d3 , and m5branes , respectively . the d3brane case is @xmath13 , @xmath14 supersymmetric @xmath8 yang mills theory , while the others are exotic superconformal field theories @xcite . we remark on the field theory interpretation of our new results in the light of holography . this paper is also of relevance beyond mere considerations of holography . some of the black hole solutions and their properties ( thermodynamic or otherwise ) are presented here for the first time . in particular , the lagrangian action calculations and subsequent determination of the phase structure are presented in their entirety here . in section [ sec : emads ] , we present an ansatz for obtaining the einstein maxwell truncation of gauged ads supergravity with appropriate compactifications of @xmath15 supergravity on @xmath16 , and @xmath17 type iib supergravity on @xmath18 . in the planar or infinite volume limit , the charged black holes in einstein maxwell anti desitter correspond to the near horizon limits of rotating m2 and d3branes . in section [ sec : holes ] , we display the solutions and note some of their properties . the computation of the action of the solutions using a euclidean section is performed in section [ sec : action ] , and their thermodynamic properties are uncovered in section [ sec : thermo ] . as the einstein maxwell anti desitter truncation is naturally associated to rotating branes , ( at least in the case of em ads@xmath10 and em ads@xmath11 , see section [ sec : emads ] ) it is very natural to suppose that there is an associated dual field theory , arising on the world - volume of some branes . these would be the familiar conformal field theories the @xmath13 , @xmath14 yang mills theory ( for coincident d3branes ) and the conformal field theory on the world volume of coincident m2branes . the case of em ads@xmath12 ( _ i.e. , _ without additional scalars ) is not related to a rotating brane truncation of the ads@xmath19 gauged supergravity ( because @xmath20 is even dimensional ) and so we can not declare that the dual field theory is the theory on the world volume of a rotating m5brane . however , we regard ads holography as a phenomenon which exists independently of string and m theory contexts@xcite . hence , in other dimensions beyond @xmath13 and 5 , we expect that there is a dual theory . in particular , for em the dual field theory is probably a close cousin of the m5brane theory . the dual field theories have their supersymmetry ( if they had any to start with ) broken due to coupling to a global background @xmath22 current ( as well as turning on a non zero temperature ) . the cft is in a thermal ensemble for which a certain @xmath22 charge density has also been `` turned on '' . in the ensemble , the expectation value of this charge breaks the global @xmath23 r symmetry of the cft . on the ads side , the electromagnetic charge carried by the black holes is in the same @xmath22 of corresponding @xmath23 gauge group . we find very interesting phase structures at intermediate temperatures ( in finite field theory volume ) , as a result of studying two complementary thermodynamic ensembles : we study thermodynamic ensembles with fixed background potential in which case the background is ads with a constant fixed potential and we also study a fixed localized charge ensemble , for which the background is an extremal black hole with that charge . in all cases , at sufficiently high temperature the physics is dominated by highly non extreme black holes , and we therefore recover the `` unconfined '' behaviour characteristic of the associated field theories@xcite . the finite horizon size of the black holes controls the behaviour of the expectation value of spatial wilson lines accordingly , yielding the area law behaviour , as follows from ref.@xcite . at intermediate temperatures , in the fixed charge ensemble , the presence of charge allows a new branch of black hole solutions to modify the qualitative phase structure in the low charge regime , resulting in a very interesting phase structure about which we will have more to say later in this section . intriguingly , as there is an extremal but non supersymmetric black hole with non zero entropy even at zero temperature , we must conclude something interesting about the field theory in the presence of the global background @xmath22 current : there must still be at @xmath24 a large number of states ( with the given charge ) available to the field theory in order to generate this entropy . for the case where we hold the potential ( _ i.e. , _ _ not _ the charge ) fixed , we do not expect that this is the ground state , because the extremal black hole can decay into kaluza klein particles , leaving ads . this is because the extremal black hole is not supersymmetric . this subtlety does not arise in the standard gibbons hawking calculus of the thermodynamics of black holes which we use here because the calculations are not sensitive to the ability of the black holes to emit charged particles . that the extremal black hole can decay by emitting charged kaluza klein particles here follows from the fact that the charge descends from rotation in higher dimensions . there are well known classical processes for reducing the rotation of objects like black holes by scattering@xcite , and therefore in the context of quantum field theory , one has the analogous processes of emission in superradiant modes@xcite . the same superradiant emission was considered in the context of charged black holes in ref.@xcite . thus one should expect the extremal black hole in the emads truncation to decay _ via _ such supperradiant emission . of course , the usual thermal hawking radiation may also tend to discharge nonextremal black holes@xcite . in the fixed potential ensemble , as the charge of the black hole is allowed to fluctuate while it is in contact with the thermal reservoir , superradiant and hawking emission processes can occur to reduce the charge of the black hole , allowing it to decay back to ads ( plus charge ) . however , in the fixed charge thermodynamic ensemble ( with varying potential ) , the extremal black hole is expected to be the long lived state at zero temperature . translating the formula for the entropy to the field theory we find , for example , that the four dimensional yang mills theory ( in the presence of the global background @xmath22 current ) has a zero temperature entropy which goes like @xmath25 for large black holes , where @xmath26 measures the total charge in units of the minimal charge of kaluza klein excitations ( _ i.e. , _ @xmath27 ) , and is proportional to the volume , @xmath28 , of the field theory . notice that the result for the four dimensional field theory is consistent with confinement at @xmath24 , as the result is independent of @xmath7 . confinement also follows from the fact that at @xmath24 , the euclidean section of the solution has no bolt , and therefore temporal wilson lines will always be homotopic to zero , and therefore have zero expectation value . meanwhile , spatial wilson lines can not interact with the horizon to produce an area law dependence , because at extremality the horizon recedes infinitely far away down a bertotti robinson throat . the phase structure which we obtain in each thermodynamic ensemble is summarized in figure [ fig : phases ] . 0.5 cm 2.0 cm the astute reader will recognize the figure on the right as the classic phase diagram of the liquid gas system ! to translate , our @xmath29 is like the temperature @xmath30 of the fluid while @xmath31 is like the pressure @xmath32 . the non extreme black holes of type ( 1 ) ( `` small '' ) and ( 3 ) ( `` large '' ) ( see section [ sec : action ] and [ sec : thermo ] for explanation ) are like the liquid phase and the gaseous phase , respectively . the critical line ( `` vapour pressure curve '' ) represents the place at which a first order phase transition between the liquid and gas occurs . as is well known , there is a critical temperature at which the vapour pressure curve terminates , representing the fact that above a critical temperature , one can convert a liquid to a gas continuously . this translates here into a critical charge above which the two types of black hole can be continuously converted into one another with no discontinuity in their size . that this system ( first modeled by van der waals@xcite , with a crucial modification by maxwell ) appears in this ads black hole thermodynamics is fascinating , and would not have been possible ( at least in this way ) without the presence of the extra branches of solutions which appear when there is negative cosmological constant . we discuss this further in sections [ sec : thermo ] and [ sec : action ] . further fascination may be found in the fact that the explicit shape of the free energy surface ( as a function of @xmath29 and @xmath30 ) is that of the classic `` swallowtail '' catastrophe , familiar from the study of bifurcations@xcite . the control surface of the `` cusp '' catastrophe also appears , which ( of course ) follows from the well known fact that it is the shape of the van der waals equation of state , viewed as a surface in @xmath33 space . that these shapes appear in this context suggests that there is some exciting universality to be explored here : catastrophe theory is largely a classification of the possible distinct types of bifurcation shapes that can occur in a wide variety of complex systems . this classification ( which , for the common `` elementary '' cases is of a e type ) is equivalent to the ( perhaps more familiar ) classfication of singularities@xcite . it is of considerable interest to discover just what circumstances might give rise to the other members of the classification . recalling that this all translates _ via _ holography into properties of a dual field theory , we would learn a great deal about universal phase structures which can occur there also . physics near the horizon of supergravity branes can be described in terms of spontaneous compactification of supergravity . in the case of non dilatonic branes which will be the focus of the paper when the compactification takes place on a _ round _ @xmath4sphere the low energy degrees of freedom are described by an effective theory of einstein gravity with a negative cosmological constant coupled to @xmath23 gauge fields . the schwarzschild anti desitter black hole solutions of this theory have been used in the context of the ads / cft correspondence to infer thermal properties of the dual field theories @xcite . a natural extension of this program is to study ads black holes which are charged under a subgroup of the gauge symmetry of the gauged supergravity . solutions of einstein maxwell anti desitter in some dimensions are known , but in the context of string / m theory , it is also interesting to determine how to make a truncation of the type iib supergravity , or of 11 dimensional supergravity , which gives the emads effective action . in other words , we must make certain higher dimensional choices which will result in the removal of the generic coupling of the @xmath34 term to scalars resulting from the kaluza klein reduction . amusingly , one simple way to introduce ( gauge ) charge on the black holes is by simply spinning or twisting the transverse ( angular ) sphere that becomes the compact space . decoupling of the scalars is accomplished by choosing the spins in a maximally symmetric way . to be concrete , take ten dimensional iib supergravity , with the metric ansatz @xmath35 , \label{ansatz}\ ] ] where @xmath36 is a five - dimensional metric , @xmath37 , the variables @xmath38 are direction cosines on @xmath18 ( and therefore are not independent , @xmath39 we follow the notation of@xcite ) , and the @xmath40 are rotation angles on @xmath18 . the ansatz for the rr 5-form field strength has `` electric '' components @xmath41 while the dual `` magnetic '' components are given by @xmath42 . in eqn . ( [ forms ] ) , @xmath43 is the volume form on the reduced five - dimensional space , and @xmath44 denotes hodge duality on this space . the parameter @xmath45 measures the size of the @xmath18 and is given by the flux of the 5form field across the @xmath18 . notice that a component @xmath46 in the time direction is interpreted as rotation of the @xmath18 in its three independent rotation planes , in equal amounts . components in the spatial direction would instead be ` twists . ' for the sake of brevity , and since in this paper we will be mainly considering @xmath46 components , one can not define magnetic ( vector ) charges on the black holes . ] , we will refer collectively to them as `` rotations '' . with this ansatz ( [ ansatz ] ) , the effective action in the five non compact dimensions becomes @xmath47.\ ] ] this is precisely the einstein maxwell anti desitter ( emads ) effective action we seek , with a chern simons term . the latter is indeed required by supersymmetry in @xmath48 five dimensional gauged supergravity @xcite , whose bosonic sector is precisely described by the action above . note that the gauge coupling is proportional to @xmath49 . the ads@xmath50 gauged supergravity theory in five dimensions has an @xmath51 gauge symmetry , associated with the group of isometries of @xmath18 . this is the r symmetry group of the dual four dimensional @xmath14 superconformal yang mills field theory living on the d3branes from which this near horizon geometry arose . the above spinning compactification corresponds to introducing rotation in the diagonal @xmath22 of the maximal abelian subgroup @xmath52 . correspondingly , there must be a dual field theory to the emads truncation , which is simply the field theory on the world volume of the rotating brane . from the field theory point of view , the rotation corresponds to considering states or ensembles in which the dual global @xmath22 current ( a subgroup of the @xmath51 r symmetry group ) has a nonvanishing expectation value . studying emads gravity and its solutions will therefore be equivalent to studying properties of the conformal field theory in the presence of this background current vector fields , each associated with the three different independent rotations of @xmath18 , and two scalars that , roughly , measure the relative sizes of the distortions of the @xmath18 caused by rotation . for simplicity , we will restrict ourselves to the case where all three rotations have the same magnitude , since it is only in this case that the scalars decouple and we find emads gravity . this framework provides the cleanest interpretation in terms of the dual cft , since the number of spin parameters or charges precisely matches the number of field theory operators which are `` excited '' . see refs . @xcite for discussion of more general actions and solutions related to this . ] . a similar construction can be obtained by starting from eleven dimensional supergravity . the compactification in this case is equivalent to focusing on the near horizon region of m2branes . in this case , take @xmath53,\ ] ] leading to the ads@xmath10 theory with a maxwell term @xmath54.\ ] ] the reduction ansatz for the 4-form field strength is @xmath55 where @xmath56 is the volume form on the reduced four dimensional space , and @xmath57 denotes hodge duality on this space . chern simons terms are absent in four dimensions . appropriate inclusion of fermions leads to four dimensional @xmath48 gauged supergravity . the more general @xmath58 theory with four independent gauge fields ( _ i.e. _ , four different rotation parameters ) 3 scalars and @xmath59 supersymmetry , as well as its black hole solutions , have been recently studied in ref . @xcite . we note here that there is no analogous construction for the ads@xmath19 gauged supergravity theory . this is because @xmath20 is even dimensional and therefore we can not have a symmetric split between @xmath22 rotations , as @xmath60 does not have an even torus for its cartan subalgebra . this means that we can not relate the physics of the black hole solutions ( which we write later ) of the em ads@xmath12 system to the physics of rotating m5branes of eleven dimensional supergravity . nevertheless , as ads holography is a phenomenon which is expected to exist independently of string or m theory realizations , we expect that the physics does have a holographic interpretation in terms of a field theory closely related to that which resides on m5brane world volumes . the black hole solutions of the above supergravity theories in @xmath61 were originally studied in the past in refs . @xcite and @xcite more recent investigations appear in refs . @xcite . as we have seen in the previous section , such theories can be regarded as compactifications of the type iib and @xmath15 supergravities , where the gauge symmetry groups of the gauged supergravities are broken by a specific choice of rotation planes in the transverse compact spheres . given these considerations , it is natural to study the reissner nordstrom anti desitter ( rnads ) black holes within the context of the ads / cft correspondence . even if the bosonic einstein maxwell anti desitter ( emads ) theories admit supersymmetric extensions only in certain dimensions , it is easy and convenient to perform the analysis of their black hole solutions for arbitrary dimension . for space time dimension @xmath62 , the action can be written as so as to absorb the prefactors in the action . ] @xmath63 , \label{actionjackson}\ ] ] with @xmath64 the cosmological constant associated with the characteristic length scale @xmath45 . then the metric on rnads may be written in static coordinates as @xmath65 where @xmath66 is the metric on the round unit @xmath67sphere , and the function @xmath68 takes the form @xmath69 here , @xmath4 is related to the adm mass of the hole , @xmath70 ( appropriately generalized to geometries asymptotic to ads@xcite ) , as @xmath71 where @xmath72 is the volume of the unit @xmath67sphere . the parameter @xmath73 yields the charge @xmath74 of the ( pure electric ) gauge potential , which is @xmath75 where @xmath76 and @xmath77 is a constant ( to be fixed below ) . if @xmath78 is the largest real positive root of @xmath68 , then in order for this rnads metric to describe a charged black hole with a non singular horizon at @xmath79 , the latter must satisfy @xmath80 finally , we choose @xmath81 which then fixes @xmath82 . the physical significance of the quantity @xmath77 , which plays an important role later , is that it is the electrostatic potential difference between the horizon and infinity . if the inequality in eqn . ( [ extbound ] ) is saturated the horizon is degenerate and we get an extremal black hole . this inequality imposes a bound on the black hole mass parameter of the form @xmath83 . in the cases where the theory admits a supersymmetric embedding one could naively expect to approach a supersymmetric state as we saturate this mass bound . however , the bound that results from the supersymmetry algebra is instead @xcite : @xmath84 , with the @xmath85 solution being a bps state , where the black hole can have magnetic charge @xmath86 , there is a magnetic ( or dyonic ) bps solution as well @xcite with @xmath87 , @xmath88 . ] . now , it is easy to see that the mass of the extremal black hole , @xmath89 is , for finite @xmath45 , always strictly larger than @xmath90 and therefore the extremal solution is non supersymmetric . on the other hand , for the supersymmetric solution one has @xmath91 which is strictly positive everywhere and therefore one finds a naked curvature singularity at @xmath92 . in fact , all the solutions violating the bound ( [ extbound ] ) are nakedly singular . in the context of the ads / cft correspondence it is interesting to consider the limit where the boundary of ads@xmath0 is @xmath93 instead of @xmath94 as was the case above . this can be regarded as an `` infinite volume limit '' , with particular relevance to the discussion of the dual field theory . it should be noted that the existence of black hole solutions in this limit is possible only due to the presence of a negative cosmological constant . in fact , black holes ( and other bolts ) in ads spaces with varied topologies ( even other than spherical and toroidal ) have been extensively studied in recent years @xcite , including in the m theory @xcite . here we will only focus on the planar ( toroidal ) solutions , which we will obtain by scaling the `` finite volume '' solutions above , as done in @xcite . to this effect , introduce a dimensionless parameter @xmath95 ( which we will shortly take to infinity ) and set @xmath96 while at the same time blowing up the @xmath97 as @xmath98 . one finds , after taking @xmath99 , @xmath100 with @xmath101 for the supersymmetric solution , the scaling is as above except for the scaling of @xmath4 . to preserve supersymmetry , one must fix @xmath85 and so @xmath102 yielding @xmath103 notice that , compared with eqn . ( [ scaledu ] ) , the parameter @xmath4 is zero in this limit . the resulting solution can be seen to be supersymmetric as well ( _ i.e. _ , the killing spinors remain finite in the limit @xmath104 , after appropriate rescaling ) , and nakedly singular . in this `` infinite volume '' limit , the solutions asymptote to ads space with the horospheric slicing . these planar solutions can be constructed with the appropriate decoupling limit@xcite of spinning d3 or m2branes , as mentioned previously . we refer the reader to ref.@xcite for the details . the study of the euclidean section ( @xmath105 ) of the solution , identifying the period , @xmath106 , of the imaginary time with inverse temperature , will define for us the grand canonical thermodynamic ensemble ( for fixed electric potential ) or the canonical ensemble ( for fixed electric charge ) . we interpret this in terms of immersing the system into a thermal bath of quanta at temperature @xmath107 . for pure ads , the background consists of both charged and uncharged quanta free to fluctuate in the presence of fixed potential @xmath77 . later , we consider the fixed @xmath29 ensemble . in that case we localize all of the charge at a specific region and keep it fixed . for such a background , as ads with a localized charge is _ not _ a solution of the emads equations , we use the extremal black hole solution as the background , and retain only neutral quanta in the thermal reservoir , in order to keep the charged fixed . this makes sense , even though the extreme limit has zero temperature , since the euclidean section has no bolt and so can be assigned an arbitrary periodicity@xcite . hence , the metrics and gauge fields can be matched in the asymptotic region . with all of this in mind we now turn to the action calculations . with our conventions the full euclidean action is given by analytically continuing eqn . ( [ actionjackson ] ) , where , as usual when the space is asymptotically ads the gibbons hawking boundary term gives a vanishing contribution . the boundary terms from the gauge field will vanish if we keep the potential @xmath46 fixed at infinity . any possible chern simons term will not contribute when we restrict ourselves to purely electric solutions . imposing the equations of motion we can eliminate the factors of @xmath108 in order to obtain the on shell action @xmath109.\ ] ] we obtain for the action ( subtracting the ads background while remembering to match the geometries of the background and black hole in the asymptotic region ) : @xmath110 here , @xmath106 denotes the period of the euclidean section of the black hole space time . using the usual formula for the period , @xmath111 , a little algebra yields the explicit form @xmath112 this may be rewritten in terms of the potential as : @xmath113 note that the temperature is zero when the black hole is extremal . this is because the horizon is degenerate there , and @xmath106 diverges , together with the fact that one can smoothly approach the extremal limit from non zero temperature . from the form of the equation for @xmath106 , it is apparent that there are qualitatively two distinct types of behaviour , determined by whether @xmath77 is less than or greater than the critical value @xmath114 . in particular , for @xmath115 , @xmath106 diverges ( @xmath30 vanishes ) at @xmath116 , while for @xmath117 , @xmath106 goes smoothly towards zero as @xmath118 . it is instructive to plot the temperature as a function of horizon radius ( size of black hole ) for these two regimes ( see fig . [ betaplot3 ] ) . -0.5 cm as can be seen from the figure , the regime of large potential ( _ i.e. , _ @xmath119 ) has a unique black hole radius associated with each temperature . we will see later that this branch dominates the thermodynamics for all temperatures . meanwhile the small potential regime has two branches of allowed black hole solutions , a branch with larger radii and one with smaller . this is qualitatively similar to the familiar case of the uncharged schwarzschild black holes analyzed in ref . @xcite , ( or the structure of the taub bolts discovered in the thermodynamic studies of ref . @xcite ) , which is the @xmath120 limit of the situation here . correspondingly , the smaller branch of holes is unstable , having negative specific heat . they do not play any role in the physics . ( generally , the sign of the specific heat for a black hole of radius @xmath78 can be inferred from the local slope of the @xmath121 curve . see also the discussion in section [ cats ] . ) if we wish to consider a situation where instead of the potential at infinity , we fix the charge of the black hole , then the action ( [ actionone ] ) is not appropriate . upon variation of the gauge field in the latter action , a boundary term results that vanishes only if we keep @xmath122 fixed . that is , the on shell action of the previous subsection is @xmath123 $ ] . if , instead , we want to keep the charge fixed , then we must add a boundary term to @xmath124 @xcite , @xmath125 where @xmath126 is a radial unit vector pointing outwards ( notice that this boundary term is determined by the terms coming from the variation of the off shell action ( [ actionjackson ] ) , and not ( [ master ] ) , which is on gravity shell . this distinction is only relevant for @xmath127 ) . then we get a thermodynamic function @xmath128 $ ] , in terms of the variables we wish to control . to compute the action for the fixed charge ensemble , using as background the extremal black hole , we evaluate eqn . ( [ master ] ) for a black hole of mass @xmath129 ( and radius @xmath130 ) , and then subtract the contribution from the extremal background . remembering to match the geometries of the background and black hole in the asymptotic region , a straightforward calculation yields the final result @xmath131 . \label{actiontwo}\ ] ] the inverse temperature , @xmath106 , is given by eqn . ( [ betaform ] ) . it is useful to plot the temperature as a function of horizon radius ( size of black hole ) for future use . there are two basic scales in this expression for @xmath132 , set by @xmath73 and @xmath45 , and so we expect that there will be two distinct regimes which may display distinct phase structure : @xmath133 and @xmath134 . for comparison , we also show the case of @xmath135 ( see fig . [ betaplot1 ] ) . the critical charge @xmath136 is the value of @xmath73 at which the turning points of @xmath121 appear or disappear . with @xmath137 , the periodicity @xmath138 will have a point of inflection with respect to @xmath78 derivatives . hence we can simultaneously satisfy @xmath139 with @xmath140 and @xmath141 . a little algebra then yields @xmath142 therefore we have for @xmath143 , @xmath144 and for @xmath145 , @xmath146 -0.5 cm in this case , the figures show that for small charge ( _ i.e. _ , below @xmath136 ) , there can be _ three _ branches of black hole solutions , to which we will refer later . the middle branch is unstable ) , @xmath147 while the branch with the smallest radii is new , and will play an interesting role in the thermodynamics . for zero charge , we return to the familiar two branch situation of schwarzschild , while for large charge , we have a situation analogous to that seen for the large fixed potential . this is the grand canonical ensemble , at fixed temperature and fixed potential . the grand canonical ( gibbs ) potential is @xmath148 using the expression in eqn . ( [ actionone ] ) , we may compute the state variables of the system as follows : @xmath149 together , they indeed satisfy the first law : @xmath150 . in order to study the phase structure and stability , we must observe the free energy @xmath151 as a function of the temperature . it is shown in figure [ freeplot1 ] . -.5 cm the interpretation of this is as follows . at any non zero temperature , for large potential ( @xmath152 ) the charged black hole is thermodynamically preferred , as its free energy ( relative to the background of ads with a fixed potential ) is strictly negative for all temperatures . this behaviour differs sharply from the small potential ( @xmath117 ) situation , which is qualitatively the same as the uncharged case : in that situation , in finite volume , the free energy is positive for some range @xmath153 , and it is only above @xmath154 that the thermodynamics is dominated by schwarzschild black holes ( the larger , stable branch ) , after their free energy is negative . ( see the center graphs in figure [ freeplot1 ] . ) so for high enough temperature in all cases the physics is dominated by non extremal black holes . in this case , ( after converting gravitational to field theory quantities , @xmath155 and @xmath156 ; for @xmath145 , @xmath157 and @xmath158 ; for @xmath159 , @xmath160 and @xmath161 . ] ) the free energy and entropy behave at ultra high temperature as @xmath162 where @xmath163 is the @xmath67dimensional spatial volume upon which the field theory lives . this is the `` unconfined '' behaviour appropriate to the dual @xmath6dimensional field theory . the function @xmath164 is @xmath165 when @xmath145 ; @xmath166 when @xmath143 ; @xmath167 when @xmath159 . the resulting power of @xmath7 shows how the number of unconfined degrees of freedom of the theory goes with @xmath7 , by analogy with the case of @xmath145 where @xmath168 counts the dependence on the number of degrees of freedom on @xmath7 for an @xmath8 gauge theory . at low temperatures , and for @xmath152 , we have something very new . notice that as we go to @xmath24 , the free energy curve approaches a maximum value which is less than zero . this implies that even at zero temperature the thermodynamic ensemble is dominated by a black hole . from the temperature curve ( [ betaplot3 ] ) it is clear that it is the extremal black hole , with radius @xmath169 . for @xmath170 , at @xmath24 we recover ads space . so this suggests that even at zero temperature the system prefers to be in a state with non zero entropy ( given by the area of the black hole ) . notice that this @xmath24 situation displays the `` confined '' behaviour characteristic of the ordinary conformally invariant zero temperature phase , despite the presence of the black hole . this follows from the fact that the temporal wilson lines will still have zero expectation value , as the fundamental strings which define them can not wind the horizon which has infinite period at zero temperature . similarly , spatial wilson lines will not display the area law behaviour , because the fundamental string world sheets can not be obstructed by the horizon , because at extremality , it is infinitely far away down a throat . having pointed out this intriguing possible zero temperature behaviour , we expect that for the case of _ fixed potential _ considered here , this is not the complete story . we must allow for the possibility that the extremal black hole might decay due to processes involving kaluza klein particles charged under the @xmath22 . ( see the discussion near the end of section i. ) this possibility can not be discounted because the extremal black hole is not supersymmetric , as pointed out before , and therefore not guaranteed to be stable by the supersymmetry algebra . we expect that calculations which include the effects of charge emission will shift the free energy back to zero , representing the true , equilibrium situation . alternatively if we consider the action ( [ actionjackson ] ) on its own merit outside of string or m theory compactifications , it may be regarded as part of a theory without fundamental charged particles . the resulting thermodynamic phase structure for the fixed potential ensemble is summarized in the left diagram of figure [ fig : phases ] . we have seen that we may consider a @xmath24 background containing an extremal black hole of charge @xmath29 . let us now keep this charge fixed and allow the potential at infinity to vary . this is the canonical ensemble , and the corresponding thermodynamic potential , the free energy , is @xmath171 the energy , entropy and electric potential are computed as @xmath172 in this case @xmath173 measures the energy above the ground state , which is the extremal black hole . together , they satisfy the first law , which in this case should be written as @xmath174 . the free energy as a function of temperature is shown below for the cases of small and large charge , respectively ( compare to the third graph in figure ( [ freeplot1 ] ) for the uncharged case ) . -0.5 cm that there are three branches for the small charge case follows from the second graph in figure [ betaplot1 ] , which is magnified and labeled in fig . [ freeplot3 ] , on the right . from there , it is clear that for low temperature there can only be one solution ( `` branch 1 '' ) for the black hole radius . at some temperature @xmath175 , the origin of two new branches ( `` branches 2 and 3 '' ) of solutions appears ( @xmath176 for the chosen parameters in the plot . ) . above this temperature ( below @xmath177 ) , there are therefore three distinct branches of solution until at temperature @xmath178 , ( @xmath179 in the plot ) two of the branches ( 1 and 2 ) coalesce and disappear , leaving again only a single branch ( 3 ) , which persists for all higher temperatures . returning to the free energy plot , the meaning is now clear . starting to the extreme left of the plot , ( low temperature ) we see that there is a single branch of free energy , corresponding to the branch 1 solutions . at @xmath180 , branches 2 and 3 appear on the graph and separate from each other at higher temperatures . at @xmath181 , branches 1 and 2 coalesce and disappear , while branch 3 persists for all higher temperatures , continuing to the left . so from zero temperature the negative free energy of branch 1 means that those non extreme black holes dominate the thermodynamic ensemble . at temperature @xmath154 , ( @xmath182 in the plot ) the free energy of branch 3 is actually more negative than that of branch 1 , and so that branch of non extremal black holes takes over the physics and continue to do so for all higher temperatures . the situation at @xmath154 is a genuine finite temperature phase transition , of first order . ( notice from the first graph in figure [ freeplot3 ] that the free energy is continuous , but its first derivative is discontinuous . ) this results from the jump ( along the dotted line in the final graph in figure [ freeplot3 ] ) from branch 1 to branch 2 , from small to large @xmath78 black holes , as the temperature increases . as the entropy is proportional to @xmath183 , there is a jump in the entropy , or a release of `` latent heat '' . as we approach the critical value , @xmath136 , of the charge representing the crossover into the large charge regime , the kink in the free energy and therefore the transition vanishes , as branches 1 and 3 merge ( and branch 2 disappears ) . the difference in horizon radii between the two branches , @xmath184 , may be thought of as an order parameter for the transition , as it vanishes above @xmath136 , where the transition goes away . as noted before in the case of fixed potential ensemble , branches 2 and 3 are the exact analogues of the small and large schwarzschild black holes of hawking and page@xcite , or the small and large taub bolts discovered in the thermodynamic studies of ref . @xcite . in those papers , above a certain temperature @xmath180 , there were two allowed solutions at a given temperature , the smaller ( branch 2 ) being unstable and the larger ( branch 3 ) being stable , which persists to dominate the thermodynamics above some critical temperature @xmath154 . the existence of a stable branch 1 , and its merger with branch @xmath165 to disappear at @xmath181 is a new feature when we add a small fixed charge to the story . conversely , if we start from a situation where charge is present on the black hole but the cosmological constant vanishes , then we find branches 1 and 2 , and it is only when the negative cosmological constant is turned on that branch 3 appears . for large charge , there is only a single branch allowed , ( see figure [ freeplot3 ] , the cusps collide and disappear ) and the associated thermodynamic story is correspondingly simpler . the free energy shows that the non extreme charged black holes dominate from @xmath24 . in all cases ( large or small @xmath29 ) , the ultra high temperature phases are dominated by a black hole and the free energy and entropy have the characteristic `` unconfined '' field theory behaviour shown in eqns . ( [ behave ] ) . one might examine the approach to the critical point more closely . in particular , consider the behaviour of the specific heat @xmath185 with @xmath141 , as the temperature approaches the critical value , one finds a singularity with @xmath186 . this behavior may be contrasted with the @xmath187 singularity found in ref.@xcite . the essential difference is of course that near the critical point we have a point of inflection with @xmath188 , while ref.@xcite considers a minimum with @xmath189 . the evolution of the free energy of the system as a function of charge is particularly interesting as one goes from zero charge to large charge . the single cusp of the uncharged ( schwarzschild ) system is joined by a second cusp which comes in from infinity , forming ( with the original one ) a section of the well known `` swallowtail '' shape , familiar as a bifurcation set or `` catastrophe '' in singularity or catastrophe theory . the significance of this is discussed in the next section . as we cross over into the large charge regime at some critical value of @xmath73 , the cusps merge and the free energy becomes a simple monotonic function . for completeness , we include a series of plots showing this evolution . ( we do not put them on the same axes , as we did for the fixed potential case , for the sake of clarity . ) -0.5 cm 0.5 cm -0.5 cm the resulting thermodynamic phase structure for the fixed charge ensemble is summarized in the diagram on the right in figure [ fig : phases ] . we can not refrain from further general comments upon the meaning and structure of the curves that we have uncovered in the previous sections . although we plotted only the cases for the @xmath145 case , representing ads@xmath11 ( and hence four dimensional field theory ) , the same universal structures appear in the cases @xmath143 and 6 as well , giving the same pleasing phase structure for the fixed charge ensemble . the phase structure that we uncovered for the fixed charge ensemble should remind the reader of the classic van der waals maxwell behaviour , modeling the liquid gas system . indeed , they are isomorphic . the @xmath121 curve ( the middle graph of figure [ betaplot1 ] ) should recall the graph of the @xmath190 van der waals equation of state , where @xmath32 ( the pressure ) is replaced here by @xmath106 and @xmath191 ( volume ) by @xmath78 . the instability of branch 2 is then simply the familiar instability of the corresponding section of the van der waals curve . the jump from branch 1 to branch 3 which we deduced from the form of the free energy is the precise analogue of the maxwell construction . in the isomorphism between our parameters and those of the van der waals maxwell system , our charge @xmath29 is equivalent to their temperature @xmath30 . the instability of branch 2 in both languages makes intuitive sense : as one increases the pressure , the volume should _ decrease _ , and therefore the positively sloped branch is not stable . a similar statement holds for the black holes after making the translation to the current situation : for black holes in equilibrium with the heat bath , an increase in the temperature results in an increase in the black hole radius and hence mass , for stable black holes . notice that this also follows from the first law , recalling that the entropy is a positive power of the radius . so the positive slope branch of the @xmath121 curve is generally unstable . in the language of catastrophe theory@xcite the study of jumps in some `` state variables '' as a result of smooth changes in `` control variables'' the physical solutions of the @xmath192 curve , viewed as a two dimensional surface in @xmath193 space , is the `` control surface '' of the `` cusp catastrophe '' . the cusp shape is the union of points in the @xmath194 plane ( the control variables ) where the state variable ( the allowed value of @xmath78 ) jumps from branch 1 to branch 3 , as branch 2 is unstable . after applying the minimum free energy condition to determine the allowed branches ( the `` maxwell criterion '' ) , the cusp catastrophe appears in the @xmath195 plane , ( or equivalently the @xmath196 plane ) collapsed to the critical line ( see figure [ fig : phases ] ) ( or `` vapour pressure curve '' ) along which the two types of black hole can coexist , and across which there is a phase transition . the end of the line , at the critical value @xmath136 , where branch 2 disappears , is the point where the distinction between branch 1 and 3 goes away . the order parameter , @xmath197 for this critical point is the radius difference of the branches @xmath198 . beyond the critical charge there is no phase transition ( @xmath199 ) in going from branch 1 black holes to branch 3 by increasing the temperature . this is of course the familiar statement that above a critical temperature , there is no phase transition in going from a gas to a liquid by increase of pressure . intriguing is the fact that the two dimensional free energy surface @xmath200 forms the shape of the swallowtail catastrophe ( see figure [ freeplots ] ) . ( note that for @xmath201 and @xmath202 the shape is the same . ) this naturally follows from the ability of the @xmath121 curve to produce three branches , and the resulting shape for the free energy curve is the union of three branches . here , the swallowtail does not have the usual interpretation as a bifurcation surface ( like the cusp does above ) but it is natural to wonder whether its appearance tells us that there is some universality at work here . this is because the language of catastrophe theory is largely a classification of the possible distinct types of bifurcation shapes that can occur . this classification ( which , for the common `` elementary '' cases is of a e type ) is equivalent to the ( perhaps more familiar ) classfication of singularities@xcite . a natural question is whether or not the inclusion of more control parameters will always result in a free energy curve of a shape ( and corresponding phase structure ) which falls into the classification . it would certainly be amusing to find yet another case of the a e classification appearing in string and m theory physics . the study of the thermodynamics of black holes in einstein maxwell anti desitter is highly relevant to the thermodynamics of certain superconformal field theories with a background global current switched on . this follows from the logic of the ads / cft correspondence , and the fact that the emads system can arise as the near horizon physics of rotating m2 and d3branes , and it should therefore be regarded as the effective theory of the strongly coupled field theory living on the rotating branes world volume . the phase structures of the charged black hole systems studied here , and summarized in figure [ fig : phases ] , are markedly different from those of the uncharged systems studied before in this context @xcite . the addition of charge revealed a rich phase structure , with precise analogues to classic thermodynamic systems . the physics is consistent with a dual field theory interpretation . in all cases , the infinite volume limit can be found by taking the limits given in eqn . ( [ scaled ] ) . this scaling may be applied to the expressions for the actions ( eqns . ( [ actionone ] ) and ( [ actiontwo ] ) ) and the period ( eqns . ( [ betaform ] ) and ( [ betaformtwo ] ) ) . in all cases , the result is that there is only one branch of black hole solutions ( like the large charge and potential situations had in finite volume ) , and the free energy is negative definite , showing that the thermodynamics is dominated by black holes for all temperatures . of course , this is what we should expect , from the field theory point of view . as we commented before , the gauge field in the ads space naturally couples to a cft current @xmath203 , following the prescription of ref . @xcite . from the asymptotic variation of the gauge field ( [ pure ] ) or its corresponding field strength , one then has an expectation value @xmath204 . thus one might think of the cft state as containing a plasma of ( globally ) charged quanta . the precise nature of the cft state depends on the ensemble , which we were studying . for the case of the fixed potential , the dual statement is that a chemical potential conjugate to the global charge has been introduced leading to the expectation value . the fixed charge calculations correspond to an ensemble of cft states with a fixed global charge . thus the difference between the two calculations is analogous to that between the canonical ( fixed @xmath30 ) and microcanonical ( fixed @xmath173 ) ensembles . in the context of d3 branes with @xmath145 , the @xmath51 gauge fields couple to the r symmetry currents in the super yang mills theory . this aspect of the duality has been used to great advantage to produce nontrivial consistency tests by comparing correlators protected by supersymmetry@xcite . of course in the present case , with the truncation to emads theory , we are focussed on a particular diagonal @xmath22 generator of the @xmath51 symmetry . in this context , we can translate the results of the supergravity calculations to quantitative statements about the strong coupling behavior of the super yang mills theory . up to numerical factors , we have as usual@xcite : @xmath205 and @xmath206 , ( where @xmath207 is the type iib string coupling ) as well as @xmath208 . it remains fix how the black hole charge should be characterized in the cft . the most natural approach is to measure the physical charge ( [ thecharge ] ) in terms of the fundamental charge of the kaluza klein excitations in the ads space , _ i.e. _ , with @xmath209 . in the field theory then , @xmath210 ( where @xmath28 is the spatial volume of the field theory ) essentially counts the number of fundamentally charged quanta per unit volume in a given state . given this framework , we can consider the field theory content of our results . for example , one might wonder what the critical charge ( [ thecritics ] ) appearing in the fixed charge phase diagram corresponds to : @xmath211 in general , translating the entropy , mass or free energy to a field theory expression produces a complicated function of both the temperature @xmath30 and the charge @xmath26 . one relatively simple case is the high temperature limit , where the charge essentially plays no role ( see eqn . ( [ behave ] ) ) . another interesting case to consider is that of the extremal black holes for which @xmath24 . by demanding that @xmath212 and @xmath213 have a consistent solution , one finds that the mass and charge parameters are related by the following expression : @xmath214 where @xmath215 and @xmath216 . a simple case to consider is that of a large black hole with @xmath217 , for which @xmath218 . further in this limit , one has that @xmath219 and so @xmath220 notice that implicitly here we are considering a regime where @xmath221 . the lack of dependence of the entropy on @xmath7 is a signal of confined behaviour at zero temperature , despite the presence of the black hole . it would certainly be interesting if this entropy result could be recovered by considering partitioning of the charge @xmath26 amongst the charged excitations of the cft . we have left aside the case of compactification of six dimensional supergravity on @xmath222 to get ads@xmath223 . by setting the @xmath222 in rotation in its two independent rotation planes , in a symmetric fashion , we get an electric potential in ads@xmath223 . doing so , notice that if we start from the solution describing a rotating six dimensional black string ( such as the one obtained from the d1d5 bound state ) , then , in the throat limit the rotation of the @xmath222 can be undone by a diffeomorphism@xcite . in other words , the effective gauge field in three dimensions is pure gauge . nevertheless , as shown in ref.@xcite , there do exist charged black hole solutions in emads theory in three dimensions . these have an electric potential that diverges logarithmically at infinity , which prevents one from defining the ensemble at fixed potential . nevertheless , if the extremal black hole background is subtracted , then the fixed charge ensemble can be appropriately defined . for non rotating black holes ( in ref.@xcite , the full kerr newman solution is constructed ) there is only one branch , just like we have found for large fixed charge ( see figure [ betaplot1 ] , left ) , with corresponding simple thermodynamic structure given by figure [ freeplot1 ] ( left ) . finally , it is also worth remarking that the close similarity that we have observed with familiar structures from equilibrium thermodynamics , and expectations from a dual field theory is further encouragement ( for those who need it ) that the quantum mechanics of black holes is not unlike that of other situations . ac is supported by pembroke college , cambridge . re is supported by epsrc through grant gr / l38158 ( uk ) , and by grant upv 063.310eb187/98 ( spain ) . support for cvj s research , and other support of this project , was provided by an nsf career grant , # phy9733173 ( uk ) . rcm s research was supported by nserc ( canada ) and fonds fcar du qubec . this paper is report # s damtp199929 , ehu ft/9902 , dtp99/9 , uk/9902 and mcgill/9907 . we would like to thank neil constable , mike crescimanno , susan gardner , harry lam , guy moore , malcolm perry , and alfred shapere for comments and useful conversations . ac , re and rcm would like to thank the department of physics and astronomy of the university of kentucky for its hospitality , and we would all like to thank the staff of the william t. young library ( university of kentucky ) for the use of excellent meeting facilities and other services . 99 j. maldacena , adv . theor . math * 2 * ( 1998 ) 231 , hep - th/9711200 s. s. gubser , i. r. klebanov , and a. m. polyakov , phys . b428 * ( 1998 ) 105 , hep - th/9802109 . e. witten , adv . * 2 * ( 1998 ) 253 , hep - th/9802150 . r. penrose , nuovo cimento 1 ( 1969 ) 252 . starobinsky , j.e.t.p . 37 ( 1973 ) 28 ; + w. unruh , phys . d10 ( 1974 ) 3195 ; + n. deruelle and r. ruffini , phys . lett . 52b ( 1974 ) 437 ; _ ibid . _ , 57b ( 1975 ) 248 . g. denardo and r. ruffini , phys . lett . 45b ( 1973 ) 259 ; + t. damour and r. ruffini , phys . 35 ( 1975 ) 463 . gibbons , commun . 44 ( 1975 ) 245 . s.w . hawking , taylor - robinson , phys.rev . * d55 * ( 1997 ) 7680 , hep - th/9702045 . d.n . page , phys . d14 ( 1976 ) 3260 . see , for example r. m. wald , ann . 82 ( 1974 ) 548 ; + r.m . `` general relativity '' _ , the university of chicago press , chicago ( 1984 ) ; + r.m . `` quantum field theory in curved space time and black hole thermodynamics '' _ , the university of chicago press , chicago ( 1994 ) ; + n.d . birrell and p.c.w . davies , _ `` quantum fields in curved space '' _ , cambridge university press , cambridge ( 1982 ) ; + p. r. brady , i. g. moss , r. c. myers , phys . * 80 * ( 1998 ) 3432 , gr - qc/9801032 . k. behrndt , a.h . chamseddine and w.a . sabra , phys . lett . b442 ( 1998 ) 97 , hep - th/9807187 ; + k. behrndt , m cvetic and w.a . sabra , hep - th/9810227 . russo , hep - th/9808117 ; + c. csaki , y. oz , j.g . russo and j. terning , phys . d59 ( 1999 ) 065008 , hep - th/9810186 ; + p. kraus , f. larsen and s.p . trivedi , hep - th/9811120 ; + j.g . russo and k. sfetsos , hep - th/9901056 . gubser , hep - th/9810255 ; + r.g . cai and k.s . soh , hep - th/9812121 . mann , class . * 14 * , l109 ( 1997 ) gr - qc/9607071 . + d.r . brill , j. louko and p. peldan , phys . rev . * d56 * , 3600 ( 1997 ) gr - qc/9705012 . + l. vanzo , phys . * d56 * , 6475 ( 1997 ) gr - qc/9705004 . r. emparan , phys . lett . * b432 * , 74 ( 1998 ) hep - th/9804031 . s. sethi and l. susskind , phys * b400 * , 265 ( 1997 ) hep - th/9702101 . + t. banks and n. seiberg , nucl . * b497 * , 41 ( 1997 ) hep - th/9702187 . + e. witten , hep - th/9507121 . + a. strominger , phys . lett . * b383 * , 44 ( 1996 ) hep - th/9512059 + for a review , see n. seiberg , nucl . . suppl . * 67 * , 158 ( 1997 ) hep - th/9705117 . gibbons and r.e . kallosh , phys . * d51 * ( 1995 ) 2839 , hep - th/9407118 ; + s.w . hawking , g.t . horowitz and s.f . ross , phys . * d51 * , 4302 ( 1995 ) gr - qc/9409013 ; + s.w . hawking and gary t. horowitz , class . * 13 * ( 1996 ) 1487 , gr - qc/950101 ; + c. teitelboim , phys * d51 * ( 1995 ) 4315 , hep - th/9410103 .

einstein , podolsky and rosen ( epr ) formulated their famous argument about the completness of quantum mechanics in the framework of continuous variable systems @xcite . however , after that bohm gave a dichotomized version of it @xcite , the debate concerning nonlocality moved to systems described by discrete variables , leading bell to formulate his celebrated inequalities in a dichotomized fashion @xcite . recently , the increasing importance of continuous variable systems leads many authors to explore the nonlocality issue in its original setting , where dichotomic observables to test bell s inequalities are not uniquely determined . the attempts to translate bell s inequalities to continuous variable systems clarified the fact that crucial in a nonlocality test is the existence of a set of dichotomized bounded observables used to perform the test itself , from which the so called `` bell operator '' is derived . the more debated question has dealt with the nonlocality of the normalized version of the original epr state , _ i.e. _ the twin - beam ( twb ) state of radiation produced by spontaneuos downconversion in a parametric amplifier @xcite . nonlocality of the twb state was not clear for a long time . using the wigner function approach , bell argued that the original epr state , and as a consequence the twb too , does not exhibit nonlocality because its wigner function is positive , and therefore represents a local hidden variable description @xcite . more recently , banaszek and wodkiewicz @xcite showed instead how to reveal nonlocality of the epr state through the measurement of displaced parity operator . furthermore a subsequent work of chen et al . @xcite showed that twb s violation of bell s inequalities may achieve the maximum value admitted by quantum mechanics upon a suitable choice of the measured observables . indeed , the amount of violation crucially depends on the kind of bell operator adopted in the analysis , ranging from no violation to maximal violation for the same ( entangled ) quantum state . systems which involves only two parties are the simplest setting where to study violation of local realism in quantum mechanics . a more complex scenario arises if multipartite systems are considered . studying the peculiar quantum features of these systems is worthwhile in view of their relevance in the development of quantum communication technology , _ e.g. _ to manipulate and distribute information in a quantum communication network @xcite . although the study of multipartite nonlocality has originated without the use of inequalities @xcite , an approach to derive bell inequalities has been developed @xcite also for these systems and applied to characterize their entanglement properties @xcite . being originally developed in the framework of discrete variables , these multiparty bell inequalities have found application also in the characterization of continuous variable systems @xcite . the aim of this paper is to apply the various approaches hitherto developed to test nonlocality of two- and three - mode continuous variable systems . we will consider tripartite gaussian states as well as non - gaussian bipartite states . in the first case strong violation of bell inequalities is found , allowing the bell factor to reach values of @xmath0 , while in the second case enhancement of nonlocality is obtained in comparison with the twb case . the paper is organized as follows . in sec . [ s : rev ] we review the different approaches to test nonlocality in the framework of continuous variables and introduce notation that will be used throughout the paper . the three - mode states we are interested in are introduced in sec . [ s:3 m ] , and their violation of local realism is analyzed in [ ss : dp3 ] and [ ss : ps3 ] . in sec . [ s:2 m ] , the tripartite states are considered as sources for conditional generation of non - gaussian bipartite states , whose nonlocal proprties are then studied in [ ss : dp2 ] , [ ss : ps2 ] and [ ss : h2 ] . finally , the main results obtained are summarized in sec . [ s : conclusions ] , which closes the paper with some concluding remarks . in this section we will briefly recall the inequalities imposed by local realism in the cases of our interest . let us start by focusing our attention on a bipartite system . let @xmath1 and @xmath2 denote two possible outcomes of two possible measurements on the first subsystem and similarly @xmath3 and @xmath4 for the second subsystem . the essential feature of this measurements is that they are local , dichotomic and bounded . the bell s combination @xmath5 under the assumption of local realism gives rise to the well known bell - chsh inequality @xcite : @xmath6 where @xmath7 is the correlation function between the measurement results , _ i.e. _ , the expectation value of the products of the results of the experiments @xmath8 and @xmath9 . in the case of a n - partite system , a nonlocality test is possible using the bell - klyshko inequalities @xcite which provides a generalization of inequality ( [ bi2 ] ) . these inequalities are based on the following recursively defined linear combination : @xmath10f_{n-1}+\frac12 \left[m(\alpha_n)-m(\alpha_n')\right]f'_{n-1 } \label{fn}\;,\ ] ] where @xmath11 and @xmath12 refer to measurements on the n - party of the system , and @xmath13 denote the same expression as @xmath14 but with all the @xmath15 and @xmath16 inverted . in the case of a three - partite system , local realism assumption imposes the following inequality from combination ( [ fn ] ) : @xmath17 where again @xmath18 is the correlation function between the measurement results . quantum mechanical systems can violate inequalities ( [ bi2 ] ) and([bi3 ] ) by a maximal amount given by , respectively , @xmath19 and @xmath20 ( see , _ e.g. _ , ref . @xcite ) . we now briefly review three different strategies to reveal quantum nonlocality in the framework of continuous variables systems . recall that in the case of a discrete bipartite system , for example a spin-@xmath21 two particle system , the local dichotomic bounded observable usually taken into account is the spin of the particle in a fixed direction , say @xmath22 . hence the correlation between two measurements performed over the two particles is @xmath23 , where the operator @xmath24 is decomposed on the pauli matrices base and @xmath25 are two unit vectors . the bell operator is then given by the expression : @xmath26 consider now a n - partite continuous variable system identified by the creation operator @xmath27 and the annihilation operator @xmath28 ( @xmath29 ) with boson commutation relations associated . following the original argument by epr it is quite natural attempting to reveal the nonlocality of this system trying to infer quadratures of one subsystem from those of the others . from now on , we will refer to this procedure as a `` homodyne nonlocality test '' , as quadrature measurements of radiation field are performed through homodyne detection . here we identify the quadrature @xmath30 according to the definition @xmath31 . as they are local but neither bounded nor dichotomic , quadrature observables are not immediately suitable to perform a nonlocality test based on bell s inequalities . the procedure to make them bounded and dichotomic is quite arbitrary and consist in the assignment of two domains @xmath32 and @xmath33 to each observable @xcite . when the result of a quadrature measurement falls in the domain @xmath34 the value @xmath35 is associated to it . usually the choice @xmath36 is considered , though a choice suitable to the system under investigation may be preferable . considering a bipartite system we can introduce the following quantities : @xmath37 where @xmath38 is the joint probability distribution of the quadratures @xmath39 and @xmath40 . we can now identify the homodyne correlation function @xmath41 as @xmath42 which can be straightforwardly used to construct the bell combination @xmath43 of eq.([bi2 ] ) and to perform the nonlocality test . the main problem of pursuing such a nonlocality test is that it is not suitable in case of systems described by a positive wigner function , as the twb state of radiation defined as @xmath44 , where @xmath45 and @xmath46 is the squeezing parameter . indeed , a positive wigner function can be interpreted as a hidden phase - space probability distribution , preventing violation of bell - chsh inequality unless the measured observables have an unbounded wigner representation , which is not the case of the dichotomized quadrature measurement described above . considering in fact that @xmath38 can be determined as a marginal distribution from the wigner function one can write from eqs . ( [ ph ] ) and ( [ eh ] ) : @xmath47 where the integration is performed over the whole phase - space and without loss of generality we have considered @xmath48 . eq.([ehwpos ] ) itself is indeed a local hidden variable description of the correlation function , hence obeying inequality ( [ bi2 ] ) . in order to overcome this obstacle different strategies have been considered by many authors , based essentially on parity measurements . banaszek and wodkiewicz @xcite demonstrated the nonlocality of the twb considering as local observable on subsystem @xmath49 the parity operator on the state displaced by @xmath15 ( hence we will refer to this procedure as a `` displaced parity ( dp ) nonlocality test '' ) , which is dichotomic and bounded : @xmath50 in the above formula , @xmath51 , while @xmath52 and @xmath53 $ ] denote the number operator and the phase space displacement operator for the subsystem @xmath49 . hence the correlation function reads : @xmath54 from which bell s combinations @xmath55 in eq.([bi2 ] ) and @xmath56 in eq.([bi3 ] ) can be easily reconstructed in the cases @xmath57 . the reason why this procedure would be able to reveal nonlocality also in case of quantum states characterized by a positive wigner function is clear using the following relation : @xmath58 indeed , the analog of eq.([ehwpos ] ) is : @xmath59 being the dirac-@xmath60 distribution unbounded , ineqs . ( [ bi2 ] ) and ( [ bi3 ] ) are no more necessarily valid for @xmath55 and @xmath56 . the maximal violation found with this procedure for a epr state is @xmath61 @xcite , still far from the maximum violation admitted by quantum mechanics . another strategy , developed by chen et al . @xcite , shares a similar behavior as the one described above , allowing to reveal nonlocality for quantum states with positive wigner function . interestingly , this type of nonlocality test , which we will refer to as `` pseudospin ( ps ) nonlocality test '' , admits a maximum violation for the epr state . it can be seen as a generalization to continuous variable systems of the one introduced by gisin and peres for the case of discrete variable systems @xcite , hence , for the case of a pure bipartite system , it is equivalent to an entanglement test @xcite . let us consider the following set of operators , known as _ pseudospins _ in view of their commutation relations , @xmath62 acting on the @xmath49-th subsystem @xmath63 where @xmath64 and @xmath65 is a unit vector associated to the angles @xmath66 and @xmath67 . in analogy to the spin-@xmath21 system and defining @xmath68 the correlation function is simply given by : @xmath69 from which the bell combinations @xmath70 and @xmath71 are evaluated . also different representations of the spin-@xmath21 algebra have been discussed in the recent literature @xcite . in particular in ref . @xcite it has been pointed out that different representations lead to different expectation values of the bell operators . hence , the violation of bell inequality for continuous variable systems turns out to depend , besides to orientational parameters , also to configurational ones . in the following sections we will also consider the pseudospin operators @xmath72 taken into account in ref . @xcite , which have the following wigner representation : @xmath73 where @xmath74 , @xmath75 and @xmath76 stands for the `` principal value '' . the correlation function obtained using operators @xmath77 will be indicated as @xmath78 . in this section we will analyze the nonlocal properties of tripartite gaussian states . in particular we will consider two classes of states , the first one proposed by van loock and braunstein @xcite , the second one proposed in @xcite . the reason why we consider this two classes is that the first is a very natural and scalable way to produce multimode entanglement using only passive optical elements and single squeezers , while the second one is the simplest way to produce three mode entanglement using a single nonlinear optical device . indeed , both states can be achieved experimentally @xcite . as concern the first class of states , it is generated with the aid of three single mode squeezed states combined in a `` tritter '' ( a three mode generalization of a beam - splitter ) . the evolution is then ruled by a sequence of single and two mode quadratic hamiltonians . as a consequence , being generated from vacuum , the three - mode entangled state is gaussian and its wigner function is given by : @xmath79 \label{wvlb}\;,\ ] ] where @xmath80 , @xmath81 are the positions and momenta of the three modes and @xmath82 is the inverse of the covariance matrix , whose explicit expression reads : @xmath83 where @xmath84 , @xmath85 , @xmath86 and @xmath46 is the squeezing parameter ( with equal squeezing in all initial modes ) . the second class of tripartite entangled states is generated in a single non linear crystal through the following interaction hamiltonian : @xmath87 @xmath88 describes two interlinked bilinear interactions taking place among three modes of the radiation field coupled with the support of two parametric pumps . it can be realized in @xmath89 media by a suitable configuration exposed in ref . notice that the same dynamics can be implemented in different physical systems , including optomechanical couplers and bose - einstein condensates in the linear regime @xcite . the effective coupling constants @xmath90 , @xmath91 , of the two parametric processes are proportional to the nonlinear susceptibilities and the pump intensities . if we take the vacuum @xmath92 as the initial state , the evolved state @xmath93 belongs to the class of the coherent states of @xmath94 and it reads @xcite : @xmath95 where @xmath96 represent the average number of photons in the @xmath49-th mode and @xmath97 are phase factors . the explicit expressions of @xmath98 are : @xmath99 ^ 2 \ ; , \nonumber \\ n_3 & = & \frac{|\gamma_1|^2}{\omega ^2 } \sin^2(\omega t ) \ ; , \label{ndit}\end{aligned}\ ] ] with @xmath100 and @xmath101 . also for this second class , being the initial state gaussian and the hamiltonian quadratic , the evolved states will be gaussian . the wigner function reads as follows @xcite : @xmath102 \label{wt}\;,\ ] ] where @xmath103 is the inverse of the covariance matrix , whose explicit expression is : @xmath104 where @xmath105 both classes of states are fully inseparable for any value of the coupling constants , namely can not be written as a factorized state for any grouping of the modes . therefore they are good candidates to reveal true tripartite nonlocality . being gaussian states , however , nonlocality can not be revealed by homodyne detection . in the following we analyze the results for displaced parity and pseudospin tests . let us start the study of tripartite systems nonlocality using the `` displaced parity test '' . considering the correlation function @xmath106 given by eq . ( [ edp ] ) , the state ( [ wvlb ] ) was found in @xcite to give a maximal violation of @xmath107 in the limit of large squeezing and small displacement . the study in @xcite however was performed for a particular choice of displacement parameters : @xmath108 and @xmath109 . a numerical optimization of the displacement parameters lead us to identify a number of parameterizations that allow a significantly higher violation of bell s inequality . as an example , consider the one given by @xmath110 and @xmath111 from which follows that @xmath112 hence the remarkably high asymptotic value of @xmath113 is found for large @xmath46 and @xmath114 ( see fig . [ f : b3dpvlbgen ] ) . the importance of a suitable choice of the displacement parameters is apparent if this asymptotic value is compared to the violations obtained in the nonlocality study performed in ref . @xcite . in that work in fact generalizations to more than three modes of state ( [ wvlb ] ) were also considered , giving an increasing violation of bell inequality as the number of modes increases , but never founding a violation greater than @xmath115 . determining the optimal choice of the displacement parameters for a given state is in general a challenging task . to our knowledge indeed there exists no general prescription to find it out , and ultimately one must rely onto a numerical analysis ( see , _ nevertheless , a careful inspection of the symmetries of the state under consideration may be helpful . in order to clarify this observation let us consider the explicit expression of the correlation function @xmath116 for state ( [ wvlb ] ) : @xmath117 \nonumber \\ -\frac{2}{3}e^{-2\,r}\left [ ( x_1+x_2+x_3)^2+(y_2-y_3)^2+(y_2-y_1)^2+(y_1-y_3)^2 \right ] \bigg\}\end{aligned}\ ] ] hence , from eq . ( [ bi3 ] ) it follows that the bell combination @xmath56 is given by the sum of three positive and one negative term . it is reasonable to expect that the maximal violation of nonlocality will be achieved for large @xmath46 . we see from eq . ( [ vlbopt1 ] ) that , in this limit , all the correlation functions in @xmath56 vanish for nonzero displacements , unless the coefficients of @xmath118 are zero . hence we impose the following system of equations , which allows for the three positive terms in @xmath56 not to vanish ( we consider @xmath119 and @xmath120 for @xmath121 ) : @xmath122 we see that the parameterization used to obtain eq . ( [ b3dpvlbgen ] ) is a solution of this system . clearly , any other solution will give , in the limit of large @xmath46 , the same violation given by eq . ( [ b3dpvlbgen ] ) , namely @xmath123 . in order to compare the violation of bell s inequality admitted by the state ( [ wvlb ] ) with the one that will be obtained below considering the state ( [ t ] ) , it is useful to rewrite eq.([b3dpvlbgen ] ) as a function of total mean photon number @xmath124 . given that @xmath125 and optimizing the displacement @xmath126 we obtain the result shown in fig [ f : b3dpn ] . the asymptotic expression of the optimized displacement as a function of @xmath127 is @xmath128 , hence very small angles are required . ) . only values violating bell inequality ( [ bi3 ] ) are shown.,scaledwidth=50.0% ] consider now the tripartite state generated by the hamiltonian ( [ inth ] ) . the correlation function is now given by eq . ( [ edp ] ) through the wigner function ( [ wt ] ) . the symmetry of the state suggests a maximum violation of bell inequality for @xmath129 ( recalling eq . ( [ ndit ] ) ) , while the fact that the separability of the state does nt depend on the phases @xmath130 and @xmath131 @xcite suggests that they are not crucial for the nonlocality test . if we consider the same parametrization that led us to eq . ( [ b3dpvlbgen ] ) and fix @xmath132 , we find : @xmath133 from which follows an asymptotic violation of bell s inequalities of @xmath134 , for large @xmath127 and small @xmath126 . a slightly better result is found if a parametrization , more suitable and numerically optimized for state ( [ t ] ) , is considered : @xmath135 and @xmath136 . the bell combination @xmath56 for this choice of parameters is depicted in fig . [ f : b3dpt ] . we note that in this case a larger choice of angles allows the violation of bell inequality if compared with fig . [ f : b3dpvlbgen ] . ) ( see text for details ) . only values violating bell inequality ( [ bi3 ] ) are shown.,scaledwidth=50.0% ] for state ( [ wvlb ] ) , dotted line , and ( [ t ] ) , solid line . the displacement parameters @xmath126 have been optimized to give the maximum value of @xmath137 for @xmath127 fixed.,scaledwidth=50.0% ] as before , optimizing the displacement @xmath126 for each @xmath127 , it is possible to find the maximum violation of bell inequality as a function of @xmath127 . we find that the asymptotic relation between the optimized displacement and the total photon number is now @xmath138 , confirming that not too small displacements are required . the asymptotic violation of bell s inequality is now @xmath139 . to compare the results obtained for the two states ( [ wvlb ] ) and ( [ t ] ) we have plotted in fig . [ f : b3dpn ] the bell combination @xmath56 as a function of the mean total energy @xmath127 , while the displacement @xmath126 has been chosen in order to maximize @xmath56 at fixed energy . notice that even if the two states have quite the same asymptotic violation , state ( [ wvlb ] ) reaches it for lower energies . consider now the pseudospin nonlocality test . let us calculate the expectation value of the correlation function ( [ eps ] ) for the state @xmath140 ( for simplicity we consider @xmath141 ) . the only non vanishing contributes are given by : @xmath142 and by @xmath143 . the correlation function then , according to eqs . ( [ ps ] ) and ( [ eps ] ) , reads as follows : @xmath144 hence , without loss of generality , we can fix for example @xmath145 and @xmath146 and look for the maximum violation of bell inequality ( [ bi3 ] ) constructed from eq . ( [ eps3ms ] ) . notice that if the coefficients @xmath147 ( @xmath148 ) were equal to one then the maximum violation admitted , @xmath149 , should be reached . considering eqs . ( [ c1c2c3 ] ) two limiting cases can be studied . first , for large @xmath150 and small @xmath151 ( or vice - versa ) a numerical evaluation of the coefficients @xmath147 shows that @xmath152 , while the other two vanish . hence , considering @xmath153 , the correlation function ( [ eps3ms ] ) reduces to that of a twb subjected to a pseudospin nonlocality test ( see eq . ( [ etwbps ] ) below ) , hence allowing an asymptotic violation of @xmath154 . this result should be expected , since in this limiting case the state ( [ t ] ) reduces to a twb for modes @xmath155 and @xmath156 , while mode @xmath157 remains in the vacuum state and factors out . consider now the case in which @xmath158 . a numerical evaluation shows that the coefficients @xmath159 for large @xmath127 , hence also in this case the maximum violation can not be attained . the asymptotic violation turns out to be @xmath160 . as already mentioned in sec . [ s : rev ] other representations for the pseudospin operators can be considered . using eqs . ( [ wpi ] ) and ( [ wt ] ) it is possible to calculate the correlation function @xmath161 . setting again the azimuthal angles @xmath162 , the latter shows the same structure as @xmath163 where now the coefficient @xmath147 are replaced by : @xmath164 an appropriate choice of angles leads to a maximal violation of bell s inequality given by @xmath165 ( see fig . [ f : b3ps ] ) , which is now reached for @xmath166 , value for which the coefficients @xmath167 are approximately near their maxima . as already pointed out , we see that different representations of the pseudospin operators give rise to different expectation values for the bell operator . applying now the same procedure to state ( [ wvlb ] ) we find the same structure for the correlation function @xmath168 , where the coefficients are now given by : @xmath169 after an optimization of the angles @xmath170 we obtain a maximal violation of @xmath171 ( see fig . [ f : b3ps ] ) , for @xmath172 ( @xmath173 ) that maximizes the coefficients @xmath147 . for states ( [ wt ] ) and ( [ wvlb ] ) , solid and dashed line respectively . @xmath127 is the total number of photons as in fig . [ f : b3dpn ] , scaledwidth=50.0% ] in this section we consider the tripartite state ( [ t ] ) as a source of two - mode states . in particular , we study the nonlocality of a two - mode non - gaussian state obtained by a conditional measurement performed on state ( [ t ] ) . gaussian states are at the heart of quantum information processing with continuous variables . the reason for this is that gaussian is the character of the vacuum state of quantum electrodynamics . this observation , in combination with the fact that the quantum evolutions achievable with current technology are described by hamiltonian operators at most bilinear in the quantum fields , accounts for the fact that the states commonly produced in laboratories are gaussian . in fact , bilinear evolutions preserve the gaussian character of the vacuum state . in addition , it is worth noticing that the operation of tracing out a mode from a multipartite gaussian state preserves the gaussian character too , and the same observation is valid when the evolution of a state in a standard noisy channel is considered . indeed , the only feasible way to `` degaussify '' a state is through a conditional measurement , or by statistically mixing it with another gaussian state . the reason to study non - gaussian states is that when the gaussian character is lost , then immediately the wigner function of the state becomes negative , for pure states , hence stronger nonclassical properties should emerge . actually , various authors have recently investigated the nonlocality properties of two - mode non - gaussian states . in particular , a twin - beam state subjected to inconclusive photon subtraction ( ips state ) has been considered in refs . @xcite and @xcite , while in ref . @xcite it has been pointed out that if the entangled coherent states @xcite could be produced experimentally they would allow for the maximal violation ( _ i.e. _ , @xmath174 ) both in case of a dp test as well as a ps test . the most natural way to obtain a non - gaussian state from a gaussian one is by elimination of its vacuum component . in fact , such a state is necessarily described by a negative wigner function ( in fact @xmath175 ) . do to the structure of state ( [ t ] ) its vacuum component can be subtracted by a conditional measurement on mode @xmath157 , the same observation being valid for mode @xmath156 . consider a photodetector able to distinguish only the presence or the lack of photons , _ i.e. _ , an on / off photodetector , and the state @xmath176 conditioned to the presence of at least one photon . the probability operator measure ( povm ) is two - valued @xmath177 , @xmath178 , with the element associated to the `` no photons '' result given by @xmath179 where @xmath180 is the efficiency of the photodetector . the probability of the outcome is given by @xmath181 = \frac{\eta n_3}{(1+\eta n_3 ) } \:,\end{aligned}\ ] ] while the conditional output state reads as follows @xmath182 \nonumber \\ \hspace{-50pt } \eql \frac{1+\eta n_3}{(1+n_1)\eta n_3 } \sum_{p=1}^\infty \left(\frac{n_3}{1+n_1}\right)^p \frac{1-(1-\eta)^p}{p!}(a^\dagger)^p \sum_{n , n ' } \left(\frac{n_2}{1+n_1}\right)^{n+n ' } |n\,n\rangle \langle n'\,n'| a^p \nonumber \\ \hspace{-50pt}\eql \frac{1+\eta n_3}{(1+n_1+n_2)\eta n_3 } \sum_{p=1}^\infty \left(\frac{n_3}{1+n_1}\right)^p \frac{1-(1-\eta)^p}{p ! } ( a^\dagger)^p |x\rangle\langle x| a^p \:,\end{aligned}\ ] ] where we have identified the twb with @xmath183 . to calculate the wigner function associated with state @xmath184 , consider that the characteristic function of the povm @xmath185 reads as follows : @xmath186(\mu)=\pi\delta^2(\mu)-\frac1\eta\exp\left[-|\mu|^2 \frac{2-\eta}{2\eta}\right]\;,\ ] ] hence the characteristic function of @xmath176 is given by : @xmath187(\lambda_1,\lambda_2)=\frac{1}{p_1 } \left\ { \chi[|t\rangle\langle t|](\lambda_1,\lambda_2,0 ) -\frac1\eta\int \frac{d^2\mu}{\pi}\chi[|t\rangle\langle t|](\lambda_1,\lambda_2,\mu ) \exp\left[-|\mu|^2\frac{2-\eta}{2\eta}\right ] \right\ } \;.\ ] ] after some algebra the wigner function associated with state @xmath184 can now be calculated . it reads as follows : @xmath188 \nonumber \\ \lo- \frac{1}{\eta } \left(\frac{2}{\pi}\right)^2\frac{2}{\sqrt{\det d } } \exp\left [ -({\bf x},{\bf y } ) \left({\bf d}^{-1}\right ) ' \left ( \begin{array}{c } { \bf x } \\ { \bf y } \end{array } \right)\right ] \bigg\rbrace \label{w1}\;,\end{aligned}\ ] ] where , from now on , @xmath189 , @xmath190 , and @xmath191 . in order to simplify the notation we have indicated with @xmath192 the @xmath193 matrix obtained from the @xmath194 matrix @xmath195 deleting the elements corresponding to the third mode ( 3-th row / column and 6-th row / column ) , due to the trace over the 3-th mode . of course , the easiest way to obtain a bipartite state from state ( [ t ] ) is to discard a mode , say the third one , by tracing over it . the state @xmath196 then obtained is simply given by the following wigner function : @xmath197 \label{wtraced}\;.\ ] ] being the wigner function @xmath198 gaussian , we expect that this state will exhibit weaker nonlocality with respect to state ( [ rho1 ] ) . in the rest of the section the nonlocal properties of the usual twb state and of the states ( [ rho1 ] ) and ( [ wtraced ] ) will be compared . notice that state ( [ wvlb ] ) can be considered as an extension to three modes of the twb . all the three nonlocality tests introduced in sec . [ s : rev ] will be taken into account . we first study the violation of inequality ( [ bi2 ] ) in the case of a `` displaced parity test '' . as already mentioned , in ref . @xcite banaszek and wodkiewicz found for the first time that the twb state exhibit a violation of local realism . they obtained the following asymptotic violation for infinite energy : @xmath199 . generalizing their procedure this result can be improved , yielding to a maximum asymptotic violation of @xmath200 @xcite . we have considered the following parametrization to obtain the maximum violation for a twb : @xmath201 and @xmath202 . the asymptotic relation between the squeezing parameter and the displacement angles is @xmath203 . using the same parametrization and considering the bell combination @xmath55 for the state @xmath196 , it turns out that the same asymptotic value of the twb is reached for large @xmath150 and small @xmath151 . in fact , as already noticed , when this limit is considered the original tripartite state ( [ t ] ) reduces to a factorized state composed by a twb and the vacuum state . consider now the conditional state @xmath184 and again the case of large @xmath150 and small @xmath151 , say @xmath204 . as in the tripartite case the phase coefficients @xmath130 and @xmath131 play no rule in the characterization of nonlocality . a stronger violation of bell inequality is then found and it is depicted in fig . ( [ f : b2dptwba ] ) , where the parametrization @xmath205 and @xmath206 has been adopted . ( see text for details ) . only values violating inequality ( [ bi2 ] ) are shown.,scaledwidth=50.0% ] indeed the asymptotic violation is higher then the previous , namely @xmath207 . it can be found , for large @xmath150 , when @xmath208 . a comparison with the violation of nonlocality attained with a ips state shows an identical asymptotic behavior @xcite . nevertheless , the scheme proposed here offers the advantage that the production rate of state @xmath176 [ _ i.e. _ , the conditional probability ( [ p1 ] ) ] is much greater then the production rate of ips state [ see ref . @xcite , eq.(14 ) ] . this is due to the fact that only a single on / off detection is required to produce @xmath176 , rather than the coincidence of two on / off detections for the case of ips state . this could be useful from a practical viewpoint . let us now focus on the `` pseudospin nonlocality test '' . considering a twb state , it is known that the correlation function ( [ eps ] ) has the following expression ( setting to zero the azimuthal angles ) @xcite : @xmath209 where , denoting with @xmath127 the total photon number , @xmath210 it turns out that the violation of bell inequality in this contest increases monotonically to the maximum value of @xmath211 as the function @xmath212 goes to unity . a straightforward calculation shows that an expression identical in form to eq . ( [ etwbps ] ) can be found both in case of state @xmath176 and @xmath196 , where the following functions @xmath213 and @xmath214 can be identified : @xmath215 in order to compare the violations in the three different cases , let us fix as in the previous subsection a small value for @xmath151 . a plot of the functions @xmath212 , @xmath213 and @xmath214 versus the total number of photons of the twb for the former and of the initial three - partite state for the latter two is given in fig . ( [ f : b2ps ] ) . it can be seen that state @xmath176 achives large violations for smaller energies with respect to the other two states . finally , a comparison with the violation attained with the ips state may be found in ref . @xcite . ( solid line ) , @xmath216 ( dotted line ) and @xmath214 ( dashed line ) defined in the text ( the summation has been numerically performed for @xmath217 and @xmath218).,scaledwidth=50.0% ] the negativity of the wigner function ( [ w1 ] ) may suggest to perform a nonlocality test based upon a homodyne detection scheme . while the positivity of a wigner function avoid to violate bell inequality ( [ bi2 ] ) with such a test , its negativity is yet not sufficient in general to ensure a violation . quantum states with negative wigner function that does nt violate local realism with a homodyne test are given for example in refs . @xcite . considering state ( [ rho1 ] ) it is necessary to calculate the correlation function ( [ ehwpos ] ) . substituting the wigner function ( [ w1 ] ) into eq . ( [ ehwpos ] ) and performing the integral we obtain the following result : @xmath219 \nonumber \\ \fl -\frac{1}{\eta } \left(\frac{2}{\pi}\right)^2\frac{2}{\sqrt{\det d } } \left [ \frac{2\,\pi \,\left ( -1 + n_3\,\left ( -2 + \eta \right ) \right)}{1 + n_3\,\eta } \ , \arctan \left(\frac{2\,\cos\psi } { { \sqrt{\frac{\left ( 1 + 2\,n_1 - n_3\,\eta \right ) \ , \left ( 1 + 2\,n_2 + n_3\,\eta \right ) } { \left ( 1 + n_1 \right ) \,n_2 } - 4\,\cos^2\psi}}}\right ) \right ] \bigg\ } \label{ehrho1}\;,\end{aligned}\ ] ] where @xmath220 . a plot of the correlation function ( [ ehrho1 ] ) is depicted in fig . ( [ f : e2h ] ) for unitary efficiency @xmath180 , together with the classical correlation function of two spin-@xmath21 particles @xcite ( see caption for details ) . unfortunately , the comparison shows clearly that the correlation given by eq . ( [ ehrho1 ] ) is always lower then the classical one , hence despite the negativity of wigner function ( [ w1 ] ) no violation of bell inequality is achievable with this scheme . particles classically correlated ( solid line ) and from eq . ( [ ehrho1 ] ) : @xmath221 ( dotted line),@xmath222 ( dot - dashed line),@xmath223 ( dashed line ) . in all cases we have fixed @xmath224 and @xmath225.,scaledwidth=50.0% ] [ f : e2h ] a detailed analysis of the nonlocality properties of multipartite continuous variables systems obtained by parametric optical systems has been presented , using the more recent approaches developed to this aim . we have considered in particular two classes of tripartite gaussian states that seems promising for quantum communication purposes in order to implement multipartite quantum protocols . the results show that for these states a nonlocality test based on displaced parity measurements is more suitable to reveal violation of local realism than one based on pseudospin operators . this results are just the opposite of what have been obtained for the bipartite case . notice , however , that a systematic approach to pseudospin operators for continuous variables havent been developed yet , hence we have only used the two inequivalent configurational parameterizations more suitable for calculations . for displaced parity test we obtained a remarkably high asymptotic value for the bell parameter , @xmath226 . in this case the choice of a proper parametrization , suitable for the state under investigation , have revealed crucial . we have also explored the possibility to enhance nonlocality in bipartite systems considering states endowed with a nonpositive wigner function . we investigated a method to conditionally produce such a state from the tripartite systems considered above . as expected , the bell parameter reaches a value higher than for a twb , namely @xmath227 . also in the case of a pseudospin test an enhancement of nonlocality has been demonstrated , while a violation of local realism using a dichotomic quadrature measurement can not be achieved . the authors are grateful to s. olivares and m. s. kim for fruitful discussions . 99 a. einstein , b. podolsky and n. rosen , phys . rev . * 47 * , 777 ( 1935 ) . d. bohm , _ quantum theory _ ( dover , nj , 1989 ) . j.s . bell , physics ( long island city , n.y . ) * 1 * , 195 ( 1964 ) . d.f . walls and g.j . milburn , _ quantum optics _ ( berlin , springer , 1995 ) . j.s . bell , _ speakable and unspeakable in quantum mechanics _ ( cambridge university press , cambridge , 1987 ) . k. banaszek and k. wodkiewicz , phys . a * 58 * , 4345 ( 1998 ) . chen , j.w . pan , g. hou , and y.d . zhang , phys . . lett . * 88 * , 040406 ( 2002 ) . p. van loock and s.l . braunstein , phys . lett . * 84 * , 3482 ( 2000 ) . h. yonezawa , t. aoki , and a. furusawa , nature * 431 * , 430 ( 2004 ) . greenberger , m.a . horne , a. shimony , and a. zeilinger , am . j. phys . * 58 * , 1131 ( 1990 ) . mermin , phys . lett . * 65 * , 1838 ( 1990 ) ; d. n. klyshko , phys . a * 172 * , 399 ( 1993 ) . n. gisin and h. bechmann - pasquinucci , phys . a * 246 * , 1 ( 1998 ) . p. van loock and s.l . braunstein , phys . a * 63 * , 0222106 ( 2001 ) . z.b . chen and y.d . zhang , phys . a * 65 * , 044102 ( 2002 ) . clauser , m.a . horne , a. shimony and r. a. holt , phys . * 23 * , 880 ( 1969 ) . a. gilchrist , p . deuar , m.d . reid , phys . lett . * 80 * , 3169 ( 1998 ) . h. jeong , w. son , m.s . kim , d. ahn , and . brukner , phys . a * 67 * , 012106 ( 2003 ) . n. gisin and a. peres , phys . a * 162 * , 15 ( 1992 ) . l. mita , r. filip , and j. fiuraek , phys . a * 65 * , 062315 ( 2002 ) ; l. praxmeyer , b.g . englert , and k. wodkiewicz , eur . j. phys . d * 32 * , 227 ( 2005 ) . g. gour , f.c . khanna , a. mann , m. revzen , phys . a * 324 * , 415 ( 2004 ) ; m. revzen , p.a . mello , a. mann , l.m . johansen , phys . a * 71 * , 022103 ( 2005 ) . a. ferraro , m.g.a . paris , m. bondani , a. allevi , e. puddu , and a. andreoni , j. opt . b * 21 * , 1241 ( 2004 ) . t. aoki , n. takei , h. yonezawa , k. wakui , t. hiraoka , a. furusawa , and p. van loock , phys . rev . lett . * 91 * , 080404 ( 2003 ) . m. bondani , a. allevi , e. puddu , a. andreoni , a. ferraro , and m.g.a . paris , opt . * 29 * , 180 ( 2004 ) . n. piovella , m. cola , r. bonifacio , phys . rev . a * 67 * , 013817 ( 2003 ) . s. pirandola , s. mancini , d. vitali , and p. tombesi , phys . rev . a * 68 * , 062317 ( 2003 ) . rodionov , a.s . chirkin , pisma zh . . fiz . * 79 * , 311 ( 2004 ) [ jetp lett . * 79 * , 253 ( 2004 ) ] . puri , phys . rev . a * 50 * , 5309 ( 1994 ) . m. m. cola , m. g. a. paris , and n. piovella , phys . a * 70 * , 043809 ( 2004 ) . k. banaszek , a. dragan , k. wdkiewicz , and c. radzewicz , phys . a * 66 * , 043803 ( 2002 ) . h. nha , and h. j. carmichael , phys . lett . * 93 * , 020401 ( 2004 ) ; r. garca - patrn snchez _ et al . _ , * 93 * , 130409 ( 2004 ) . s. olivares , and m. g. a. paris , phys . rev . a * 70 * , 032112 ( 2004 ) . b. c. sanders , phys . a * 45 * , 6811 ( 1992 ) . s. olivares , m. g. a. paris , and r. bonifacio , phys . a * 67 * , 032314 ( 2003 ) . a. ferraro , s. olivares and m. g. a. paris , _ gaussian states in c.v . quantum information _ ( bibliopolis , napoli , 2005 ) . w.j . munro , phys . rev a * 59 * , 4197 ( 1999 ) ; s. mancini , v.i . manko , e.v . shchukin , and p. tombesi , j. opt . b : quantum s o * 5 * , s333 ( 2003 ) . a. peres , _ quantum theory : concepts and methods _ ( kluwer , dordrecht , 1993 ) .

in the classical graph coloring problem we assign a color to each vertex of a given graph such that the graph is properly colored , i.e. , no two adjacent vertices have the same color . the chromatic number @xmath2 of a graph @xmath0 is the smallest @xmath3 such that @xmath0 can be colored with @xmath3 distinct colors . deciding whether the chromatic number of a graph is at most @xmath3 is well known to be np - complete , even in the case with three colors . the online variant of graph coloring can be defined as follows : the vertices of @xmath0 arrive one by one , and an online algorithm must color vertices as they arrive so that the revealed graph is properly colored at all times . when a vertex arrives , the algorithm sees edges to previously colored vertices . to give the online algorithm slightly more power , a copy of @xmath0 is sent at the start of the input . however , the exact correspondence between the incoming vertices and the vertices of the copy of @xmath0 is not known to the algorithm . this problem is called online graph coloring . in this paper we focus on a graph parameter called the _ online chromatic number _ @xmath4 of a graph @xmath0 . this parameter is analogous to the standard chromatic number of a graph : it denotes the smallest number @xmath3 such that there exists a deterministic online algorithm which is able to color the specified graph @xmath0 using @xmath3 colors . the online chromatic number has been studied since 1990 @xcite . one of the main open problems in the area is the computational complexity of deciding whether @xmath5 for a specified simple graph @xmath0 , given @xmath0 and @xmath3 on input ; see e.g. kudahl @xcite . we denote this decision problem as online chromatic number . in this paper , we fully resolve this problem : [ thm : main ] the decision problem online chromatic numberis pspace - complete . as is usual in the online computation model , we can view online graph coloringas a game between two players , which we call (representing the online algorithm ) and ( often called adversary in the online algorithm literature ) . in each round chooses an uncolored vertex @xmath1 from @xmath0 and sends it to without telling him to which vertex of @xmath0 it corresponds , only revealing the edges to the previously sent vertices . then must properly color ( `` paint '' ) @xmath1 , i.e. , cannot use a color of a neighbor of @xmath1 . we stress that in this paper is restricted to be deterministic . the game continues with the next round until all vertices of @xmath0 are colored . deciding the outcome of many two - player games is pspace - complete ; among those are amazons , checkers or hex , to name a few . however , in most of these games both players have roughly the same power . this does not hold for online graph coloringwhich is highly asymmetric , since has perfect information ( knows which vertices are sent and how they are colored ) , but does not . may only guess to which part of the graph does the colored subgraph really belong . this is the main difficulty in proving pspace - hardness . * examples . * consider a path @xmath6 on four vertices . initially , sends two nonadjacent vertices . if assigns different colors to them , then these are the first and the third vertex of @xmath6 , thus the second vertex must get a third color ; otherwise they obtained the same color @xmath7 and they are the endpoints of @xmath6 , therefore the second and the third vertex get different colors which are not equal to @xmath7 . in both cases , there are three colors on @xmath6 and thus @xmath8 , while @xmath9 . note also that we may think of deciding where an incoming vertex belongs at some time _ after _ it is colored provided that his choice still allows for at least one isomorphism to the original @xmath0 . this is possible only for a deterministic . a particularly interesting class of graphs in terms of @xmath10 is the class of binomial trees . a binomial tree of order @xmath3 is defined inductively : the binomial tree of order 0 is a single vertex ( the root ) and the binomial tree of order @xmath3 is created by taking two disjoint copies of binomial trees of order @xmath11 , adding an edge between their roots and choosing one of their roots as the root for the resulting tree . thus @xmath6 is a binomial tree of order 2 with root on the second vertex of @xmath6 . it is not hard to generalize the example of @xmath6 and show that the online chromatic number of the binomial tree of order @xmath3 is @xmath12 @xcite . this shows that the ratio between @xmath10 and @xmath2 may be arbitrarily large even for the class of trees . * history and related work . * the online problem online graph coloringhas been known since 1976 @xcite , originally studied in the variant where the algorithm has no extra information at the start of the input . bean @xcite showed that no online algorithm that is compared to an offline algorithm can perform well under this metric . the notion of online chromatic number was first defined in 1990 by @xcite . for the online problem , lovsz , saks and trotter @xcite show an algorithm with a _ competitive ratio _ @xmath13 , where the competitive ratio is a ratio of the number of colors used by the online algorithm to the ( standard ) chromatic number . this was later improved to @xmath14 by kierstad @xcite using a deterministic algorithm . there is a better @xmath15-competitive randomized algorithm against an oblivious adversary by halldrsson @xcite . a lower bound on the competitive ratio of @xmath16 was shown by halldrsson and szegedy @xcite . our variant of online graph coloring , where the algorithm receives a copy of the graph at the start , was suggested by halldrsson @xcite , where it is shown that the lower bound @xmath16 also holds in this model . ( note that the previously mentioned algorithmic results are valid for this model also . ) kudahl @xcite recently studied online chromatic numberas a complexity problem . the paper shows that the problem is conp - hard and lies in pspace . later @xcite he proved that if some part of the graph is precolored , i.e. , some vertices are assigned some colors prior to the coloring game between and and also reveals edges to the precolored vertices for each incoming vertex , then deciding whether @xmath17 is pspace - complete . we call this decision problem online chromatic number with precoloring . the paper @xcite conjectures that online chromatic number(with no precolored part ) is pspace - complete too . interestingly , it is possible to decide @xmath18 in polynomial time @xcite . keep in mind that while online graph coloringis an online problem , online chromatic numberis an ( offline ) decision problem of checking whether @xmath19 . * proof outline . * the fact that online chromatic numberbelongs to pspace is not hard to see : the online coloring is represented by a game tree which is evaluated using the minimax algorithm . this can be done in polynomial space , since the number of rounds in the game is bounded by @xmath20 , i.e. , the number of vertices , and possible moves of each player can be enumerated in polynomial space : has at most @xmath20 possible moves , because it either uses a color already used for a vertex , or it chooses a new color , and has @xmath21 moves where @xmath22 is the number of colored vertices , since it chooses which colored vertices shall be adjacent to the incoming vertex . must ensure that sent vertices form an induced subgraph of @xmath0 , but this can be checked in polynomial space . inspired by @xcite , we prove the pspace - hardness of online chromatic numberby a reduction to q3dnf - sat , i.e. , the satisfiability of a fully quantified formula in the 3-disjunctive normal form ( 3-dnf ) . an example of such a formula is @xmath23 the similar problem of satisfiability of a fully quantified formula in the 3-conjunctive normal form is well known to be pspace - complete . since pspace is closed under complement , q3dnf - sat is pspace - complete as well . note that by an easy polynomial reduction , we can assume that each 3-dnf clause contains exactly three literals . we show the hardness in several iterative steps . first , in section [ sec : largeprecol ] , we present a new , simplified proof of the pspace - hardness of online chromatic number with precoloringin which the sizes of both precolored and non - precolored parts of our construction are linear in the size of the formula . then , in section [ sec : smallprecol ] , we strengthen the result by reducing the size of the precolored part to be logarithmic in the size of the formula . this is achieved by adding linearly many vertices to our construction . finally , in section [ sec : noprecol ] , we show how to remove one precolored vertex and replace it by a non - precolored part , while keeping the pspace - hardness proof valid . the cost for removing one vertex is that the size of the graph is multiplied by a constant , but since we apply it only logarithmically many times , we obtain a graph of polynomial size and with no precolored vertex . this will complete the proof of theorem [ thm : main ] . we remark that removing the last precolored vertex is the most difficult part of proving pspace - hardness of online chromatic number . still , our technique for removing a precolored vertex can be used for any graph satisfying a few assumptions . in our analysis , often uses the natural greedy algorithm , which is ubiquitous in the literature ( see @xcite , @xcite ) : the online algorithm colors an incoming vertex @xmath24 using the smallest color not present among colored vertices adjacent to @xmath24 . our first construction will reduce the pspace - complete problem q3dnf - sat to online coloring with precoloring with a large precolored part . given a fully quantified formula @xmath25 in the 3-disjunctive normal form , we will create a graph @xmath26 that will simulate this formula . we assume that the formula contains @xmath20 variables @xmath27 and @xmath28 clauses @xmath29 , @xmath30 , and that variables are indexed in the same order as they are quantified . our main resource will be a large precolored clique @xmath31 on @xmath3 vertices and naturally using @xmath3 colors ; the number @xmath3 will be specified later . using such a precolored clique , we can restrict the allowed colors on a given uncolored vertex @xmath1 by connecting it with the appropriate vertices in @xmath31 , i.e. , we connect @xmath1 to all vertices in @xmath31 which do not have a color allowed for @xmath1 . for simplicity we use the precoloring in the strong sense , i.e. , is able to recognize which vertex in @xmath31 is which . we use this to easily recognize colors . however , it is straightforward to avoid the strong precoloring by modifying the precolored part ; for example by creating @xmath32 independent and identical copies of the @xmath32-th vertex in @xmath31 , each having the same color and the same edges to other vertices in @xmath31 and the rest of the graph . with such a modification , would able to recognize each color by the number of its vertices in @xmath31 . each vertex in @xmath31 thus corresponds to a color . colors used by are naturally denoted by numbers 1 , 2 , 3 , ... @xmath3 , but we shall also assign meaningful names to them . we want to construct a graph @xmath26 that has the online chromatic number @xmath3 if and only if the quantified 3-dnf formula can be satisfied . see figure [ fig : gadgetexample ] for an example of @xmath26 and an overview of our construction . we use the following gadgets for variables and clauses : for a variable @xmath33 which is quantified universally , we will create a gadget consisting of _ universal _ vertices @xmath34 and @xmath35 , connected by an edge . the vertex @xmath34 represents the positive literal @xmath33 , while @xmath35 represents the negative literal @xmath36 . both vertices have exactly two allowed colors : @xmath37 and @xmath38 . if @xmath34 is assigned the color @xmath39 , it corresponds to setting the variable @xmath33 to @xmath40 , and vice versa . note that if presents a vertex @xmath41 to , is able to recognize that it is a vertex corresponding to the variable @xmath42 , but it is not able to recognize whether it is the vertex @xmath41 or @xmath43 . for a variable @xmath42 which is quantified existentially , we will create a gadget consisting of three _ existential _ vertices @xmath41 ( for the positive literal @xmath42 ) , @xmath43 ( for the literal @xmath44 ) and @xmath45 ( the helper vertex ) , connected as a triangle . coloring of the first two variables also corresponds to setting the variable @xmath42 to true or false , but in a different way : @xmath41 has allowed colors @xmath46 and @xmath47 , while @xmath43 has allowed colors @xmath48 and @xmath47 . we want to avoid both @xmath41 and @xmath43 to have the color of type @xmath49 , and so the `` helper '' vertex @xmath45 can be colored only by @xmath46 or @xmath48 . note that the color choice for the vertices of @xmath42 means that if is presented any vertex of this variable , can recognize it and decide whether to set @xmath42 to @xmath40 ( and color accordingly ) or to @xmath50 . we call existential and universal vertices together _ variable vertices_. for each clause @xmath29 , we will add four new vertices . first , we create a vertex @xmath51 for each literal in the clause , which is connected to one of the vertices @xmath34 and @xmath35 corresponding to the sign of the literal . for example if @xmath52 , then @xmath51 is connected to @xmath34 , @xmath53 is connected to @xmath43 and @xmath54 to @xmath55 . the allowed colors on a vertex @xmath51 are @xmath56 . finally , we add a fourth vertex @xmath57 connected to the three vertices @xmath58 . this vertex can be colored only using the color @xmath59 or the color @xmath60 . the color @xmath60 is used to signal that this particular clause is evaluated to be @xmath50 . if the color @xmath59 is used for the vertex @xmath57 , this means that the clause is evaluated to @xmath40 , because @xmath59 is not present on any of @xmath58 , thus they have colors of type @xmath61 and their neighbors corresponding to literals have colors of type @xmath49 . the last vertex we add to the construction will be @xmath62 , a final vertex . the vertex @xmath62 is connected to all the vertices @xmath57 corresponding to the clauses . the allowed colors of the vertex @xmath62 are @xmath63 . this final vertex corresponds to the final evaluation of the formula . if all clauses are evaluated to @xmath50 , the vertex @xmath62 has no available color left and must use a new color . we have listed all the vertices and colors in our graph @xmath26 and the functioning of our gadgets , but we will need slightly more edges . the reasoning for the edges is as follows : if presents any vertex of the type @xmath64 or @xmath62 before presenting the variable vertices , or in the case when the variable vertices are presented out of the quantifier order , we want to give an advantage to so it can finalize the coloring . this will be achieved by allowing to treat all remaining universal vertices as existential vertices , i.e. , can recognize which of the two universal vertices @xmath65 corresponds to setting @xmath42 to @xmath40 . to be precise , we add the following edges to @xmath26 : every existential vertex @xmath66 is connected to all previous universal vertices @xmath34 , that is to all such @xmath34 for which @xmath67 . every universal vertex @xmath65 is connected to all previous universal vertices @xmath34 such that @xmath67 . every vertex of type @xmath51 is connected to all the universal vertices @xmath68 for @xmath69 . note that @xmath51 is connected either to @xmath34 , or to @xmath35 ; we do not add an edge to such vertices . every vertex of type @xmath57 is connected to all the universal vertices @xmath34 for all @xmath32 . the vertex @xmath62 is connected to all the universal vertices @xmath34 for all @xmath32 . we call all non - precolored vertices the _ gadgets _ for variables and clauses . the full analysis can be found in appendix [ app : section2 ] . we now make a step to the general case without precoloring by reducing the size of the precolored part so that it has only logarithmic size . our construction is based on the one with a large precolored part ; namely , all the vertices @xmath70 ( the gadgets for variables and clauses ) and the whole color clique @xmath31 will be connected the same way . let @xmath26 denote the gadgets for variables and clauses and @xmath31 . since @xmath31 is now not precolored and may send it after the gadgets , we help by a structure for recognizing vertices in @xmath26 or for saving colors . we remark that there is also a simpler construction with a logarithmic number of precolored vertices . if we just add precolored vertices to recognize vertices in @xmath26 , the following proof would work and be easier . however , when we replace a precolored vertex @xmath1 by some non - precolored graph in section [ sec : noprecol ] , we will use some conditions on the graph @xmath71 that this construction would not satisfy . our structure will consist of many small _ nodes _ , all of them have the same internal structure , only their adjacencies with other vertices vary . each node consists of three vertices and a single edge ; vertices are denoted by @xmath72 and the edge leads between @xmath73 and @xmath74 . we call the vertices @xmath75 and @xmath73 the _ lower partite set _ of the node , @xmath74 form the _ upper partite set_. see figure [ fig : node ] for an illustration of a node . r2.8 cm clearly , the online chromatic number of a node is two . the intuition behind the nodes is as follows : if presents vertices of a node in the correct way , needs to use two colors in the lower partite set of the node . no color can be used in two different nodes . each vertex @xmath76 ( in the gadgets and in @xmath31 ) has its own node @xmath77 that identifies @xmath1 . if the vertex @xmath74 from @xmath77 does not arrive before @xmath1 is sent , can color @xmath74 and @xmath1 with the same color , thus save a color . otherwise , can use the node to recognize @xmath1 . universal vertices @xmath78 for each universally quantified variable @xmath33 should be distinguishable only by the same vertices as in the previous section . therefore they are identified by the same two nodes . let @xmath79 be the number of vertices in @xmath26 . we create @xmath79 nodes , denoted by @xmath80 , one for each vertex in @xmath26 . for any two distinct nodes @xmath81 and @xmath82 ( @xmath83 ) , there is an edge between each vertex in @xmath81 and each vertex in @xmath82 . therefore , no color can be used in two nodes . we now define which nodes identifies which vertices and add edges between the original construction @xmath26 and nodes . let @xmath84 be the vertices in @xmath26 ( in an arbitrary order ) . then @xmath81 identifies the vertex @xmath85 . moreover , if @xmath85 is a vertex @xmath55 or @xmath86 for a universally quantified variable @xmath87 and @xmath88 is the other vertex , then @xmath82 also identifies @xmath85 and @xmath81 also identifies @xmath88 . thus each node identifies one or two vertices and each vertex is identified by one or two nodes . for a vertex @xmath76 and for a node @xmath77 , if @xmath77 identifies @xmath1 , then we connect only the whole lower partite set of @xmath77 to @xmath1 , i.e. , we add two edges from @xmath1 to both @xmath75 and @xmath73 of @xmath77 . otherwise , we add three edges one between @xmath1 and every vertex in @xmath77 . the only precolored part @xmath89 of the graph is intended for distinguishing nodes . since there are @xmath79 nodes in total , we have @xmath90 precolored vertices @xmath91 with no edges among them . precolored vertices have a color that may be used later for coloring @xmath26 ( the gadgets and @xmath31 ) . for simplicity , we again use the precoloring in the strong sense , i.e. , is able to recognize which precolored vertex is which . we connect all vertices in the node @xmath81 to @xmath92 if the @xmath93-th bit in the binary notation of @xmath32 is 1 ; otherwise @xmath92 is not adjacent to any vertex in @xmath81 . clearly , the node to which an incoming vertex belongs can be recognized by its adjacency to the precolored vertices . note that a vertex from nodes is connected to at least one precolored vertex and there is no edge between @xmath26 and precolored vertices . so far , we have introduced all vertices and edges in our construction of the graph @xmath71 . the full analysis can be found in appendix [ app : section3 ] . in this section we show how to replace one precolored vertex by a large nonprecolored graph whose size is a constant factor of the size of the original graph , while keeping s winning strategy in the case of a satisfiable formula . s winning strategy in the other case is of course preserved also and easier to see . we prove the following lemma which holds for all graphs with precolored vertices satisfying a few assumptions . [ l : removeprecolvertex ] let @xmath0 be a graph with precolored subgraph @xmath94 created from a fully quantified formula @xmath95 , and let @xmath96 be its precolored vertex . let @xmath97 be the induced subgraph with all nonprecolored vertices that are _ not _ connected to @xmath98 and let @xmath99 be the induced subgraph with all nonprecolored vertices that are connected to @xmath98 . let @xmath3 be an integer . assume that the following holds : @xmath5 if and only if @xmath95 is satisfiable , in the winning strategy of in the case if @xmath95 is satisfiable , can color @xmath99 using before two nonadjacent vertices from @xmath97 arrive . moreover , in this case if assigns the same color to a vertex in @xmath97 and to a vertex in @xmath99 before two nonadjacent vertices from @xmath97 arrive , can still color @xmath0 using @xmath3 colors . then there exists an integer @xmath100 and a graph @xmath101 with the following properties : @xmath101 has only @xmath102 precolored vertices , and @xmath103 , @xmath101 can be constructed from @xmath0 in polynomial time with respect to the size of @xmath0 , it holds that @xmath104 if and only if @xmath95 is satisfiable . the proof of theorem [ thm : main ] using lemma [ l : removeprecolvertex ] is in appendix [ app : proofoftheorem ] . * construction of @xmath101 . * let @xmath79 be the total number of vertices in @xmath97 and @xmath99 and let @xmath105 . our graph @xmath101 consists of precolored part @xmath106 , graphs @xmath97 and @xmath99 and three huge cliques @xmath107 and @xmath108 of size @xmath109 ; cliques @xmath107 and @xmath108 together form a _ supernode_. we keep the edges inside and between @xmath97 and @xmath99 and the edges between @xmath110 and @xmath111 as they are in @xmath0 . we add a complete bipartite graph between cliques @xmath112 and @xmath108 , i.e. , @xmath113 forms a clique of size @xmath114 . no vertex in @xmath77 is connected to @xmath112 or @xmath108 . in other words , the supernode is created from a node by replacing each vertex by a clique of size @xmath109 and the only edge in the node by a complete bipartite graph . there are no edges between the supernode ( cliques @xmath77 and @xmath113 ) and a precolored vertex in @xmath110 . it remains to add edges between the supernode and @xmath115 . there is an edge between each vertex in @xmath99 and each vertex in the supernode , while every vertex in @xmath97 is connected only to the whole @xmath77 and @xmath112 , but not to any vertex in @xmath108 . the fact that @xmath97 and @xmath108 are not adjacent at all is essential in our analysis . our construction is depicted in figure [ fig : supernode ] . where parts @xmath107 and @xmath108 form a supernode and parts @xmath97 and @xmath99 form the original graph @xmath0 ( the remaining precolored vertices are not shown ) . a thick line connecting two parts of @xmath101 corresponds to a complete bipartite graph between them . the edges between @xmath97 and @xmath99 remain unchanged from the original graph @xmath0 . if there is no line between two parts , then there is no edge between them in @xmath101.,width=151 ] let @xmath101 be the graph defined as above . note that the number of vertices in @xmath101 is at most @xmath116 , @xmath101 can be constructed from @xmath0 in polynomial time and @xmath101 has only @xmath102 precolored vertices . therefore , it remains to prove @xmath104 if and only if @xmath95 is satisfiable for some @xmath100 . we set @xmath100 to @xmath117 , since there will be at most @xmath114 colors used in the supernode . assuming that @xmath95 is not satisfiable , it is straightforward to design a winning strategy for on @xmath101 ; we only need to adapt the approach of sections [ sec : largeprecol ] and [ sec : smallprecol ] . see appendix [ app : noprecolwinstratd ] for a full description of the strategy . in the rest of this section we focus on the opposite direction : assuming that @xmath95 is satisfiable , we show that can color @xmath101 with @xmath100 colors regardless of the strategy of . in the following , when we refer to the colored part of @xmath101 , we do not take precolored vertices into account . actually does not look at precolored vertices unless it uses its winning strategy for coloring @xmath0 , which exists by the assumptions . * intuition . * at the beginning has too little data to infer anything about the vertices . therefore , shall wait for two nonadjacent vertices from @xmath97 and for two large cliques ( larger than @xmath118 ) with a small intersection . before such vertices arrive , it will color greedily . note that the greedy coloring algorithm eventually stops before everything is colored . having two large cliques , one mostly from @xmath77 and the other mostly from @xmath113 , and two nonadjacent vertices from @xmath97 , is able to recognize where an incoming vertex belongs . therefore , can use the supernode like a precolored vertex and colors the remaining vertices from @xmath97 and @xmath99 by its original winning strategy on @xmath0 . this approach may fail if a part of @xmath97 is already colored by s application of the greedy strategy . to remedy this , we prove that colors used on @xmath97 so far are also used in @xmath108 or @xmath99 , or will be used on @xmath108 later . the other obstacle is that might not be able to distinguish between one clique from @xmath97 and vertices in @xmath77 if nothing from @xmath112 arrives . nevertheless , each vertex @xmath24 in such a `` hidden '' clique is connected to all other colored vertices in @xmath97 and to the whole colored part of @xmath99 , otherwise it would be distinguishable from vertices in @xmath77 . hence , it does not matter on the color of @xmath24 . in summation , the sheer size of the supernode should allow the player to be able to use it as if it would be precolored . still , we need to allow for some small margin of error . this leads us to the following definition : [ dfn : almost ] let @xmath79 be the number of vertices of @xmath111 as in the construction of @xmath101 . for subgraphs @xmath119 , we say that @xmath120 is _ practically a subgraph _ of @xmath121 if @xmath122 , and @xmath120 is _ practically disjoint _ with @xmath121 if @xmath123 . we also say that a vertex @xmath1 is _ practically universal _ to a subgraph @xmath124 if it is adjacent to all vertices in @xmath120 except at most @xmath79 of them . similarly , we say a vertex @xmath1 is _ practically independent _ of a subgraph @xmath124 if @xmath1 has at most @xmath79 neighbors in @xmath120 . at first , the player uses the following algorithm for coloring incoming vertices , which may stop when it detects two useful vertices @xmath125 and @xmath126 : while the algorithm may seem to use a huge amount of computation for one step , we should realize that we are not concerned with time complexity when designing the strategy for . in fact , even a non - constructive proof of existence of a winning strategy would be enough to imply existence of a pspace algorithm for finding it we have observed already in section [ sec : intro ] that online chromatic numberlies in pspace . let @xmath1 be the incoming vertex @xmath24 when stops ; note that @xmath1 is not colored by the algorithm and @xmath1 can be from any part of @xmath101 . one of the cliques @xmath127 and @xmath128 is practically a subgraph of @xmath129 and we denote this clique by @xmath130 . the other clique must be practically a subgraph of @xmath77 and we denote it by @xmath131 . ( keep in mind that both cliques may contain up to @xmath79 vertices from @xmath115 . ) we remark that some vertices from @xmath108 must have arrived , as @xmath77 and @xmath112 alone are indistinguishable by step [ wfd : if ] of . by the same argument , the player knows whether @xmath132 or @xmath133 . let @xmath125 and @xmath126 be the nonadjacent vertices that caused the algorithm to stop . we observe that @xmath134 by eliminating all other possibilities : neither of @xmath125 and @xmath126 can be from @xmath99 , since any vertex of @xmath99 is practically universal to both cliques . both @xmath125 and @xmath126 can not be from @xmath113 or both from @xmath77 , as they would be adjacent . if @xmath125 is in @xmath113 and @xmath126 in @xmath77 , then we have a contradiction with the fact that @xmath125 and @xmath126 are not practically universal to the same clique . if @xmath135 and @xmath126 would be from @xmath77 or @xmath112 , then @xmath125 and @xmath126 are adjacent . finally , if @xmath135 and @xmath136 , then the clique to which they are not practically universal can not be the same for both , since @xmath125 is universal to the whole @xmath77 and @xmath126 to the whole @xmath113 . having cliques @xmath137 and @xmath130 and vertices @xmath134 , uses the following rules to recognize where an incoming or an already colored vertex @xmath24 belongs : if @xmath24 is practically universal to both @xmath130 and @xmath137 , then @xmath138 . if @xmath24 is practically universal to @xmath130 and practically independent of @xmath137 and @xmath24 is adjacent to @xmath125 , then @xmath139 . if @xmath24 is practically universal to @xmath130 and practically independent of @xmath137 , but there is no edge between @xmath125 and @xmath24 , then @xmath140 . if @xmath24 is not practically universal to @xmath130 , but it is practically universal to @xmath137 , then @xmath141 or @xmath142 . among such vertices , if there is a vertex not adjacent to @xmath24 or @xmath24 is not adjacent to a vertex in @xmath99 or @xmath24 is adjacent to a vertex in @xmath112 , then @xmath142 ; we say that such @xmath24 is _ surely in _ @xmath97 . otherwise , the player cannot yet recognize whether @xmath141 or @xmath142 . the reader should take a moment to verify that indeed , the set of rules covers all possible cases for @xmath24 . let @xmath143 be the colored parts of @xmath101 when stops . we observe that in the last case of the recognition the vertices from @xmath144 which are indistinguishable from @xmath77 form a clique ; we denote it by @xmath145 . note that all vertices in @xmath145 are connected to all vertices surely from @xmath97 that arrived and @xmath145 contains all vertices in @xmath144 that are not surely in @xmath97 . we stress that does not know @xmath145 or even its size . * intuition for the next step . * since can now recognize the parts of the construction ( with an exception of @xmath145 ) , we may basically use the winning strategy for on @xmath0 and on the rest . more precisely , creates a virtual copy of @xmath0 , adds vertices into it and simulates the winning strategy on this virtual graph . our main problem is that some part of @xmath97 ( namely @xmath144 ) is already colored . we shall prove that if @xmath144 is not a clique , can ignore colors used in @xmath144 ( but not the colors that it will use on @xmath97 ) , since they are already present in @xmath108 or @xmath99 or they may be used later in @xmath108 . if @xmath144 is a clique , it may be the case that @xmath108 and @xmath77 arrived completely and have the same colors , thus cannot ignore colors used on @xmath144 . another obstacle in the simulation is @xmath145 , the hidden part of @xmath97 . to overcome this , tries to detect vertices in @xmath145 and reclassify them as surely in @xmath97 . shall keep that all vertices in @xmath145 are connected to all currently colored vertices in @xmath97 and @xmath99 , therefore it does not matter much on colors in @xmath145 . when discovers a vertex from @xmath145 , it adds the vertex immediately to its simulation of @xmath0 . on the other hand , the size of @xmath145 increases when sends a vertex from @xmath97 which is indistinguishable from @xmath77 . the precise algorithm used by to finish the coloring of @xmath101 using @xmath100 colors and its analysis can be found in appendix [ app : noprecolwinstratp ] . * acknowledgments . * the authors thank christian kudahl and their supervisor ji sgall for useful discussions on the problem . _ effective coloration_. the journal of symbolic logic , vol . 469 - 480 ( 1976 ) . . _ first fit and on - line chromatic number of families of graphs_. ars combinatorica 29c , 168176 ( 1990 ) . . _ on - line graph coloring and finite basis problems_. combinatorics : paul erdos is eighty volume 1 . , 207214 ( 1993 ) . j. algorithms , 23 , 265280 ( 1997 ) . the electronic journal of combinatorics , 7(1 ) , r7 ( 2000 ) . theoretical computer science 130.1 : 163 - 174 ( 1994 ) . israel journal of mathematics 105 , 93 - 104 ( 1998 ) . . _ on - line graph coloring . _ master s thesis , university of southern denmark ( 2013 ) . . to appear in proceedings of _ 9th international conference on algorithms and complexity _ ( ciac 2015 ) . also arxiv 1406.1623 . annals of discrete mathematics 43 : 319 - 325 ( 1989 ) . the number of colors allowed for (the same as the size of @xmath31 ) is @xmath146 where @xmath28 is the number of clauses , @xmath147 the number of universally quantified variables and @xmath148 the number of existentially quantified variables . the next two lemmas contain the analysis of our construction . if anytime during the game uses a color not present in the color clique @xmath31 , the lemma is proven . we therefore assume this does not happen and show that the vertex @xmath62 can not be colored with any color present in the precolored clique @xmath31 . s strategy is to first present the vertices @xmath149 in the same order of the variables as in the formula . whenever sends a vertex @xmath34 or @xmath35 for a universally quantified variable @xmath33 , is not able to detect whether it is setting the value of @xmath33 to @xmath40 or @xmath50 . as the formula is not satisfiable , can therefore present these vertices in a sequence such that chooses the value of @xmath33 so that the final evaluation is false . we now know that all clauses are evaluated to @xmath50 . this means that at least one of the vertices @xmath58 of each clause @xmath7 is assigned the color @xmath59 . it follows that the vertex @xmath57 of this clause must be assigned the color @xmath60 . this holds for all clauses , thus the vertex @xmath62 has a neighbor of the color @xmath60 for each clause @xmath7 . hence , @xmath62 can not get any of the @xmath3 colors allowed and needs to be assigned a new color , which completes the proof . if presents a vertex , is able to recognize it by the set of edges to the clique @xmath31 , i.e. , by the set of allowed colors , with the exception of vertices @xmath78 for a universally quantified variable @xmath33 . if presents a vertex of type @xmath64 or @xmath62 before any of the universal vertices @xmath78 is presented for a universally quantified variable @xmath33 , will be able to distinguish the vertex @xmath34 from the vertex @xmath35 and therefore choose the assignment of @xmath33 . furthermore , for universal vertices @xmath78 and for a vertex @xmath1 for a variable @xmath42 with @xmath150 , if presents @xmath1 before any of these universal vertices is sent , is then able to distinguish the universal vertices @xmath34 and @xmath35 . the second observation is easy to see by noting that presenting @xmath1 means that it has an edge with @xmath34 , but not with @xmath35 ( or , in the case of @xmath51 , it may have an edge with @xmath35 , but not with @xmath34 , but this is symmetric ) . this means that we can distinguish @xmath34 from @xmath35 by the presence or absence of the edge . we continue with the proof of the lemma . we say that a variable @xmath33 is _ set _ , if at least one of its vertices was colored or if has assigned a value to it ( e.g. , if a vertex of a variable @xmath42 was sent before a vertex of @xmath33 for @xmath151 ) ; otherwise @xmath33 is _ unset_. until presents a vertex of type @xmath152 , or @xmath62 , colors an incoming vertex @xmath1 that corresponds to a variable @xmath33 using the following strategy : let @xmath153 be the set of all unset variables @xmath42 with @xmath154 , i.e. , all previous unset variables . chooses an assignment for all variables in @xmath153 according to satisfiability of the formula . then can color vertices for variables in @xmath153 according to this assignment , since it is now able to distinguish vertices of each universally quantified variable in @xmath153 . therefore every variable in @xmath153 becomes set . if @xmath33 is not set and it is quantified existentially , knows how to set this variable to satisfy the formula , thus colors @xmath1 according to the value of @xmath33 , since @xmath1 can be recognized . if @xmath33 is set and cannot recognize @xmath1 , then @xmath33 must be quantified universally and the other vertex for @xmath33 is colored , therefore there is a single color left for @xmath1 . ( note that in this case no vertex for a variable @xmath42 with @xmath155 arrived . ) when the vertex @xmath24 of type @xmath64 , or @xmath62 is sent , let @xmath153 be the set of all unset variables . chooses an assignment for all variables in @xmath153 according to satisfiability of the formula . then can color vertices for variables in @xmath153 according to this assignment , since it is now able to distinguish vertices of a universally quantified variable in @xmath153 . then all variables are set and decides how to color all remaining vertices in the graph using the following rules : a vertex of type @xmath57 obtains color @xmath59 if all of its adjacent vertices of type @xmath51 have colors of type @xmath156 , i.e. , the clause @xmath7 evaluates to @xmath40 ; otherwise it obtains color @xmath60 . [ o : eachnode2colors ] if reveals all the nodes @xmath81 before it sends any vertex from @xmath26 ( the gadgets and @xmath31 ) , can force to use two colors in the lower partite set of each node and these colors are different for each node . thus uses the following strategy for each node @xmath77 independently : first presents the vertex @xmath75 from @xmath77 ; let @xmath157 be its color . then sends a vertex @xmath158 that is one of vertices @xmath73 and @xmath74 . cannot deduce which of them is @xmath158 , because both are not connected to @xmath75 ( and both are connected to all vertices in other nodes ) . if assigns the color @xmath157 to @xmath158 , then @xmath159 and sends @xmath73 which must get another color than @xmath157 . otherwise if @xmath158 obtains another color than @xmath157 , then @xmath160 . [ o : nodestwocoloredbygreedy ] suppose that is using , i.e. , always assigns the smallest color not present among colored vertices adjacent to the incoming vertex . then there are at most two colors used on each node . @xmath161 : let the color of @xmath73 be @xmath157 and the color of @xmath74 be @xmath162 ( clearly @xmath163 ) . since the edges from @xmath75 and @xmath73 to other vertices in the construction except @xmath74 are exactly the same , can use @xmath157 for @xmath75 , but he can not use any color forbidden for @xmath73 when @xmath73 was colored with the possible exception of @xmath162 . hence @xmath75 obtains @xmath157 or @xmath162 , but in both cases the node @xmath77 has two colors . @xmath164 : let the color of @xmath75 be @xmath157 and the color of @xmath74 be @xmath162 . if @xmath163 , uses @xmath157 for @xmath73 for the same reason as in the previous case . otherwise @xmath165 and must color @xmath73 with another color , but there are two colors on @xmath77 . @xmath166 : we show that @xmath75 and @xmath73 have the same color . without loss of generality , @xmath75 arrives first and obtains a color @xmath157 . when @xmath73 arrives , it can be colored by @xmath157 , but by no other color @xmath167 , since @xmath168 would be available for @xmath75 also when @xmath75 was colored , because the edges from @xmath75 and @xmath73 to other vertices except @xmath74 are the same . hence uses another color for @xmath74 and @xmath77 is 2-colored . has @xmath169 colors for the @xmath79 nodes and @xmath3 colors for @xmath26 ( with the same names as in the previous section ) , thus overall is allowed to use @xmath170 colors . the precolored vertices have a color that may be used later in @xmath26 . if at any point of the game there are @xmath171 colors in the graph , the lemma is proven . s strategy is to first present all nodes ( in any order ) and force to use two colors in the lower partite set of each node . moreover , has to use different colors in distinct nodes . forcing such a coloring is possible by observation [ o : eachnode2colors ] . then sends @xmath31 and the situation is similar to the one with the precolored @xmath31 , since none of two colors used in the lower partite set of a node is allowed for a vertex in @xmath26 , thus must use the @xmath3 colors in @xmath31 for coloring the gadgets . is able to recognize vertices in @xmath31 by nodes , but nodes does not give an additional knowledge compared to @xmath31 . in particular , is not able to distinguish vertices @xmath78 for a universally quantified variable @xmath33 when one of them arrives . hence we conclude the proof by applying lemma [ l : precolnotsat ] . in the case of a satisfiable formula , nodes and precolored vertices become important . s strategy to win is basically the following : color vertices in nodes greedily until a vertex @xmath24 from @xmath26 arrives . at this time , some ( but maybe not all ) vertices in the gadgets can be recognized by their nodes . uses the winning strategy from lemma [ l : precolsat ] on the gadgets for vertices that it can recognize , even if @xmath31 has arrived only partially . for each vertex @xmath1 that can not be recognized , is able to color ( or already colored ) the lower partite set of the @xmath1 s node @xmath77 with only one color , therefore it can use the same color for @xmath1 and the vertex @xmath172 , since they are not adjacent . in the following proof , waits for two nonadjacent vertices in @xmath26 , even although one vertex from @xmath26 suffices . the reason is the we shall need such a condition in section [ sec : noprecol ] when we remove precolored vertices . let @xmath173 be the set of @xmath169 colors for nodes and let @xmath174 be the set of @xmath3 colors for @xmath26 ( the gadgets and @xmath31 ) including the color used on precolored vertices ; both @xmath173 and @xmath174 are ordered arbitrarily and @xmath175 . let @xmath176 and @xmath177 be the two nonadjacent vertices from the stopping condition of . observe that the nodes identifying @xmath176 and @xmath177 are different , since these nodes can be the same only for both universal vertices for one variable , but these vertices are adjacent . if @xmath24 is from @xmath26 and can recognize it by a nonadjacent vertex @xmath1 from nodes ( line [ alg - ln : win - precol - g1vertrecognized ] ) then coloring it with the winning strategy is correct , since vertices that have a color from @xmath174 in @xmath71 have the same color in @xmath178 . otherwise , if cannot recognize @xmath24 from @xmath26 ( line [ alg - ln : win - precol - g1vertnotrecognized ] ) , the vertex @xmath74 from the node @xmath77 that serves for identifying @xmath24 is not yet sent . it follows that there is at most one color used in the lower partite set of @xmath77 by observation [ o : nodestwocoloredbygreedy ] . if @xmath24 is from a node @xmath77 , a vertex @xmath1 from @xmath26 identified by @xmath77 is colored by @xmath157 , there is no edge between @xmath1 and @xmath24 , @xmath179 and @xmath157 is not present among nodes ( line [ alg - ln : win - precol - nodessavecolor ] ) , then @xmath24 is the vertex @xmath74 of the node @xmath77 . there is at most one color used in the lower partite set of @xmath77 before coloring @xmath24 by observation [ o : nodestwocoloredbygreedy ] and that @xmath157 is not used anywhere except @xmath1 . an unrecognized vertex @xmath1 from @xmath26 which is not a universal vertex has the same color as the vertex @xmath74 from the single node identifying @xmath1 , since @xmath77 identifies only @xmath1 and other vertices are adjacent to @xmath74 . an unrecognized universal vertex @xmath1 identified by two nodes @xmath77 and @xmath180 obtains some color @xmath179 . let @xmath162 be the color of the other vertex @xmath181 for the same universal variable . vertices @xmath74 from @xmath77 or @xmath180 arrive after @xmath1 and one of them obtains @xmath157 , since the algorithm prefers to use colors of @xmath1 and @xmath181 if their colors are in @xmath173 . recognized vertices from @xmath26 are colored by the wining strategy on @xmath178 using @xmath3 colors from @xmath174 . as we have shown , each unrecognized vertex @xmath1 has the same color as a vertex @xmath74 in one of the nodes identifying @xmath1 . there are at most @xmath169 colors used in nodes by observation [ o : nodestwocoloredbygreedy ] and all of them are from @xmath173 . thus uses at most @xmath170 colors . suppose that @xmath95 is not satisfiable and consider @xmath101 constructed from @xmath95 . starts by sending @xmath77 and then @xmath113 such that no color is both in @xmath77 and in @xmath112 ; this is possible , since cliques @xmath77 and @xmath108 have the same size . more precisely , if would assign a color from @xmath77 to a vertex from @xmath113 , decides that the vertex is in @xmath108 ; otherwise the vertex is in @xmath112 . this forces to use @xmath114 different colors on @xmath77 and @xmath112 and no such color can be used for @xmath97 and @xmath99 . now the colored part of @xmath101 ( precolored vertices and the supernode ) may be viewed as a precolored part in @xmath0 except that may not be able to use the supernode in the same way as a precolored vertex . since this only helps , we use the `` only if '' direction of the first assumption , i.e. , sends vertices according to its winning strategy for @xmath0 . this proves the `` only if '' direction of the first proposition of lemma [ l : removeprecolvertex ] . after the algorithm finishes , the player applies an algorithm which simulates its winning strategy on @xmath0 and maintains disjoint sets of colors @xmath173 ( for the supernode and @xmath97 ) and @xmath182 ( mainly for @xmath99 ) . recall that is not able to distinguish between @xmath77 and @xmath145 , thus vertices in @xmath145 are treated in the same way as vertices in @xmath77 . more precisely , we define three color sets @xmath173 , @xmath183 and @xmath182 as follows : if colored vertices surely in @xmath97 form a clique , let @xmath183 be the set of colors that are used on a colored vertex surely in @xmath97 and that are not used in @xmath184 . otherwise , if there are two nonadjacent colored vertices surely in @xmath97 , let @xmath183 be an empty set . let @xmath173 be the set of colors present currently in the supernode or in @xmath145 and among vertices surely in @xmath97 ( i.e. , in @xmath185 , or @xmath144 ) except colors from @xmath183 and colors present also in @xmath186 . we will update the color sets when we apply our algorithms and as more vertices arrive on input ; the precise updating procedure is specified throughout the algorithms . the winning algorithm for is split into two parts , initialization and coloring : when adding a colored vertex @xmath24 that is surely in @xmath97 to the virtual graph @xmath178 , if @xmath187 , coloring @xmath24 with @xmath157 in @xmath178 does not harm the winning strategy by the assumptions of lemma [ l : removeprecolvertex ] if @xmath183 is nonempty , colored vertices surely in @xmath97 form a clique and may use any proper coloring for them in the winning strategy . moreover , some colors may be both in @xmath144 and in @xmath186 , but the winning strategy still works by the assumptions . otherwise , if @xmath188 , @xmath24 may obtain a different color in @xmath178 than in the actual @xmath101 . we remark that the first vertex @xmath24 colored by the algorithm is @xmath1 , the vertex on which stops . in a sense , if @xmath144 , the part of @xmath97 colored by , is not a clique , pretends that in @xmath144 there are colors from the simulation , not the colors assigned by . otherwise , if @xmath144 is a clique , uses colors from @xmath101 in @xmath178 which may result in renaming the colors in the winning strategy on @xmath178 . observe that any color used in @xmath101 is in @xmath173 or @xmath182 and that these sets of colors are disjoint . thus the colors used for @xmath115 are different from those in the supernode except colors used in @xmath144 or colors of vertices that were in @xmath145 when they arrived . observe that if @xmath24 is from @xmath99 or surely from @xmath97 , coloring with @xmath157 is sound ; this follows from the winning strategy on @xmath0 and from the fact that colors in @xmath182 are not used in the supernode ( recall that colors in @xmath183 are not used in @xmath184 ) . it remains to show the following claim : [ clm : colorsfromdinc ] for each vertex @xmath24 from @xmath144 ( either surely from @xmath97 or from @xmath145 when starts ) colored with @xmath157 in @xmath101 , the color @xmath157 is present in @xmath178 , or in @xmath108 when the whole @xmath101 is colored . we assume that @xmath157 is not used in @xmath178 , otherwise the claim holds for @xmath24 . colors in @xmath144 and not in @xmath178 ( thus also not in @xmath183 ) are in @xmath173 . observe that @xmath157 can be used only for a vertex in @xmath108 . let @xmath189 be the time of coloring @xmath24 . if at time @xmath189 there is some color in @xmath108 not used in @xmath77 , let @xmath168 be the smallest such color in the ordering of colors . we observe that uses the color @xmath168 for @xmath24 , since other colors allowed for coloring @xmath24 are some colors used in @xmath99 at time @xmath189 ( which we assume that @xmath24 does not get ) , or a new color not yet used anywhere , but the new color is greater than @xmath168 . hence the greedy algorithm prefers @xmath168 and @xmath190 . otherwise , when colors @xmath24 , all colors used in @xmath108 are also present in @xmath77 which also means that there are at least as many colored vertices in @xmath77 as in @xmath108 at time @xmath189 . in this case we show that @xmath157 will be used in @xmath108 after time @xmath189 . suppose that at least @xmath79 vertices from @xmath108 are colored by after @xmath24 arrives . then , must assign @xmath157 to a vertex in @xmath108 at some point , since it prefers colors used in @xmath144 before colors in @xmath77 or new colors , plus we know at least @xmath191 vertices from @xmath108 will arrive after @xmath24 . assuming the opposite , more than @xmath192 vertices arrived from @xmath108 before time @xmath189 , which means that also more than @xmath192 vertices arrived from @xmath77 . since @xmath24 is colored by , there can not be two nonadjacent vertices from @xmath97 at time @xmath189 , because otherwise the algorithm would stop . observe that in this case stops immediately after there are two nonadjacent vertices in @xmath97 , thus @xmath144 is a clique and in particular , @xmath157 is in @xmath178 or @xmath157 was used in @xmath184 which concludes the proof . it follows that all colors in @xmath173 are used in the supernode . there are always @xmath114 colors used in the supernode , since otherwise there must be a color @xmath157 in @xmath77 that is not in @xmath113 , but both and prefer coloring @xmath112 using colors in @xmath77 and coloring @xmath77 using colors from @xmath113 . thus @xmath193 . since all colors in @xmath182 are used in the virtual graph @xmath178 ( this includes also colors in @xmath183 ) and uses the winning strategy on @xmath178 , we know that @xmath194 . overall , uses at most @xmath195 colors and we proved @xmath104 for @xmath101 constructed from a satisfiable formula @xmath95 . this concludes the proof of lemma [ l : removeprecolvertex ] . let @xmath95 be a formula of size @xmath20 . we first construct a graph @xmath71 with @xmath196 precolored vertices @xmath91 as described in section [ sec : largeprecol ] and [ sec : smallprecol ] . then for each precolored vertex , we apply lemma [ l : removeprecolvertex ] iteratively until we obtain a graph @xmath197 with no precolored vertex . we check that the graph satisfies both assumptions of lemma [ l : removeprecolvertex ] in each iteration @xmath32 . the first assumption , @xmath5 if and only if @xmath95 is satisfiable , holds for the first iteration by lemmas [ l : smallprecolsat ] and [ l : smallprecolnotsat ] , and for every other iteration by the previous use of lemma [ l : removeprecolvertex ] . recall the second assumption : in the winning strategy of in the case if @xmath95 is satisfiable , can color @xmath99 using before two nonadjacent vertices from @xmath97 arrive . moreover , if assigns the same color to a vertex in @xmath97 and a vertex in @xmath99 before two nonadjacent vertices from @xmath97 arrive , can still color @xmath0 using @xmath3 colors . let @xmath198 be @xmath71 and for each iteration @xmath32 let @xmath199 be the graph before removing the @xmath32-th precolored vertex @xmath200 . let the nonprecolored vertices of @xmath199 be partitioned into two disjoint induced subgraphs @xmath201 and @xmath202 such that all vertices from @xmath202 are connected to @xmath200 and no vertex in @xmath201 is connected to @xmath200 . let @xmath203 be the graph created from @xmath199 by the construction in lemma [ l : removeprecolvertex ] ; we denote by @xmath204 and @xmath205 the cliques in the supernode . thus @xmath203 consists of parts @xmath206 , and @xmath202 ( and possibly some precolored vertices which we do not take into account ) . note that @xmath202 contains only some nodes and @xmath201 contains the whole @xmath26 ( i.e. , @xmath31 and the gadgets for variables and clauses ) together with supernodes from previous iterations and nodes that are not in @xmath202 . the first part of the second assumption holds for the first iteration , since the algorithm from the winning strategy from lemma [ l : smallprecolsat ] colors the nodes by before two nonadjacent vertices from @xmath26 arrive and @xmath207 contains @xmath26 . for the `` moreover '' part in the first iteration , if assigns the same color @xmath157 to a vertex @xmath162 in @xmath207 and a vertex @xmath208 in @xmath209 before two nonadjacent vertices from @xmath207 arrive , then @xmath208 must be from a node that identifies @xmath162 and in this case can just pretend that @xmath162 is an unrecognized vertex . in other words , the color @xmath157 is intended for nodes and not used in @xmath26 . for an iteration @xmath210 , recall that @xmath203 consists of parts @xmath206 , and @xmath202 ( except precolored vertices ) and that @xmath201 and @xmath202 form the graph @xmath199 from the previous iteration . similarly , @xmath199 consists of @xmath211 , and @xmath212 . note that @xmath213 , and @xmath214 are in @xmath201 and that there may be some nodes both in @xmath201 and in @xmath212 and also some nodes both in @xmath202 and in @xmath215 . the first part of the second assumption holds for the iteration @xmath32 , because in the winning strategy on @xmath199 colors @xmath199 using before two large and practically disjoint cliques from the supernode arrive , one mostly from @xmath216 and the other mostly from @xmath217 . since there are many pairs of nonadjacent vertices between these cliques and these cliques are contained in @xmath201 , can color @xmath202 using before two nonadjacent vertices from @xmath201 arrives . for the `` moreover '' part , if assigns the same color @xmath157 to a vertex @xmath162 in @xmath201 and a vertex @xmath208 in @xmath202 before two nonadjacent vertices from @xmath201 arrive , then @xmath208 is from nodes ( as @xmath202 contains only nodes ) and @xmath162 is either from @xmath26 as in the first iteration , or from the supernode created in the previous iteration ; in the latter case , @xmath208 is in @xmath215 and @xmath162 is in @xmath214 , since otherwise they would be connected . ( note that @xmath162 can not be from nodes , since there is complete bipartite graph between every two nodes and a node is never split between @xmath99 and @xmath97 . ) in both cases , this does not harm the winning strategy of on @xmath199 , because the algorithm assumes that may use the same colors for more vertices in @xmath218 or in @xmath219 . this concludes that the assumptions of lemma [ l : removeprecolvertex ] are satisfied in each iteration . since @xmath220 iff @xmath95 is satisfiable , it holds that @xmath221 iff @xmath95 is satisfiable by lemma [ l : removeprecolvertex ] ( for some @xmath100 ) . the number of vertices in @xmath197 is polynomial in @xmath20 , because @xmath71 has linearly many vertices and the number of vertices is multiplied by a constant with each of logarithmically many applications of lemma [ l : removeprecolvertex ] . the constructions in sections [ sec : largeprecol ] and [ sec : smallprecol ] and lemma [ l : removeprecolvertex ] yield a polynomial - time algorithm for computing @xmath101 from @xmath95 .

in recent years , the study of entanglement in quantum many - body systems has become a common issue in condensed matter physics , statistical mechanics , and quantum information theory . entanglement is a purely quantum mechanical phenomenon in which quantum systems are linked together even when they are spatially separated so that one system can not be correctly described without full mention of its counterpart . this concept was first introduced by schrdinger in @xcite . from the viewpoint of quantum information science , entanglement is a fundamental measure of how much quantum effects we can observe and use to control one quantum system by another , and it is the primary resource for quantum information processing and communication @xcite . on the other hand , in condensed matter physics and statistical mechanics , the concept of quantum entanglement has recently been used to investigate quantum phase transitions @xcite ( see @xcite for a review ) , topological order @xcite , and macroscopic properties of solids @xcite . a fundamental and practical problem common in all the areas is how to detect entanglement @xcite and quantify the degree of it . a general approach is to study the reduced density matrix of a certain subsystem in an entangled state . the concept of the reduced density matrix was first introduced by dirac in @xcite . there are many kinds of measures of entanglement related to the reduced density matrix ( see @xcite and references therein ) . among them , one of the most popular characteristic functions is the von neumann entropy ( entanglement entropy ) of a reduced density matrix @xcite . it serves as the measure of entanglement for a pure state . generally speaking , to elucidate quantum entanglement in many - body systems is a formidable task since we need a precise description of many - body quantum states . so far , much insight has been obtained by studying exactly solvable models in which there is a possibility to obtain the reduced density matrix and calculate the von neumann entropy exactly or approximately . representative models are the harmonic models @xcite , the xy spin chain @xcite , the xxz spin chain @xcite , the calogero - sutherland model @xcite , one - dimensional critical models @xcite , conformal field theories @xcite , the affleck - kennedy - lieb - tasaki ( aklt ) model @xcite , the kitaev model @xcite , and the quantum dimer ( rokhsar - kivelson ) model @xcite . among them , the aklt model @xcite has some important meanings since this model is directly related to one of the schemes of quantum computation , measurement based quantum computation @xcite . historically , the aklt model and its exact ground state known as the valence - bond - solid ( vbs ) state were proposed to understand ground - state properties of the haldane gap systems @xcite . the authors ( aklt ) rigorously proved that the vbs state is a unique ground state and there is an energy gap immediately above the ground state . exponential decay of correlation functions in the vbs ground state has also been shown . there is an intimate relation between the vbs states and fractional quantum hall states which has been revealed by using a spin coherent state representation @xcite . a hidden topological order in the vbs state is characterized by string order parameters @xcite and existence of edge states which are degenerate ground states in a chain with open boundary conditions @xcite . the vbs picture and the edge states were also experimentally supported by electron spin resonance ( esr ) in the @xmath1 haldane chain ni(c@xmath2h@xmath3n@xmath2)@xmath2no@xmath2(clo@xmath4 ) ( nenp ) @xcite . although most of the exactly solvable models are limited to one dimension , the aklt model and vbs ground state can be generalized to higher dimensions and/or inhomogeneous ( non - translational invariant ) lattices @xcite . this feature is especially important in connection with the measurement based quantum computation as we discussed later . for studying nature of the vbs states , the non - trivial observation has been done by klmper _ et al_. , through the @xmath5-deformation of the aklt model @xcite . they found that the one - dimensional ( 1d ) spin-1 vbs state can be expressed in a form of the matrix product state ( mps ) . this representation gives an efficient method to calculate correlation functions . mathematical theory and generalization of the vbs states , i.e. , finitely correlated states ( fcs ) , were essentially developed by fannes _ et al_. , in @xcite . another interesting generalization of the vbs state is to replace su(2 ) spins with other lie algebra generators . in this direction , there have been constructed many versions of the aklt model such as su(3 ) @xcite and su(@xmath6 ) modes @xcite , su(@xmath7)-extended model @xcite , symplectic model @xcite , so(5 ) model @xcite , so(@xmath8 ) model @xcite , supersymmetric model @xcite , and several @xmath5-deformations @xcite . the vbs state has recently been attracting renewed interest from the viewpoint of quantum information theory . in particular , it has been proposed that the vbs state can be used as a resource state in measurement - based quantum computation ( mqc ) . mqc is one of the schemes of quantum computation in which we can perform universal quantum computation using only measurements as computational steps ( see @xcite for an introduction ) . there are two kinds of measurement based models . one is the teleportation quantum computation ( tqc ) @xcite . the other is the so - called one - way quantum computation ( 1wqc ) or cluster state computation @xcite . in this model , any quantum gate array can be implemented as a pattern of single - qubit measurement on a highly entangled cluster state . note that 1wqc with one - dimensional cluster states is insufficient for universal quantum computation and thus two - dimensional cluster states are inevitably needed . it has then been proved in @xcite by verstraete and cirac that tqc and 1wqc are equivalent using the vbs picture . gross , eisert and collaborators have introduced novel schemes for mqc based on finitely correlated or tensor network state beyond the cluster state @xcite . another recent important development by brennen and miyake is that the vbs state ( more precisely dynamically coupled aklt chains ) can be used as a resource state in mqc instead of the cluster state @xcite . the vbs state is more advantageous to 1wqc than the cluster state since it is robust against noise due to the energy gap . implementations of the aklt hamiltonian using bosonic atoms @xcite or spin-1 polar molecules @xcite in optical lattices have also been proposed . in this paper , motivated by considerable current interest from several viewpoints , we study entanglement properties of vbs states on two - dimensional ( 2d ) graphs which is relevant to mqc . before discussing the 2d vbs states , let us briefly summarize the obtained results in 1d vbs states . the reduced density matrix and entanglement properties in the 1d spin-1 vbs state was first studied in @xcite . the authors found that the von neumann entropy of the block with the rest of the chain approaches a constant value exponentially fast , as the size of the block increases . the constant value is given by @xmath9 . then , the 1d integer-@xmath10 vbs states were studied in @xcite . in this case , the von neumann entropy of the large block with the rest is given by @xmath11 and this value is interpreted in terms of the edge states , the degenerate ground states of the open aklt chain . this picture leads to an idea that the reduced density matrix of the block in the aklt chain is _ exactly _ spanned by the degenerate ground states of the block hamiltonian ( see eq . ( [ block_ham ] ) for the definition ) . this idea has been confirmed in the series of papers @xcite and was summarized in the review @xcite . another interesting quantities such as the entanglement length @xcite , the geometric entanglement @xcite , the single - copy entanglement @xcite , and the boundary effect @xcite in the 1d spin-1 vbs state have also been studied . more generally , the entanglement of formation in the 1d fcs was estimated in @xcite and an area law in 1d gapped quantum system was proven in @xcite . compared to the 1d vbs states , few results have been obtained in 2d vbs states so far . entanglement properties of the vbs state on the cayley tree @xcite was studied in @xcite . the authors showed that the von neumann entropy of the block does not depend on the whole size of the system and its asymptotic value is linearly proportional to the number of valence bonds crossing the boundary . this gives an explicit proof of the area law in this system ( see @xcite for a review of area laws and the entanglement entropy ) . general entanglement properties of the vbs state on an arbitrary graph @xcite have been studied in @xcite . the authors showed that the eigenspace of the reduced density matrix of the block is spanned by the degenerate ground states of the block hamiltonian . the authors have shown that the von neumann entropy of a large block with the rest approaches a value less than @xmath12 , where @xmath13 is the ground - state degeneracy of the block hamiltonian . although some general results for 2d vbs states have been obtained so far , a quantitative analysis of entanglement was missing since a powerful analytical method such as a transfer matrix technique does not work sufficiently in the analysis of 2d systems . therefore , entanglement properties of the 2d vbs states must be analyzed with the aid of numerics . in this paper , we study entanglement properties of the 2d vbs states on symmetric graphs . symmetries such as reflection and inversion enable us to develop an efficient method to study the reduced density matrix . due to the symmetry , one can find the relation between the eigenvalues of the reduced density matrix and those of the _ overlap _ matrix ( see eqs . ( [ overlap1]-[overlap2 ] ) for the definition ) . the overlap matrix is a real symmetric and its elements are given by the inner product of the degenerate ground states of the block hamiltonian which are linearly independent but may not be orthonormal . the inner product can be , for example , calculated by monte carlo integration using the schwinger boson representation of the vbs state . even without numerical integration , one can formally write the overlap matrix as a matrix - valued correlation function in the vbs state restricted to one of the subsystems . from this , we conjecture that the von neumann entropy per valence bond ( boundary site ) is strictly less than @xmath0 even in the 2d infinite system . holographic _ interpretation of the deviation from @xmath0 is given in terms of the hidden spin chain along the boundary between two subsystems ( reflection axis ) . to confirm this conjecture , we numerically study the eigenvalues of the overlap matrix by exact diagonalization . since the dimension of the overlap matrix does not depend on the size of the whole system but is given by @xmath13 , we can study relatively large systems . from the eigenvalues of the reduced density matrix , we calculate the von neumann entropy . we find that the von neumann entropy per valence bond on the boundary is strictly less than @xmath0 for finite size systems . we also find a function which fits the obtained numerical results well . the long - distance behavior of this function strongly supports our conjecture . these numerical results are analytically or algebraically confirmed for the graphs with a ladder geometry , where transfer matrix technique can be applied . the organization of this work is as follows . in section 2 , we review the construction of the aklt model on an arbitrary graph and introduce the schwinger boson representation of the vbs state . in section 3 , we provide a general argument on the schmidt decomposition and discuss a relation between the reduced density matrix and the overlap matrix . we show how to apply this method to study the von neumann entropy of the vbs state . then we make a conjecture on the von neumann entropy in the 2d infinite system . in section 4 , numerical results of the vbs states on the square and hexagonal lattices are shown . in section 5 , the obtained numerical results are partly confirmed analytically for graphs in a ladder geometry . conclusions are given in the last section . in appendix , we provide a graphical interpretation of the overlap matrix in terms of closed loops and open strands . let us first define the _ basic _ aklt model on a connected graph . a graph consists of two types of elements , namely vertices ( sites ) and edges ( bonds ) . as shown in fig . [ fig : vbs_basic_model ] , a vertex is drawn as a ( large ) circle and an edge is as a solid line . has the spin @xmath14 in this graph . ] a graph is called _ connected _ if every pair of distinct vertices in the graph can be connected through some path . let @xmath15 be a connected graph , where @xmath16 and @xmath17 are the sets of vertices and edges , respectively . henceforth , we assume @xmath18 , otherwise there is no interaction between spins . the spin operator @xmath19 is located at vertex @xmath20 and its spin value is denoted by @xmath21 . in the basic model , we require that @xmath22 , where @xmath23 is the coordination number , i.e. , the number of incident edges connected to @xmath20 . to guarantee uniqueness of the ground state , this relation must be true for any vertex @xmath20 , including boundaries . let us now define the spin hamiltonian for the basic aklt model . the hamiltonian is written as a sum of interactions on all the edges : h= _ k , l e a(k , l ) _ s_k+s_l ( k , l ) . [ basic_aklt_ham ] here , @xmath24 is an arbitrary positive real coefficient ( it may depend on the edge @xmath25 ) and the operator @xmath26 projects the total spin on the edge @xmath25 , @xmath27 , on the subspace with the highest possible spin value @xmath28 . since the hamiltonian in eq . ( [ basic_aklt_ham ] ) is a linear combination of projection operators with positive coefficients , @xmath29 is positive semi - definite . the explicit form of the projector @xmath30 in terms of @xmath31 and @xmath32 is given by _ s_k+s_l(k , l)=_|s_k - s_l| j s_k+s_l-1 . [ projector_pi ] note that one can expand @xmath33 and hence the projector defined in eq . ( [ projector_pi ] ) is a polynomial in the scalar product @xmath34 of the degree @xmath35 , where @xmath36 is the minimum of @xmath21 and @xmath37 . as an example , consider a case where @xmath38 , which corresponds to the @xmath1 homogeneous aklt chain . the projector @xmath39 is written as _ 2(k , l)= ( s_k s_l)^2 + ( s_k s_l)+ . the projectors @xmath40 and @xmath41 are related to the hamiltonians on the hexagonal and the square lattices , respectively : @xmath42 in general , explicit expressions in terms of @xmath34 become more complicated . let us now discuss the ground state of @xmath29 in eq . ( [ basic_aklt_ham ] ) with condition s_k = z_k . the ground state is unique and known as the valence - bond - solid ( vbs ) state . although there are several possible representations for the vbs state ( see @xcite ) , the most convenient one for our purposes is the schwinger boson representation @xcite . in this representation , we introduce a pair of bosonic creation and annihilation operators at each vertex to realize su(2 ) lie algebra . we define a pair of bosonic operators @xmath43 and @xmath44 for each vertex @xmath20 with the canonical commutation relations : = [ b_k , b^_l ] = _ kl with all other commutators vanishing : = [ b_k , b_l]=[a_k , b_l]=[a_k , b^_l]=0 , ^(k , l ) . spin operators are represented by the schwinger bosons as s^+_k = a^_k b_k , s^-_k = b^_k a_k , s^z_k = ( a^_k a_k - b^_k b_k ) . to reproduce the dimension of the spin-@xmath21 hilbert space at each vertex @xmath20 , the following constraint on the total boson occupation number is required : a^_k a_k + b^_k b_k = 2 s_k . with this constraint in mind , the vbs ground state is written as |vbs = _ k , le ( a^_k b^_l - b^_k a^_l ) |vac , [ vbs_gs ] where the product runs over all edges and the vacuum @xmath45 is defined as the direct product of vacuums of each vertex , i.e. , |vac= _ k |vac^[k ] . note that @xmath46}\ra$ ] is destroyed by any annihilation operators @xmath43 or @xmath44 . using the schwinger boson representation , one can generalize the vbs state which is the ground state of the _ generalized _ aklt model . by associating a positive integer @xmath47 to each edge @xmath48 of @xmath49 , the generalized vbs state is written as |vbs_gen = _ k , le ( a^_k b^_l - b^_k a^_l)^m_k , l |vac . for the construction of the generalized aklt hamiltonian and the condition of the uniqueness of the ground state , please see original references @xcite . in this paper , we shall focus on the basic model , i.e. , @xmath50 for any @xmath25 . let us now prove that the vbs state in eq . ( [ vbs_gs ] ) is the zero - energy groud state of the basic aklt hamiltonian @xmath29 . to prove this , we have only to show for any vertex @xmath20 and edge @xmath25 ( i ) the total power of @xmath51 and @xmath52 is @xmath53 so that the spin value at the vertex @xmath20 @xmath21 , and ( ii ) the maximum value of the total spin of the edge @xmath25 is @xmath54 . ( i ) can be shown by expanding r.h.s . ( [ vbs_gs ] ) and finding that the total power of @xmath51 and @xmath52 is @xmath55 . let us prove ( ii ) . by counting the power of @xmath56 s minus @xmath57 s on the bond @xmath25 , we find that the maximal eigenvalue of @xmath58 is @xmath59 , i.e. , @xmath60 . since the state @xmath61 is invariant under global rotations , the maximum value of the total spin of the edge @xmath25 is @xmath54 . therefore , from ( i ) and ( ii ) , the vbs state @xmath62 in eq . ( [ vbs_gs ] ) is a zero - energy ground state of the basic hamiltonian @xmath29 . it was shown in @xcite that this ground state is unique . then , eq . ( [ pre_schmidt ] ) can be written as @xmath80}_\tau d^{[b]}_\eta } ( { \tilde x}^{\rm t}{\tilde y})_{\tau \eta } |e_\tau\ra \otimes |f_\eta\ra , \label{pre_schmidt2}\ ] ] where @xmath81 denotes matrix transpose . so far , we have considered a general case . let us now focus on the special situation where @xmath69}$ ] and @xmath70}$ ] are the same , i.e. , @xmath69}=m^{[b]}=m$ ] . furthermore , if @xmath82 is a real symmetric matrix , it can be diagonalized by some orthogonal matrix @xmath83 as @xmath84 . in such a case , eq . ( [ pre_schmidt2 ] ) becomes a simpler form : @xmath85 note that @xmath86}_\tau = d^{[b]}_\tau$ ] . one may think that the above situation is not generic . however , as we will see later , it is enough for our purpose to restrict ourselves to such a special case . let us now consider the density matrix for the total system consisting of @xmath65 and @xmath66 . it is defined by @xmath87 . then the reduced density matrix for @xmath65 is given by @xmath88 . in our case , it is given by @xmath89 therefore , the von neumann entropy in this bipartite partitioning is given by @xmath90 therefore , we can obtain the von neumann entropy in terms of the overlap matrix @xmath82 . to obtain @xmath91 , we need all the eigenvalues of @xmath82 . let us now apply the schmidt decomposition method to study entanglement in vbs states on graphs with symmetries such as reflection and/or inversion . for simplicity , we shall focus on the graphs with reflection symmetry . however , one can easily generalize our argument to graphs with inversion symmetry . we define reflection symmetry such that any site ( vertex ) in a subgraph @xmath65 has a reflection partner in @xmath66 and vice versa . a graphical example is shown in fig . [ fig : ref_sym ] . and @xmath92 denote the sets of sites ( vertices ) on the boundaries of subgraphs @xmath65 and @xmath66 . any site @xmath93 has a unique reflection symmetric partner @xmath94 . ] in the schwinger boson language , the vbs state on a reflection symmetric graph can be written as @xmath95 where @xmath96 and @xmath97 are sets of edges ( bonds ) in subsystems @xmath65 and @xmath66 while @xmath98 and @xmath92 are sets of vertices ( sites ) on the boundaries of @xmath65 and @xmath66 . we note that we only consider the _ basic _ vbs state which is the ground state of the basic aklt model . in this state , there is one valence bond on any edge @xmath99 . the generalization of our argument to generic vbs states is straightforward . to apply the schmidt decomposition , we slightly change the above expression by a local gauge transformation : @xmath100 and @xmath101 for @xmath102 . ( [ ref_vbs ] ) becomes @xmath103 since the graph is reflection symmetric , @xmath93 necessarily has a reflection symmetric partner @xmath104 ( see fig . [ fig : ref_sym ] ) . for simplicity , we henceforth assume that any @xmath105 is nearest neighbor to @xmath106 in the graph @xmath49 . therefore , the above expression can be written as follows : @xmath107}_{\{\al \ } } \ra \otimes |\phi^{[b]}_{\ { \al \}}\ra,\end{aligned}\ ] ] where @xmath108 @xmath109 corresponds to the spin state of the boundary state , i.e. , degenerate ground state of the _ block hamiltonian _ defined within the subsystem @xmath65 or @xmath66 . the block hamiltonians are explicitly defined as @xmath110 where @xmath111 is defined in eq . ( [ projector_pi ] ) . the states @xmath112}_{\{\al \ } } \ra$ ] and @xmath113}_{\ { \al \}}\ra$ ] are given by @xmath114 respectively , where @xmath115}\ra$ ] ( @xmath116}\ra$ ] ) is the vacuum for bosons in @xmath65 ( @xmath66 ) and @xmath117}\ra \otimes |{\rm vac}^{[b]}\ra$ ] . equations in ( [ phi_a_b ] ) immediately yield that overlap matrices @xmath69}$ ] and @xmath70}$ ] are the same . the matrix element of @xmath118}=m^{[b]})$ ] is given by @xmath119}|\prod_{\la i , j \ra\in{\cal b}_a}(a_i b_j - b_i a_j ) \prod_{k \in \lambda_a } ( a_k)^{1/2+\al_k } ( b_k)^{1/2-\al_k } \nonumber \\ & \times \prod_{k \in \lambda_a } ( a\dag_k)^{1/2+\be_k } ( b\dag_k)^{1/2-\be_k } \prod_{\la i , j \ra\in{\cal b}_a}(a\dag_i b\dag_j - b\dag_i a\dag_j)|{\rm vac}^{[a]}\ra . \label{integral0}\end{aligned}\ ] ] the rank of @xmath82 is @xmath120 , where we denote the number of elements in a set @xmath10 by @xmath121 . the matrix element of @xmath82 is real , i.e. , @xmath122 , because the commutations of bosons do not produce any complex phases . therefore , one can easily show the symmetric property @xmath123 . we now rewrite the above expression in terms of the spherical angles . this can be done by introducing spin coherent state representation . let us first introduce spinor coordinates @xmath124 at each vertex ( site ) @xmath20 : @xmath125 where @xmath126 and @xmath127 . then for a point @xmath128 on the unit sphere , the spin-@xmath21 coherent state is defined as follows : @xmath129}\ra.\ ] ] here we have fixed the overall phase ( u@xmath130 gauge degree of freedom ) since it has no physical content . the coherent state is not orthogonal but complete and the resolution of identity is given by @xmath131 where @xmath132 denotes the simultaneous eigenstate of @xmath133 and @xmath134 , and @xmath135 is the @xmath136-dimensional identity matrix . by inserting this resolution of the identity , eq . ( [ integral0 ] ) can be recast in an integral form : @xmath137 where we have used the fact @xmath138 . therefore , if we can obtain the matrix element of @xmath82 by some method such as monte carlo integration , the only thing to do is to diagonalize the matrix @xmath82 . once the eigenvalues of @xmath82 , i.e. , @xmath139 ( @xmath140 ) , are obtained , the von neumann entropy can be calculated through eq . ( [ ee_in_terms_of_m ] ) . let us now estimate the upper bound on the von neumann entropy . suppose that all the eigenvalues of @xmath82 are the same . this means that the subsystems @xmath65 and @xmath66 are _ maximally entangled_. then , the von neumann entropy is given by @xmath141 . this value gives an upper bound on the von neumann entropy , i.e. , @xmath142 . however , as we will see in the next section , numerical results indicate a deviation from this naive estimate and the von neumann entropy @xmath91 is strictly less than @xmath141 in 2d vbs . it is interesting to note that there is a _ holographic _ interpretation of the overlap matrix @xmath82 . the matrix @xmath82 can be regarded as an operator acting on the auxiliary one dimensional chain which is along the boundary of @xmath65 , i.e. , @xmath98 . from the fact that @xmath143 , one can rewrite @xmath82 as @xmath144 here we have used the fact that @xmath145 . note that the pauli matrices @xmath146 ( @xmath147 ) act on the auxiliary space , @xmath148}\ra , |\downarrow^{[i]}\ra \}$ ] . then using the following relation : @xmath149 the overlap matrix can be expressed as a matrix - valued correlation function : @xmath150}|\prod_{i \in \lambda_a } \left ( s_i+\frac{1}{2}+{\vec s}_i \cdot { \vec \sigma}_i \right)|{\rm vbs}^{[a]}\ra , \label{overlap_and_cor}\ ] ] where @xmath151}\ra \equiv \prod_{\la i , j \ra \in { \cal b}_a}(a\dag_i b\dag_j - b\dag_i a\dag_j ) |{\rm vac}^{[a ] } \ra \label{half_vbs}\ ] ] note that the matrix rank of @xmath152 for @xmath93 is @xmath153 while that for @xmath154 is @xmath155 . one can interpret the overlap matrix as a hamiltonian acting on @xmath156-spin space . the matrix element of this hamiltonian is governed by the boundary correlation in the half - vbs state given in eq . ( [ half_vbs ] ) . the naive upper bound value of the von neumann entropy , @xmath141 , corresponds to the strongly disordered limit where any boundary correlation among @xmath10-spins vanishes . however , as shown numerically in @xcite , the spin - spin correlation functions at small distances are nonzero in the vbs - type state even though they decay exponentially fast . therefore , the leading term in eq . ( [ overlap_and_cor ] ) is given by a constant plus a term proportional to the spin-@xmath157 heisenberg hamiltonian whose coupling @xmath158 is given by the nearest neighbor @xmath10-spin correlation function . the eigenvalues of this matrix can be different and hence the entanglement spectrum may not be flat . therefore , it is plausible that the von neumann entropy is strictly less than @xmath159 even in the 2d infinite system . to support this conjecture , we perform numerical and analytical calculations for the vbs states on square and hexagonal lattices in the following sections . in the previous section , we derived the integral formula for the overlap matrix ( eq . ( [ integral1 ] ) ) which helps us to reduce computational cost . let us explain this point in more detail . suppose that the graph @xmath49 has @xmath160 vertices and every vertex has the same spin @xmath10 . then , the needed dimension for representing the vbs state on @xmath49 is @xmath161 without consideration of symmetry . however , if we focus on the von neumann entropy of reflection symmetric vbs state , the needed dimension is greatly reduced from @xmath161 to @xmath120 , where @xmath162 denotes the number of sites on the boundary between @xmath65 and @xmath66 . therefore , we can study the von neumann entropy of 2d vbs state on a relatively large system by combining monte carlo integration and exact diagonalization . after obtaining the overlap matrix and all its eigenvalues , the von neumann entropy can be calculated according to eq . ( [ ee_in_terms_of_m ] ) . in this section , we calculate von neumann entropies of vbs states on square and hexagonal lattices with open boundary conditions using monte carlo integration . we first study the von neumann entropy of the vbs state on square lattices with open boundary conditions . the reflection axis of the square lattice is taken to be the boundary between @xmath65 and @xmath66 as shown in fig . [ fig : squarelattice ] . @xmath163 in fig . [ fig : squarelattice ] is the number of sites along @xmath164-axis while @xmath165 is that along @xmath166-axis . for square lattice , the number of boundary sites @xmath162 is the same as @xmath165 , i.e. , @xmath167 . and @xmath168 denote the numbers of sites along @xmath164- and @xmath166-axes in @xmath65 , respectively . ] we first study the @xmath163-dependence of the von neumann entropy . figure [ graph : ee - square - nx ] shows the von neumann entropy per boundary site , i.e. , valence bond , for various @xmath165 as a function of @xmath163 . the von neumann entropy per valence bond , @xmath169 , rapidly converges to certain values depending on @xmath165 . the case of @xmath170 reduces to the linear aklt chain along @xmath164-axis . in this case , @xmath171 monotonically decreases with increasing @xmath163 and approaches to @xmath172 exponentially fast . this correctly reproduces the exact result in @xcite . the obtained results show that @xmath169 for @xmath173 and @xmath174 approach to @xmath175 and @xmath176 , respectively . these values are consistent with the analytical results shown in the next section . for the square vbs states with various @xmath165 . the dotted lines indicate the exact results in @xmath177 limit . ] next we consider the @xmath165-dependence of the von neumann entropy . figure [ graph : ee - square - ny ] shows the von neumann entropy per boundary site as a function of @xmath165 . we find that @xmath169 is well fitted by the following function : @xmath178 where @xmath179 , @xmath180 and @xmath181 are fitting parameters . the constant term @xmath181 means the von neumann entropy per valence bond in the limit of @xmath182 . note that @xmath179 , @xmath180 and @xmath181 depend on @xmath163 . the blue curves in fig . [ graph : ee - square - ny ] represent the fitting curves . the obtained numerical data show good agreement with the power law behavior assumed in eq . ( [ eq : fitting ] ) . for the square lattice vbs states with various @xmath163 . blue curves represent the fitting curves ( eq . ( [ eq : fitting ] ) ) . ] .the coefficients of the fitting function eq . ( [ eq : fitting ] ) for square lattices . [ cols="^,^,^,^",options="header " , ] next we shall study the vbs states on the 3-leg and 4-leg hexagonal ladders . as an illustration , in fig . [ fig : hexagonal_vbs_ladder2 ] , we show an example of 3-leg hexagonal ladders . we first study the 3-leg ladders . in this case , the @xmath183 matrix is defined similarly to eq . ( [ z_n_3leg ] ) : @xmath184 with @xmath185 and @xmath186 . the recursion relation becomes slightly complicated compared to the square cases due to the wiggly nature of the hexagonal ladders . it is composed of 2 steps as follows : @xmath187 where two matrices @xmath188 and @xmath189 are given by @xmath190 respectively , with @xmath191 . using the above recursion relation , one can obtain the coefficients @xmath192 . by diagonalizing the matrix in eq . ( [ z_n_3leg_h ] ) with these coefficients , we can calculate the von neumann entropy along the same lines as sec . [ sec : 3-leg square ladder ] . the obtained results are summarized in table [ tab : hexagonal ladders ] , which are compared with numerical results obtained in sec . [ sec : mc hexagonal ] . we now consider the large - block - size limit . let @xmath193 be the perron - frobenius vector of @xmath194 . this vector is numerically given by @xmath195 using the above values , we obtain the von neumann entropy in the large - block - size limit as @xmath196 and hence the von neumann entropy per valence bond is given by @xmath197 , which is strictly less than @xmath0 . as is obvious , the argument so far can be straightforwardly generalized to @xmath198-leg ladders with @xmath199 @xcite . in table [ tab : hexagonal ladders ] , we show the results for the 4-leg hexagonal ladders . all the analytically obtained results indicate that the boundary correlation does not vanish even in the large - block - size limit and support the conjecture in sec . [ sec : schmidt ] , i.e. , the eigenvalues of the reduced density matrix can be different . a numerical implementation of the transfer matrix technique for the @xmath198-leg ladders with @xmath200 will be discussed elsewhere @xcite . so far , we have studied square or hexagonal ladders in a horizontal geometry . in this section , we shall study the opposite limit , i.e. , the ladders in a vertical geometry . let us first consider the square ladder with @xmath201 and the height @xmath165 shown in fig . [ fig : vertical_ladder](a ) . + then the @xmath183 matrix is given by @xmath202 similarly to the horizontal ladder cases , we can use the integral formula eq . ( [ 3.3 ] ) and rewrite eq . ( [ n mat 1 ] ) as @xmath203 where @xmath204 . here we have abbreviated @xmath205 . the above formula can also be obtained using loops and strands shown in [ sec : loop model ] . note that @xmath206 is @xmath207 identity matrix . we now look for a recursion relation between @xmath208 and @xmath209 . we first decompose @xmath209 into four matrices as @xmath210 where @xmath211 trivially acts on the @xmath165-th vector space @xmath212 while @xmath213 acts on it as @xmath214 . this decomposition is graphically shown in fig . [ fig : decomp ] . . ] using the graphical representation , we find the following set of recursion relations : @xmath215 \otimes \sigma^\alpha , \\ \fl~~ z(n_y+1 ) & = & z^0 ( n_y+1 ) + \sum^3_{\alpha=1 } n^\alpha(n_y+1),\end{aligned}\ ] ] where ` . ' denotes matrix multiplication while ` @xmath216 ' denotes tensor product , and ` @xmath217 ' is the @xmath207 identity matrix . solving the above recursions by _ mathematica _ , we obtain the overlap matrix @xmath209 exactly up to @xmath218 . then we numerically diagonalize @xmath209 and calculate the von neumann entropy as a function of @xmath165 . figure [ fig : algebraic_nx1](a ) shows the result . the obtained values and also fitting parameters in eq . ( [ eq : fitting ] ) show good agreement with the monte carlo results . @xmath219{ee_l_s_b_o_nx1_exact.eps } & \hspace{5 mm } & \includegraphics[width=0.45\columnwidth]{ee_l_h_b_o_nx1_exact.eps } \\ ( { \rm a } ) & \hspace{5 mm } & ( { \rm b } ) \end{array}$ ] let us next consider the hexagonal ladder with @xmath201 and the height @xmath165 shown in fig . [ fig : vertical_ladder](b ) . it is convenient to set @xmath220 since @xmath165 is always odd in this case . in this case , from eq . ( [ eq : def of z ] ) , the @xmath183 matrix is given by z(m ; _ 1 , ... , _ m)= ( ^n_y_i=1 ) ^m_j=1(1+_2j-1_j ) ^n_y-1_k=1(1-_k _ k+1 ) . along the same lines as the square ladder case , we obtain the set of recursion relations for the @xmath183 matrix : @xmath221 \otimes \sigma^\alpha , \\ \fl~~ z(m+1)=z^0(m+1)+\sum^3_{\alpha=1}z^\alpha ( m+1).\end{aligned}\ ] ] here we have abbreviated @xmath222 . again , solving the above recursions , we find the @xmath183 matrix up to @xmath223 ( @xmath224 ) . the numerically calculated von neumann entropy by this method is shown in fig . [ fig : algebraic_nx1](b ) . the obtained values and the fitting parameters also show good agreement with the monte carlo results . the obtained results for square and hexagonal ladders with @xmath201 strongly support that the assumed fitting function eq . ( [ eq : fitting ] ) correctly describes the behavior of large-@xmath165 systems . in conclusion , we have studied entanglement properties of the vbs states on two - dimensional graphs with reflection symmetry . we have shown that the reflection symmetry permits us to develop an efficient method to study the reduced density matrix of the subsystem which is a mirror image of the other one . we found the relation between the reduced density matrix and the overlap matrix whose element is given by the inner product between degenerate ground states of the block hamiltonian defined in the subsystem . this relation enables us to conjecture that the von neumann entropy per valence bond on the boundary between the subsystems is strictly less than @xmath0 in 2d systems . this conjecture means that the entanglement spectrum is not flat in this system and is in contrast to the case of 1d vbs where the reduced density matrix is proportional to the identity matrix in the large - block limit . we have given a _ holographic _ interpretation of the reduced density matrix in terms of the spin chain along the boundary between the subsystems , which describes a hidden correlation among the degenerate ground states of the block hamiltonian . to confirm the conjecture , we have numerically studied the eigenvalues of the reduced density matrix for finite square and hexagonal graphs by combining monte carlo integration with exact diagonalization . from the eigenvalues of the reduced density matrix , we have calculated the von neumann entropy and have found that the von neumann entropy per valence bond is well fitted by a constant term with a function decaying algebraically in the length of the boundary . the constant is strictly less than @xmath0 and supports our conjecture . we have also analytically and algebraically studied the quasi-1d cases where graphs are on ladders . the analytical results are totally consistent with the numerical results and strongly support our conjecture . although our conjecture is very plausible , we still lack a rigorous proof of it in the infinite 2d system . therefore , it would of course be interesting to study the problem by a completely different approach based on inequalities and perhaps reflection positivity . the authors are grateful to zhengcheng gu , akimasa miyake , masaki oshikawa , yutaka shikano , xiaoliang qi , and ying xu for their valuable comments and discussions . the creative atmosphere of erwin schrdinger institute was also important for advancing the work . the present work was supported in part by nsf grant dms 0905744 , by mext grant - in - aid for scientific research ( b ) ( 19340109 ) and for scientific research on priority areas `` novel states of matter induced by frustration '' ( 19052004 ) , and by next generation supercomputing project , nanoscience program , mext , japan . is partly supported by the jsps postdoctral fellow for research abroad . s.t . is partly supported by grant - in - aid for young scientists start - up ( 21840021 ) from the jsps . the computation is executed on computers at the supercomputer center , issp . in this appendix , we illustrate a graphical interpretation of the overlap matrix @xmath183 ( see eq . ( [ eq : integral=2leg_square ] ) ) in terms of loops and strands . as we have already mentioned in sec . [ sec : square ladders ] , the terms containing odd number of @xmath225 s in @xmath226 vanish after integration over @xmath225 since the measure of integration is invariant under the local change of variable , @xmath227 . this can be graphically explained as follows . suppose that one vertex @xmath228 is connected to two vertices . they are labeled 1 and 2 ( see fig . [ fig : klt](a ) ) . then we consider the following integral : @xmath229.\end{aligned}\ ] ] from the lemma 3.3 in ref . @xcite , we can show @xmath230 and hence we obtain @xmath231 where @xmath191 . this procedure is graphically shown in fig . [ fig : klt](a ) . next , suppose that one vertex @xmath106 is surrounded by three vertices labeled by 1 , 2 , and 3 . then we encounter with the following integral : @xmath232 using eq . ( [ eq : integral_formula_1 ] ) , again , we obtain @xmath233 \nonumber \\ & = & 1 + q\ , \oh_1 \cdot \oh_2 + q\ , \oh_2 \cdot \oh_3 + q\ , \oh_3 \cdot \oh_1 . \label{loop2}\end{aligned}\ ] ] this procedure can also be graphically interpreted as shown in fig . [ fig : klt](b ) . ) . the mark @xmath216 denotes the integration over the spherical angle @xmath225 . the broken and colored ( thick ) lines have weight @xmath234 and @xmath191 , respectively . ( b ) graphical representation for eq . ( [ loop2 ] ) . notations are the same as ( a ) . ] therefore , as we have seen , the integrand containing odd number of @xmath225 s will vanish . in terms of the graphical representation , the matrix @xmath235 for the 2-leg square ladder can be written as @xmath236 where @xmath237 and @xmath238 denote the numbers of colored edges ( bonds ) and loops in the configuration @xmath239 , respectively . the minus sign stems from the product like @xmath240 . in the first sum over @xmath241 , only the closed loops are allowed . on the other hand , in the second sum over @xmath242 , open ends from @xmath234 and @xmath243 are allowed ( see fig . [ fig : config ] ) . and open strand @xmath244 configurations for the length @xmath245 , and @xmath246 . the weight of each configuration is assigned at the bottom right of each figure . ] therefore , we can obtain the coefficients @xmath247 and @xmath248 in eq . ( [ eq : zmatrix_2leg_square ] ) by a completely graphical or combinatorial way . from fig . [ fig : config ] , one can read @xmath249 , @xmath250 , @xmath251 and @xmath252 , @xmath253 , @xmath254 . although we only show the case of 2-leg square ladders for simplicity , one can easily generalize this approach to any aklt ladders . 10 url # 1#1urlprefix[2][]#2 schrdinger e 1935 _ naturwissenschaften _ * * 23 * * 807

the simplest way to describe kinetic phenomena is by the boltzmann equation ( be ) . there are however cases when the be can not be used , and one has to use quantum kinetic equations ( qkes ) that are in the simplest case equations for the one particle green functions . if the keldysh technique @xcite is used , the one particle green functions that participate in the description of kinetic processes are the retarded green function ( and its complex conjugate ) , and the `` statistical '' green function , which is simply the field correlator . for free particles these green functions are diagonal in the basis of the free particle states , and are described by the free particle energies @xmath0 , and their occupation numbers @xmath1 . these are the quantities that appear in the be for the _ interacting _ particles _ in external fields_. the description of kinetic phenomena by qkes instead of bes results in two effects . the first effect is state mixing - due to interactions and external fields the green functions are nondiagonal in the basis of the free particle states . this effect is responsible for coherence phenomena , of which a well known example is interband polarization described by semiconductor bloch equations @xcite . the second effect is level shift and broadening - the green functions are nonzero off the energy shell within some width defined by the interactions and the external fields . level broadening effects in qkes for phonons have been studied in refs . @xcite . the effects of level broadening are of importance in two cases . the first case is that in which level broadening allows scattering processes that are forbidden by energy and momentum conservation . an example of such a case is that of electron - electron scattering in a 2deg , in the presence of a quantizing magnetic field @xcite . if one tries to calculate the auger transition rates responsible for anti - stokes luminescence , that was observed in ref . @xcite , using fermi s golden rule , one encounters delta functions with an argument equal to zero , due to singularities in the density of states , see ref . @xcite . the second case in which level broadening effects should be considered is that of narrow energy distributions : when there are small energy scales in the problem that are narrower than the quantum width of the state . a simple example of such a case , that has one small energy scale , is that of an electron distribution , excited by a narrow band - width laser . the problem becomes more complex when another small energy scale is added , for example : the electrons are excited to the close vicinity of some threshold . a problem of this nature will be dealt with in this paper . in the time domain , level shift and broadening lead to non - markovian effects in the scattering processes @xcite . these effects are important in the case of short excitation pulses @xcite . if state mixing induced by coherent excitation can be neglected , and only shift and broadening effects are considered , the qkes can be presented in a form similar to the bes . we will give a general recipe how to write such equations , and will use this recipe when considering the following problem : hot luminescence from a quantum wire ( qwr ) due to electrons that are excited by a narrow band , noncoherent laser , to the close vicinity of the threshold for optical lo phonon emission . optical and transport phenomena in qwrs have been of great interest lately . the great interest in transport phenomena @xcite is due to the possibility that a 1deg in a qwr will have the properties of a luttinger liquid . the fermi edge singularity @xcite and the possibility of stimulated emission @xcite enlarged interest in optics of qwrs . the important role played by lo phonons in the trapping of electrons @xcite from three - dimensional extended states , into one - dimensional localized states , has been reported in ref . @xcite . theoretical studies of qwrs were concentrated around elementary excitations of a 1deg @xcite , fermi edge singularity @xcite , excitonic effects @xcite , electron - phonon scattering rates @xcite and interactions @xcite , relaxation of photoexcited carriers @xcite , calculation of envelope states @xcite , and conductance of qwrs with self - consistent broadening effects @xcite . the retarded green function and the `` statistical '' green function which appear in the kinetic equations are defined as @xmath2_{\pm } \rangle,\ ] ] @xmath3_{\mp } \rangle,\ ] ] where @xmath4 are the field operators in the heisenberg representation . the square brackets with a plus ( minus ) sign signify an anticommutator ( commutator ) . the upper signs refer to fermions and the lower signs refer to bosons ( this convention is used throughout the paper ) . the spin degrees of freedom @xmath5 , will be suppressed from now on , and one can think of them as included in the space coordinates . in case of interacting particles the green functions obey the dyson equation @xmath6 where @xmath7 is a matrix of green functions , and @xmath8 is a matrix of self energy functions @xcite : @xmath9 the superscript @xmath10 indicates free particle green functions . the advanced green function @xmath11 . in the case of free particles we can use the free particle eigenstates @xmath12 , and the corresponding annihilation operators @xmath13 , to write @xmath14 . in this case @xmath15 depends only on the time difference @xmath16 , and can be fourier transformed in time into @xmath17 ( the index @xmath18 stands for r , a or s ) . for an orthonormal set of states @xmath19 we can define @xmath20 such that @xmath21 from the definitions ( [ gr - def ] ) , ( [ gs - def ] ) we obtain ( near the singularities ) @xmath22,\ ] ] and @xmath23,\ ] ] where @xmath0 is the energy of state @xmath19 , and @xmath24 is the principal part function . the upper ( lower ) term in the column corresponds to fermions ( bosons ) . the occupation number of state @xmath19 , @xmath25 , is not necessarily the equilibrium occupation number . in the case of interacting particles in external fields , if one neglects coherence effects , the green functions are still diagonal in @xmath19 , but the energy levels are broadened and shifted . let us first look at the time invariant case ( systems under d.c . conditions ) . the level shift @xmath26 , and the level broadening @xmath27 , are defined as @xmath28 { \rm i m } \sigma^{\rm r}_{\alpha}(\epsilon),\ ] ] where @xmath29 is the retarded self energy of particles in state @xmath19 . the column has the same meaning as in expression ( [ gr0-ae ] ) . the retarded green function is an outcome of the dyson equation ( [ dyson ] ) , that is written as @xmath30 in the representation of the @xmath19 states . using the defintions ( [ s - w - def ] ) we obtain from equation ( [ dyson - gr ] ) ( for positive frequencies ) @xmath31 where @xmath32 is the `` smeared '' and `` shifted '' principal value , and @xmath33 is the `` smeared '' and `` shifted '' delta function . comparing expression ( [ gr - ae ] ) with expression ( [ gr0-ae ] ) we see that @xmath34 is in fact a generalization of @xmath35 , where the level shift and level width have been introduced into the principal value and delta functions . we therefore write @xmath36 as a generalization of @xmath37 , thus defining the occupation functions @xmath38 @xmath39 expressions ( [ gr - ae ] ) and ( [ gs - ae ] ) were written for the case of positive frequencies , which will be assumed from now on . in case of negative boson frequencies one has to multiply the retarded green function and @xmath40 in these expressions by minus one . the definitions above agree with the relation between the statistical green function and the retarded green function at thermal equilibrium @xmath41 where @xmath42 is the equilibrium distribution function ( fermi or bose - einstein ) . the smeared delta function obeys the following normalization @xmath43 and the occupation numbers are obtained from the occupation functions in the following manner @xmath44 relations ( [ equil ] ) - ( [ on - of ] ) can be obtained from the lehmann representation of the green functions . these relations are valid only if the one particle green functions can be considered as diagonal in @xmath19 , in other words , in the absence of coherence effects . as a result of definitions ( [ s - w - def ] ) and ( [ gs - ae ] ) , the three unknown functions @xmath45 and @xmath46 have been replaced by a different set of three unknown functions : @xmath47 and @xmath38 . if the system is not under d.c . conditions , one can define a new pair of time variables : the relative time @xmath48 , and the `` center of mass '' time @xmath49 . only slow processes will be considered so that all quantities vary in @xmath50 with some time scale that is large compared to the inverse characteristic energy of the particles . in such a case one can still use the formalism given above for the green functions , but with @xmath50 as a parameter of the problem , in addition to @xmath51 ( see for example ref . @xcite ) . for slow processes @xmath52 again @xmath53 can be written as a generalization of @xmath35 , in the same manner as before , but now @xmath50 is a parameter which appears in all functions . thus , expressions ( [ gr - ae ] ) and ( [ gs - ae ] ) for @xmath54 and @xmath55 , can be used with the three unknown functions @xmath56 and @xmath57 that depend on @xmath50 as well as on @xmath51 . the known procedure for obtaining a kinetic equation for @xmath57 is to apply the operator @xmath58 to the dyson equation for @xmath59 @xcite , where the operator @xmath60 is equal to @xmath61 for electrons , and @xmath62 for photons and phonons ( we use the convention @xmath63 throughout the paper ) . the slow variation in @xmath50 should be considered and then the equation should be fourier transformed in the time difference coordinates . this procedure leads to the qke @xmath64 from now on the `` bar '' over @xmath65 will be dropped , with the understanding that all time variables are in fact center of mass time variables . equation ( [ qke ] ) for photons and phonons includes a factor of @xmath66 that multiplies the time derivative on the left hand side . since we are interested in off - shell energies that are close to the on - shell energy , this factor can be taken to be one . in order to obtain equations for @xmath67 and @xmath68 one has to substitute an expression for the retarded self energy function , in terms of the one particle green functions , into the definitions of @xmath27 and @xmath26 ( [ s - w - def ] ) . these equations are in fact the imaginary and the real parts of equation ( [ dyson - grte ] ) . this yields coupled , self - consistent equations for @xmath27 and @xmath26 , that are also @xmath69 dependent . it is important to stress that the occupation function depends on time explicitly , while the dependence of @xmath27 and @xmath26 on time is only through their dependence on @xmath69 . therefore in order to find @xmath69 one has to write a kinetic equation , that will include time evolution through time derivatives . the equations for the level shift and width will include their dependence on time only through the appearance of @xmath69 , and will not include time derivatives . if one uses the self - consistent - born approximation for the self energy functions , one finds that the equations obtained for @xmath70 , and @xmath68 could have been obtained easily by applying a few generalizations to the be for @xmath1 . in the transfer from the be to the qke the number of unknowns increases from one ( @xmath69 ) , to three ( @xmath71 and @xmath26 ) , therefore apart from the kinetic equation we require two more equations for the unknown functions @xmath67 and @xmath68 , that appear in the equation for @xmath72 . we present here a recipe which allows one to go from the be for @xmath73 , to the qke for @xmath72 , and from the decay term in the boltzmann equation ( which is defined below ) to an equation for @xmath67 . an equation for @xmath68 can be obtained from the equation for @xmath67 using kramers - kronig relations . the be , in its most general form , can be written as @xmath74 2 \pi \delta ( \epsilon_{\alpha } - \epsilon_{\beta } + \epsilon_{\gamma} ... )\equiv s.\ ] ] this equation describes the evolution in time of the occupation number of particles in state @xmath19 , due to interactions with particles in states @xmath75 , with matrix elements @xmath76 . different states can also mean different particles . the right hand side of the equation includes delta functions that are responsible for energy conservation . there are many cases in which it is convenient to think of the collision integral @xmath77 as composed of `` scattering in '' events and `` scattering out '' events : @xmath78 . the first term on the right hand side of equation ( [ boltzmann ] ) is the scattering in term : it describes processes in which particles enter state @xmath19 due to the interaction with other particles . this term can be written as @xmath79 , where @xmath80 includes all the terms on the right hand side of the equation that do not multiply @xmath1 . we refer to @xmath80 as the `` generation '' term . the second term is the scattering out term : it describes processes in which particles leave state @xmath19 . this term is usually written as @xmath81 , where @xmath82 is the scattering _ out _ time from state @xmath19 . in the context of the generalization of the be to a qke , it is more convenient to think of the collision integral @xmath77 as made up of `` generation '' and `` decay '' terms : @xmath83 , where the decay term @xmath84 includes all the terms containing @xmath1 . thus @xmath85 is the total decay rate of particles in state @xmath19 . it is the _ total _ decay rate , and not only the scattering _ out _ rate , that is related to the retarded self energy @xcite . the general form for the decay rate @xmath86 , as it is deduced from equation ( [ boltzmann ] ) , is @xmath87 2 \pi \delta ( \epsilon_{\alpha } - \epsilon_{\beta } + \epsilon_{\gamma} ... ).\ ] ] in a d.c . situation one obtains for the occupation numbers @xmath88 in order to transform the be into a qke one should apply the following rules . to the left hand side of equation ( [ boltzmann ] ) apply * @xmath89 , where @xmath51 is the off - shell energy of particles in state @xmath19 . to the right hand side of the equation apply the following steps * the occupation numbers for particles in states @xmath90 are replaced with the occupation functions , that depend on the off - shell energies @xmath91 , corresponding to the on - shell energies @xmath92 respectively : @xmath93 , @xmath94 ... * the summand under the sum over all states @xmath75 is multiplied by the product of the smeared delta functions for particles in these states with the corresponding off - shell energies @xmath95 , and integrated over the off - shell energies : @xmath96 * the energy conservation delta function of the on - shell energies @xmath92 is replaced with an energy conservation delta function of their corresponding off - shell energies @xmath91 : @xmath97 . * multiply the entire right hand side by @xmath98 . in order to obtain an equation for @xmath67 one should apply the first three steps of the generalization rules for the right hand side of the equation , to the expression for @xmath86 ( [ decay ] ) . the qke can be written generally in the form @xmath99 the function @xmath100 contains all the terms on the right hand side of the qke , that do not multiply @xmath72 . it is the generating term - the term in the kinetic equation that is responsible for the particle generation rate . equation ( [ qke - gg ] ) shows clearly that the particle level width @xmath67 is in fact the total particle decay rate . note that due to ( [ norm ] ) the integration of equation ( [ qke - gg ] ) over @xmath51 results in the disappearance of the @xmath101 factor from the left hand side . the occupation function in the time independent case is given by @xmath102 the representation of the qke in terms of @xmath71 and @xmath26 is more convenient than the representation in terms of green functions , since one can clearly see all physical processes , and use the physical intuition that one gained from the be in order to simplify the qke . ( 36,32)(0,0 ) ( 0,0 ) ( 3,1)(0,0)@xmath103 ( 7,29)(0,0)@xmath104 ( -1,27)(0,0)@xmath105 ( -1.2,19)(0,0)@xmath106 ( 0,35)(0,0)[b]@xmath107 ( -1.8,12)(0,0)@xmath108 ( -1.5,9)(0,0)@xmath109 ( 16.5,23)(0,0)[r]@xmath105 ( 20,17)(0,0)[br]@xmath110 ( 22.5,17)(0,0)[bl]@xmath111 ( 30,19)(0,0)@xmath112 we now use the formalism given in section [ qke ] in order to deal with the specific problem of hot luminescence from a qwr . the case in which electrons are excited from the highest valence subband to the lowest conduction subband via photons , and then relax via lo phonon emission ( see fig . [ qwr ] ) is considered . all other conduction and valence subbands are neglected ( in contrast to ref . we consider a situation in which electrons are excited by a narrow band , noncoherent laser , just above the threshold for optical phonon emission @xmath113 , and describe the hot luminescence due to the recombination of electrons with holes in an impurity level . it is assumed there is some mechanism due to which the electrons leave the bottom of the conduction subband , so that there is no electron accumulation there , but this recombination can be neglected in the balance equations near the threshold . the spin degree of freedom will be completely disregarded . unless specified otherwise we will assume that the electrons are excited above the threshold , since we are mainly interested in luminescence due to these electrons . below the threshold the electrons relax to the bottom of the band by emitting acoustical phonons . the relaxation rate due to acoustical phonons @xmath114 , is much smaller than that due to optical phonons , and can be neglected above the threshold . the qwr runs along the @xmath115 direction so that the electron wave functions are localized in the @xmath116 plane , and free waves in the @xmath115 direction . the wave function in the conduction band can be written as @xmath117 , where @xmath118 is the wire normalization length , and @xmath119 is a vector in the @xmath116 plane . the electron energy is @xmath120 . all energies are measured from the bottom of the conduction band . the holes in the valence band have wave functions of the form @xmath121 and energy @xmath122 , with @xmath123 . the exciting photons are taken to be plane waves and the photon frequency is @xmath124 ( @xmath125 is the photon momentum and @xmath126 is the light velocity ) . the lo phonons are three - dimensional and have a flat dispersion law , @xmath127 . the impurities are distributed randomly in the bulk . the wave function of a hole in impurity @xmath128 is of the form @xmath129 , where @xmath130 are the coordinates of the impurity position . in order to analyze the hot luminescence we neglect light polarization effects for simplicity , and assume that the wire and the crystal are cylindrically symmetric ( an assumption that would not fit v - grooved wires nor the assumption of ref . thus , the wave functions of the emitted photons are @xmath131 where @xmath132 is the normalizing radius and @xmath118 the normalizing length of the crystal , and @xmath133 and @xmath134 are the cylindrical polar coordinates . @xmath135 are the bessel functions , and @xmath136 is the @xmath137 zero of @xmath135 . the luminescence photon frequency is given by @xmath138 ( where the prime signifies this frequency is of luminescence , and not of excitation ) . we neglect all excitonic effects ( these are treated for the case of optical absorption in one - dimensional semiconductors in refs . @xcite ) . we begin by writing down the bes for the electrons in the conduction band , and for the luminescence photons . it is shown why this description fails and one has to turn to the qkes . we will then employ the generalization rules in order to obtain the qkes from the bes , and use the qkes in the analysis of the luminescence . we assume low temperature and a weak excitation level . at equilibrium , when there is no excitation , all the electrons are in the valence band , and there are no phonons ( @xmath139 ) , due to the low temperature . in other words , there are no electrons in the conduction band ( @xmath140 ) , no holes in the valence band ( @xmath141 ) , and the impurity level is fully occupied by holes ( @xmath142 ) . the excitation creates electrons in the conduction band , and holes in the valence band . most of the excited electrons relax to the bottom of the conduction band , emitting lo phonons , but a small fraction of them recombine with holes in the impurity level , producing hot luminescence . by the assumption of weak excitation we mean that @xmath143 , @xmath144 and @xmath145 are small . the luminescence would be even weaker than the excitation , therefore @xmath146 remains close to one . due to the weak excitation the luminescence is spontaneous , and the occupation numbers for the luminescence photons are also small . in the derivation of the bes , terms that are nonlinear in the electron occupation numbers will be neglected . we will also neglect nonequilibrium contributions to the phonon and the hole ( both in the valence band and in the impurity level ) occupation numbers . the interaction of the electrons with the luminescence field is neglected in the balance equation for the electrons , in comparison with their interaction with the excitation field , due to the weakness of the excitation . in the case of c.w . excitation the balance equation for the electrons is @xmath147 where @xmath148 is the matrix element for the electron - exciting - photon interaction , and @xmath149 is the matrix element of the electron - lo phonon interaction . let us first look at the electron - exciting - photon matrix element . in this interaction one photon of wave vector @xmath150 is absorbed , and a hole of momentum @xmath151 and an electron of momentum @xmath152 are created . the interaction , in second quantized form , is given by @xmath153 the mass @xmath154 is the bare electron mass and the constant @xmath155 is the `` bare '' electron momentum operator ( in the direction of the photon polarization , assumed linear for simplicity ) , sandwiched between the valence band top and the conduction band bottom bloch wave functions . @xmath156 and @xmath157 are the electron and hole field operators , respectively . the photon field operator is given by @xmath158 where @xmath159 , @xmath160 are the components of @xmath150 in the @xmath116 plane , and in the @xmath115 direction , respectively , @xmath161 is the annihilation operator , and @xmath162 is the normalization volume of the crystal in which the quantum wire is embedded . in order to find @xmath163 one should sandwich interaction ( [ int - exc ] ) between the states @xmath164 and @xmath165 . this is equal to @xmath166 the photon momentum is small and so is @xmath119 , since it is limited by the wire cross section , therefore @xmath167 . integrating over the @xmath115 coordinates the square of the matrix element is @xmath168 in the electron - lo phonon interaction one phonon of momentum @xmath169 is emitted , an electron @xmath152 is annihilated , and an electron @xmath170 is created . the interaction of the electrons with the polarization created by the lo phonons is given by @xmath171 where @xmath172 is some interaction constant ( see for example ref . @xcite , @xcite ) , and the phonon field operator is @xmath173 the components of @xmath169 are defined in the same manner as those of @xmath150 , @xmath174 is the annihilation operator , and @xmath175 is the reduced mass per unit cell in the lattice . the matrix element @xmath149 can be found by sandwiching interaction ( [ int - lo ] ) between the states @xmath176 and @xmath177 . carrying out the integration over the @xmath115 coordinates , the square of the matrix element is @xmath178 where we exchanged @xmath172 for the known frlich constant @xmath19 , using the relation @xmath179 ( see ref . @xcite ) . substituting @xmath180 and @xmath181 into equation ( [ e - be ] ) , we can sum over @xmath151 and @xmath169 . taking the normalizing volumes to infinity and thus exchanging the sums with integrals we obtain @xmath182 where @xmath183 @xmath184 and @xmath185 the zeroth order bessel function @xmath186 is a result of the integration over @xmath187 . since the matrix elements ( [ |m - e - ep|^21 ] ) and ( [ |m - e - ph|^21 ] ) are smooth functions of their arguments , and we are interested only in processes which involve electrons that were excited close to the threshold , we may take them at the threshold values . the threshold value of @xmath188 is @xmath189 , where @xmath190 . this is the frequency that will excite electrons from the valence band to the threshold in the conduction band . the threshold value of @xmath191 is @xmath192 . the electron wave functions that appear in @xmath5 limit the spatial integration to @xmath193 , where @xmath194 is the wire width , therefore we can estimate that @xmath195 for @xmath196 and @xmath197 for @xmath198 . for gaas @xmath199 , thus for most wire widths we are in the regime of @xmath198 , in which @xmath5 is expected to decrease with increasing wire cross section . the constant @xmath200 is the effective frhlich constant in the one - dimensional case . according to ref . @xcite @xmath201 ( which is of the same order as the bulk frhlich constant ) for a wire of cross section @xmath202 , while according to ref . @xmath203 for a wire of cross section @xmath204 . we now perform the integrations which appear in equation ( [ e - be1 ] ) . we begin with the generating term @xmath205 , the free term ( one that does nt include @xmath144 ) in the balance equation ( [ e - be1 ] ) for the electron occupation number . this term is physically the generation rate of electrons in the conduction band . since @xmath206 we can approximate @xmath207 . using @xmath208 , where @xmath209 stands for the solid angle increment , we obtain @xmath210 where @xmath211 , is the photon density of states , and @xmath212 is the angular average of @xmath213 , for photons of frequency @xmath214 , that create electrons with energy @xmath215 . we consider a narrow band photon excitation field such that the angular average of the occupation number is given by @xmath216 where @xmath217 is the excitation field energy density , @xmath218 is the spectral width , and @xmath219 is the central frequency . we are interested in a narrow band excitation close to the threshold , therefore @xmath220 , and the detuning ( the difference between the central excitation frequency and the threshold frequency ) is small : @xmath221 . the integrations that appear in the expression for the decay rate @xmath222 , the term that includes all coefficients of @xmath223 in the kinetic equation ( [ e - be1 ] ) , will now be carried out . the first contribution to @xmath222 is that of the photons , and is equal to the generating term ( [ e - be1-ep ] ) . the second contribution to @xmath222 is due to lo phonons , @xmath224 the photon contribution to the decay rate is negligible compared to the phonon contribution since radiative processes are slow compared to non - radiative processes . therefore @xmath222 can be written as @xmath225 where @xmath226 the meaning of the energy scale @xmath227 will be explained at the end of this section . for @xmath228 @xcite one obtains @xmath229 , thus @xmath230 . below the threshold , where @xmath231 , the acoustical phonon contribution to @xmath27 is important . since the latter is a smooth function near the threshold , a good approximation for @xmath27 is @xmath232 at @xmath233 . the occupation number of the electrons that were excited above the threshold is given by expression ( [ n - a ] ) , where @xmath234 and @xmath27 are given by expressions ( [ e - be1-ep ] ) and ( [ decay - ek ] ) respectively . it is clear from the expressions above that @xmath144 depends on @xmath152 through @xmath215 only . we now turn our attention to the kinetic equation for the luminescence photons , @xmath235 , \end{aligned}\ ] ] which for @xmath142 and in case of spontaneous luminescence is @xmath236 @xmath237 is the matrix element for the luminescence photon - electron interaction . the time derivative in the equation above is kept in order to clarify what is the luminescence source . let us first look at the matrix element that describes the recombination of an electron with a hole from a specific impurity . we will then have to sum over all impurities and average over all impurity configurations . in this process a photon of quantum numbers @xmath238 is emitted , and an electron @xmath152 and a hole in impurity @xmath128 are destroyed . the interaction is given by @xmath239 where @xmath240 is the field operator of the luminescence photons . the matrix element of the interaction above between the states @xmath241 and @xmath242 for holes in impurity @xmath128 is @xmath243 we now sum the matrix element squared over all impurities and average over all impurity configurations . this yields @xmath244 carrying out the integrations over the @xmath115 coordinates we obtain @xmath245 where @xmath246 . the second argument of @xmath247 , @xmath248 was replaced by @xmath249 because @xmath206 . since the impurities are randomly distributed in the bulk , the average over all impurity configurations of the product of the hole wave functions , that appear in ( [ |m - lp - e|^21 ] ) , can depend only on the difference between their coordinates . thus we define @xmath250 the summation over impurities divided by the normalizing volume renders a factor of @xmath251 - the impurity spatial density . the averaged matrix element squared can then be written as @xmath252 the matrix element squared is much larger for @xmath253 than for other values of @xmath254 , since @xmath134 and @xmath255 are constrained to a small region in the @xmath116 plane ( due to the electron wave functions ) , and @xmath256 is the only bessel function which is finite at @xmath257 , @xmath258 . thus the matrix element squared is proportional to @xmath259 . from now on the index @xmath254 is omitted with the understanding that we are dealing only with @xmath253 . the matrix element squared is a smooth function of @xmath260 and @xmath152 , therefore these can be substituted by their threshold values . the threshold values of @xmath260 and @xmath152 are @xmath261 and @xmath262 respectively . the kinetic equation for the luminescence photons can then be written as @xmath263 where the matrix element squared is @xmath264 substituting @xmath265 into equation ( [ lp - be1 ] ) and integrating over @xmath152 we obtain @xmath266 the luminescence source can be characterized by the spectral dependence of @xmath267 , the energy of the emitted field , of all spectral modes within the interval @xmath268 , per unit length of the wire , per unit time . multiplying the generating term of the luminescence photons ( the right hand side of equation ( [ lp - source ] ) ) by @xmath269 and summing over @xmath160 and @xmath69 , we obtain @xmath270 the variable @xmath260 in the square root can be replaced with its threshold value @xmath271 , due to the smoothness of this function . taking the normalization volume ( that appears in the matrix element ) to infinity the sums over @xmath160 and @xmath69 are transformed into integrals . performing the integration we obtain the final result @xmath272 where @xmath273 is a constant , and @xmath274 is essentially the product of the electron generation rate and the electron life - time at the corresponding energy . since @xmath275 is a smooth function of @xmath276 , the spectral distribution of the luminescence is given by @xmath277 . let us first consider a narrow band excitation that does not overlap the threshold , i.e. @xmath278 . in this case @xmath277 is a peaked function of @xmath276 , reproducing the shape of the excitation . it is of a lorentzian shape of width @xmath279 , centered at @xmath280 , where @xmath281 ( see fig . [ xi1 ] ) @xmath282 the rescaling of @xmath276 compared to @xmath283 follows from the obvious relations ( see fig . [ qwr ] ) @xmath284 and @xmath285 . ( 10,7)(0,0 ) ( 0,0 ) ( 6,-0.1)(0,0)[t]@xmath286 ( 0,6.5)(0,0)[b]@xmath287 ( 10.3,0)(0,0)[lb]@xmath288 ( 0,0)(0,0)[tr]@xmath106 ( 5.6,2.75)(0,0)[r]@xmath289 ( 6.4,2.75)(0,0)[l]@xmath290 ( 7.2,2.75)(0,0)[l]@xmath291 ( 7.3,4.7 ) ( 7.3,6.5)(0,0)[b]@xmath287 ( 10,4.7)(0,0)[lb]@xmath288 ( 7.3,4.7)(0,0)[tr]@xmath106 only close to the threshold , when @xmath292 , the spectral behavior of the luminescence differs from that of the excitation and is given by ( see the inset of fig . [ xi1 ] ) @xmath293 ( 4,3)(0,0 ) ( 0,0 ) ( 1.25,0)(0,0)[t]@xmath294 ( 0,2.6)(0,0)[b]@xmath287 ( 4.1,0.1)(0,0)[l]@xmath288 ( 0,0)(0,0)[tr]@xmath106 when the narrow band excitation overlaps the threshold , i.e. @xmath295 , the spectral distribution of the luminescence is very different from that of the excitation ( see fig . [ xi3 ] ) : @xmath296 and @xmath297 if one is interested in luminescence due to electrons that were excited close to but _ below _ the threshold , via an excitation that is centered _ above _ the threshold , i.e. @xmath298 and @xmath299 , then @xmath300 in expression ( [ xi - be ] ) should be replaced by @xmath114 . in this case @xmath277 has the same lorentzian form as ( [ xi - lor ] ) with the amplitude @xmath301 replaced by @xmath302 . one can expect that the luminescence distribution @xmath277 , derived from the be , is not correct for frequencies @xmath276 close to the threshold @xmath303 . luminescence of such frequencies is due to electrons with energies @xmath215 near the threshold , that have a large width @xmath304 ) , ( see equation ( [ decay - ek ] ) ) , while the be assumes that the width of the states is small compared to other energy scales of the problem . in our case one of these energy scales is @xmath305 . equating the width with this energy scale , @xmath306 , one finds an additional energy scale of the problem ( and a critical decay rate ) , that is given by @xmath227 that has been defined before ( [ gamma - c ] ) . when @xmath307 , we have @xmath308 , and the assumption of the be breaks down . one would therefore expect that for @xmath307 in the description of the excited electrons , and for @xmath309 in the description of the hot luminescence , the results predicted by the boltzmann equation would fail . we will learn from the qkes that in fact the situation is more complicated . as we have seen in the previous section , the boltzmann description may fail for electrons excited to the vicinity of the threshold and hence for luminescence photons emitted by these electrons . therefore , close to the threshold , the electrons and the luminescence photons have to be treated using qkes . these qkes will be written by applying the generalization rules to equations ( [ e - be1 ] ) and ( [ lp - be1 ] ) . it is assumed that the energies of the holes , the phonons and the excitation photons are not broadened . in the generalization to the qkes only those terms that were kept in the boltzmann description will be retained . threshold values of the arguments of the matrix elements will be substituted as was done in the boltzmann description , due to the smoothness of the matrix elements . the qke for the electrons is @xmath310 where @xmath51 is the electron off - shell energy , @xmath69 is the electron occupation function , and @xmath27 and @xmath311 are the electron energy level width and shift , respectively . the coefficients of @xmath312 in equation ( [ e - qke ] ) can be recognized as the electron level width due to lo phonons @xmath313 and the total electron width is given by @xmath314 the photon contribution to the level width was neglected ( as it was neglected in the boltzmann description of the decay term ) . one can see that the generating term in equation ( [ e - qke ] ) is equal to expression ( [ e - be1-ep ] ) , the generating term of the electrons be , with the on - shell energy @xmath215 replaced by the off - shell energy @xmath51 ( in the exciting photon occupation number ) @xmath315 the occupation function @xmath316 is found by substituting @xmath234 and @xmath27 into ( [ n - ae ] ) . the generating term @xmath317 depends on @xmath152 through @xmath318 only , therefore @xmath319 . it follows from ( [ gamma - e ] ) that @xmath27 is independent of @xmath152 ( and so is the level shift ) . therefore , @xmath320 . equation ( [ gamma - e ] ) is a self consistent equation for the level width . integrating the right hand side of the equation we find that @xmath321^{-1/2 } + \right . \\ \left . \left [ \epsilon - \omega_{o } - \delta \epsilon ( \epsilon - \omega_{o } ) - \frac { i}{2 } \gamma ( \epsilon - \omega_{o } ) \right]^{-1/2 } \right\}.\end{aligned}\ ] ] when @xmath322 the arguments of @xmath323 and @xmath324 are close to the bottom of the band . in this case the level shift can be neglected since it is just a small renormalization . the level width close to the bottom of the band @xmath325 , is due to scattering with acoustic phonons and due to thermal recombination . the latter contribution to @xmath325 can be neglected , while the acoustic phonon contribution can be considered as a constant ( though not the same as the width due to acoustic phonons at the threshold ) . we assume that @xmath325 is small compared to all other energy scales of the problem . thus close to the threshold @xmath326^{-1/2 } + \left [ \epsilon - \omega_{o } - \frac { i}{2 } \gamma_{o } \right]^{-1/2 } \right\}.\ ] ] the behavior of @xmath327 when @xmath51 is close to @xmath328 is shown in fig . [ gamma ] . ( 22,15 ) ( 0,0 ) ( 1,-0.5)(0,0)[t]@xmath329 ( 10,-0.5)(0,0)[t]@xmath330 ( 12.3,-0.5)(0,0)[t]@xmath331 ( 21.3,-0.5)(0,0)[t]@xmath332 ( 11,14.5)(0,0)[b]@xmath333 ( 22.5,0)(0,0)[lb]@xmath334 ( 11.9,1.2)(0,0)[l]@xmath335 from expression ( [ gamma - th ] ) we see that above the threshold for @xmath336 , @xmath337 can be written as @xmath338 comparing expression ( [ gamma- < ] ) to expression ( [ decay - ek ] ) , it is evident that above the threshold @xmath337 is equal to @xmath222 when the on - shell energy @xmath215 is replaced by the off - shell energy @xmath51 . since @xmath325 is the smallest energy scale in the problem , expression ( [ gamma- < ] ) can be used for the electron level width above the threshold . contrary to the boltzmann description , the electron - lo phonon scattering contributes to the electron level width below the threshold as well . when @xmath339 the electron width due to optical phonons decays like @xmath340 , much faster than the decay above the threshold ( see fig . [ gamma ] ) . the contribution of the acoustic phonons to the electron width below the threshold @xmath114 , will be negligible compared to @xmath341 for @xmath342 . since @xmath343 and @xmath344 the right hand side of the inequality above is much smaller than @xmath227 , and much larger than @xmath325 ( see the crossing point of @xmath345 and @xmath114 in fig . [ gamma ] ) . the qke that is obtained by applying the generalization rules to the be for the luminescence photons ( [ lp - be1 ] ) is @xmath346 where @xmath276 is the photon off - shell energy , @xmath347 is the photon occupation function , and @xmath172 and @xmath348 are the photon energy level width and shift , respectively . all the photon functions are written for @xmath253 . by applying the generalization rules to the right hand side of the full be ( [ lp - be ] ) one can find the photon level width @xmath349 note that @xmath350 corresponds to photon generation . it is evident from expression ( [ gamma ] ) that @xmath172 does not depend on @xmath160 , and therefore neither does @xmath348 . using ( [ gamma ] ) equation ( [ lp - qke ] ) can be written as @xmath351 the right hand side of equation ( [ lp - qke1 ] ) is the photon generating term , the term from which the luminescence spectral distribution will be obtained . as in the classical case , the luminescence source can be characterized by the spectral dependence of @xmath352 , that is obtained by multiplying the right hand side of equation ( [ lp - qke1 ] ) by @xmath353 and summing over @xmath160 and @xmath69 @xmath354 the main contribution to @xmath355 comes from @xmath160 and @xmath69 such that @xmath260 is close to @xmath276 . it is convenient to write @xmath356 in the following form @xmath357 where @xmath358 is given by ( [ c ] ) , and @xmath359 when the electrons energy is taken to be on shell , that is the electron smeared delta function is reduced to a singular delta function , @xmath277 is reduced to expression ( [ xi - be ] ) . in writing @xmath172 in the form ( [ gamma-1 ] ) , the normalization volume has been separated from the rest of the function , that is independent of the index @xmath69 , and can therefore be taken out of the summation in @xmath360 . taking the normalization volume to infinity the sums over @xmath160 and @xmath69 are transformed into integrals . from equation ( [ gamma-1 ] ) it is evident that @xmath361 . this is due to the fact that the photons are emitted into an `` infinite '' space , while their interaction with the electrons is confined to the finite volume of the wire . in this case @xmath172 and @xmath348 are negligible compared to the other widths , and the smeared photon delta function that appears in the expression for @xmath360 becomes a singular delta function , giving @xmath362 since @xmath363 is a smooth function it is clear at this point that in order to understand the dependence of @xmath360 on @xmath276 , the behavior of @xmath364 should be analyzed . performing the integration over @xmath51 in ( [ bar - gam ] ) , and substituting the expression we got for @xmath365 , we get @xmath366 since @xmath367 , we substitute @xmath328 for @xmath215 in the one - dimensional density of states . this integral contains in fact two lorentzians . the first is of width @xmath368 : the electron level width at energy @xmath369 . the second lorentzian is of width @xmath370 : a width that is proportional to the spectral width of the exciting photon source . since the excitation is such that @xmath215 is close to @xmath328 , and all the energy scales that can characterize the level width are much smaller than @xmath328 , the integrand goes to zero when @xmath215 approaches zero from above . thus , the lower boundary of the integral can be taken to minus infinity . as a result the spectral distribution of the luminescence is given by the function @xmath371^{2 } + \delta \nu_{o}^{2}/4 } \right).\end{aligned}\ ] ] the region of validity of the be is evident from equation ( [ bar - gam2 ] ) . in order to obtain the result of the be ( [ xi - be ] ) , one has to replace the first lorentzian with a delta function . this can be done only when its width is smaller than that of the second lorentzian , i.e. @xmath372 . it follows from this inequality that the boltzmann description of the luminescence spectra is correct only far from the threshold , when @xmath373 . the quantum interval @xmath374 differs from the naive estimate @xmath227 ( see the discussion at the end of subsection [ b - des ] ) : it is larger for `` narrow '' band excitation ( @xmath375 ) , and smaller for `` wide '' band excitation ( @xmath376 ) . for an excitation that is mostly of frequencies above @xmath377 , the spectral dependence of the energy of the emitted photons will behave according to the predictions of the be ( see figs . [ xi1]-[xi3 ] ) . however , as long as the excitation is above the threshold , very close to the threshold @xmath277 will increase linearly with @xmath276 , and not as a square root , see ( [ sqrt ] ) . non - boltzmann behavior is obtained when @xmath378 , that is @xmath379 , and the second lorentzian can be treated as a delta function . in this case one obtains @xmath380^{-1}.\ ] ] we bring here two specific examples of extreme non - boltzmann behavior of @xmath277 . in both examples the excitation is centered within the quantum interval , i.e. @xmath381 , and `` narrow '' band , i.e. @xmath375 . due to the latter inequality @xmath382 . in the first case the detuning is large , @xmath383 , therefore the energy width of an electron excited by the central excitation frequency , @xmath384 , that is of the order of @xmath385 , is small compared to its distance from the threshold . this is true for most of the electrons . in such a case the electron states are `` well defined '' , but the prediction of the be for the electron distribution is wrong , since its width @xmath386 is much smaller than the width of the states . as a result it follows from ( [ xi - q ] ) that the luminescence spectra is symmetric and centered at the classical position @xmath280 , with @xmath387 , but its width is given by the quantum - mechanical width of the electron states ( see fig . [ xi - q1 ] ) @xmath388^{1/2}.\ ] ] ( 20,10)(0,0 ) ( 0,0 ) ( 0,11.5)(0,0)[b]@xmath287 ( 18.5,0)(0,0)[lb]@xmath288 ( 11.3,-0.1)(0,0)[t]@xmath389 ( 0,0)(0,0)[tr]@xmath106 ( 10.5,4.3)(0,0)[r]@xmath289 ( 12,4.3)(0,0)[l]@xmath290 ( 13.5,4.3)(0,0)[l]@xmath291 ( 13,8 ) ( 13,11.5)(0,0)@xmath287 ( 13,8)(0,0)[tr]@xmath106 ( 18.5,8)(0,0)[lb]@xmath288 ( 30,16)(0,0 ) ( 0,0 ) ( 0,16)(0,0)[b]@xmath287 ( 25.5,0)(0,0)[lb]@xmath288 ( 15,0)(0,-0.5)[t]@xmath332 ( 0,0)(0,0)[tr]@xmath106 in the second case the detuning is small , @xmath390 , and the electrons are excited to `` badly defined '' states , since the broadening of these states is larger than their distance from the threshold . as a result the luminescence spectral distribution differs greatly from that predicted by the be . one finds from ( [ xi - q ] ) that ( see fig . [ xi - q3 ] ) @xmath391 and @xmath392 in conclusion let us note that the hole dispersion plays an essential role in determining the quantum behavior of the luminescence spectra . the width of the second lorentzian in ( [ bar - gam2 ] ) @xmath370 , is not equal to the width of the classical luminescence line @xmath393 , and the integral is not a simple convolution of the classical luminescence profile with the spectral function of an electron state . one can see from ( [ bar - gam2 ] ) that for a flat hole band ( @xmath394 ) the second lorenzian does not depend on the integration variable and the first lorenzian is integrated to one , restoring the be result , unexpectedly . the luminescence due to electrons that were excited _ below _ the threshold ( @xmath395 ) by an excitation centered _ above _ the threshold ( @xmath299 ) is not obtained by the simple exchange of @xmath341 by @xmath114 , as in the classical case . as long as @xmath396 one should take @xmath397 from expression ( [ gamma - th ] ) since @xmath398 , and therefore although the electrons are below the threshold , it is the optical phonon contribution to the electron width that is dominant . we have presented a recipe that allows one to generalize the be to a qke . the qke obtained by employing this recipe is the same as that obtained by the keldysh green function technique in the self - consistent born approximation . the advantage of this method for writing the qke is that it provides a physical understanding of the terms in the qke , and it allows one to neglect those terms that were negligible in the boltzmann description . this is due to the fact that the equations are written for quantities that are similar to the quantities described by the boltzmann equation . we considered the specific example of hot luminescence from a qwr . the recipe described above was used in order to generalize the set of bes that describe the problem to a set of qkes . solving these equations we were able to describe the luminescence spectral distribution . we have shown that there is a domain of luminescence frequencies , that correspond to a domain of phtoexcited electron energies , for which the quantum description of the luminescence spectral distribution leads to a different behavior than that given by the boltzmann description . this quantum domain could not be easily guessed from level width considerations . two other nontrivial conclusions were obtained . the first is the role played by the hole mass in the definition of the quantum domain . when the hole dispersion relation is flat the quantum domain shrinks to zero and the results of the boltzmann description are retrieved . the second nontrivial conclusion is that there is an energy domain below the threshold for lo phonon emission , in which the lo phonon contribution to the electron level width is dominant .

observations and modeling of supernova ( sn ) light indicate that the two most common power sources of the emission we observe are the radioactive decay of and the internal energy deposited in the envelope by the shock that unbinds the star ( * ? ? ? * and references therein ) . the emission powered by the latter is known as the cooling envelope emission and it dominates the photospheric phase in most type ii sne @xcite that mark the explosion of a red supergiant ( rsg ) ( * ? ? ? * ; * ? ? ? * and references therein ) . these are sub - classified into two sub - types based on the luminosity decline rate during that phase . if the photospheric phase shows no ( or at most a moderate ) decline the sn is classified as ii - p while if the decline is faster it is classified as ii - l @xcite . this classification is not well defined and different authors use different decline rates in different bands in order to separate the two classes ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? earlier studies suggested that the decline rate separates type ii sne into two distinctive populations ( e.g. , * ? ? ? * ; * ? ? ? * ) . however , recent studies ( @xcite , see however @xcite ) suggest that there is a continuous distribution of decline rates , thus challenging the usefulness of this classification . in this paper we do not attempt to separate sne into these two categories , using instead the measured decline rate to characterize each sn . interestingly , there is no clear theoretical explanation for the origin of the difference in the decline rates ( be it continuous or not ) , or to the correlation of the decline with other sne properties such as the peak brightness @xcite or spectral features @xcite . the most common , yet unconfirmed , suggestion is that steeper decline rates are generated by smaller envelope masses ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? numerical modeling of rsg explosions that include suggest that although cooling envelope emission dominates the photospheric phase , the power deposited by may have an observable effect also during that phase ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the main effect of on the light curve is via the additional radiated energy , which results in a brighter emission . a second order effect is the increase in ionization fraction ( and thus the opacity ) of the envelope , thereby delaying the release of internal energy deposited both by the shock and by the itself . for any realistic distribution of , its contribution to the observed emission increases with time during most of the photospheric phase ( this is true even for a uniform mixing through the entire star ; e.g. , @xcite ) . this may make the final stages of the photospheric phase brighter , and possibly more extended . the purpose of this paper is to use observations of type ii sne to study the impact of decay on the photospheric phase emission . in particular we are interested in a quantitative measurement of the importance in observed sne and in separating contribution from that of the cooling envelope emission , which is powered by shock deposited energy . we pay special attention to the effect has on the light curve decline rate . the paper structure is as follows . in [ sec : theory ] we use exact energy conservation arguments to derive a robust measurement of the contribution of during the photospheric phase , using the time - weighted integrated bolometric light curve . we also derive a measure of the cooling envelope emission alone ( i.e. , the one that would have been observed if there were no ) , which is directly related to its light curve shape . in section [ sec : nicontribution ] we apply these measurements to the bolometric light curves of a sample of sne that were compiled from the literature and are described in [ sec : sample ] . we show that is important in many sne , and constrain the effect that has on the light curve shape . in [ sec : correlations ] we test for correlations between the observables we introduce here and other observables we obtain for our sample . finally , in [ sec : sn_properties ] we use the time - weighted light curve to obtain a new global measurement that constrains the progenitor radius , envelope mass , and explosion energy , by rigorously subtracting the contribution . we show that , contrary to previous suggestions , a faster decline rate is most likely not a result of a lower envelope mass . we summarize our results in [ sec : summary ] . we are interested in a well - defined observationally - based measure that can separate the contributions of cooling envelope and decay to the photospheric emission for a given bolometric light curve @xmath0 . in type ii sne there is a sharp drop in the light curve once the photosphere ends crossing the envelope , allowing a clear distinction between the photospheric phase and the tail . we denote this time as @xmath1 , and in well observed sne it can be determined to within 10d accuracy or better . at @xmath2 the diffusion time through the envelope is much shorter than @xmath3 and therefore @xmath4 , where @xmath5 is the instantaneous injection of energy into the ejecta due to the decay chain . @xmath6 is time since the explosion in days . escape of gamma - rays generated by the decay of is minor in type ii sne for some time after @xmath1 , as evident from the tail observed decay rates , which are typically consistent with eq . [ eq : qni ] . therefore , observations of the tail provide a rather accurate estimate of the total mass in the ejecta ( e.g. , * ? ? ? ] , @xmath7 . once @xmath7 is found , the injection rate of energy @xmath8 at any given time is determined using ( up to late times where gamma - ray escape becomes significant ) . now we are faced with the following question at early times @xmath9 , how much of @xmath0 was due to the cooling envelope and how much was due to @xmath10 ? there are a few challenges that do not allow a simple separation of the two contributions : * energy which is deposited at some time @xmath11 is released at an unknown later time @xmath3 . * the opacity depends on the temperature and composition and is thus affected in a non linear way by the contribution of the two energy sources . * there are continuous adiabatic losses so that the emitted energy is smaller than the injected energy . let us imagine first that the ejecta were not expanding , so that there were no adiabatic losses . in such a case , we would know that the total emitted energy @xmath12 has to be the sum of the shock deposited energy and the integrated injected energy @xmath13 . since we know @xmath0 and @xmath8 , we would be able to determine each of the contributions . note that this separation would be based on total energy conservation and would thus not depend on the difference between deposited and released times nor on the effect of decay on the opacities . if we were to apply the same energy conservation argument to an expanding ejecta it would fail due to the significant and unknown adiabatic losses . a way to follow the conservation of total energy in an expanding ejecta was recently realized in the context of type ia sne @xcite and we apply the same method here . the idea is that for non - relativistic homologous expansion , where the energy is dominated by radiation , adiabatic losses do not affect the product @xmath14 , where @xmath15 is the energy in radiation that is trapped in the ejecta and @xmath3 is the time since explosion . thus we can separate the contributions by using the time - weighted emission @xmath16 . for clarity we repeat here these arguments and extend them to include the cooling envelope emission which is negligible in sne ia . during the homologous phase , which starts in the case of rsg progenitors about a day after the explosion , the derivative of the internal energy ( which is dominated by radiation ) in the outflow satisfies : @xmath17 the term @xmath18 is the total rate of adiabatic losses . after rearranging the equation , multiplying both sides by @xmath3 and integrating over @xmath3 from @xmath19 to @xmath1 one obtains : @xmath20 where @xmath21 is the time at which the homologues phase begins , @xmath22 and we used the fact that the contribution at @xmath23 to the integral on the r.h.s of the equation is negligible . we also used the fact that at late enough time @xmath24 the diffusion time through the envelope is much shorter than @xmath3 and therefore @xmath25 . on the left hand side of eq . , we have the total ( time weighted ) energy released . on the right side we have the total ( time weighted ) energy deposited , where we separated contributions from the decay chain @xmath26 and from the shock deposited energy @xmath27 ( just like in the hypothetic static case , but with time weights ) . the measurable parameter @xmath27 is of interest by itself as it is related directly to the progenitor structure and mass and to the explosion energy . we explore this relation in @xcite . we use their results here ( section [ sec : sn_properties ] ) to study the sample of sne that we compile in section [ sec : sample ] . we use here the notation of @xcite for @xmath27 : @xmath28 next , we define @xmath29 as the luminosity that would have been seen if there were no present in the ejecta . since without @xmath30 , eq . ( [ eq : e0t01 ] ) implies @xmath31 . this in turn implies that when does exist ( but @xmath29 is still the hypothetical luminosity when it does not ) : @xmath32 an observable measure of the contribution to the emission during the photospheric phase ( @xmath9 ) can be defined as the ratio of the ( time weighted ) energy deposited by decay to the shock deposited energy ( multiplied by @xmath21 ) : @xmath33 @xmath34 implies that shock deposited energy is the main power source and cooling envelope emission dominates during the entire photospheric phase . @xmath35 indicates that decay dominates the energy output during most of the photospheric phase . when @xmath36 the two power sources have a comparable contribution where cooling envelope dominates at early time and is more dominant near the end of the photospheric phase . finally , we define another dimensionless observable : @xmath37 where @xmath38 and @xmath39 are the hypothetical @xmath29 and the observed @xmath40 on day 25 , respectively . as the effect on @xmath40 is expected to be negligible on day 25 we assume that @xmath41 . the rational behind this definition is that @xmath42 is an observable that depends purely on @xmath29 . being the ratio between @xmath38 and the time - weighted integrated @xmath29 it measures a combination of the decline rate and duration of the light curve had been absent . in section [ sec : nicontribution ] we discuss its interpretation in more detail . the constant @xmath43 is inserted to make @xmath42 a dimensionless parameter of order unity , where the specific choice of @xmath44 is explained in section [ sec : nicontribution ] . we therefore have defined three observables , @xmath45 , @xmath42 , and et , that will allow us to separate and evaluate the relative contributions of the cooling envelope and to a given light curve . p1cmcccccccccccc 1999em & 0.54 & 0.55 & 0.17 & 1.34 & 14 & 12 & 12 & 136 & 4.7 & bol & 3280 & 1 + 1999gi & 0.63 & 0.66 & 0.33 & 1.37 & 16 & 13 & 12 & 130 & 5.8 & bol & 3700 & 1,2 + 2003hn & 0.31 & 1.3 & 0.85 & 1.16 & 23 & 13 & 11 & 110 & 3.2 & bol & & 1 + 2004et & 0.43 & 0.77 & 0.53 & 1.83 & 23 & 16 & 14 & 136 & 5.0 & ubvrijhk & 4020 & 3 + & ( 0.32)&(0.85)&(0.70)&(1.23 ) & ( 17 ) & ( 10 ) & ( 8.8)&(136 ) & ( 2.5 ) & ( ubvri ) & + 2005cs@xmath46 & 0.26 & 0.08 & -0.09 & 0.28 & 1.9 & 2.0 & 2.1 & 130 & 0.5 & bol & 1980 & 4 + & ( 0.15)&(0.11)&(0.01)&(0.17)&(1.2 ) & ( 1.1 ) & ( 1.2 ) & ( 130)&(0.17 ) & ( ubvri ) & + 2007od@xmath46 & 0.12 & 0.97 & 0.87 & 2.12 & 32 & 22 & 15 & 112 & 2.0 & uvoir & 3130 & 5 + 2009n & 0.61 & 0.67 & -0.04 & 0.40 & 4.6 & 5.1 & 4.8 & 112 & 2.0 & bol & 2580 & 6 + & ( 0.45)&(0.68)&(0.09)&(0.21)&(2.5 ) & ( 2.5 ) & ( 2.3 ) & ( 112)&(0.8 ) & ( bvri ) & & + 2009ib & 2.6 & 1.1 & 0.1 & 0.29 & 5.2 & 4.8 & 4.7 & 142 & 4.6 & uvoir & 3090 & 7 + 2009md & 0.28 & 0.59 & 0.26 & 0.23 & 2.4 & 2.2 & 1.9 & 121 & 0.5 & ubvrijhk & 2000@xmath47 & 8 + 2012a & 0.29 & 0.91 & 0.56 & 0.42 & 6.1 & 4.6 & 3.6 & 117 & 0.9 & uvoir & 2840@xmath47 & 9 + 2012aw@xmath48 & 0.49 & 0.56 & 0.31 & 1.52 & 16 & 13 & 12 & 135 & 4.9 & uvoir & 3890 & 10 + 2012ec & 0.57 & 0.88 & 0.18 & 0.73 & 10 & 9.3 & 8.8 & 112 & 3.5 & ubvrijhk & 3370 & 11 + 2013by & 0.2 & 1.2 & 1.19 & 1.66 & 33 & 25 & 11 & 104 & 2.9 & bol & & 12 , 13 [ table : bolsample ] p1cmcccccccccccc 1992h@xmath46 & 0.71 & 1.1 & 0.70 & 17 & 11 & 9.0 & 142 & 4.2 & bvr & 1 + 1995ad@xmath48 & 0.4 & 1.4 & 0.83 & 9.1 & 6.2 & 4.2 & 98 & 1.5 & ubvri & 2 + 2001dc & 0.43 & 0.82 & 0.35 & 0.90 & 0.73 & 0.65 & 120 & 0.22 & bvri & 3 + 2003z@xmath49 & 0.27 & 0.57 & 0.24 & 0.95 & 0.84 & 0.76 & 130 & 0.21 & bvri & 4 + 2004a & 0.63 & 0.68 & 0.06 & 4.4 & 4.3 & 4.2 & 120 & 1.75 & bvri & 5 + 2008 in & 0.39 & 0.96 & 0.53 & 2.8 & 2.1 & 1.7 & 115 & 0.55 & bvri & 6 + 2009bw & 0.17 & 0.64 & 0.77 & 13 & 7.8 & 6.6 & 140 & 1.2 & ubvri & 7 + 2009dd & 0.46 & 1.1 & 0.83 & 12 & 7.3 & 5.5 & 130 & 2.1 & ubvri & 2 + 2010aj & 0.09 & 1.3 & 1.05 & 11 & 7.5 & 4.3 & 94 & 0.5 & ubvri & 2 + 2013ab & 0.47 & 1.2 & 0.71 & 13 & 8.1 & 6.8 & 116 & 2.6 & uvo & 8 + 2013ej & 0.16 & 1.4 & 1.27 & 19 & 8.8 & 6 & 112 & 1.1 & uvo & 9 . [ table : optsample ] we compiled from the literature 24 well observed type ii sne for which detailed bolometric or pseudo - bolometric light curves are available . the light curves are taken from the references in tables [ table : bolsample ] and [ table : optsample ] and are based on the distances and extinctions used in each reference . an exception is sn2013by for which we computed the bolometric light curve ourselves based on blackbody fits to the photometry from @xcite assuming no significant extinction and taking a distance modulo @xmath50 . for a sn to be included in our sample , we required a reasonable coverage of the photospheric phase , namely first detection less than 30d after the estimated time of explosion , and several data points along the photospheric phase that enable a reasonable constraint on the luminosity evolution during that phase . we further required at least two measurements along the tail so that the mass of could be constrained . the analysis in this paper is sensitive to the light curve starting about 20 d after the explosion ( we verify that below ) and up to the tail . at these times most of the observed light is in the optical and ir . all the light curves we compile include the light emitted in the optical bands but some are missing the ir light . we therefore divide the 24 sne to two samples . the first is composed of 13 sne for which the published light curves provide a good approximation to the true bolometric light at the time of interest . these are light curves that include bolometric correction or pseudo - bolometric light curves that are base on uv / opt / ir or opt / ir light . we refer to this sample as the bolometric sample . the sne in this sample and their properties are given in table [ table : bolsample ] . the second sample has 11 sne where ir luminosity is missing , and thus their light curves may miss a significant fraction of the emitted luminosity . we refer to this sample as the optical sample , and its sne are listed in table [ table : optsample ] . as described below we first derive the observables of the bolometric sample and then , based on the results of the bolometeric sample , we derive properties of the optical sample that can be used for our analysis . for each sn in the bolometric sample we estimated first the mass based on the luminosity during the tail , and on equation [ eq : qni ] , assuming that @xmath51 . in all but two sne , the tail evolved as expected , namely a decay at a rate consistent with 0.98 mag/100d . the light curves of sne 2005cs and 2007od do not follow the expected evolution . we cautiously kept them in the sample , using the observed tail to get a rough estimate of the mass for sn2005cs and a value taken from the literature for sn 2007od @xcite , where dust may strongly affect the observed tail luminosity . the masses that we find are given in table [ table : bolsample ] . the mass found for each sne is very similar to the value derived in the paper from which we took the sne light curves ( always within the uncertainly and up to 10 - 20% from the best estimated value ) . next we determined @xmath1 for each sn . this can be done accurately ( to within @xmath52 d ) in all sne but one ( sn 2012aw ) , where @xmath1 was taken between the last data point of the photospheric phase and the first data point on the tail ( see table [ table : bolsample ] for details ) . for each sn we found the luminosities at days 25 , 50 , and 75 denoted as @xmath39 , @xmath53 and @xmath54 respectively , by linear interpolation ( in @xmath55 $ ] space ) . as a measure of the decline rate we define @xmath56 . we then calculate @xmath45 , @xmath42 and et for each sn . for all sne with available spectra we measure the ejecta velocities , and in a single case ( sn 2009md ) we take it from the literature ( spectra for the majority of the sne were obtained from wiserep ; @xcite ) . these velocities are used in section [ sec : sn_properties ] . traditionally , the feii @xmath57 absorption line velocity , as measured in mid - plateau , is considered a good proxy for the velocity of the photosphere ( e.g. , @xcite ; @xcite ) . briefly , as often done before @xcite , we cross correlate the spectra with a library of high signal - to - noise spectra for which the velocity of the @xmath57 line has been measured directly . the velocity from the cross - correlation and its uncertainty are then propagated to day 50 past explosion , following @xcite , who showed that photospheric velocities of sne ii - p follow a tight power law relation . we use the improved determination of the phase dependance of the velocity by @xcite . we note that @xcite found that sne ii - l have a different , slower , velocity evolution , with some scatter . as a result , our calculations may slightly underestimate the velocity for such sne . we can use the bolometric sample to evaluate the accuracy of observables that are derived based on light curves that lack ir luminosity . we repeat the derivation described above for three of the sne in the bolometric sample , where ubvri ( or bvri ) and bolometric light curve are provided in the same paper ( i.e. , using the same data , estimated distance and extinction ) . the derivation of all the observables is done as if the optical light curve is bolometric . namely , @xmath7 is derived by comparison of the optical light in the tail to equation [ eq : qni ] , while @xmath45 and @xmath42 are calculated by time integration of the optical light . comparison of the quantities derived based on optical light alone to those derived based on bolometric light show that in all observables that are linear with the luminosity ( i.e. , @xmath58 , @xmath7 and et ) the usage of optical light only leads to an underestimate by about a factor of 2 , and in some cases even by a factor of 3 . however , the difference in the dimensionless observables is much smaller a moderate factor of @xmath59 in the measurement of @xmath45 , a difference of 0.010.08 in @xmath60 and a difference of @xmath61 mag/50d in @xmath62 . the reason that these dimensionless quantities are rather accurate also when only optical light is considered is that they are basically a ratio of the luminosities observed at different times and therefore they are insensitive to the absolute value of the missing bolometric correction factor . instead they are only sensitive to the time evolution of the correction factor , which turns out to be not too large . based on these results we restrict the usage the optical sample in our analysis to dimensionless quantities . these are given in table [ table : optsample ] . we also provide , for completeness , the value of the mass we use for the derivation of @xmath45 and @xmath42 , although these are not the true masses in these sne as they are derived using the optical light curves as if they were bolometric . finally , in order to test the effect of missing data at early time ( in about half of the sample the first data point is around day 20 ) , as well as the effect of missing uv light , we used 8 sne in our two samples , for which data starts no more than 10d after the estimated time of explosion . we compared @xmath45 and et ( the only two quantities affected by early data ) of these sne when measured with and without the data before day 20 . the typical difference in @xmath45 is @xmath63 and in et it is @xmath64 . we therefore conclude that the effect of missing early data and/or uv light is not a major source of uncertainty . this result also implies that the uncertainty in the estimated time of explosion ( which is typically better than 10 days ) also does not introduce a significant uncertainty to the quantities we measure . the values of @xmath45 in our sample are all , except one , in the range 0.090.71 , with most sne having @xmath65 ( tables [ table : bolsample ] and [ table : optsample ] ) . the exception is sn 2009ib with @xmath66 . the fraction of the time weighted integrated luminosity that is contributed by is @xmath67 . thus , the observed values of @xmath45 in our sample ( 0.092.6 ) imply that the range of contribution to the time weighted integrated luminosity is 8%-72% with a typical value of 30% . these values indicate that has a non - negligible contribution to the photospheric emission for most of the sample . but what is this effect on the observed light curve and can it be quantified ? @xmath45 and @xmath42 have the advantage of being observables that are independent of radiative transfer . but , for that they must be integrals and as such they do not hold information on the exact shape of the light curve . namely , knowing their values does not enable us to determine exactly how the light curve would have been looked like if there were no . the reason is that in order to remove the contribution we need to know the exact distribution and to calculate its effect on the radiative transfer . yet , @xmath45 and @xmath42 can provide quantitative constraints on the effect of on the observed light curve . to obtain that , we use the expectation that is produced in the core and that its fraction drops , or at most remains constant , with radius . this implies that the relative contribution of increases with time , reaching a peak near the end of the photospheric phase . thus , the effect of on the light curve can be either flattening ( i.e. , reducing the decline rate ) or extending the duration of the photospheric phase ( or a combination of both effects ) . in fig . ( [ fig : le_sketch ] ) we sketch the two extreme possibilities of effect on the light curve . the first is that is extending the plateau duration without affecting the decline rate . in this case the cooling envelope contribution underlying the combined ` observed ' light curve would have looked like the dashed red line in fig . ( [ fig : le_sketch ] ) . alternatively , could make the last stages of the plateau brighter ( i.e. , making it flatter ) without significantly affecting the plateau duration , and the cooling envelope emission would look like the dotted blue line in fig . ( [ fig : le_sketch ] ) . the true effect is a combination of the two , but what is the actual possible effect in each scenario and which one is more likely to be the dominant ? to address this question we consider each scenario in some detail and estimate ( quantitatively ) its possible effect on the sne in our sample . first , consider the possibility that only extends the plateau without significantly affecting its decline rate . namely , does not affect the light curve up to the time that @xmath29 starts fading , at which point it becomes the main power source , thereby extending the plateau . this behavior is expected to dominate when is concentrated in the core and is not well mixed into the envelope . then , the luminosity generated by power diffuses to the observer only after almost all the shock deposited energy has already been leaked out of the envelope fades and the power becomes the dominant energy source . ] . if the plateau is flat ( i.e. , @xmath40 is constant ) then the relation between @xmath45 and the plateau extension in this scenario is analytic . denoting @xmath68 and @xmath69 as the durations of the plateaus of @xmath40 and @xmath29 , respectively , then in this scenario @xmath70 , where from eqs . ( [ eq : et2 ] ) and ( [ eq : etani ] ) we find @xmath71 . thus , in this case @xmath72 , where the final equality is for @xmath65 . assuming instead @xmath40 with a constant decline rate of 1 [ 2 ] mag/100d and @xmath73 d , the observed range @xmath65 corresponds to @xmath74 [ @xmath75 . next , we consider the other extreme possibility , namely that the plateau duration is not affected by ( or @xmath76 ) and that instead the changes only the luminosity along the plateau . since is always negligible at early times the effect must be an increase in the luminosity at late times , namely more results in a flatter ( or even rising ) plateau . this is the effect which is expected to dominate if is well mixed throughout a significant fraction of the envelope . efficiently mixed generates luminosity that diffuses to the observer at earlier stages , simultaneously with the cooling envelope emission . without the cooling envelope emission , the generated luminosity would have looked in this case like a broad peak ( similar to that observed in sn 1987a ) . the main effect on the plateau duration is then via increase in the envelope opacity . the flattening effect on is seen clearly in the numerical light curve of the fully mixed case in figure 12 of @xcite . the change in the decline rate that induces in this case can be estimated using @xmath42 , which is a ratio between the luminosity at early time ( day 25 ) and the time - weighted average luminosity . thus , if the duration of the plateau is roughly constant in all sne then we expect @xmath42 to be a measure of the decline rate of @xmath29 . to verify this , and to calibrate @xmath42 , we define its bolometric luminosity counterpart : @xmath77 fig . ( [ fig : lambdavsdm ] ) shows the tight linear correlation between @xmath62 and @xmath78 , which is consistent with @xmath79 ( the constant @xmath43 in the definition of @xmath80 and @xmath42 was chosen to obtain this equality ) . thus , @xmath78 is a good estimator of the observed decline rate ( in units of mag/50d ) and if the does not affect the plateau duration then @xmath42 is a good estimator of the decline rate of @xmath29 . fig . ( [ fig : lambdavsdm ] ) also shows @xmath81 as a function of @xmath78 . the measured values are narrowly distributed around @xmath82 ( with the exception of sn 2009ib ) . this implies two interesting points . in this scenario the reduces the decline rate in flat sne ( with small @xmath80 ) by about 1 mag/100 d , while sne with fast decline ( large @xmath80 ) are less affected . for the majority of the sne in our sample the flattening in this scenario is in the range of 0.5 - 1 mag/100 d. second , the tight correlation between @xmath80 and @xmath83 indicates that while affects the observed decline rate , it is not the only source for the difference between sne in the observed decline rates . namely , different observed decline rates often reflects different decline rates of the cooling envelope emission . the two scenarios discussed above provide bounds to the effect of on the light curve . in reality it is a combination of both ( i.e. , both a flattening and an extension of the plateau ) , where the relative importance depends on the level of mixing in the envelope . yet , there are several lines of evidence that support that flattening is significant , and possibly the dominant effect on the light curve : \(i ) flattening is expected if is well mixed through the envelope . the best clue regarding mixing in sn explosions of progenitors with massive hydrogen - rich envelopes is sn 1987a and similar events ( e.g. * ? ? ? * ) , which have progenitors with similar masses and composition but smaller radii than those of regular type ii sne . the handful of sne of this type that were well observed are all showing rather similar light curves , which often look like a clone of sn 1987a up to a normalization factor ( * ? ? ? * and references therein ) . detailed light curve modeling of sn 1987a favor an efficient mixing throughout the hydrogen envelope ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? \(ii ) the first scenario no mixing and a significant extension of the plateau requires a fine tuning of the amount of . the reason is that the photospheric phase can be separated in this case into two phases first a cooling envelope phase where is negligible and later a dominated phase . the luminosity of the two phases depends on different explosion parameters ( the first phase is determined mostly by the progenitor radius and to some extent by the explosion energy and progenitor mass , while the second phase depends only on the amount of mass ) . therefore , without finely tuning the amount of , a transition between the two phases is expected to be observed during late stages of the plateau , either as a bump if there is too much or a dip if there is not enough . this can be seen in @xcite ( their fig . 2 ) where @xmath7 is varied by a factor of 4 ( and so does @xmath45 between 0.5 and 2 ) . this variation is enough to change the light curve from a dip near the end of the plateau to a bump . in our sample the plateaus do not show any transition between two phases , even though the contribution varies significantly ( @xmath84 ) including a case where dominates the photospheric light curve . \(iii ) in the second scenario , where is well mixed , its contribution becomes significant earlier , near the middle of the plateau ( if sn1987a is a reasonable example for contribution then around day 50 ) , and from that point , both and cooling envelope are significant until the plateau ends . thus , instead of a transition between the two phases , a flattening is expected to be seen around the middle of the plateau . such a flattening is indeed observed in many sn light curves ( e.g. , @xcite ; the transition from s1 to s2 in their notation ) . \(iv ) each scenario predicts different correlations between observables . if results in flattening of the light curve then an anti - correlation is expected between the importance of contribution ( @xmath45 ) and the decline rate . on the other hand , if extends the plateau no such anti - correlation is expected and instead a correlation between @xmath45 and @xmath1 is predicted . as we show in the next section we find a highly significant anti - correlation between @xmath45 and @xmath62 and at most a weak correlation between @xmath45 and @xmath1 . these arguments are all suggesting that @xmath81 is not only an upper limit on the decline rate of the light curve without , @xmath29 , but that it also provides a rough estimate of this decline rate . to conclude , contribution to the photospheric emission is significant in type ii sne , although in most it is not the main source of the observed luminosity . typically it contributes around 30% of the time weighted integrated luminosity , but in some sne it contributes less than 10% and in the extreme case of sn2009ib it contribute 72% . this contribution flattens and extends the plateau . in typical sne , the maximal possible extension of the plateau is by about 25% while the maximal possible flattening is by about 1 mag/100 d. the actual effect is a combination of both , however , several line of evidence suggest that flattening is significant . finally , while most likely affect the decline rate of the plateau , the observed range of decline rates is clearly dominated by the decline rates of the cooling envelope emission and can not be a result of alone . p1cmccccccccccccc @xmath45 & & n / a & @xmath85 & 0.05 & n / a & n / a & n / a & 0.02 & n / a & n / a & n / a & n / a + @xmath42 & & & @xmath86&@xmath870.04 & 0.03 & 0.06 & n / a & n / a & n / a & n / a & n / a & n / a + @xmath62 & & & & @xmath870.04 & @xmath88&0.002 & 0.06 & n / a & 0.06 & 0.005 & 0.02 & 0.04 + @xmath1 & & & & & n / a & n / a & n / a & 0.07 & n / a & n / a & n / a & n / a + @xmath39 & & & & & & & & 0.06 & @xmath88&@xmath86&@xmath86 & @xmath86 + @xmath53 & & & & & & & & 0.04 & @xmath89&@xmath86&@xmath86 & @xmath86 + @xmath54 & & & & & & & & 0.008 & @xmath90&@xmath86&@xmath86 & @xmath86 + @xmath7 & & & & & & & & & n / a & 0.03 & 0.01 & 0.05 + [ table : correlations ] table [ table : correlations ] list the significance of correlations ( and anti - correlations ) between every two observables listed in tables [ table : bolsample ] and [ table : optsample ] . here we focus on the correlations of the two new dimensionless observables that we introduce , @xmath45 and @xmath42 , which are relevant for the effect on on the light curve . @xmath45 does not show a significant correlation to most of the observables . specifically , it seems to be uncorrelated with @xmath42 , which is a measure of the cooling envelope shape , and with the plateau luminosity @xmath39 , @xmath53 and @xmath54 ( see illustration in figure [ fig : etanivsx ] ) . it also shows no correlation with @xmath91 , @xmath92 and @xmath93 ( see discussion in section [ sec : sn_properties ] ) . a marginally significant correlation ( probability of no correlation @xmath94 ) is seen with the plateau duration and a slightly more significant correlation ( probability of no correlation @xmath95 ) with @xmath7 ( see figure [ fig : etanivsx ] ) . the only highly significant finding ( probability of no correlation @xmath96 ) is an anti - correlation with the bolometric decline rate between days 25 and 75 , @xmath62 . ( [ fig : dmvsetani ] ) shows a scatter plot of @xmath45 vs. @xmath62 . a roughly linear anti - correlation is clearly seen . namely , the role of during the photospheric phase is more prominent in sne with a smaller decline rate ( a ` flatter ' plateau ) . the scatter around the correlation is considerable and it is most likely a combination of intrinsic scatter and the inhomogeneity of our sample . the anti - correlation between contribution and the decline rates can also be seen when the well known correlation between the plateau luminosity and the mass is plotted . ( [ fig : l25vsmni ] ) shows the luminosity at 25 days after explosion , @xmath39 , as a function of the mass . an approximate linear relation is seen , similarly to the finding of previous works ( e.g. , * ? ? ? in order to show the relation to the contribution and decline rate , we use different symbols for sne with different values of @xmath62 . sne that decline fast are significantly above those that decline slowly . namely , for the same amount of fast declining sne are brighter and thus have a smaller @xmath45 . this is consistent with the results of @xcite . ( [ fig : l75vsmni ] ) is similar to fig . ( [ fig : l25vsmni ] ) , but with @xmath54 instead of @xmath39 . while fast declining sne are still brighter on average at day 75 ( for the same amount of ) , the difference from slow declining sne is reduced , as expected . we have seen in section [ sec : nicontribution ] that for sne with a high value of @xmath45 , must have an observable effect either via flattening the plateau or by extending it or , most likely , both . the significant anti - correlation between @xmath45 and @xmath62 supports a picture were flattening is a major effect . this in turn requires to be mixed into a significant part of the envelope . the marginal correlation with @xmath1 suggest that also extends the plateau , although not significantly . next we examine the correlations of @xmath42 , some of which are illustrated in figure [ fig : lambda_evsx ] . @xmath42 probes the shape of @xmath29 , the light curve that we would have seen if there were no . it is larger when @xmath29 declines faster and/or when its duration is shorter . @xmath42 shows a very strong correlation with @xmath62 . this indicates that the observed decline rate is dominated by the cooling envelope emission . namely , while most likely affect the decline rate it is not the source of the range of decline rates that are observed . instead this range is dominated by the cooling envelope emission . @xmath42 also shows a marginal anti - correlation with @xmath1 suggesting that the variance that we see in plateau durations is at least partially determined by @xmath29 . finally @xmath42 is also correlated with the plateau luminosity . this correlation is most likely related to the correlation between the decline rate and the luminosity @xcite , whose origin is still unknown . other than that @xmath42 shows no correlation to any other observable we examined . interestingly , this includes @xmath7 and @xmath45 , suggesting that the shape of @xmath29 is not correlated with the amount in the ejecta or its importance in shaping the observed light . as discussed in [ sec : theory ] the total contribution of the cooling envelope to the ( time - weighted ) energy release , et , can be directly obtained from observation without major uncertainties related to the details of radiative transfer . this observable is directly related to the explosion energy and progenitor structure . this relation was explored recently by @xcite . here we use their results to study our bolometric sample ( table [ table : bolsample ] ) , for which we obtained a reliable measure of et . @xcite used a large set of numerically calculated red supergiants to study what et can teach us on the progenitor and explosion energy . in general @xmath98 where @xmath99 , @xmath100 , @xmath101 and @xmath102 are energy , mass , radius and velocity that characterize the explosion and the progenitor . specifically they find three approximations , each for a different characteristic e , m and r : @xmath103 where @xmath104 is the total explosion energy carried by the entire ejecta to infinity , @xmath105 is the energy carried by the envelope to infinity , @xmath106 [ @xmath107 is the ejecta [ envelope ] mass , @xmath108 the progenitor radius and @xmath109 is the mass weighted average radius of the envelope ( @xmath110 is mass element of the progenitor at radius @xmath111 ) . the quality of the approximations vary between the three . the reasons are , first , that the cooling envelope emission depends only on the envelope properties , and therefore approximations that use only envelope characteristics , the first two in equation [ eq : et ] , are more accurate than the third one , which use global properties of the ejecta . second , @xmath91 also depends on the internal structure ( density profile ) of the progenitor , thus the first approximation which takes it into account is better than the second one . quantitatively , the first , which depends on @xmath105 , @xmath112 and @xmath113 , is accurate to within 20% for all the progenitors considered by @xcite , including those that lost almost all of their hydrogen envelope . the second approximation , which depends on @xmath105 , @xmath108 and @xmath113 is accurate to within 20% for progenitors that retain most of their envelope , and to within a factor of 3 for progenitors that lost most of their envelope . the third approximation , which depends on @xmath104 , @xmath108 and @xmath106 is applicable only to progenitors where @xmath114 , in which case it is accurate to within about 30% . additional information can be extracted from et for sne with spectral measurements of the photospheric velocity . @xcite show that @xmath97 provides a good estimate of @xmath115 , and when the ejecta is dominated by the envelope it is also a good approximation of @xmath116 . thus , @xmath117 and @xmath118 . in order to apply equation [ eq : et ] to observations a reliable estimate of et is needed , which in turn requires good bolometric light curve . therefore we use here only the 13 sne in our bolometric sample ( table [ table : bolsample ] ) . ( [ fig : e0m0r0vsl25 ] ) depicts scatter plots of @xmath39 with @xmath91 , @xmath92 and @xmath93 , normalized by @xmath119 , @xmath120 and @xmath121 , respectively , where @xmath122 erg , @xmath123 and @xmath124 . the logarithmic mean of all three is almost identical , 0.8 , suggesting that the canonical progenitor envelope values we used for the normalization are representative for this sample . the spread in the values is largest for @xmath92 , about an order of magnitude , and smallest for @xmath93 , about a factor of 3 . this suggests that unless there is an anti - correlation between @xmath113 and @xmath108 ( rsgs calculated by the stellar evolution code mesa does not show such anti - correlation ; @xcite ) , the spread is dominated by variation of @xmath105 which in most progenitors is similar to @xmath125 . this spread is consistent with the findings of @xcite who found @xmath126 in a sample of sne ii - p with detected progenitors . the typical values that we find are similar to those found in estimates of @xmath125 , @xmath127 and @xmath108 via detailed hydrodynamical modeling of individual events in our sample ( see the references in table [ table : bolsample ] ) . however , @xcite finds systematically larger explosion energies and ejecta masses , and lower progenitor radii . the estimates in @xcite are based on general numerical models @xcite that solve for @xmath125 , @xmath127 and @xmath108 based on three observables , @xmath53 ( in v band ) , @xmath97 , and the plateau duration . these models ignore the contribution and this can explain at least part of the discrepancy . as we find here , is affecting the plateau properties in most type ii sne . our results show that either flatten the plateau , in which case it is most likely contributes significantly to @xmath53 , and/or extends the plateau . both effects , when ignored , cause overestimates of @xmath125 and @xmath127 and an underestimate of @xmath108 . table [ table : correlations ] lists the correlation significance of @xmath91 , @xmath92 and @xmath93 with other observables . no correlation is found with @xmath45 suggesting the relative contribution is most likely uncorrelated to the envelope properties and explosion energy . we do find however highly significant correlations with all luminosities ( @xmath39 , @xmath53 and @xmath54 ; see for example fig . [ fig : e0m0r0vsl25 ] ) . this implies , again assuming no anti - correlation between @xmath113 and @xmath108 , that brighter sne are more energetic and more massive and/or with larger radius . this is consistent with the estimates based on hydrodynamical modeling for individual sne ( see the references in table [ table : bolsample ] ) . finally , all three observables show a significant correlation with @xmath62 . this suggests that faster declining sne have larger @xmath105 and larger @xmath113 and/or @xmath108 . some of the implications of this correlation are discussed in the next sub - section . the progenitor properties of type ii - l sne is still unknown . the most common suggestion is that the fast decline is a result of a small envelope mass . this possibility was first speculated by @xcite . later , numerical simulations have shown that envelopes masses of @xmath128 can produce the observed linear declines ( e.g. , * ? ? ? * ; * ? ? ? * ; see however @xcite ) . our results of positive correlation between @xmath129 and @xmath62 does not support this picture . this correlation is evident in fig . ( [ fig : e0m0r0vsdm ] ) . it shows that @xmath130 of fast declining sne are larger by a factor of 2 - 3 compared to slow declining sne . thus , unless there is a strong anti - correlation between @xmath113 and @xmath108 , which contradicts predictions of stellar evolution models , the envelope of fast declining sne in our sample are at least as massive as those of slow declining sne . moreover , for 6 out of the 11 sne with estimate of @xmath130 there are pre - explosion images of the progenitor @xcite from which @xmath108 can be estimated ( see table 1 in @xcite ) . the range of radii is small , about a factor of 2 , and they show no correlation with @xmath62 . another indication that faster decline is unlikely to be the result of a lower envelope mass is that if that were the case , the photospheric phase of fast declining sne should have been much shorter . the reason is that for a given ejecta velocity the duration of the photospheric phase depends strongly on the ejecta mass ( fast declining sne have similar @xmath97 to slow declining ones with similar luminosity ) . this can bee seen for example in @xcite and @xcite who model type ii - l sne with low mass ( @xmath131 ) envelopes , where for all these models @xmath132 d. in our observed sample , on the other hand , @xmath1 is in the range of 90140d and very weakly correlates with @xmath62 if at all in the sample is narrowly clustered around the mean ( 120 d ) where almost all sne are in the range 110 - 140 d and no correlation with @xmath62 is observed . the exception are three fast declining sne , 1995ad , 2010aj and 2013by , which have @xmath1 shorter by @xmath133 than the average . ] . if , indeed , the envelope mass is not the main driver of the observed range of decline rates then what can it be ? we show here that while most likely affects the observed decline rate its effect is only secondary to that of the intrinsic cooling envelope emission , @xmath29 . the temporal evolution of @xmath29 , in turn , is determined to a large extent by the pre - explosion density profile of the progenitor , we therefore speculate that the decline rate is determined mostly by the envelope structure and not by its mass . we explore the effect of on the photospheric emission of type ii sne . we use energy conservation in a spherical outflow that expands homologously to derive a measure for the importance of ( eq . [ eq : etani ] ) . this measure is obtained by time weighted integrals over the bolometric light curve and decay energy deposition rate . as such it is insensitive to the complicated physical processes of radiative transfer and its accuracy depends mostly on the quality of the bolometric light curve . we use a similar method to derive a measure that depends purely on the light curve shape of the cooling envelope emission that would have been seen if there were no in the ejecta ( eq . [ eq : lambda_e ] ) . we compile from the literature a sample of 24 type ii sne with detailed bolometric and quasi - bolometric light curves of the photospheric phase and tail . the sample is heterogeneous , ranging over a factor of 30 in plateau luminosity and mass , and over a range of 2.5 mag/100d in decline rate . namely , the sample includes luminous , intermediate and low - luminosity type ii sne as well as sne that were classified as ii - p and as ii - l . we analyze this sample , calculating the new observables that we derive here and find the following results : * contribution is significant in most of the sample . in many sne it is the source of about 30% of the time weighted integrated luminosity during the photospheric phase . but , there is a single sn ( 2009ib ) in the sample where contributes 72% and several sne where it powers about 10% . * contribution can possibly flatten and/or extend the photospheric light curve . we find that if flattening were the only effect , then sne that are observed to have a flat plateau would have shown a decline rate of about 1 mag/100 d in the absence of . if on the other hand the only effect of were to extend the plateau , then without the photospheric phase would have been typically shorter by a factor of about 1.151.35 and in most extreme cases by almost a factor of 2 . * contribution can not explain the entire range of observed decline rates , which is mostly an intrinsic property of the cooling envelope . several independent lines of evidence suggest that the dominant effect of is flattening of the light curve . this includes a significant anti - correlation between the level of contribution and the decline rate ( i.e. , is more important in sne with a flatter plateau ) . this is expected if is mixed throughout a significant fraction of the envelope ( as suggested by @xcite for sn1999em ) . with this interpretation the picture that arises from our results is that flat plateaus are not a generic feature of cooling envelope emission . instead many observed plateaus includes a significant contribution of , and without it many of the type ii - p sne would have shown a decline rate of up to 1 mag/100d . note that while not being expected a - priori , there is no need for a ` conspiracy ' so the combination of cooling envelope and emission would result in plateaus . first , type ii sne exhibit a range of decline rates , thus in some cases the combined emission produces a plateau and in others it does not . second , it seems that even without the cooling envelope emission often shows a moderate decline rate where the luminosity drops by less than a factor of two over 100 d. in these cases the contribution of is enough to make the light curve even flatter . the coincidence in type ii sne seems to be the fact that the contribution of cooling envelope and powered emission are comparable over a wide range of light curve luminosities . this is a result of the well known tight correlation between the luminosity and mass @xcite , which most likely reflects a correlation between the explosion energy and production . finally , we do not find correlations between the level of contribution and most of the observed light curve properties , as well as other properties of the cooling envelope emission . on the other hand , @xmath42 , a property of cooling envelope emission alone , does correlate with the plateau luminosity and decline rate . this suggests that the many correlations seen between properties of type ii light curve ( luminosity , velocity , decline rate , etc . ) seem to be intrinsic properties of the cooling envelope emission . we use the method to remove the effect of on the integrated light curve to derive an observable measure of a combination of the progenitor radius , envelope mass and explosion energy ( equation [ eq : et ] ; see @xcite for more details ) . this measure is robust in the sense that , being a result of an integrated energy conservation equation , it is insensitive to the radiation transfer of the observed light . applying this measure to our sample , together with a measurement of the photospheric velocity at day 50 , we find , in agreement with previous studies , that brighter sne are most likely generated by more energetic explosions with larger ejecta mass and/or progenitor radius . we also find that the decline rate is positively correlated with @xmath134 , which contradicts the popular view that fast decline in type ii sne is a result of significant envelope stripping and thus a low value of @xmath135 . instead it is more likely to be related to the mass distribution of the ejecta , rather than to its total mass . we thank iair arcavi , doron kushnir , viktoriya morozova and tony piro for useful comments . we thank ofer yaron for pointing out the extreme light curve of 2009ib . we thank the anonymous referee for a thorough report that helped us improving the paper . this research made use of the weizmann interactive supernova data repository ( ` www.weizmann.ac.il/astrophysics/wiserep ` ) , as well as the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory , california institute of technology , under contract with nasa . this research was partially supported by the i - core program ( 1829/12 ) . en was was partially supported by an erc starting grant ( grb / sn ) , isf grant ( 1277/13 ) and an isa grant . bk was partially supported by the beracha foundation .

the spin - orbit coupled quantum systems are described by the hamiltonian which is realized as the differential operator in the tensor product space @xmath0 : @xmath1 in ( [ eq : diffop ] ) , @xmath2 stands for the spin - orbit - coupling strength , @xmath3 is the strength of the magnetic zeeman field @xcite . the symbol @xmath4 denotes the three - dimensional laplace operator , @xmath5 ( @xmath6 ) is the gradient in the @xmath7th component of a three - dimensional position - vector . the standard pauli matrices are represented by @xmath8 . the imaginary unit @xmath9 . here and elsewhere we write @xmath10 for an identity operator . the space in which @xmath10 acts is expected to be understood from the context . the main purpose of the current exposition is to study the spectrum of self - adjoint extensions of the symmetric operator constructed from ( [ eq : diffop ] ) . let @xmath11 be defined on a set of compactly supported smooth functions , with the support outside the origin @xmath12 : @xmath13 then @xmath11 is symmetric with respect to the scalar product in the hilbert tensor product @xmath14 . in applications , the self - adjoint extensions of so defined @xmath11 are referred to as the rashba spin - orbit coupled hamiltonians considered in the presence of the _ out - of - plane _ magnetic field , with the impurity scattering treated via the spin - dependent contact interaction . a good review on various types of disorder in condensed - matter systems is given in @xcite . for the analysis of long - range interactions in electronic systems , as opposed to the present discussion where the zero - range interactions are studied , the reader may refer to @xcite . the special case @xmath15 was discussed recently in @xcite , where , among other things , the authors examined the effects due to the _ in - plane _ magnetic field . the corresponding term in ( [ eq : diffop ] ) is then replaced by a unitarily equivalent one @xmath16 . in their theorem 2 in @xcite the authors showed under what conditions the point spectrum is empty . here we prove that under exactly the same conditions there is a subfamily of self - adjoint extensions whose point spectrum is @xmath17 . the case @xmath18 is a classic one . the book @xcite is much used as a reference book for a standard ( or von neumann ) operator extension theory ; see also an extensive list of references therein . the extensions constructed using spaces of boundary values are discussed , for example , in @xcite . a modern and comprehensive review of various operator techniques is given in @xcite . in one and two spatial dimensions , the spectral properties of self - adjoint extensions of @xmath11 , for @xmath19 , were studied in @xcite . a mean field interpretation , which is commonly accepted in physics literature , can be found in @xcite . a recently proposed technique @xcite ( see also @xcite ) for producing the rashba - type spin - orbit coupling for a three - dimensional ultracold atom serves as our main motive to examine a general case @xmath20 . many more motivating aspects concerning the dynamics of quantum particles can be found in @xcite , and we do not repeat them here . contrary to @xcite , where the modified krein resolvent formula @xcite has been used as a starting point for calculating the resolvent of @xmath11 ( with @xmath15 ) , here , we exploit the theory of singular finite rank perturbations @xcite . on the one hand , the two approaches lead to an identical description of the boundary conditions that ensure self - adjointness @xcite . on the other hand , the singularities of green function ( see _ e.g. _ @xcite and references therein ) are successfully eliminated using the so - called renormalization procedure @xcite . such a procedure is implemented in the theory of singular perturbations in its most natural way @xcite . as an illustration , let us consider a formal `` operator '' @xmath21 , where @xmath11 in ( [ eq : diffop ] ) is written in matrix representation , and where @xmath22 is the dirac distribution concentrated at the origin ; for @xmath18 , this type of operator , when restricted to one spatial dimension but extended to the many - body case , is studied in @xcite . the matrix @xmath23 , which is hermitian and invertible , plays the role of the coupling parameter of contact interaction . the lipmann schwinger equation for the perturbed operator shows that the solution of the integral equation is inconsistent , in that the green function for @xmath11 is singular at @xmath24 ( see ( [ eq : frg ] ) for the details ) , meaning that the equation does not have a unique solution . the phenomenon is usually called the ultraviolet divergence . for , the fourier integral of the green function in the momentum representation diverges when @xmath24 . to avoid divergences , typically one introduces an auxiliary uv momentum cut - off and regularizes the interaction parameter . the details can be found in ( * ? ? ? 3.7 ) , see also @xcite . in the theory of singular perturbations , however , one deals with a regularization and a renormalization procedure applied to @xmath22 . in this case the regularization is meant in the sense that the dirac distribution is extended to some subspace of a hilbert space . roughly speaking the action of @xmath22 on the function which is undefined at the origin is replaced by the action of some regularized @xmath22 . the present discussion is justified rigorously in sec . [ sec : self ] . next , the regularized and renormalized @xmath22 is determined by the choice of an hermitian matrix @xmath25 , which is usually called _ admissible _ ; the matrix realization of @xmath25 is due to an extra @xmath26 in ( [ eq : diffop ] ) . it turns out that the theory also deals with an additional parameter ( _ i.e. _ the matrix @xmath25 ) . however , the precise operator realization of @xmath21 by extending @xmath22 gives one an advantage over the cut - off practice in an obvious way . one shows that the self - adjoint realizations of @xmath21 are parametrized in terms of @xmath27 , with the exception of the trivial extension corresponding to @xmath28 and the friedrichs extension corresponding to @xmath29 . for a general @xmath19 , the functional @xmath22 is of the class for which @xmath25 is not unique , but in some cases @xmath25 can be found exactly ; _ @xmath30 for @xmath18 . for some special classes of singular finite rank perturbations , the uniqueness of @xmath25 is discussed in great detail in @xcite . another reason for choosing the approach of singular perturbations is that the scattering theory for singular finite rank perturbations is fully established ( * ? ? ? 4 ) : using our results , the scattering matrix can be found directly from ( * ? ? ? * eq . ( 4.34 ) ) . for this particular reason we do not examine scattering states but rather concentrate on analytic properties of the spectrum . to the best of our knowledge , the model studied in the current paper is investigated by means of the theory of singular perturbations for the first time . the lack of rigorous results concerning the spectral analysis of the perturbed @xmath11 may be explained by the complexity of the green function for the free hamiltonian ( or else the trivial extension of @xmath11 ) . a typical strategy is to write the green function in the momentum representation . then the dispersion relation and the essential spectrum are easy to deduce . however , the computation of the inverse fourier transform of green function is a rather troublesome task . recently @xcite , a hypergeometric series representation for the green function in the coordinate representation has been derived . in contrast , the analogous result in two spatial dimensions possesses a closed form @xcite . the details are listed in theorems [ thm : without ] , [ thm : existence ] , [ thm : existence3].,title="fig : " ] 0.35 . the details are listed in theorems [ thm : without ] , [ thm : existence ] , [ thm : existence3].,title="fig : " ] 0.35 . the details are listed in theorems [ thm : without ] , [ thm : existence ] , [ thm : existence3].,title="fig : " ] the absolute convergence of hypergeometric series causes the limitation on the parameters @xmath19 . the result is that the infimum of the continuous spectrum @xmath31 must satisfy the condition @xmath32 . here @xmath33 equals @xmath3 if @xmath34 and @xmath35 otherwise . for @xmath2 arbitrarily small or zero , the restriction can be removed by analytic continuation . using the resolvent formula for a perturbed hamiltonian ( or else nontrivial self - adjoint extension of @xmath11 ) , we find that the eigenvalues solve the transcendental equation . we then derive the solutions analytically with the accuracy up to @xmath36 . in this particular case we observe a tendency that the spin - orbit term moves the eigenvalues down from the threshold @xmath37 . in general , the main conclusions following from our results that concern eigenvalues can be schematically described as shown in fig . [ fig : abc ] : for @xmath38 small , there are no eigenvalues above the threshold @xmath37 no matter the form of a nonzero coupling parameter of contact interaction ( @xmath23 ) . for @xmath2 large , there exists a subfamily of hamiltonians having eigenvalues in @xmath39 $ ] , for some lucid @xmath40 , but none of hamiltonians has eigenvalues in @xmath41 . let @xmath42 be a lower semi - bounded self - adjoint operator in a separable hilbert space @xmath43 . let @xmath44 be the scale of hilbert spaces associated to @xmath42 . the scalar product and the induced norm in @xmath45 are denoted by @xmath46 and @xmath47 , respectively . consider the orthonormal system @xmath48 and define the duality pairing @xmath49 , for @xmath50 , in a usual way ( * ? ? ? the orthogonality , which is assumed with respect to the scalar product @xmath51 in @xmath52 , implies the one with respect to @xmath53 . we write @xmath54 for the column matrix @xmath55 ; @xmath56 is the column matrix @xmath57 . let @xmath58 be a symmetric and densely defined restriction of @xmath42 to the domain @xmath59 for @xmath60 ( the resolvent set of @xmath42 ) , we define the functions @xmath61 and the column matrices @xmath62 . the functions @xmath63 are orthonormal with respect to the scalar product @xmath64 in @xmath43 , for it holds @xmath65 ( @xmath66 ) . let @xmath67 be the adjoint of @xmath58 . then @xmath68 and @xmath67 has deficiency indices ( 2,2 ) . considering @xmath69 as a banach space @xmath70 with the norm @xmath71 ( @xmath72 ; @xmath73 ; @xmath74 ) , we write @xmath75 the column matrix @xmath76 , and we regard @xmath77 as a functional @xmath78 . let @xmath79 be an extension of @xmath80 to @xmath69 @xcite : @xmath81 where @xmath82 and @xmath83 are as above , and where @xmath84 . the matrix @xmath85 is the admissible matrix for the functionals of class @xmath86 . then @xmath79 is an element of the topological dual @xmath87 equipped with the supremum norm @xmath88 ; the supremum is taken over @xmath89 and @xmath90 such that @xmath91 . using another decomposition @xmath92 , we have that , for @xmath90 , @xmath93 ( @xmath94 ; @xmath95 ) . but then , using ( [ eq : fdomh * ] ) and ( [ eq : fdomh*0 ] ) , @xmath96 ( @xmath97 is regarded as a functional @xmath78 ; the column matrix @xmath98 ) and @xmath99 using the normalization of @xmath63 , equations ( [ eq : fdomh*0 ] ) and ( [ eq : bbp ] ) , and @xmath100 one shows that @xmath101 is a boundary triple for @xmath67 . the self - adjoint extensions @xmath102 of @xmath58 in @xmath43 are in one - to - one correspondence with an hermitian matrix @xmath103 , and they are defined as restrictions of @xmath67 to the domain @xcite @xmath104 ( @xmath27 ; @xmath105 ; @xmath106 ) provided that @xmath107 . when the coupling parameter @xmath108 , we set @xmath109 . in view of ( [ eq : fdomh * ] ) , ( [ eq : hadjf ] ) , ( [ dom : ak ] ) , @xmath102 is regarded as @xmath110 so that @xmath111 is a singular rank two perturbation . the resolvent operator of @xmath102 is given by @xmath112 ( @xmath113 ) . here @xmath114 is the krein @xmath115-matrix function @xmath116 let us show that ( [ dom : ak ] ) does not apply to the case @xmath29 . let formally @xmath29 . then @xmath117 for all @xmath118 . but then , since there always exists @xmath90 such that @xmath119 , it follows from the hahn banach theorem that @xmath120 is not dense in @xmath69 , and hence in @xmath43 . this contradicts the fact that @xmath58 is densely defined and thus its all self - adjoint extensions are operators . let @xmath121 . using ( [ eq : h0 ] ) , ( [ eq : tf ] ) , and the orthogonality of @xmath63 , @xmath122 . using ( [ eq : psiext ] ) , @xmath123 , @xmath124 . [ prop : infty2 ] @xmath125 is the friedrichs extension of @xmath58 . let @xmath126 be the friedrichs extension of @xmath58 . we show that @xmath127 . let @xmath128 , with @xmath129 , be the form associated with @xmath58 : @xmath130:=\braket{h^{0}f , g}_{0}\quad(f , g\in { \mathop{{\mathrm{dom}}}}h^{0}).\ ] ] then its closure @xmath131 is defined on @xmath132 , _ i.e. _ @xmath133 , and , by definition , @xmath134 . put ( [ eq : fdomh * ] ) in @xmath135 , apply @xmath136 , and deduce that @xmath137 and @xmath138 . but @xmath122 and hence @xmath139 as required . according the theory of boundary triplets ( see _ e.g. _ ( * ? ? ? * chap . 14 ) ) , the self - adjoint extensions of @xmath58 are in one - to - one correspondence with the self - adjoint relations on @xmath26 . one can show that , for @xmath107 and @xmath140 , the extensions correspond to the relations whose range is of dimension @xmath141 and @xmath142 . when @xmath28 , the range of the corresponding relation on @xmath26 is @xmath143 , _ i.e. _ @xmath144 . when @xmath29 , the relation is @xmath145 , _ i.e. _ @xmath146 . throughout , @xmath147 ( @xmath148 ) is the sobolev space of @xmath149-functions whose distributional derivatives of order @xmath150 belong to @xmath149 . the closure of @xmath151 in @xmath147-norm is @xmath152 . the closure of the operator @xmath11 defined on the domain ( [ eq : diffop1 ] ) is a densely defined symmetric operator @xmath153 equivalently , @xmath154 where @xmath155 the operator @xmath42 in ( [ eq : freeh2 ] ) is the maximal operator associated with @xmath11 , and it is the closure of the operator @xmath156 . by gauss and green formulas ( see _ e.g. _ ( * ? ? ? * appendix d ) ) one deduces that the formal adjoint of @xmath156 coincides with itself . hence the closure of @xmath156 is the minimal operator associated with @xmath11 . now that the adjoint of the minimal ( resp . maximal ) operator coincides with the maximal ( resp . minimal ) operator , it follows that @xmath42 is self - adjoint . we call @xmath42 the _ free hamiltonian_. by using exactly the same arguments one concludes that the adjoint operator @xmath67 of @xmath58 , ( [ eq : freeh])([eq : freeh1 ] ) , is defined as follows : @xmath157 the free hamiltonian @xmath42 has lower bound @xmath158 , where @xmath33 equals @xmath3 if @xmath34 and @xmath35 otherwise . the green function for @xmath42 , or the _ free green function _ , possesses a representation @xmath159 provided that @xmath160 meets at least one of the following : 1 . [ item : a ] @xmath161 and @xmath162 2 . [ item : b ] @xmath163 ; the equality is available only if @xmath164 3 . [ item : c ] @xmath165 and @xmath166 ( @xmath167 ) . we always assume that @xmath160 satisfies at least one of ( [ item : a])([item : c ] ) unless explicitly stated otherwise . in particular , @xmath168 possesses a series representation ( [ eq : frg ] ) if @xmath169 ( put @xmath170 in ( [ item : a])([item : c ] ) ) . the functions @xmath171 ( @xmath6 ) obey a hypergeometric series representation , and they are studied in @xcite . let @xmath172 when @xmath24 , one assumes the limit @xmath173 on the right . we have that [ eq : g1x10renx10 ] @xmath174 when @xmath175 , the inverse hyperbolic tangent @xmath176 is defined entirely on @xmath177 . let @xmath178 be an @xmath179-neighborhood of the origin @xmath12 . for @xmath180 and for @xmath181 small , @xmath182 let @xmath22 be the dirac distribution concentrated at @xmath12 . define the singular distribution @xmath183 the normalization constant @xmath184 is defined as @xmath185 here we write , for simplicity , @xmath186 for @xmath18 , @xmath187{2}\sqrt{\pi}$ ] is as in ( * ? ? ? 2.3 ) , where the authors examine the laplace operator in dimension three ( recall that @xmath188 for @xmath18 ) with the interaction determined using the model of generalized perturbations . in general , the relation @xmath184 can be shown as follows . observe that [ eq : stp5 ] @xmath189 here we define , for convenience , @xmath190 { 2+\beta^{2}-2\sqrt { 1+\beta^{2 } } } \quad ( c\geq0),\ ] ] and @xmath191 is the principal value of the argument ; the range of @xmath191 is in @xmath192 $ ] . then @xmath193 seeing that @xmath194 we conclude from ( [ eq : phinorm * ] ) that the condition @xmath195 is equivalent to the condition @xmath196 but , for a fixed @xmath197 , @xmath198 { 2+\beta^{2}-2\sqrt { 1+\beta^{2 } } } < \sqrt{1+\sqrt{1+\beta^{2}}}.\ ] ] therefore , @xmath195 and this last estimate implies that we can always choose @xmath184 . next , using ( [ eq : defel0 ] ) , @xmath199 with arguments and parameters as in ( [ eq : defel0 ] ) . here @xmath200 . below we prove two equivalent relations : * @xmath63 is an orthonormal system in @xmath201 * @xmath202 is an orthonormal system in @xmath86 ( @xmath203 ; @xmath204 ) . the important conclusion is that , if the above relations hold true , then the operator @xmath58 in ( [ eq : freeh])([eq : freeh1 ] ) may be treated similar to the operator @xmath58 in ( [ eq : h0 ] ) ; subsequently , the functions @xmath205 in ( [ eq : fsigma ] ) form deficiency subspaces of the adjoint operator @xmath67 defined in ( [ eq : hadjf ] ) , ( [ eq : adjh ] ) . using ( [ eq : defel0 ] ) and @xmath206 we have that @xmath207 but @xmath208 then @xmath209 ( _ cf . _ * proof of theorem 3.1 ) ) implies that @xmath136 . next , we show that @xmath210 ( @xmath211 ; @xmath60 ; @xmath212 ; @xmath213 is the kronecker symbol ) . when @xmath214 , take the limit @xmath215 in ( [ eq : phizorth ] ) : @xmath216 ( @xmath211 ; @xmath217 ) . since @xmath218 it follows from ( [ eq : defel0 ] ) that @xmath219 then , using ( [ eq : fsigma ] ) , @xmath220 but @xmath221 and , by ( [ eq : nablajg1x0 ] ) , @xmath222 put ( [ eq : stp3 ] ) and ( [ eq : stp4 ] ) in ( [ eq : stp2 ] ) and deduce ( [ eq : phizorth ] ) . since @xmath223 densely , we conclude that @xmath202 is an orthonormal system in @xmath86 . here we apply the resolvent formula ( [ eq : resolvak ] ) to the spectral analysis of the operator @xmath102 regarded as a perturbed hamiltonian @xmath224 and where the perturbation @xmath111 is referred to as the spin - dependent contact interaction ( recall ( [ eq : hv ] ) ) . we assume that @xmath225 . when @xmath226 , the corresponding operator is @xmath125 ( proposition [ prop : infty2 ] ) . since the spectrum of @xmath125 coincides with that of @xmath42 , and the spectrum of @xmath42 is @xmath31 and it is absolutely continuous , we mainly concentrate on the spectrum of @xmath102 . we adopt the classification of the spectrum in ( * ? ? ? 9 ) , ( * ? ? ? vii.3 ) . the _ continuous spectrum _ of @xmath102 is given by @xmath31 . this follows from an invariance of the continuous component of the spectrum under singular finite rank perturbations ( * ? ? ? * theorem 4.1.4 ) and from the resolvent formula . the real and isolated singularities of the resolvent operator of @xmath102 coincide with the points @xmath227 that solve @xmath228 . therefore , the _ singular spectrum _ of @xmath102 consists of the points @xmath227 such that @xmath229 the krein @xmath115-matrix function , which is defined in ( [ eq : w3 ] ) , is a diagonal matrix @xmath230 whose entries are given by @xmath231 ( @xmath232 ) . to prove ( [ eq : kreinq ] ) , notice that @xmath233 then , using ( [ eq : defel0 ] ) and ( [ eq : w3 ] ) , @xmath234 by ( [ eq : fsigma ] ) , it follows that @xmath235.\end{aligned}\ ] ] but @xmath236 and @xmath237 which proves ( [ eq : kreinq ] ) . in particular , @xmath238 and @xmath239 . for an arbitrary @xmath240 , but for @xmath18 , @xmath241 coincides with the @xmath115-function obtained in ( * ? ? ? * sec . 2.3.1 ) . [ lem : eigenf ] the eigenspace @xmath242 consists of eigenfunctions @xmath243 where the coefficients @xmath244 are found from the boundary condition defining the domain of @xmath102 . for @xmath245 , @xmath246 . for @xmath247 , @xmath181 and @xmath248 is understood in the generalized sense , _ i.e. _ @xmath248 approaches @xmath249 strongly in @xmath43 as @xmath250 . the proof of the lemma for @xmath245 is a natural generalization of results posted in ( * ? ? ? 2.3.1 ) , so we concentrate on the case @xmath247 . for convenience , define @xmath251 . . then @xmath249 possesses representations ( [ eq : fdomh * ] ) , ( [ eq : fdomh*0 ] ) with an appropriate boundary condition . plugging ( [ eq : fdomh * ] ) and ( [ eq : hadjf ] ) with @xmath253 into the eigenvalue equation , we obtain the equation @xmath254 by hypothesis , the operator @xmath255 is unbounded in @xmath43 and it can not be applied to ( [ eq : eigen1 ] ) . we apply @xmath256 instead . define [ eq : eigen2a ] @xmath257 put ( [ eq : eigen1 ] ) in ( [ eq : eigen2a-1 ] ) : @xmath258 let @xmath259 be as in ( [ eq : eigenf ] ) . similar to ( [ eq : eigen2a-1 ] ) , define @xmath260 use @xmath249 in ( [ eq : fdomh * ] ) and plug the representation into ( [ eq : eigen4a ] ) . then apply ( [ eq : defel0 ] ) and ( [ eq : eigen2a ] ) : @xmath261 let @xmath262 be the resolution of the identity for @xmath42 , and let @xmath263 be the spectral measure . using ( [ eq : eigen4a ] ) and @xmath264 , @xmath265 or equivalently , @xmath266 by ( * ? ? ? * theorem 11.6-(iii ) ) , the borel transform @xmath267 of @xmath268 fulfills @xmath269 since @xmath270 for all @xmath247 , we conclude that @xmath271 . put this value in ( [ eq : eigen5d ] ) , and then deduce the result stated in the lemma from ( [ eq : eigen5 ] ) . [ thm : sc ] the singular continuous part is absent from the spectrum of @xmath102 . consider the singular spectrum as a closed set of points @xmath272 that solve ( [ eq : singularsp ] ) for @xmath273 , for some @xmath274 . let @xmath275 , @xmath276 , and @xmath277 be the reducing subspaces of @xmath43 corresponding to the singular , discontinuous , and singular continuous parts of @xmath102 . then @xmath278 . for each function @xmath279 , one finds functions @xmath280 and @xmath281 such that @xmath282 . then the proof is accomplished if one shows that @xmath283 for each such @xmath82 . using the resolvent formula of @xmath102 , the integral representation of the resolution of the identity @xmath284 ( * ? ? ? * lemma xviii.2.31 ) , and the residue theorem , we find that @xmath285 for arbitrary @xmath286 . here @xmath287 is the characteristic function of the singular spectrum . when deriving ( [ eq : e ] ) we used @xmath270 for the resolution of the identity @xmath262 of @xmath42 . we write @xmath288 to ensure that the right hand side of ( [ eq : e ] ) exists for @xmath289 . by lemma [ lem : eigenf ] , one safely replaces @xmath288 with @xmath290 in the spectral measure @xmath291 . for @xmath292 , @xmath293 . the characteristic function in ( [ eq : e ] ) indicates for which points @xmath227 the function @xmath294 is nonzero . these are the points that belong to the singular spectrum . since out task is to examine the functions in @xmath275 , we identify @xmath295 with an arbitrary @xmath296 . equation ( [ eq : e ] ) is valid for an arbitrary @xmath286 , and now we take @xmath279 . for each @xmath296 , there exists @xmath279 such that @xmath297 . if otherwise , since @xmath279 , there must exist a lebesgue null set @xmath298 such that @xmath299 is absent in the singular spectrum and @xmath300 . but @xmath301 for all @xmath302 , and hence @xmath303 by ( [ eq : e ] ) ; this contradicts the definition of @xmath275 . next , for each @xmath279 , there exist @xmath280 and @xmath281 such that @xmath282 . by lemma [ lem : eigenf ] , @xmath304 . therefore , since @xmath277 is the orthogonal complement of @xmath305 , one has that @xmath306 for all @xmath307 . but then , using ( [ eq : e ] ) , @xmath308 . recalling that @xmath297 and @xmath282 , it follows from the latter that @xmath309 . hence @xmath310 , and the proof of the theorem is accomplished . it follows from theorem [ thm : sc ] that the _ point spectrum _ of @xmath102 coincides with the singular spectrum . using the integral representation of the resolution of the identity and the resolvent formula , one shows that , for @xmath299 in the set of borel subsets of @xmath311 , and for @xmath286 , the spectral measures of @xmath102 and @xmath42 agree up to sets of lebesgue measure zero . since the absolutely continuous part of a measure is supported by the set of @xmath227 such that @xmath312 ( see _ e.g. _ @xcite ) , we conclude that the _ absolutely continuous _ parts of spectral measures of @xmath102 and @xmath42 coincide . if we let @xmath313 and @xmath314 be the reducing subspaces of @xmath43 corresponding to the continuous and absolutely continuous parts of @xmath102 , then we arrive at the same conclusion by noting that @xmath315 and @xmath316 , hence @xmath317 . according to the results of the previous paragraph , the notions of singular spectrum and point spectrum ( eigenvalues ) are used interchangeably . as a rule , we prefer the notion of singular spectrum to that of eigenvalues , because the singular spectrum consists of the singular points @xmath318 of the resolvent operator of @xmath102 and because the singular points are solutions of equation ( [ eq : singularsp ] ) , which is our main target in the present section . define @xmath319 the matrix @xmath320 is hermitian . let also @xmath321 when @xmath322 , one takes the limit @xmath323 on the right hand side : @xmath324 . using ( [ eq : gammaomega ] ) and ( [ eq : xif ] ) , ( [ eq : singularsp ] ) obeys the following form : @xmath325 the upper ( resp . lower ) sign is taken if @xmath326 ( resp . @xmath327 ) . owing to @xmath176 , equation ( [ eq : sing ] ) is transcendental and does not possess analytic solutions . for the reason that we are unable to solve equation ( [ eq : sing ] ) as it stands , we examine ( [ eq : sing ] ) in three specific cases : 1 . [ item : a ] case without spin - orbit coupling : @xmath15 , @xmath328 . [ item : b ] case with small spin - orbit coupling : @xmath329 small , @xmath330 . [ item : c ] case with large spin - orbit coupling : @xmath331 , @xmath332 , @xmath247 . although the @xmath115-function itself is valid for all @xmath19 , the limitation on the parameters is caused by the series representation for the free green function ( [ eq : frg ] ) . in fact , the validity of ( [ eq : phizorth ] ) is governed by ( [ item : a])([item : c ] ) . in particular , putting @xmath333 in ( [ eq : phizorth ] ) , the condition @xmath169 follows from ( [ item : a])([item : c ] ) ; this latter condition also ensures the existence of the normalization constant @xmath184 . for @xmath2 zero or arbitrarily small ( cases ( [ item : a ] ) or ( [ item : b ] ) , respectively ) , conditions ( [ item : a])([item : c ] ) can be safely removed without affecting the absolute convergence of the series that defines the free green function . this follows from the fact that for @xmath38 small , the confluent horn function , which defines the functions @xmath334 in ( [ eq : frg ] ) , can be analytically continued to the region @xmath335 ; see @xcite for more details . as a result one writes @xmath336 in ( [ item : a ] ) and @xmath337 in ( [ item : b ] ) . for @xmath247 in ( [ item : c ] ) , the eigenfunctions in lemma [ lem : eigenf ] are restricted to ( [ item : a ] ) or ( [ item : c ] ) . using the asymptotic formula @xmath338 as @xmath339 , and picking the first term from the right hand side of ( [ eq : sr ] ) , we rewrite ( [ eq : sing ] ) as follows : @xmath340 it takes little effort to conclude from ( [ eq : xif ] ) and ( [ eq : singa ] ) the following result . [ thm : without ] let @xmath15 and @xmath341 . the part of the singular spectrum of @xmath102 below the threshold @xmath37 is given by the set @xmath342 . for @xmath343 , the singular spectrum embedded in the continuous spectrum is described by the following singletons : @xmath344 iff @xmath345 ; @xmath346 iff @xmath347 and @xmath348 ; @xmath349 iff @xmath350 . for @xmath322 , there are no embedded eigenvalues . it follows that @xmath102 may have maximum three eigenvalues in the interval @xmath351 $ ] . with regard to @xcite , assume that @xmath352 . if @xmath347 , then theorem 2 in @xcite says that the point spectrum is empty for @xmath353 , while it follows from theorem [ thm : without ] that the point spectrum consists of the points @xmath354 when @xmath355 . if @xmath356 and @xmath357 , then theorem 3 in @xcite says that _ there exists _ such that , for @xmath359 , @xmath102 does not have embedded eigenvalues . by theorem [ thm : without ] , however , there are no embedded eigenvalues _ for all _ @xmath357 ; recall that @xmath360 and @xmath361 together suit @xmath42 . here we study the behavior of eigenvalues of @xmath362 in the limit @xmath339 . the main conclusion following from the results posted below is that , for arbitrarily small but nonzero @xmath2 , the singular spectrum is empty above the threshold @xmath37 . our strategy is to expand @xmath363 and @xmath364 in ( [ eq : gammaomega ] ) as a series with respect to @xmath339 , and then , using the asymptotic formula ( [ eq : sr ] ) , to solve ( [ eq : sing ] ) . using ( [ eq : phinorm * ] ) , we have that @xmath365 as @xmath339 , where [ eq : nseries01 ] @xmath366{2}\sqrt{\pi } \sqrt[4]{\sqrt{1+\beta^{2}}+s\beta } , \\ \mathcal{n}_{s}^{(1)}:= & \frac{\sqrt{\pi}}{6\sqrt[4]{2}\beta^{2 } } \left(3\beta+s\left(1-\sqrt{1+\beta^{2}}\right)\right ) \sqrt[4]{2+\beta^{2}-2\sqrt{1+\beta^{2 } } } \left(\sqrt{1+\beta^{2}}+s\beta\right)^{\frac{3}{4}}.\end{aligned}\ ] ] using ( [ eq : g1x10renx10 ] ) and ( [ eq : gammaomega ] ) , @xmath367 as @xmath339 , where [ eq : lambdaseries01 ] @xmath368{2+\beta^{2}-2\sqrt{1+\beta^{2}}}.\end{aligned}\ ] ] if we let @xmath369 ( @xmath370 ; @xmath371 ; @xmath211 ; @xmath372 ) , then @xmath373 as @xmath339 ( @xmath211 ) . it is clear that @xmath374 , @xmath375 coincide with @xmath363 , @xmath364 in theorem [ thm : without ] . put ( [ eq : omegaseries ] ) , ( [ eq : omegaseriesb ] ) , and ( [ eq : sr ] ) in equation ( [ eq : sing ] ) and get that @xmath376 as @xmath339 . here @xmath377 when @xmath15 , ( [ eq : singb ] ) reduces to ( [ eq : singa ] ) . we look for the solutions of the form @xmath378 for some small @xmath379 ; @xmath380 is in the singular spectrum of @xmath381 , which is described in theorem [ thm : without ] . it follows that @xmath382 . let us pick @xmath383 such that @xmath384 and @xmath385 , where @xmath386 [ lem : existence ] let @xmath329 be arbitrarily small and @xmath387 . the following holds : @xmath388 iff @xmath389 ; @xmath390 @xmath391 ; @xmath392 @xmath393 . by ( [ eq : xif ] ) , @xmath394 . put @xmath395 in ( [ eq : singb ] ) and get a zero - valued quadratic polynomial in @xmath38 . the monomial of degree @xmath396 is precisely the condition in theorem [ thm : without ] ensuring that @xmath384 . the monomial of degree @xmath141 is absent , and the monomial of degree @xmath142 reads @xmath397 ( @xmath89 ) . using ( [ eq : gammaomega0 ] ) , @xmath398 let @xmath399 , @xmath400 . it follows from ( [ eq : deg1 ] ) and ( [ eq : sssss2 ] ) that @xmath401 since @xmath402 and @xmath403 , it follows from ( [ eq : sssss4 ] ) that @xmath404 , which is equivalent to @xmath385 . this proves the first statement of the lemma . let @xmath405 , @xmath181 small . by ( [ eq : xif ] ) and ( [ eq : qsigma ] ) , @xmath406 ( @xmath89 ) . put these expressions in ( [ eq : singb ] ) and get that @xmath407 then , by the second equality in ( [ eq : deg1 ] ) , @xmath408 . using the latter and ( [ eq : sssss2 ] ) , @xmath409 which is false , since by ( [ eq : nseries01 ] ) , ( [ eq : lambdaseries01 ] ) , ( [ eq : gammaomega0 ] ) , the function on the left has a single maximum @xmath410 at @xmath411 . when @xmath412 , ( [ eq : singb ] ) also leads to ( [ eq : cnd ] ) , hence @xmath390 for small nonzero @xmath179 . next @xmath413 implies that @xmath414 ( theorem [ thm : without ] ) . when @xmath415 , we have that @xmath416 and @xmath417 ( @xmath89 ) . put these expressions in ( [ eq : singb ] ) and deduce that ( [ eq : singb ] ) fails for @xmath181 . when @xmath246 , we obtain the system @xmath418 and @xmath419 . the system does not have solutions @xmath420 , @xmath421 in @xmath311 , hence @xmath392 for @xmath415 small . applying the above procedure for @xmath412 we arrive at the same conclusion . as previously , let @xmath422 . by lemma [ lem : existence ] , @xmath423 unless @xmath15 . we also conclude from theorem [ thm : without ] and lemma [ lem : existence ] that @xmath326 . next we look for @xmath424 of the form @xmath425 for some @xmath426 ( @xmath427 ) . put this @xmath295 in ( [ eq : xif ] ) : @xmath428 as @xmath339 , where @xmath429 and [ eq : xiseries0123 ] @xmath430 \nonumber \\ & -e^{(0)\,2 } [ 8e^{(0)}e^{(1)}e^{(2 ) } ( 1+\xi_{0}^{2}e^{(0)})- 8e^{(0)\,2}e^{(3 ) } ( 1 + 2\xi_{0}^{2}e^{(0 ) } ) \nonumber \\ & -e^{(1)\,3}(3 + 2\xi_{0}^{2}e^{(0 ) } ) ] + \beta^{2}[8e^{(0)}e^{(1)}e^{(2 ) } -16e^{(0)\,2}e^{(3)}(1 + 2\xi_{0}^{2}e^{(0 ) } ) \nonumber \\ & + e^{(1)\,3}(1 + 6\xi_{0}^{2}e^{(0)})]\bigr\}. \label{eq : xiseries0123 - 3}\end{aligned}\ ] ] notice that each denominator in ( [ eq : xiseries0123 ] ) is nonzero since @xmath423 . put ( [ eq : xiseries ] ) in ( [ eq : qsigma ] ) and get that @xmath431 as @xmath339 , where @xmath432 and [ eq : qseries0123 ] @xmath433 put ( [ eq : xiseries ] ) and ( [ eq : qseries ] ) in ( [ eq : singb ] ) , and then collect the terms with the same powers @xmath434 of @xmath2 . since a cubic polynomial in @xmath38 is zero , set each monomial to zero : a zero - valued monomial of degree @xmath396 coincides with ( [ eq : singa ] ) , and the rest zero - valued monomials of degree @xmath435 are of the form [ eq : systemb ] @xmath436 equations ( [ eq : systemb-1 ] ) and ( [ eq : systemb-2 ] ) imply that [ prop : lambda1 ] @xmath437 . let @xmath438 . then ( [ eq : systemb-1 ] ) simplifies thus : @xmath439 since @xmath440 ( @xmath89 ) , it follows that @xmath441 for both @xmath89 . but then , either @xmath442 or @xmath443 . if the former , then , by ( [ eq : xiseries0123 - 1 ] ) , @xmath437 . if the latter , then @xmath444 under these conditions , ( [ eq : systemb-2 ] ) fails unless @xmath442 . therefore , @xmath437 in either case . if @xmath445 , then @xmath446 . but then ( [ eq : systemb-1 ] ) implies either @xmath442 or @xmath447 and @xmath322 . since @xmath343 , @xmath437 necessarily . finally , when @xmath448 , ( [ eq : systemb-1 ] ) also implies that @xmath442 necessarily . by ( [ eq : qseries0123 - 1 ] ) and proposition [ prop : lambda1 ] , @xmath449 , and the system ( [ eq : systemb ] ) reduces to ( [ eq : systemb-2])([eq : systemb-3 ] ) . it follows that @xmath450 and @xmath451 are uniquely defined by the latter system unless ( [ eq : cnd2 ] ) holds . otherwise we need more terms of expansion . thus , @xmath36 in ( [ eq : sr ] ) further reads @xmath452 by definition , the normalization constant @xmath453 depends on even powers of @xmath2 . then @xmath36 in ( [ eq : omegaseries ] ) is given by @xmath454 although @xmath455 and @xmath456 can be found similar to @xmath457 in ( [ eq : gammaomega0 ] ) , their explicit representation is inessential since our final goal is to express @xmath295 with the accuracy up to @xmath36 . next , if we supplement @xmath424 with terms indexed by @xmath458 , then the values of functions @xmath459 and @xmath460 in ( [ eq : xiseries ] ) and ( [ eq : qseries ] ) , respectively , are also supplemented with terms indexed by @xmath458 . put these values along with ( [ eq : sr2 ] ) and ( [ eq : omegaseries2 ] ) in ( [ eq : sr ] ) , ( [ eq : omegaseries ] ) , and then in ( [ eq : sing ] ) . then collect the terms with the same powers @xmath461 of @xmath2 , and deduce a zero - valued polynomial in @xmath38 of degree @xmath462 . set each monomial to zero , apply proposition [ prop : lambda1 ] , the relation @xmath449 , and condition ( [ eq : cnd2 ] ) , and get that the zero - valued monomials of degree @xmath463 are of the form : [ eq : systemb2 ] @xmath464 [ prop : lambda3 ] @xmath465 . by proposition [ prop : lambda1 ] and ( [ eq : xiseries0123 - 3 ] ) , @xmath466 therefore , @xmath467 . since @xmath449 and @xmath442 , we see that ( [ eq : systemb-3 ] ) coincides with ( [ eq : systemb-1 ] ) but with @xmath468 replaced by @xmath451 . if ( [ eq : cnd2 ] ) does not hold , then we can redo all the steps made up during the proof of proposition [ prop : lambda1 ] to conclude that @xmath469 . hence @xmath465 , as claimed . if ( [ eq : cnd2 ] ) holds , then by ( [ eq : systemb2 - 2 ] ) , either @xmath469 or @xmath470 substitute ( [ eq : cnd3 ] ) in ( [ eq : systemb2 - 1 ] ) and get that @xmath471 takes the value @xmath472 put this value in ( [ eq : cnd3 ] ) and get that @xmath473 . now put this @xmath450 and ( [ eq : ppppp ] ) in ( [ eq : systemb2 - 3 ] ) and deduce that either @xmath469 or @xmath474 . if the latter , then @xmath475 , since the inequality in ( [ eq : ppppp ] ) fails for @xmath476 . if @xmath475 , then @xmath477 . but @xmath478 and therefore @xmath447 , which is true iff @xmath322 , hence false . [ prop : lambda2 ] for @xmath438 , @xmath479 further , if @xmath480 , @xmath481 , and @xmath482 , then @xmath483 is twofold , and it takes the values @xmath484 @xmath485 . it can be shown that , for @xmath486 diagonal and @xmath438 , the following holds : @xmath487 for @xmath488 , @xmath478 ; @xmath489 for @xmath490 , @xmath481 . by proposition [ prop : lambda2 ] , this means the spin - orbit term moves the eigenvalues down from the threshold @xmath37 . using proposition [ prop : lambda1 ] , one finds from ( [ eq : xiseries0123 - 2 ] ) that @xmath491 when ( [ eq : cnd2 ] ) does not hold , @xmath483 is found from ( [ eq : systemb-2 ] ) and ( [ eq : aaaaaa-1 ] ) . if ( [ eq : cnd2 ] ) holds , then @xmath483 follows from ( [ eq : systemb2 - 1 ] ) and ( [ eq : aaaaaa-1 ] ) . [ thm : existence ] let @xmath329 be arbitrarily small and @xmath387 . then the part of the singular spectrum of @xmath492 that is below the threshold @xmath37 consists of the points @xmath493 where @xmath438 belongs to the singular spectrum of @xmath381 . let @xmath494 be in the singular spectrum of @xmath492 . then @xmath395 and @xmath389 necessarily ; otherwise @xmath295 does not belong to the singular spectrum . by propositions [ prop : lambda1 ] , [ prop : lambda3 ] , [ prop : lambda2 ] , the singular spectrum of @xmath492 consists of the points @xmath495 up to @xmath36 . to prove the first part of the theorem , we need only show @xmath496 . let @xmath497 , @xmath498 small . then @xmath405 where @xmath499 . by proposition [ prop : lambda2 ] , @xmath500 for some real @xmath501 whose explicit representation is inessential here . but then @xmath502 , as claimed . to prove the second part , we need to show @xmath503 , as it follows from lemma [ lem : existence ] . by theorem [ thm : without ] and ( [ eq : systemb-2 ] ) , @xmath504 and we show that equation ( [ eq : empty ] ) does not have solutions with respect to @xmath505 . let @xmath506 . the real and imaginary parts of ( [ eq : empty ] ) imply that [ eq : l2 ] @xmath507 it follows that ( [ eq : l2-a ] ) @xmath508 ( [ eq : l2-b ] ) iff @xmath509 , which contradicts the initial hypothesis @xmath510 ( theorem [ thm : without ] ) . if @xmath511 , then @xmath512 and @xmath513 . put this @xmath514 along with @xmath515 in ( [ eq : sssss2 ] ) and get ( [ eq : cnd0 ] ) with @xmath516 replaced by @xmath517 . in this case the left - hand side approaches @xmath518 as @xmath519 , hence false . when @xmath448 , @xmath483 satisfies ( [ eq : l2 ] ) , and we conclude that @xmath520 for all @xmath521 . here we wish to post some most important properties of solutions @xmath247 of ( [ eq : sing ] ) under hypothesis in ( [ item : c ] ) . for this we find it convenient to define the functions @xmath522\to{\mathbb{r}}$ ] as @xmath523 the parameter @xmath524 is defined as @xmath525 , and it takes the values @xmath526 ( solid curves ) and @xmath527 ( dashed curves ) at @xmath528 . the curve labeled by 1 shows the values of @xmath529 ; the ( solid and dashed ) curves labeled by 2 show the values of @xmath530 and @xmath531 ; the ( solid and dashed ) curves labeled by 3 show the values of @xmath532 and @xmath533 ; the ( solid and dashed ) curves labeled by 4 show the values of @xmath534 and @xmath535 , _ i.e. _ @xmath536 and @xmath527 as @xmath537 . the function @xmath534 takes the value @xmath538 everywhere except at @xmath539 , which is also a zero of @xmath535 . for @xmath540 , a zero of @xmath527 coincides with the first zero , @xmath541 , of @xmath536 , and @xmath536 also has the second zero , @xmath542 , if @xmath543 . in figure , @xmath544 and @xmath545.,scaledwidth=100.0% ] the parameter @xmath524 is predetermined by the relations @xmath546 and @xmath169 , and by the definition of @xmath33 . some typical plots of the functions @xmath536 and @xmath527 are illustrated in fig . [ fig : fig ] . in particular , if @xmath343 is arbitrarily small , then @xmath524 varies from @xmath141 to @xmath547 depending on @xmath548 . therefore , the curves in fig . [ fig : fig ] labeled by 4 represent the case when the magnetic field @xmath3 approaches zero from above and the spin - orbit - coupling strength @xmath2 is large enough . the function @xmath536 has at most two zeros , which we denote by @xmath541 and @xmath542 . we assume that @xmath549 is the zero of @xmath527 . then the second zero @xmath550 solves the equation @xmath551 ( @xmath552 ) provided that @xmath543 . the relation @xmath553 is meaningless because the domain of @xmath536 is @xmath522 $ ] , by definition . if @xmath554 , then the function @xmath555 has only one zero @xmath556 . if @xmath537 , we write @xmath557 , and @xmath558 formally behaves like an ill - defined dirac delta function concentrated at a zero @xmath539 of @xmath535 . in this latter case , the two zeros @xmath559 and @xmath560 coincide . in all other cases , _ i.e. _ for @xmath561 , the function @xmath536 has either two zeros , if @xmath543 , or a single zero , if otherwise . in general , @xmath562 , and the equality holds only if @xmath563 . yet another quantity which we use to describe our main results is a function @xmath564\to [ \beta,\infty)$ ] defined as @xmath565 the relations @xmath566 imply that @xmath567 . in particular , we have that @xmath568 . in the limit @xmath323 , the function @xmath569 ranges from @xmath396 to @xmath570 when @xmath524 ranges from @xmath141 to @xmath547 . choosing @xmath571 ( _ i.e. _ @xmath572 , @xmath323 ) one deduces @xmath573 . [ thm : existence3 ] let @xmath574 , @xmath575 . the singular spectrum of @xmath102 , which is embedded in the continuous spectrum @xmath31 , is given by the set @xmath576{\colon } & x\in[x_{\nu,1},\nu],\ , \gamma=\omega_{+}\omega_{-}+(\beta/2)v_{\nu}(x ) , \\ & 2\omega_{-}=x^{2}u_{\nu}(x)\sum_{s}\omega _ { s } \left(\nu^{2}+ s\right)\bigr\}.\end{aligned}\ ] ] for example , if @xmath3 is arbitrarily small and @xmath2 large , say @xmath572 , then @xmath524 approaches @xmath577 from below , and theorem [ thm : existence3 ] tells us that there is an eigenvalue , @xmath578(@xmath579 ) , above the threshold @xmath580 iff @xmath581 . this particular eigenvalue fails to satisfy ( [ item : b ] ) and ( [ item : c ] ) , so the corresponding eigenfunction in lemma [ lem : eigenf ] has to be analytically continued applying the methods similar to those used in @xcite . since @xmath582 , @xmath583 does not solve ( [ eq : sing ] ) . if @xmath584 , then @xmath585 , @xmath586 , and @xmath587 is of the form @xmath588 , where @xmath589 is the real part . put this form in ( [ eq : sing ] ) and deduce the system of equations @xmath590 where @xmath591 now that @xmath592 and @xmath593 , it follows that @xmath594 , in contradiction to @xmath595 . next , suppose that @xmath596 . in this case @xmath597 where the real part @xmath598 while the imaginary part @xmath599 for @xmath600 , and @xmath601 for @xmath602 . if we let @xmath603 then @xmath604 for @xmath600 , and @xmath605 for @xmath602 . put @xmath597 and @xmath606 in ( [ eq : sing ] ) and deduce the system ( [ eq : subsystem ] ) where now @xmath607 we see that @xmath592 and @xmath608 for @xmath600 , and @xmath593 for @xmath602 . but then ( [ eq : subsystem ] ) leads to the previous @xmath594 . finally , if @xmath327 , then @xmath609 , where @xmath610 reads @xmath611 then @xmath612 , and ( [ eq : sing ] ) reduces to ( [ eq : subsystem ] ) where @xmath613 and @xmath614 . using the definition of @xmath615 , we write @xmath616 where @xmath617 ( @xmath618 ) . when written in terms of @xmath524 , @xmath619 , @xmath527 , @xmath536 , the system ( [ eq : subsystem ] ) coincides with that defining the singular spectrum in the theorem . it remains to show that @xmath620 , _ i.e. _ @xmath621 . let @xmath554 . since @xmath555 , it follows from ( [ eq : subsystem ] ) that @xmath622 , hence @xmath623 . but then @xmath624 $ ] and @xmath625 $ ] . let @xmath552 and @xmath626 . in this case @xmath627 and @xmath628 . on the other hand , the system ( [ eq : subsystem ] ) implies that @xmath629 , @xmath630 , and thus @xmath631 . we conclude that @xmath632 $ ] . in the current paper , the main results concerning the eigenvalues ( or else the singular points ) are summarized in theorems [ thm : without ] , [ thm : existence ] , [ thm : existence3 ] . other parts of the spectrum , including the eigenfunctions , are discussed in sec . [ sec : sp ] and particularly in lemma [ lem : eigenf ] and theorem [ thm : sc ] . although we said nothing about the eigenvalues below the threshold @xmath158 when @xmath546 , their existence can be seen from a general equation ( [ eq : sing ] ) . for example , taking @xmath27 diagonal with the entries such that @xmath352 , one finds that @xmath633 0.48 of the perturbed hamiltonian ( [ eq : hv ] ) as a function of impurity scattering strength @xmath634 when the admissible matrix @xmath635 with @xmath636 . the spin - orbit - coupling strength @xmath572 . when @xmath637 , the perturbed hamiltonian coincides with the free hamiltonian whose spectrum @xmath638 is absolutely continuous . when @xmath639 , the perturbed hamiltonian is described by the friedrichs extension whose spectrum is that of the free hamiltonian . in ( a ) the threshold approaches @xmath580 and the dashed line shows the eigenvalue @xmath640 corresponding to the coupling parameter @xmath501 whose absolute value is arbitrarily large but finite . in ( b ) the threshold is @xmath641 , the black ( upper ) solid curves represent the eigenvalue for @xmath642 in ( [ eq : sing ] ) , the blue ( lower ) solid curves represent the eigenvalue for @xmath643 , and the black dashed ( upper ) and blue dashed ( lower ) lines show the eigenvalues @xmath644 ( for @xmath642 ) and @xmath645 ( for @xmath643 ) corresponding to @xmath501 whose absolute value is arbitrarily large but finite.,title="fig : " ] 0.48 of the perturbed hamiltonian ( [ eq : hv ] ) as a function of impurity scattering strength @xmath634 when the admissible matrix @xmath635 with @xmath636 . the spin - orbit - coupling strength @xmath572 . when @xmath637 , the perturbed hamiltonian coincides with the free hamiltonian whose spectrum @xmath638 is absolutely continuous . when @xmath639 , the perturbed hamiltonian is described by the friedrichs extension whose spectrum is that of the free hamiltonian . in ( a ) the threshold approaches @xmath580 and the dashed line shows the eigenvalue @xmath640 corresponding to the coupling parameter @xmath501 whose absolute value is arbitrarily large but finite . in ( b ) the threshold is @xmath641 , the black ( upper ) solid curves represent the eigenvalue for @xmath642 in ( [ eq : sing ] ) , the blue ( lower ) solid curves represent the eigenvalue for @xmath643 , and the black dashed ( upper ) and blue dashed ( lower ) lines show the eigenvalues @xmath644 ( for @xmath642 ) and @xmath645 ( for @xmath643 ) corresponding to @xmath501 whose absolute value is arbitrarily large but finite.,title="fig : " ] provided that the magnetic field @xmath3 is arbitrarily close to @xmath396 . in this regime , the eigenvalue @xmath245 exists if @xmath486 is of the form @xmath646 for some real @xmath647 , where @xmath648 is the identity matrix ; this follows from the conditions @xmath649 and @xmath650 and the definition in ( [ eq : gammaomega ] ) , which in turn gives a one - to - one correspondence between @xmath651 and @xmath647 . note that , when @xmath15 , ( [ eq : last ] ) gives the solution @xmath652 , @xmath653 , which is ( by no accident ) in exact agreement with that described in theorem [ thm : without ] for @xmath322 . also , @xmath654 , because @xmath655 may correspond only to the friedrichs extension ( proposition [ prop : infty2 ] ) whose spectrum is only absolutely continuous . however , if one takes the coupling parameter of contact interaction of the form @xmath634 with @xmath501 nonzero and real , then necessarily @xmath635 with @xmath656 real . it then follows that @xmath657 for @xmath658 arbitrarily large but finite . for a given @xmath656 , there exists @xmath651 corresponding to @xmath657 , so ( [ eq : last ] ) shows that the perturbed hamiltonian with arbitrarily large but finite impurity scattering strength @xmath658 has an eigenvalue below the threshold . for example , when @xmath572 and @xmath355 , the parameter @xmath659 approximates @xmath656 , and the eigenvalue @xmath640 ( fig . [ fig : fig2](a ) ) . when @xmath355 and @xmath650 , there also exists an eigenvalue @xmath660 above the threshold ( theorem [ thm : existence3 ] ) , but the same does not apply to the case when @xmath322 . this explains why we chose @xmath3 arbitrarily small but nonzero . however , the above considerations remain valid for @xmath322 as well , with the exception that now the hypothesis of theorem [ thm : existence3 ] fails . for comparison reasons , in fig . [ fig : fig2](b ) we plot the eigenvalues corresponding to @xmath89 in ( [ eq : sing ] ) as functions of @xmath501 when the magnetic field is nonzero . the present discussion is easy to extend to a general case when @xmath661 and @xmath662 . moreover , the considerations apply to the case when @xmath663 . one can show similar plots either solving equation ( [ eq : sing ] ) numerically or applying analytic expressions in theorem [ thm : existence ] for @xmath2 sufficiently small . for the mean field analysis of eigenvalues at large @xmath658 , but in dimensions one and two , the reader may also refer to @xcite , where the case when @xmath663 is studied . we recall that in our model , when @xmath664 , @xmath142 is the maximum possible value of @xmath2 , as well as @xmath141 is the maximum possible value of @xmath3 . this is due to the restriction @xmath665 caused by the absolute convergence of the hypergeometric series which defines the green function for the free hamiltonian . however , an additional procedure of analytic continuation , similar to that when @xmath663 is small , may probably lead to larger @xmath2 as well . when applied to concrete schemes that propose the creation of spin - orbit coupling in cold atomic gases , the parameters @xmath2 and @xmath3 are expressed in terms of the quantities which are measurable experimentally either directly or indirectly . for example , using @xmath666 and @xmath667 ( @xmath668 is the mass of the particle ) , the spin - orbit coupling for the rashba hamiltonian constructed in @xcite is characterized by @xmath669 , where @xmath670 ( @xmath671m@xmath672 in si units ) points to the strength of the magnetic field gradient . without the additional procedure of continuation our model can therefore be applied straightforwardly up to @xmath673 . likewise , in the experimental setup proposed in @xcite , the spin - orbit coupling is given by @xmath674 , where @xmath675 is the two - photon recoil energy due to raman process . hence , in the regime @xmath664 , one would expect @xmath676 . in @xcite one can find some realistic parameters suggested for observing bound states when using the scheme in @xcite with @xmath677 and @xmath678 . while the point spectrum of the rashba hamiltonian with spin - dependent delta - like impurity scattering depends on the admissible matrix @xmath25 , the main conclusion , which is independent of @xmath25 , is that : 1 ) the hamiltonian does not have eigenvalues above the threshold provided that the spin - orbit - coupling strength is small enough ( @xmath663 ) but nonzero . 2 ) the hamiltonian does not have eigenvalues located in - between the threshold and the minimum of the upper branch of dispersion provided that the spin - orbit - coupling strength is large enough ( @xmath664 ) . 3 ) the maximum possible eigenvalue that the hamiltonian can reach is @xmath679 , and @xmath680 depends only on @xmath664 and @xmath343 . the author acknowledges the anonymous referees whose remarks have helped in improving the presentation . the author also thanks the referee of the previous versions of the paper . [ 2 ] http://www.ams.org/mathscinet-getitem?mr=#1[#2 ] [ 2]#2 claudio cacciapuoti , raffaelle carlone , and rodolfo figari , _ perturbations of eigenvalues embedded at threshold : i. one- and three - dimensional solvable models _ , j. phys . a : math . theor . * 43 * ( 2010 ) , 474009 . lawrence w. cheuk , ariel t. sommer , zoran hadzibabic , tarik yefsah , waseem s. bakr , and martin w. zwierlein , _ spin - injection spectroscopy of a spin - orbit coupled fermi gas _ , phys . * 109 * ( 2012 ) , 095302 .

the two - dimensional ( 2d ) ferromagnetic xy model has had a long history of studies concerning the dynamics of spinwaves at low temperatures . it was first analyzed by villain@xcite and by moussa and villain@xcite , for low temperatures , in the harmonic approximation . they found that the dynamic structure function @xmath1 has a spinwave peak of the form @xmath3 where @xmath4 is the critical exponent describing the decay of the static spin - spin correlation function , and @xmath5 is the spinwave frequency at wavevector @xmath6 . nelson and fisher@xcite treated the model in a fixed - length hydrodynamic description , obtaining the following expression around the spinwave peak , @xmath7 where for small @xmath8 , @xmath9 , with @xmath10 being the spinwave velocity . these theoretical approaches led to spinwave peaks diverging unphysically at the spinwave frequency . here we describe an alternative theoretical description in which the divergence does not occur . at higher temperatures , the spinwave or harmonic approximation begins to fail , and xy magnets are known to undergo a topological phase transition@xcite associated with the unbinding of vortices . we generally consider temperatures below this unbinding temperature @xmath11 , where a spinwave approximation will be valid . at very low temperatures any nonlinear effects should be very weak , do not lead to vortex production , and can be included as the damping or scattering of the spinwave modes , that results in a spinwave lifetime . thus our approach will be to start from the harmonic approximation to the spin hamiltonian , @xmath12 , and then include the damping that results from the terms fourth order in spin operators ( @xmath13 ) in an expansion of the original full spin hamiltonian , @xmath14 . the resulting damping strength @xmath0 will imply a finite lifetime @xmath15 for the spinwave modes . we are interested to investigate whether the presence of such lifetime effects can eliminate the divergence of the dynamic structure function @xmath1 at the spinwave frequency . we calculate the spinwave damping by a feynmann diagram technique for finite temperatures , applied to the fourth - order hamiltonian @xmath13 , using the leading diagrams to second order in @xmath13 . the only diagrams we include are those that give a contribution to the _ imaginary part _ of the spinwave energy , i.e. , to the damping or inverse lifetime . these are the terms that involve a _ dynamical _ effect on the spinwave energies . the associated integral expressions for these diagrams require the evaluation of four - dimensional integrals involving a delta - function . methods for evaluating this type of integral in three dimensions have been developed by gilat and co - workers ; here we show how to extend their work and apply the same technique in four dimensions . in principle , we could also use the feynmann technique to determine the leading changes in the _ real part _ of the spinwave energy , @xmath16 , due to the perturbation @xmath13 . this real part would result as the usual effect due to finite temperature : a shift in the spinwave frequencies to lower values as the temperature increases . it is essentially a _ static _ effect , primarily due to the renormalization of the nearest neighbor exchange interaction strength @xmath17 with temperature . calculation of these spinwave frequency shifts by the feynmann technique would require including diagrams first order in @xmath13 ; the so - called single - loop diagrams . in this paper , however , we are primarily concerned with accounting for dynamical effects . we will include the renormalization of the spinwave energies using the memory function formalism , combined with some input information from classical monte carlo ( mc ) data . in the memory function approach , the position of the spinwave peak in @xmath1 does not occur exactly at the zero - temperature dispersion relation , @xmath5 , but rather , close to the value @xmath18 , where @xmath19 is the temperature - dependent second moment . @xmath19 can be estimated from static quantities , either using low - temperature analytic expressions or results from monte carlo , the latter being necessary for temperatures approaching @xmath11 . the memory function approach already includes to leading order the exchange strength renormalization , as can be seen in the shift in the position of the spinwave peaks . thus we use it rather than making any renormalization of the exchange @xmath17 . a few comments can be made about how the damping @xmath0 is included into the memory function calculation . note that the damping calculation sketched above is completely independent of the memory function calculation ; the latter is a formalism for calculating dynamic correlations for a given hamiltonian and operators . due to the xy symmetry of the model , the natural coordinates of the spins are the in - plane angle @xmath20 and out - of - plane spin component @xmath21 . fourier transforming and using a standard canonical transformation , the hamiltonian is written in terms of creation and annihilation operators @xmath22 and @xmath23 , with time dependencies varying as @xmath24 and @xmath25 , respectively . once damping is included , the time dependencies become @xmath26 and @xmath27 , respectively . henceforth the calculation proceeds as usual in the memory function approach , except for this small modification of the operator time dependencies via the shifted operator frequencies , @xmath28 . in the memory function calculation , spinwave sum and difference processes enter . these are characterized by combinations of frequencies , @xmath29 , where @xmath6 is the wavevector at which the dynamic correlations are measured , and @xmath30 is summed over to include all possible such processes . based on the above modifications of the operator frequencies , there also results the modification for these processes , @xmath31 . the damping rates for different wavevectors always add for both types of processes . the paper is organized as follows : first , in sec . [ model ] , we present the model and develop the expansion of the hamiltonian to quartic order in creation / annihilation operators . in sec . [ diagrams ] , the expressions for the damping via the feynmann technique are presented . then we show how these are evaluated using the gilat procedure . in sec . [ memory ] , the key aspects of the memory function calculation of @xmath1 are reviewed , together with a focus on the modification due to the damping . some details of the monte carlo and spin dynamics simulations are summarized in sec . [ mc+sd ] . this is followed in sec . [ results ] by presentation of the results both for very large and relatively smaller systems , in order to discuss the dynamical finite size effects . the memory function results also are compared with standard monte carlo spin - dynamics calculations of the dynamic correlation functions in order to evaluate the validity and accuracy of the method . our conclusions are presented in sec . [ conclusions ] . the ferromagnetic spin hamiltonian under consideration is @xmath32 where the first sum is over all sites @xmath33 of a 2d square lattice , the second is over all nearest neighbor displacements @xmath34 , @xmath17 is the nearest neighbor coupling strength , and each @xmath35 is a three - component classical spin - vector . because each bond is counted twice , a factor of @xmath36 is included on @xmath17 . we primarily concentrate on the xy model , with anisotropy parameter @xmath37 , however , for completeness , formulas covering the easy - plane anisotropy range , @xmath38 , are presented . in order to develop the damping as a perturbation of the harmonic approximation , we need to expand the full spin hamiltonian , including the harmonic ( quadratic ) and quartic deviations from the ground state . the ground state corresponds to spins aligned in any direction within the xy plane . writing the spins in hamiltonian ( [ hamil ] ) using coordinates @xmath39 and @xmath40 , @xmath41 the hamiltonian becomes @xmath42 ~ , \ ] ] where we use a simplified notation for the nearest neighbors , @xmath43 . at low temperature , it is reasonable to expand for small @xmath40 and @xmath44 , keeping up to quartic order in these , obtaining , @xmath45 where the harmonic and quartic hamiltonians are @xmath46,\end{aligned}\ ] ] @xmath47 ~ .\end{aligned}\ ] ] it is convenient to carry out the calculations in wavevector space by defining fourier transforms @xmath48 together with a similar definition for @xmath49 . here @xmath6 denotes a two - dimensional wavevector , the sums are over the first brillouin zone ( bz ) , and @xmath50 is the total number of lattice sites . then we rewrite @xmath12 and @xmath13 as @xmath51 ~ , \ ] ] and @xmath52 for the square lattice , with coordination number @xmath53 , the @xmath54 function @xmath55 is an even function of @xmath6 , i.e. , @xmath56 . the sums in ( [ h1 ] ) are constrained to @xmath57 . using the following canonical transformation to creation / annihilation operators , @xmath58 @xmath59^{1/4 } , \quad\quad \beta_{\bq}=\left[\frac{1-\gamma_{\bq } } { 4s^2(1-\lambda\gamma_{\bq})}\right]^{1/4 } ~ , \ ] ] the quadratic hamiltonian is diagonalized , @xmath60 ~ , \ ] ] @xmath61 the diagonal parts of @xmath13 can then be written as @xmath62 where @xmath63 we have dropped terms in @xmath13 whose expectation values in the ground state are zero . note also that if a particular @xmath64 is zero , then @xmath65 diverges , however , when this occurs the corresponding factors multiplying @xmath65 go to zero faster , canceling the divergence . the time evolution of the creation / annihilation operators as given from the harmonic hamiltonian is @xmath66 as discussed in the introduction , the effect of the @xmath13 perturbation is to shift the spinwave frequencies , in the most general case , by some complex self - energy @xmath67 . for the moment , we ignore the real part of @xmath67 , whose effect will be included by the memory function approach , and concentrate instead on the contributions to the imaginary part of @xmath67 , which determine the spinwave damping . to find @xmath67 , the contributions of @xmath13 to the single particle green s function will be evaluated , keeping only the diagrams that have a nonzero imaginary part . the lowest order diagrams that satisfy this requirement are found to be second order in @xmath13 ; there is no contribution from first order diagrams . following the standard procedure,@xcite the time - ordered zero - temperature green s function , in second order , is @xmath68 here @xmath69 is the time ordering operator , and @xmath70 is the ground state of @xmath12 . physically , this green function represents the amplitude for a quasiparticle ( i.e. , spinwave ) created at time @xmath71 in state @xmath72 to be destroyed at time @xmath73 in state @xmath74 , after undergoing two scattering events due to @xmath13 . the free - particle green function , @xmath75 involves no scattering . due to momentum conservation , these green functions are proportional to the kronecker delta , @xmath76 , so we can drop the double momentum indices . also , because of time translational invariance , they depend only on the time difference @xmath77 . introducing the time fourier transforms , @xmath78 the free propagator becomes the well - known expression , @xmath79 where @xmath80 . we apply the rules for feynmann diagrams to evaluate the full propagator , @xmath81 , in terms of @xmath82 and @xmath83 . the free propagator @xmath84 is represented as a straight line with an associated momentum @xmath30 and frequency @xmath85 . the interaction term in ( [ h_1 ] ) is represented as a wavy line as in fig . [ aq1q2q3q4 ] , with an outgoing straight line ( free propagator ) for each creation operator and an incoming straight line for each destruction operator , both with assigned momentum and frequency . the two leading diagrams that contribute to @xmath86 are shown in fig . [ diagrams ] . following the standard feynmann rules for these diagrams , with summations over the intermediate frequencies and wavevectors , and using momentum and frequency conservation at each junction , @xmath86 can be written as @xmath87 where the self - energy function due to the scattering parts of the diagrams is @xmath88 ~ .\end{aligned}\ ] ] @xmath89 is a sum over the `` direct '' and the `` exchanged '' diagrams ; the factor of 4 is the multiplicity of each of these under rearranging the exterior lines . @xmath89 is seen to be @xmath86 with its incoming and outgoing lines removed . the full interacting propagator @xmath81 can be approximated as a sum over repeated applications of @xmath90 , i.e. , @xmath91 which , after summing , leads to the well - known dyson equation , @xmath92 the location of the pole gives the perturbed quasiparticle energy , @xmath93 , and the damping @xmath94 is given by @xmath95 to apply the result to the system at finite temperature @xmath69 , and obtain the self - energy associated with the temperature green s function , the integrals over frequency are transformed to sums according to @xmath96 @xmath97 thus the frequency integrals in ( [ sigma ] ) become @xmath98 [ n(\omega_{\bq}+\omega_{\bk+\bp-\bq})-n(\omega_{\bp } ) ] } { \omega-(\omega_{\bk+\bp-\bq}+\omega_{\bq}-\omega_{\bp } ) } ~ , \end{aligned}\ ] ] where the bose factors are @xmath99 the latter form applies to the classical limit . the imaginary part of @xmath100 is derived from the analytic relation , @xmath101 where @xmath102 takes the principal part and @xmath103 is the usual delta - function . because @xmath6 and @xmath104 each have two components , the sum in ( [ sigma ] ) is equivalent to a four - dimensional integration , once we take the continuum limit . each dimension converts as , for example , @xmath105 , where @xmath106 is the system size , in terms of the lattice constant @xmath107 , leading to @xmath108 the factors of @xmath50 here are canceled by those in the expression for @xmath109 , leading to a size - independent result . thus there finally results for the damping function , @xmath110 where the kernel @xmath111 is given by @xmath112 \times \nonumber \\ & & [ n(\omega_{\bq})-n(-\omega_{\bk+\bp-\bq } ) ] [ n(\omega_{\bq}+\omega_{\bk+\bp-\bq})-n(\omega_{\bp } ) ] ~ .\end{aligned}\ ] ] we use this result to calculate the damping at the spinwave frequency : @xmath113 , in the thermodynamic limit @xmath114 . note that in the classical limit , the bose factors are simply proportional to temperature . therefore , we only need to calculate @xmath115 at one temperature ( @xmath116 ) and then scale with @xmath117 to get its value at other temperatures . the summation / integration in ( [ gamma ] ) involving a delta function presents an interesting mathematical problem in its evaluation . general integrals of this form ( i.e. , with one delta - function ) in @xmath118 dimensions can be considered , @xmath119 where @xmath120 is a @xmath118-dimensional vector . in two dimensions , for example , the delta - function will pick off a particular contour that contributes to the integral ; in three dimensions that contour will actually be a surface , and in four dimensions it will be a three - dimensional volume . depending on the form of the function @xmath121 , the contributing region need not be simply connected , but may consist of separated distinct parts . for the higher dimensional cases , it is difficult to visualize the region or regions that contribute to the integral . thus it is necessary to have a way to evaluate the integral without appealing directly to geometrical properties . in a series of papers , gilat and co - workers@xcite have considered how to evaluate these integrals in three dimensions , in work related the calculation of spectral density functions . we have used the `` linear analytic '' ( la ) method@xcite of gilat and raubenheimer , extended to four dimensions , which gives good precision for few sample points , at little cost in complexity . suppose we calculate @xmath122 over a small range of @xmath85 and then take the limit @xmath123 . to do so , a 4d cubic grid of small cells of size @xmath124 is set in the @xmath120-space . then we consider the contribution of each cell to @xmath122 , thereafter summing over the cells . suppose the cells are small enough so that the kernel @xmath125 has only a small variation within a cell , and , therefore , we can use its value at the cell center , @xmath126 . then we can write @xmath127 @xmath128 the idea of the la method is to expand @xmath121 around the cell center , @xmath129 shifting coordinates to @xmath130 , and writing the gradient at the cell center as @xmath131 , where the unit vector @xmath132 points along the gradient evaluated at the cell center , we have @xmath133 @xmath134 the variable @xmath135 is a rescaling of @xmath85 . if we use the integral representation of the delta function , in the form , @xmath136 then the contribution from one cell is found to be @xmath137 where the @xmath138 are the values of parameter @xmath135 at the 16 corners of the cell , i.e. , @xmath139 with all 16 possible choices of the signs , and @xmath140 is the number of minus signs used in each @xmath138 . one can note that similar simple formulas result for this type of integral in other dimensions as well.@xcite in the actual application of eq . ( [ dcell ] ) , inside the cell , the parameter @xmath135 will range between minimum and maximum values : @xmath141 these values hold at two opposite corners of the cell . by eq . ( [ w - omega ] ) , there will be a corresponding range of @xmath85 over which this cell contributes to @xmath122 . then for the general problem where @xmath142 is desired over a large range of @xmath85 , what is done is to set a uniform grid of @xmath85-values , finding the contribution of each cell to @xmath122 for each value on the grid , say , @xmath143 . typically it works well to have the grid fine enough so that each cell contributes to @xmath142 at many different @xmath144 . for the damping problem , eq . ( [ gamma ] ) , we have the replacements , @xmath145 with the 4d variable @xmath146 . we want to find @xmath147 right at @xmath148 for some particular wavevector , @xmath30 . to accomplish this , we set a range @xmath149 , and define the grid of values @xmath150 , using @xmath151 . for the integrations over @xmath6 and @xmath104 , the cell sizes actually can be quite large : we used the cell width @xmath152 , where @xmath153 is a `` system size , '' which gave results consistent with those using much smaller cells ( larger @xmath154 ) . that is , each axis , @xmath155 , @xmath156 , @xmath157 , @xmath158 , was partitioned into 41 cells . first , to get good precision , a narrow range symmetrical around the spinwave frequency was used : @xmath159 , @xmath160 . second , for comparison , the behavior of the damping function was calculated over the whole range of possible spinwave frequencies , i.e. , @xmath161 , @xmath162 . in both of these approaches , interpolation on the @xmath85-grid gives @xmath163 . the scheme was found to be very robust , with both approaches agreeing to better than one percent precision . we should also note one other important calculational detail . because eq . ( [ dcell ] ) depends on the reciprocal gradient of @xmath164 , it is necessary to avoid any points in the @xmath165-space where this gradient vanishes . we accomplished this by taking wavevectors @xmath166 , ( at @xmath153 ) , and then shifting the @xmath165-cell center positions slightly from the usual values : @xmath167 , @xmath168 . here @xmath169 , etc . , are integers ranging from 0 to @xmath170 , and we used @xmath171 to shift the @xmath165-cells away from high symmetry points . the choice of an odd number for @xmath154 also helps to avoid certain high symmetry points . some slight improvement in precision can be achieved with a larger value of @xmath154 , however , at the expense of a tremendous increase in cpu time , which varies as @xmath172 . at @xmath153 , the typical relative errors in @xmath147 were found to be much less than one percent , when compared to calculations using @xmath173 . some results for the damping function , which is proportional to @xmath117 , are shown in fig . [ gam_omega - fig ] for @xmath37 . in general , it is found that @xmath147 either has a cusp or a peak at the spinwave frequency ( arrows in fig . [ gam_omega - fig ] ) . the damping strength increases with @xmath174 , as seen in fig . [ gam_k - fig ] . along the ( 11 ) direction , however , @xmath115 tends to saturate halfway out in the zone . keeping in mind that @xmath175 , the damping strength we find here is quite weak , especially when we consider its effects at @xmath176 where the spinwave approximation is valid . at small nonzero values of the anisotropy parameter @xmath177 , there will be only a very slight change in @xmath115 , because @xmath177 does not appear explicitly in @xmath13 , but only enters the calculation via the dispersion relation and via the @xmath178 and @xmath179 functions of ( [ canonical ] ) . to demonstrate this effect , fig . [ gam_lambda - fig ] shows @xmath115 at various values of @xmath177 for @xmath180 along the ( 10 ) direction . in general , the damping decreases with increasing @xmath177 , i.e. as the system becomes more isotropic . the calculation of the memory function requires the knowledge of @xmath115 over the entire bz . we can reduce the region of calculation to @xmath181 , because , by symmetry , @xmath182 , and @xmath183 . in this region @xmath115 was calculated on a uniform square grid with spacing @xmath184 . linear interpolation was used to get values between these grid points . for the calculation of the dynamic correlation function @xmath185 , we use the memory function formalism due to reiter and sjlander,@xcite who applied the procedure to the isotropic heisenberg model . the technique is based on a low - temperature expansion of the memory function.@xcite the formalism was applied to the 1d isotropic antiferromagnet in an external field by de raedt et al@xcite and to a 1d easy - plane antiferromagnet by gouva and pires.@xcite more recently , it was also applied to 2d easy - plane ferromagnets by menezes et al,@xcite although some numerical difficulties inhibited numerical evaluation of the analytic formulas . some of these problems were surpassed in the work of pereira et al.@xcite in all of these works , no mode damping was included . the presence of damping actually relieves the numerical difficulty of the problem , as we see below . the dynamic correlation function is defined in terms of the fourier transformed spin - fields as @xmath186 where @xmath178 is any component of @xmath187 , and by symmetry , @xmath188 . in this and following formulas for correlation functions , we note the distinction between continuum ( functions of @xmath6 ) and discrete ( subscripts @xmath6 ) fourier transforms , which are related by factors of @xmath2 for each dimension.@xcite this is necessary so that we can make a careful comparison between the theory and the mc - sd simulations , _ including the absolute magnitudes_. specifically , the continuum correlation function in ( [ sqw ] ) , which gives a thermodynamic weight over a _ wavevector range _ @xmath189 , is related to the discrete correlation at a _ specific _ wavevector @xmath6 , using discrete fourier transforms as defined in ( [ fourier ] ) , by@xcite @xmath190 in the analytic calculations , many intermediate quantities are discrete functions , however , for the comparison with the mc - sd data , both results will be converted and displayed as continuum quantities . due to the symmetry in the @xmath191-plane , the calculation is carried out by making averages of the rotationally symmetric quantity , @xmath192 , with @xmath193 . then in the memory function calculation , the continuum dynamical structure function can be written in terms of a second - order memory function , @xmath194 , to be described below , according to @xmath195 ^ 2 + \left [ \omega ~{\rm i m } ~ \sigma_{\bq}^{\perp}(\omega ) \right]^2 } ~ , \end{aligned}\ ] ] where @xmath196 is the liouville operator , and the second moment @xmath197 is determined from a ratio , @xmath198 the location of the spinwave peak that results from eq . ( [ sqw - memory ] ) is very sensitive to the value of @xmath197 , therefore we are especially concerned to get an accurate evaluation of it . the denominator of ( [ moment2 ] ) is a static correlation function , whose value will be taken from the monte carlo simulations , although for low temperatures it can be evaluated approximately as [ sqperp ] @xmath199 @xmath200 the numerator of ( [ moment2 ] ) has been shown@xcite to have a leading low - temperature dependence , also related to static nearest neighbor correlations , @xmath201 the correlation @xmath202 , therefore its contribution can be ignored for small @xmath177 . there being two bonds per lattice site , the in - plane near - neighbor correlation is related to the average energy per bond by @xmath203 this correlation can also be obtained from the mc simulations , although a good low - t approximation is @xmath204 a comparison of the analytic and mc static parameters will be given below . to the leading order in temperature , the time - dependent memory function has been shown to be given from @xmath205 @xmath206 where @xmath207 is projection onto the non - secular variables , see ref . , whose action is equivalent to @xmath208 using expressions ( [ mapping ] ) expanded to second order in @xmath39 , @xmath40 , and ( [ canonical ] ) , it is straightforward to write the spins @xmath209 in terms of the creation / annihilation operators . in the leading order in temperature , the resulting expressions involve expectation values of products of four creation / annihilation operators . the damping is included in the time evolution according to , @xmath210 then , there results , after some manipulations , expressions to give the memory function , @xmath211 the two terms correspond to spinwave sum ( @xmath212 ) and difference ( @xmath213 ) processes , with weights @xmath214 where @xmath215 \bigr\ } ~ , \end{aligned}\ ] ] @xmath216 \bigr\ } ~ . \end{aligned}\ ] ] the frequencies and damping rates of these processes are @xmath217 @xmath218 finally , the frequency - dependent memory function is given by the one - sided fourier - laplace transform , @xmath219 from this expression , eq . ( [ m - memory ] ) , and equations following , we obtain a sum over all the sum / difference processes , @xmath220 \nonumber \\ & & + w_{- } \bigl [ \frac{\omega+\omega_{-}-i\gamma_{-}}{(\omega+\omega_{-})^{2}+\gamma_{-}^{2 } } + \frac{\omega-\omega_{-}-i\gamma_{-}}{(\omega-\omega_{-})^{2}+\gamma_{-}^{2 } } \bigr ] \bigr\ } ~ .\end{aligned}\ ] ] where the @xmath221 dependencies of @xmath222 , and @xmath223 have been suppressed . this is the expression used to evaluate @xmath224 , and thus the dynamic structure function , via eq . ( [ sqw - memory ] ) . to evaluate ( [ sigma_sum ] ) on a discrete @xmath225 lattice , the wavevector @xmath6 must be one of the allowed values , @xmath226 where @xmath227 and @xmath228 are integers . it is also necessary to make the summations over @xmath104 such that the vectors @xmath229 are both given by integers times @xmath2 . thus , it is convenient to change from the summation over @xmath104 to a summation over a new variable @xmath30 defined by @xmath230 where @xmath231 and @xmath232 are integers . then , we also have @xmath233 equal to an allowed wavevector . this insures that the sums will be over spinwaves that are allowed on the given finite lattice . this approach is essential for calculation of the finite size dynamical effects . on the other hand , for the infinite sized system , no variable change is needed , and there is the usual replacement , @xmath234 . note that in the limit of zero damping , eq . ( [ sigma_sum ] ) is inapplicable , except for the continuum limit ( i.e. , @xmath114 ) . the resulting continuum integrals for the infinite system are singular , and the analytic relation ( [ delta - relation ] ) must be applied , leading to @xmath235 \nonumber \\ & & + p \int d^{2}p ~ w_{- } \bigl [ \frac{1}{\omega+\omega_{- } } + \frac{1}{\omega-\omega_{- } } \bigr ] \nonumber \\ & & -i \pi \int d^{2}p ~ w_{+ } \bigl [ \delta(\omega+\omega_{+ } ) + \delta(\omega-\omega_{+ } ) \bigr ] \nonumber \\ & & -i \pi \int d^{2}p ~ w_{- } \bigl [ \delta(\omega+\omega_{- } ) + \delta(\omega-\omega_{- } ) \bigr ] \bigr\ } ~ .\end{aligned}\ ] ] indeed , in the very low temperature range ( @xmath236 ) where the damping is extremely weak , it may be more efficient to use this expression rather than eq . ( [ sigma_sum ] ) , which has poorer convergence with @xmath154 . ( [ sigma_int ] ) was applied by pereira et al.@xcite ; the imaginary part can be evaluated by techniques described here ( sec . [ integration ] ) , and the principal - valued real part can be calculated by an extension@xcite of techniques used in 1d.@xcite in ref . the last expression was found to give finite amplitude spinwave peaks , without damping present . before presenting the results obtained via the memory function approach , the key aspects of the monte carlo and spin - dynamics simulations are summarized . to test the validity of the memory function technique , we compare it with numerical simulations on @xmath237 square lattices ( @xmath238 , @xmath37 ) using monte - carlo and spin - dynamics ( sd ) simulations , which include effects due to all thermodynamically allowed excitations . as mentioned above , we also use some data for static correlations from the monte carlo calculations as input to the memory function dynamical calculations . the techniques used here have been described in ref . and are based on the simulation methods developed in ref . . in the mc calculation , we applied a combination of metropolis single - spin moves and over - relaxation moves that modify all three spin components , and in addition , wolff single - cluster operations@xcite that modify _ only _ the @xmath191 spin components . the over - relaxation and cluster moves are important at low temperatures , where the @xmath191 spin components tend to freeze and single spin moves become inefficient . in the combined mc - sd simulations , a set of mc generated states from thermodynamic equilibrium at temperature @xmath69 are used as initial conditions for numerical integrations in time via a fourth order runge - kutta scheme . specifically , the first 4000 mc steps were used for equilibrating the system . then , every 400 mc steps a state was taken to initiate the sd integration . this was repeated 500 times ; the displayed @xmath1 data are averages over the 500 time integrations . the total mc sequence was then @xmath239 steps ; every 25 steps a state was used to calculate averages of the static correlation functions . in the time integration , the basic runge - kutta time step was @xmath240 , and every 11 time steps , data samples ( for @xmath241 , etc . ) were saved for @xmath6 s in the ( 10 ) , ( 01 ) , and ( 11 ) directions only ( due to machine memory restrictions ) . a total of @xmath242 data samples were saved , followed by applying a fast - fourier - transform to obtain @xmath243 . the total time of integration of @xmath244 is large enough so that no smoothing window is necessary for the data generation . then the frequency resolution of these results is good , @xmath245 , so that any fine details in the low temperature results should be visible . for the lowest temperatures ( @xmath246 ) , where we wanted to see the complex fine details due to the finite size effects , some simulations were made saving data samples every 28 times steps , leading to a total integration time @xmath247 , and frequency resolution @xmath248 . first , a comparison of the low-@xmath69 formulas and static mc results needed as input to the memory function formalism is presented . then dynamical results from the memory function and from md - sd will be presented for large systems , and finally , for smaller systems , where the finite size features will be analyzed . the continuum static structure function , @xmath249 was measured directly in the mc simulations . a comparison with the results based on the low - temperature expression ( [ sqperp ] ) is given in fig . [ sq - fig ] . there we also show a fit of the mc data to a form similar to the analytic expression , but with a fitting coefficient ( @xmath250 ) and power ( @xmath251 ) , @xmath252 from the fits , the parameter @xmath250 is somewhat larger than the low - temperature limit , @xmath253 . this can be expected due to the renormalization of the effective @xmath17 to lower values with temperature . the power @xmath251 falls with increasing @xmath69 , which must also be due to anharmonic effects . there is little difference between the low - t expression and the mc data at the lowest temperature . the deviations between the mc and low - t theory increase substantially by @xmath254 , especially at higher wavevectors . note that the fits are only shown for demonstration ; we used the mc data itself as inputs to the memory function calculations . the other static input to the memory function calculation , for determining the second moment , is the quantity @xmath255 , which is related to the nearest neighbor correlation @xmath256 by eq . ( [ lsls ] ) . estimates of that latter from the mc evaluation of the average energy per spin [ eq . ( [ s0sa - en ] ) ] are compared to the theoretical expression ( [ s0sa - theory ] ) in fig . [ s0sa - fig ] . the analytical expression is seen to give results consistent with the mc data up to @xmath257 . when combined to give the second moment , the mc data produces results as shown in fig . [ moment2-fig ] . it is essentially this quantity that has the strongest influence on the locations of the spinwave peaks in @xmath1 . these are the data used as input to the memory function calculation . we evaluate @xmath194 by using the sum over discrete wavevectors @xmath104 as displayed in eq . ( [ sigma_sum ] ) , and then apply eq . ( [ sqw - memory ] ) to determine @xmath193 . if @xmath154 is taken large enough , then the sum should be a good approximation to the infinite size limit . in general , we can see whether larger @xmath154 is necessary by the presence of finite size features in the data for @xmath224 and @xmath243 . typically we used @xmath258 to give a reasonable approximation to the continuum limit for @xmath259 and larger @xmath6-values ; at lower temperatures or wavevectors even larger @xmath154 is necessary to smooth out all the finite size features , due to the very weak damping . in fig . [ sigt3qpi4pi4-fig ] , some typical results for the real and imaginary parts of @xmath194 are shown for @xmath37 , @xmath260 , and @xmath261 , where @xmath262 was used in the sum . the resulting dynamical structure function is compared to the corresponding mc - sd data for @xmath238 in fig . [ sxxt3qpi4pi4-fig ] . at the ( zero-@xmath69 ) spinwave frequency @xmath263 , i m @xmath194 achieves its minimum value , which is not zero due to the presence of the damping , while re @xmath194 has a small nonzero value there as well . the spinwave peak in the memory data occurs at @xmath264 , which is just slightly greater than the associated value of the second moment , @xmath265 . the spinwave peak position will occur where the denominator of eq . ( [ sqw - memory ] ) achieves its minimum value , which gives , to a good approximation , @xmath266 for this case , the memory function data give @xmath267 , which is consistent with this relation . the peak position is very sensitive to the value of the second moment , which is why we have tried to estimate it as accurately as possible from the static mc data . still , the mc - sd data show the spinwave peak closer to @xmath268 , slightly lower than the memory function prediction . using eq . ( [ sqw - memory ] ) , the spinwave peak height is proportional to @xmath269^{-1}$ ] , while the peak width is approximately equal to @xmath270 . when compared to the mc - sd data , the memory function calculation usually overestimates the peak height and underestimates the peak width , although both are in reasonable agreement with the mc - sd data . in fact , these errors tend to be larger at lower temperature . in figs . [ sxxt1-fig][sxxt5-fig ] , we display some further comparisons of the memory function calculations and mc - sd results at other values of @xmath6 for temperatures ranging from @xmath271 to @xmath272 . for the lower temperatures there is very good agreement between the memory theory and mc - sd peak positions , while the peak heights and widths do not agree as well . on the other hand , at higher temperatures the opposite is true : the memory data show greater errors in the peak positions , especially at large @xmath8 , but the peak heights and widths are predicted very well . probably these discrepancies are caused by a fairly large _ underestimate _ of the damping strength at low temperature , as well as errors in the estimate of the second moment at higher temperatures . however , in general , the tails of the spinwave peaks far from @xmath5 are described very well by the memory function calculations . when we consider the results using relatively smaller values of wavevector , @xmath243 shows not only a spinwave peak , but additional features ( small peaks ) which depend on the system size . these features are obvious in the mc - sd @xmath1 data shown above . similar size - dependent dynamical features have been noticed in earlier mc - sd simulations of xy - symmetry@xcite and easy - axis symmetry@xcite magnets , but without a clear theoretical explanation . we now consider a closer look at the details of these features , making a comparison with the memory function calculations at smaller values of @xmath154 . some examples showing the resulting memory function data for @xmath243 for @xmath238 systems are given in fig . [ sxxt3q16]a [ @xmath273 , @xmath274 , fig . [ sxxt3q9]a [ @xmath275 , @xmath274 , and fig . [ sxxt1q5]a [ @xmath276 , @xmath277 . in fact , there appear various sequences of small peaks . the number of separate sequences appears to depend on the wavevector in integer units of @xmath2 . note that the logarithmic scale exaggerates the size of these subpeaks , they are typically a few orders of magnitude smaller than the spinwave peak . for sufficiently low temperatures , these effects also appear clearly in the mc - sd results : see figs . [ sxxt3q16]b , [ sxxt3q9]b and [ sxxt1q5]b , although the corresponding features there are smeared out compared to the memory function calculation . the effects are also present at larger wavevectors , however , in that case there are many more of these finite size subpeaks , which tend to merge together into a smoother background . these extra peaks can each be associated with particular spinwave sum or difference process . to see that this is the case , consider first the difference processes , whose frequencies @xmath278 are always less than the spinwave frequency @xmath5 . a plot of the weights @xmath279 [ eq . ( [ weights ] ) ] that determine the contributions to the memory function versus the corresponding frequencies @xmath280 , can show how strongly each process [ i.e. , each @xmath30 as defined in ( [ k - discrete ] ) ] contributes to i m @xmath224 and hence to @xmath1 . in fig . [ omminust3q16 ] , we made such a plot for an even integer @xmath6 , namely , @xmath281 , for @xmath238 and @xmath260 . each dot there corresponds to one allowed wavevector @xmath30 ; some of the values of @xmath30 in units of @xmath2 are indicated there . the sequence of @xmath30 s , ( 8,0 ) , ( 8,1 ) , ( 8,2 ) , etc . , leads to multiple contributions at zero frequency , that is , a weak central peak , which is a general result for any even integer @xmath6 . the sequence , ( 9,0 ) , ( 9,1 ) , ( 9,2 ) , etc . , produces a sequence of small peaks in @xmath1 out to @xmath282 , and , the next sequence , ( 10,0 ) , ( 10,1 ) , ( 10,2 ) , etc . , gives a sequence of peaks out to @xmath283 , and so on . each of these @xmath30-sequences can be identified with corresponding peak - sequences in the memory function calculation of @xmath1 , as seen in fig . . indeed , for the wavevector @xmath281 , there are @xmath284 of these sequences below the spinwave peak , including the ones that contribute only at @xmath285 . the higher sequences are more spread out , and merge with the other ones , making a complicated fine structure in @xmath1 . similar features are seen in the mc - sd data , fig . [ sxxt3q16]b , although it is clear that the individual peaks of each sequence are smeared out , resulting in one wider feature from the entire sequence . another example of these finite size features , for an odd integer wavevector , @xmath286 is shown in fig . [ ompmt3q9 ] , again for @xmath238 and @xmath260 . for this smaller integer wavevector there are now only 5 sequences of fine - structure difference process peaks . it is significant that there is no such peak at zero frequency . this is clearly caused by the absence of any allowed spinwave difference process exactly at @xmath287 , although there are many such processes with very weak weights _ close _ to @xmath287 . in general , for any odd integer wavevector , @xmath1 will have a relative minimum at @xmath285 , because of the absence of allowed difference processes there . in fig . [ ompmt3q9 ] we also show some of the weights of the sum - processes , @xmath288 , indicated by plus signs . for either sum or difference processes , numbers in parenthesis refer to the values of @xmath30 in units of @xmath2 . the sum frequencies must fall above the spinwave frequency @xmath5 . comparing figures [ ompmt3q9 ] and [ sxxt3q9 ] , we see that the finite size features above @xmath5 in @xmath1 can each be identified with a particular spinwave sum process . the identification , however , is somewhat more complex than that for the difference processes , because the different sequences of processes overlap in frequency . the match between the sum peak frequencies in the memory data and the mc - sd data is not as good as for the differences processes below @xmath5 . we can note that at these higher frequencies , any relative errors appear larger on an absolute scale , and , the spinwave sum / difference frequencies do not include any exchange softening effects . finally we show another example , for @xmath289 at @xmath290 , @xmath238 in fig . [ ompmt1q5 ] , where both the difference and sum processes are shown . in this case there are only 3 dominant sequences of difference processes leading to finite size features . although the memory function formalism applied above , strictly speaking , applies to low temperatures , it is interesting to consider what kind of predictions it makes if applied at higher temperatures . it is clear from the above comparisons between mc - sd data and the memory function data that at low temperatures , where the spinwave approximation should be valid , the damping of the modes has been underestimated . this is seen especially in the @xmath271 memory data where all the finite size features , as well as the spinwave peak , are too narrow . thus it is interesting to see whether the spinwave width and shape are better described by this theory at higher temperature , because the theoretical damping rates grow as @xmath117 . the other reason to look at higher temperature is to consider how the background intensity in @xmath1 for frequencies @xmath291 , depends on the temperature . for one thing , the increase in temperature should spread all of the finite - size effect peaks and smooth the spinwave background , but can be expected to raise its intensity as well . for comparison , as @xmath69 crosses the vortex unbinding temperature @xmath11 , previous mc - sd simulations@xcite show that this background grows with @xmath69 , until it dominates over the spinwave peak . the effect has been described by a theory employing a gas of freely moving vortices,@xcite although simulations have given some indication that vortex lifetimes@xcite are rather too short to take this theory literally , and , vortices tend to be pinned by the lattice and also have complex motions not resembling free translation . thus it is interesting to compare the memory approach to the mc - sd data at higher temperatures , where the latter will include effects due to all possible types of excitations , beyond linear spinwaves . for @xmath37 , the vortex unbinding temperature has been estimated to be @xmath292 . in fig . [ hi - t - sxx ] , we show some comparisons of the continuum ( @xmath293 ) memory and mc - sd ( @xmath238 ) data for @xmath1 , at temperatures @xmath294 for two particular values of @xmath6 . as already seen , the spinwave peak locations in the mc - sd data here also fall lower than the memory function theory prediction . also , there is a significant growth of the low - frequency background ( @xmath291 ) in the mc - sd data and in the memory function calculations . for @xmath295 the low - frequency background already dominates over the spinwave peak , if the latter can be seen at all . the effect is stronger in the mc - sd data : there is a clear excess of intensity around zero - frequency in @xmath1 in the md - sd data compared to the memory function data . this effect is absent at @xmath296 , however , where the mc - sd and memory data coincide for low frequencies ; in this low temperature region the spinwave approximations used in the memory approach are valid . for temperatures near and above @xmath11 , this extra central peak intensity , not described within the spinwave picture of the memory function approach , may possibly be attributed to nonlinear excitations , including specifically , vortices . this is consistent with the phenomenological vortex gas description of mertens et al:@xcite the excess central peak component is stronger for lower wavevectors , because the vortex pairs being generated above @xmath11 disorder the spins on long length scales . the presence of spinwave damping included into the memory function approach has been suggested here as a way in which finite amplitude spinwave peaks result in the 2d xy model , canceling the divergences in previous theories . a rough calculation of the damping strength has been made using the feynmann diagram technique applied to the quartic corrections of the harmonic hamiltonian . the diagrams used involve only two scattering events ; clearly this gives an underestimate of the damping , and it is likely that including the third order diagrams could make a substantial modification to the results . generally the damping @xmath115 obtained increases away from the center of the brilluoin zone , being close to zero near the center and being greatest along the ( 11 ) direction . in this classical calculation , the damping varies as the square of the temperature ; this results in extremely weak damping strength ( @xmath297 ) for temperatures below @xmath298 . as the temperature approaches and passes @xmath11 , the damping becomes comparable to the spinwave frequency . the comparison of the memory function data and mc - sd data for @xmath1 is good in several aspects : the theory gives finite spinwave peaks , they are positioned correctly for low temperatures , and the multiple finite - size - effect peaks are reproduced as in the mc - sd data . the most significant problem in the memory function approach is the extremely narrow peak widths clearly caused by an underestimate of the true damping strength , especially at lower temperatures and wavevectors . for very low temperatures in the present theory , the longest wavelength spinwaves will not scatter even once before crossing the entire system . of course , this is a quantum picture , applied to the classical system , so perhaps a better approach in this low - t limit would be to consider some kind of adiabatic scattering in the nonuniform potential due to the spinwave population . this would essentially be the equivalent of the quantum calculation at very high order diagrams rather than at the lowest order . when the memory calculation is carried out for large systems , that is , in the continuum limit , the finite - size peaks merge into a smooth background . indeed , in this limit , the spinwave peak itself has contributions on the high frequency side from spinwave sum processes , and on the low frequency side from spinwave difference processes . the combination of all of these contributions , essentially an infinite number of sum and difference processes , leads to an overall smooth and nondivergent spinwave peak . even in the limit of _ zero damping _ , the spinwave peak are nondivergent , according to continuum calculations [ using eqs . [ sigma_int ] ] by pereira et al.@xcite however , when damping is included , the heights of the spinwave peaks are smaller , and we expect the results to be more realistic . in the memory function theory , the spinwave peak height was seen to be proportional to @xmath269^{-1}$ ] . the function @xmath299 , to a leading approximation , is proportional to the temperature @xmath69 . from eq . ( [ sqw - memory ] ) , in regions away from the spinwave peak @xmath1 should be proportional to @xmath299 . thus we see that the memory function theory demonstrates how the background grows with temperature while the spinwave peak relatively tends to fall . for high enough temperature , the distribution of spinwave difference processes near zero frequency dominates the spectrum of @xmath1 , leading to a central peak stronger than the spinwave peak . clearly this means applying the theory well above the limiting temperature where we expect spinwave approximations to be valid . however , this gives results very similar to those in the mc - sd simulations , although the latter can be expected to include nonlinear effects such as vortices . in fact , we have seen a strong indication for nonlinear effects ( beyond spinwave difference processes ) above @xmath11 as extra intensity in @xmath1 near zero frequency in the mc - sd data . in conclusion , application of the memory function formalism , including a leading approximation to the mode damping , has been shown to accomplish three things for the xy model : i ) produce nondivergent spinwave peaks ; ii ) explain the finite size features in @xmath1 both above and below @xmath5 ; iii ) explain the background in @xmath1 below @xmath5 for the infinite system . note that strictly speaking , damping is necessary only for item ii ) , because @xmath1 can not be calculated in the memory function formalism for a finite system without damping . the continuum memory function approach gives finite size spinwave peaks and a low - frequency background even in the absence of damping . the theory also suggests that a sizable contribution to the growth of the central peak for @xmath69 above @xmath11 may be due to spinwave difference processes . the excess central peak intensity in the mc - sd data , when compared to the memory function data , must be due to nonlinear excitations , and gives evidence for dynamical effects due to vortices . gmw thanks universidade federal de minas gerais for their hospitality and support on this project . the authors thank the support by nsf - int-9502781 , nsf cda-9724289 , fapemig , and cnpq . permanent address : department of physics , kansas state university , manhattan , ks 66506 - 2601 . [ email : _wysin@phys.ksu.edu_ ] . j. villain , j. phys . ( paris ) * 35 * , 27 ( 1974 ) . f. moussa and j. villain , j. phys . c * 9 * , 4433 ( 1976 ) . d. r. nelson and d. s. fisher , phys . b * 16 * , 4945 ( 1977 ) . v. l. berezinskii , sov . jetp * 32 * , 493 ( 1970 ) ; * 34 * , 610 ( 1972 ) . j. m. kosterlitz and d. j. thouless , j. phys . c*6 * , 1181 ( 1973 ) r. d. mattuck , _ a guide to feynmann diagrams in the many - body problem _ , mcgraw - hill international book company ( 1976 ) . a. l. fetter and j. d. walecka , _ quantum theory of many - particle systems _ , mcgraw - hill , new york ( 1971 ) . a. a. abrikosov , l. p. gorkov and i. e. dzyaloshinski , _ quantum field theoretical methods in statistical physics _ , pergamon press , oxford ( 1965 ) . g. gilat and j. raubenheimer , phys . rev . * 144 * , 390 ( 1966 ) . g. gilat and z. kam , phys . rev . lett . * 22 * , 715 ( 1969 ) . g. gilat , j. comp . * 10 * , 432 ( 1972 ) . in two - dimensions , ( [ dcell ] ) becomes @xmath300 and in three dimensions , @xmath301 for arbitrary dimension @xmath118 , the generalization is @xmath302 g. reiter and a sjlander , phys . lett . * 39 * , 1047 ( 1977 ) ; j. phys . c : solid state phys . * 13 * 3027 ( 1980 ) . h. mori , prog . * 33 * , 423 ( 1965 ) . b. de raedt , h. de raedt and j. fivez , phys . b * 23 * , 4597 ( 1981 ) . m. e. gouva and a. s. t. pires , j. phys . c : solid state phys . * 20 * , 2431 ( 1987 ) . s. menezes , a. s. t. pires and m. e. gouva , phys . b * 47 * , 12280 ( 1993 ) . a. f. a. pereira , m. s. thesis , universidade federal de minas gerais , brazil ( 1999 ) . a. f. a. pereira , m. e. gouva and g. m. wysin , unpublished . the usual conversion from discrete to continuum fourier weights is to divide by the volume @xmath303 per allowed point in @xmath6-space : @xmath304 the @xmath305 normalization of the fourier transform definition ( [ fourier ] ) simplifies many formulas in this paper , including the @xmath50-independent discrete correlation ( [ sqperp ] ) . distributing this weight over the @xmath6-space volume @xmath306 , leads to an @xmath50-_dependent _ continuum correlation @xmath307 . to avoid this @xmath50-dependence in the continuum quantity , it is only necessary to use the alternative fourier transform definitions , @xmath308 this would complicate many intermediate formulas , but in the end the continuum @xmath1 will be independent of @xmath50 and also have a prefactor of @xmath309 . g. m. wysin , unpublished . m. e. gouva , unpublished . m. e. gouva and g. m. wysin , phys . b * 56 * , 14192 ( 1997 ) . c. kawabata , m. takeuchi and a.r . bishop , j. magn . mater . * 54 - 57 * , 871 ( 1986 ) ; j. stat . phys . * 43 * , 869 ( 1986 ) ; g.m . wysin and a.r . bishop , phys . b * 42 * , 810 ( 1990 ) ; u. wolff , nucl . phys . * b300 * , 501 ( 1988 ) ; phys . lett . * 62 * , 361 ( 1989 ) . evertz and d.p . landau , phys . b * 54 * , 12302 ( 1996 ) . d. p. landau , j. mag . mater . * 200 * , 231 ( 1999 ) ; m. krech and d. p. landau , phys . b * 60 * , 3375 ( 1999 ) . m. e. gouva , g. m. wysin , s. a. leonel , a. s. t. pires , t. kamppeter and f. g. mertens , phys . b * 59 * , 6229 ( 1999 ) . mertens , a.r . bishop , g.m . wysin and c. kawabata , phys . b * 39 * , 591 ( 1989 ) . d.a . dimitrov and g.m . wysin , phys . b * 53 * , 8539 ( 1996 ) ; j. phys . c * 10 * , 7453 - 65 ( 1998 ) .

non - critical string theory @xcite provides a simple and tractable toy model to understand various aspects of string theory . one of the fascinating formulation of the theory is given by the description with the matrix model . this realizes a non - perturbative definition of string theory @xcite and provides an explicit example of gauge / string duality @xcite . recently , our knowledge about the duality in non - critical string theory has extended to various kind of string theories . the first non - trivial extension was to the correspondence between the two - cut matrix models @xcite and type 0 superstring theory @xcite ; later , the multi - cut matrix models @xcite were proposed as a description of fractional superstring theory @xcite . despite of many qualitative discussions , little is known quantitatively about the nature of multi - cut matrix models . therefore , in this paper we try to give explicit answers to the following questions : how do critical points and potentials look like and what is the corresponding geometry appearing at the vicinity of critical points ? to address these issues , _ macroscopic loop amplitudes _ @xcite have played important roles . the one - point function of macroscopic loop operator , which is also known as resolvent , is defined as @xmath9 . \label{introresolvent}\end{aligned}\ ] ] geometrically , the macroscopic loop amplitudes describe the probability amplitudes for creating holes on the two - dimensional surface as induced from the feynman graph of matrix integral . analytically , we can view the macroscopic loop amplitudes as a generating function for correlation functions of local operators ( in @xmath10 ) in the matrix models . most importantly , it enables us to probe eigenvalue distribution of a matrix @xmath10 @xcite and spacetime geometry in the weak coupling region @xcite . thus , the macroscopic loop amplitude is of fundamental importance in the theory of matrix models . in order to calculate the macroscopic loop amplitudes , one powerful way is to use the method of orthogonal polynomials @xcite . through a useful identity ( see appendix [ relationresolventorthpoly ] ) , we can express the resolvent @xmath11 and the spectral parameter @xmath12 as a pair of conjugate variables acting on the space of orthogonal polynomials , whose commutations relation gives rise to the douglas ( string ) equation . in terms of the differential operators , the string equation turns into a differential equation on the macroscopic loop amplitudes and we can solve it without directly evaluating the matrix integral ( [ introresolvent ] ) . right at the critical points , the resolvent has specific critical behavior . conversely , with some proper choice of the critical resolvent , one can obtain the critical potentials . in the two - matrix - model cases , this approach is especially useful for finding critical points @xcite and has been explicitly formulated for general @xmath13 critical points @xcite . along this line , we explicitly construct the critical points and potentials in the multi - cut two - matrix models . our results provide further insight into the understanding of the relationship @xcite between the existence of critical points and the the choice of hermiticity of critical potentials . macroscopic loop amplitudes at the vicinity of critical points ( i.e. with perturbation of the cosmological constant ) have been investigated in @xcite . the seminal formula for general @xmath13 critical points with cosmological constant @xmath14 was first found by kostov @xcite and can be expressed as @xmath15 with @xmath16 , _ the chebyshev polynomials of the first kind _ , and @xmath17 . in the case of @xmath18 minimal superstring theory ,- cut cases , there are two kinds of indexes : @xmath13 and @xmath18 . the former corresponds to the standard cft labeling of critical points ; the latter corresponds to order of differential operators @xmath19 and @xmath20 . these labellings are different in general @xcite@xcite . since the liouville continuum formulation of @xmath0 symmetric critical points has not been known so far , we only use @xmath18 labeling except for the bosonic @xmath21 cases . ] the corresponding formula was also found on the liouville side @xcite . the non - trivial solution ) with @xmath22 , which is called one - cut solution . the other is essentially different from the one - cut solution which is called two - cut solution ( [ introsupersesh ] ) . ] can be expressed as @xmath23 with @xmath24 , _ the chebyshev polynomials of the second kind _ , and @xmath17 . this general formula has been utilized in various studies of @xmath13 ( or @xmath18 ) critical points @xcite , even in the d - instanton effects of the string theory @xcite . moreover , in comparison with worldsheet cft , e.g. with the fzzt - brane or zz - brane amplitudes in liouville theory @xcite , it is essentially the above formula which is realized in the calculation of @xmath13 ( or @xmath18 ) minimal ( super)string theory @xcite . one of the significances of this formula is that it provides us further checks of the string duality ( e.g. matching of correlators ) beyond matching of operator contents and critical exponents . since this comparison is necessary especially in the correspondence between the multi - cut matrix models and fractional superstring theory @xcite , discovery of the general formula for the multi - cut two - matrix models is one of the most important stage in understanding the conjectured duality as well as the system itself . in this paper , the generalization of these formulae are explored especially in the @xmath0 symmetric potential @xmath25 , @xmath26 of the general @xmath2-cut two - matrix models @xmath27 . identifying a proper ansatz for the non - polynomial part of the amplitudes ( generalizing the square - root of ( [ introsupersesh ] ) ) , we explicitly show that there is a one - parameter set of solutions @xmath28 in the `` unitary '' series @xmath29 which can be tersely expressed as @xmath30{\bigl(z-1\bigr)^l\bigl(z+1\bigr)^{k - l } } , \\ q&= \mu^{\frac{\hat q}{\hat p+\hat q-1 } } p^{(-\frac{2l - k}{k},\frac{2l - k}{k})}_{\hat q-1}(z ) \ , \sqrt[k]{\bigl(z-1\bigr)^{k - l}\bigl(z+1\bigr)^{l}},\end{aligned}\ ] ] in terms of _ the jacobi polynomials_. ] @xmath31 . this ansatz moreover enables us to explicitly describe the geometry of off - critical amplitudes . for instance , our ansatz for the @xmath32 case gives the following algebraic curve : @xmath33 which contains precisely three symmetric cuts corresponding to the eigenvalue distributions in the three - cut two matrix model . the organization of this paper is following : some basics of the multi - cut two - matrix models are reviewed in section [ subsecpreliminary ] . the scaling ansatz for the orthogonal polynomials and its relation to the @xmath34-rotated / real potentials are discussed in section [ secrotatedvsreal ] . the solutions for critical resolvents and potentials are studied in section [ seccritipotentialsresolvents ] ( some examples of critical potentials are listed in appendix [ listofcriticalpotentials ] ) . the daul - kazakov - kostov prescription @xcite is reviewed in section [ secrevofdkkoff ] and the two - cut cases are discussed in section [ twocutrevisited ] . the @xmath0 symmetric @xmath2-cut cases are studied in section [ kcutsymmetricjacobi ] . its algebraic structure is also discussed in section [ secgeometrymulticut ] . section [ secdiscussion ] is devoted to conclusion and discussion . some of the results for the critical potentials are listed in appendix [ listofcriticalpotentials ] . direct evaluation of douglas equation in the three - cut cases is given in appendix [ directevaluationthreecutstringeq ] . several useful formulae about jacobi polynomials are collected in appendix [ jacobip ] . we start with a brief review of some elementary facts about the orthogonal polynomial system in the multi - cut matrix models , which also serves to fix the notation used in this paper . as is the usual two - matrix model @xcite , the multi - cut two - matrix models are characterized by the orthonormal polynomial system , @xmath35 with an inner product , @xmath36 defined with respect to the two - matrix potential @xmath37 . we will use the normalization of the potential , @xmath38 , throughout the main text . the degrees of the potentials are chosen to be multiples of @xmath2 , @xmath39 here @xmath2 is the number of cuts and @xmath40 are arbitrary positive integers . a distinct feature of the multi - cut matrix models is that the integration domain @xmath41 is defined as a contour along radial directions specified by the @xmath42 symmetry of the system in the complex plane @xcite , @xmath43 where @xmath44 is a root of unity , which is chosen to be @xmath45 the @xmath2-cut matrix models possess a @xmath0 charge conjugation , which is a generalization of the reflection transformation for the matrix ( @xmath46 ) in the two - cut phases of the one - matrix models @xcite . the @xmath0 charge conjugation in the @xmath2-cut two - matrix models is given @xcite as @xmath47 which preserves the interaction term , @xmath48 , in the two - matrix potential . in general , if the potential @xmath25 is a @xmath0 symmetric potential , @xmath49 the system is said to be @xmath0 symmetric . the @xmath0 charge conjugation naturally defines a charge conjugation of the orthonormal polynomial as @xmath50 in the @xmath0 symmetric cases , the orthonormal polynomials are invariant under this charge conjugation , equivalently we have @xmath51 that is , the orthonormal polynomials are eigenfunctions of this @xmath0 transformation . the orthonormal polynomial system is characterized by the recursive relation , which can be written as @xmath52 where @xmath53 and @xmath54 . here @xmath55 is the index - shift operator @xcite@xcite@xcite that shifts every index @xmath56 by one unit on its right - hand side : @xmath57 and @xmath58 . note that , by construction , @xmath59 and @xmath60 can not be zero , @xmath61 including @xmath62 . in the @xmath0 symmetric case , invariance of the system asserts that the expansion coefficients @xmath63 and @xmath64 must be zero except for @xmath65 with @xmath66 and @xmath67 . from the recursive relation ( [ orthrecab ] ) and ( [ orthreccd ] ) , one can check that the operator pairs @xmath68 and @xmath69 satisfy the canonical commutation relations , @xmath70=[c , d]=n^{-1}. \label{primitiveccr}\end{aligned}\ ] ] furthermore , taking inner product ( [ orth ] ) between @xmath71 and @xmath72 and then integration by part leads to the following identity , @xmath73 combine these equations together , it can be seen that there is only one independent canonical commutation relation , @xmath74=\bigl[c,\bigl(-a^{\rm t}\bigr)\bigr ] = n^{-1}. \label{twomatccr}\end{aligned}\ ] ] one can also extract equations for @xmath75 and @xmath76 from eq . ( [ steqmatrix ] ) : @xmath77_{\geq -1 } - z^{-1}\frac{n / n}{\sqrt{r(n ) } } , \label{twomata}\\ c(n , z)&= \bigl[v_1'\bigl(a^{\rm t}(n , z)\bigr)\bigr]_{\geq -1 } - z^{-1}\frac{n / n}{\sqrt{r(n)}}. \label{twomatc}\end{aligned}\ ] ] here the bracket is defined as @xmath78_{\geq n}\equiv \left\ { \begin{array}{ll } z^s & ( s\geq n ) \cr 0 & ( s < n ) . \end{array } \right.\end{aligned}\ ] ] the system of equations was originally used to extract the string equations @xcite and can be considered as the fundamental equations dictating the double scaling limit of the two - matrix models . once we solve @xmath75 and @xmath76 , eq . ( [ steqmatrix ] ) leads to the solution of @xmath79 and @xmath80 in straightforward manner . the strategy for solving critical points in two - matrix models was proposed in @xcite . the observation is this ; the operators @xmath75 and @xmath76 have the following scaling behaviors at @xmath13 critical points ( determined by critical potentials ) : @xmath81 where the lattice spacing @xmath82 is introduced to satisfy @xmath83 & { \partial}_n = -a^{1/2 } g_{str}{\partial}_t \equiv -a^{1/2}{\partial}\to 0,\qquad ( a\to 0 ) , \label{scalinglattice}\end{aligned}\ ] ] and the operator @xmath84 is the index - derivative operator in the index - shift operator @xmath85 . conversely , if one assumes this critical behavior ( [ twomatcriti ] ) , then there exist critical points . that is , we can obtain critical potentials which satisfy eq s ( [ twomata ] ) and ( [ twomatc ] ) @xcite . explicit solutions to this problem have been given in @xcite for the general @xmath13 critical points in the one - cut cases as follows : for each pair of potentials with degrees @xmath86 ( @xmath87 of eq . ( [ twomatpotentialorder ] ) ) , the maximal order of the differential operators ( [ twomatcriti ] ) is @xmath88 and the operator @xmath75 and @xmath76 is given as @xmath89 in the large @xmath1 limit . here @xmath90 is the critical value of @xmath60 , @xmath91 then it can be shown that , with these critical forms of @xmath75 and @xmath76 , we can solve for the critical potential @xmath25 which satisfies eq . ( [ twomata ] ) and ( [ twomatc ] ) . in the rest of this section , we generalize this idea to the multi - cut setting . the procedure of section [ subsecpreliminary ] implies that the orthonormal polynomials become smooth functions of the index @xmath56 at the vicinity of the critical points , especially at eq . ( [ scalinglattice ] ) . the scaling ansatz for the multi - cut cases is to find a sequence of orthonormal polynomials @xmath92 ( and @xmath93 ) which are smooth continuous functions in not only @xmath12 but also the index @xmath56 . first of all , since the orthonormal polynomials behave under @xmath0 transformation as @xmath94 there should be @xmath2 distinct smooth scaling functions , each of which has a distinct eigenvalue of the @xmath0 transformation . this is obvious in the @xmath0 symmetric background ( [ zksympol ] ) . hence , we separate the polynomials into the following @xmath2 sets of polynomials @xmath95 with respect to mod @xmath2 . for each set labeled by @xmath96 , we assume that there exists a smooth function in @xmath97 which can be regarded as the continuum limit of the sequence . the proper choice of the smooth functions was first proposed in @xcite and we further write it as @xmath98 with @xmath99 the phase factors indicate that the polynomials are approximated by smooth functions in the directions of @xmath100 and @xmath101 : @xmath102 since the phase does not depend on @xmath97 here . these directions are actually the liouville direction @xcite@xcite@xcite . the main reason for this ansatz is that the critical point of eigenvalues , @xmath103 , is at the origin @xmath104 when we consider @xmath0 symmetric potentials @xmath105 .. ] the consequence is that the operators acting on the vector - valued polynomials @xmath106 ( and @xmath107 ) become smooth functions of @xmath97 . the operators in this basis are expressed as @xmath108 with explicit expressions : @xmath109 where @xmath110 with @xmath111 and @xmath112 here @xmath113 is the identity matrix of @xmath114 . now eq . ( [ steqmatrix ] ) becomes @xmath115 in terms of the @xmath114 matrix - valued operators @xmath116 and @xmath117 . note that the smoothness of polynomials @xmath106 and @xmath107 in the index @xmath97 only implies smoothness of sequence of recursion coefficients @xmath118 in the index @xmath97 . this means that each recursion coefficient sequence ( labeled by @xmath96 ) can be approximated by a smooth function with the following taylor expansion with respect to the scaling parameter @xmath119 ; @xmath120 here the lattice spacing @xmath82 and scaling parameter @xmath121 are defined as @xmath122 & { \partial}_n = -a^{1/2 } g_{str}{\partial}_t \equiv -a^{1/2}{\partial}\to 0,\qquad ( a\to 0 ) , \label{scalinglattice2}\end{aligned}\ ] ] for the @xmath18 critical points of the multi - cut matrix models . it has been argued in the two - cut matrix models @xcite that @xmath123 symmetry breaking critical points can only be realized in the potentials @xmath124 which has an imaginary number @xmath125 in the following way : @xmath126 here the potential @xmath127 is polynomial in @xmath12 with real coefficients . this consideration has a natural correspondence in the liouville theory because the liouville boundary cosmological constant @xmath128 is related to the eigenvalue @xmath12 with an imaginary number @xmath125 @xcite as @xmath129 a natural extension of this consideration is the @xmath34-rotated potential ( [ rotatedpotentials ] ) . from our results , . ] within the prescription reviewed in section [ subsecpreliminary ] , one can also find real - potential solutions for the @xmath123 symmetry breaking critical points . therefore , we consider both cases in this section . we will show in the next subsection that , at least in the sense of critical potentials , the map between two potentials is just an analytic continuation . the @xmath34-rotated potentials are defined as @xmath130 with a real - coefficient potential @xmath131 . if one puts @xmath132 , then this turns out to be eq . ( [ antihermitian ] ) given in @xcite . from the definition , this potential satisfies @xmath133 and consequently one can show that are real functions . ] @xmath134 therefore the uniqueness of the orthonormal polynomials implies @xmath135 here we have used the invariance of contour @xmath136 and the @xmath0 invariance of the measure , @xmath137 . this complex conjugation corresponds to the @xmath0 charge conjugation of the polynomials ( [ chargeconjpols22 ] ) , @xmath138 the hermiticity of coefficients in the operators @xmath75 , @xmath79 , @xmath76 and @xmath80 is also obtained in the same way as @xmath139 comparing this with the expression ( [ aoptilde ] ) , ( [ boptilde ] ) , ( [ coptilde ] ) and ( [ doptilde ] ) , one can see that the coefficients of the matrix - valued operators @xmath140 are all real number . in this sense , eq . ( [ tildesteqmatrix ] ) can have critical solutions with this @xmath34-rotated potential , including the @xmath0 symmetry breaking critical points . the real - potential models are defined by real two - matrix potentials @xmath25 with all coefficients real : @xmath141 the same discussion as above shows that the polynomials @xmath142 and @xmath143 , and the operators @xmath144 , and @xmath80 are all real functions . the relation between eq . ( [ steqmatrix ] ) and the smooth operators becomes simpler if one considers another convention of smooth orthonormal polynomials : @xmath145 } \alpha_{kl+j}(x),\qquad \chi_{l , j}^{(\rm real)}(y ) \equiv ( -1)^{[-\frac{kl+j}{k } ] } \beta_{kl+j}(y),\end{aligned}\ ] ] instead of eq . ( [ smoothfunpsichicomp ] ) . the corresponding smooth operators @xmath146 and @xmath147 are then given as @xmath148 here @xmath149 and @xmath150 are the counterpart of @xmath106 and @xmath107 given in eq . ( [ smoothfunpsichi ] ) . one can easily see that the operator @xmath151 can be obtained from @xmath152 by just replacing the phase @xmath153 in eq . ( [ aoptilde ] ) with @xmath154}$ ] ( the phases in the other operators @xmath155 are also similarly replaced : the phase @xmath156 is replaced by @xmath157}$ ] ) and by replacing the matrix @xmath158 with the matrix @xmath159 , @xmath160 which satisfies @xmath161 . here @xmath162 $ ] is the gauss symbol , which expresses the maximal integer less than or equal to @xmath163 . then eq . ( [ tildesteqmatrix ] ) is equivalent to @xmath164 since everything is real here , one can have critical points ( even @xmath0 symmetry breaking ones ) in the same sense as the @xmath34-rotated models . the large @xmath1 solutions to eq . ( [ tildesteqmatrix ] ) are given by critical values of the matrix - valued operators @xmath165 and @xmath166 ( defined in eq . ( [ aoptilde ] ) and eq . ( [ coptilde ] ) ) , @xmath167 and the critical potentials @xmath168 . at the vicinity of critical points , the coefficients of operators @xmath165 ( and @xmath166 ) should be approximated by the following smooth functions ; @xmath169 in principle , the leading coefficients @xmath170 ( and @xmath171 ) can depend on the index @xmath96 and the critical operators of @xmath165 ( and @xmath172 ) are expanded by these coefficients . for a while , let us see what happens in some special cases .. ] that is , we consider the @xmath173 critical points realized in the potentials of degree @xmath174 ( see eq . ( [ twomatpotentialorder ] ) ) . first we solve the condition ( [ accriticalmulticut ] ) . since @xmath165 and @xmath172 are almost similar , the operator @xmath165 is only considered here . by solving eq . ( [ accriticalmulticut ] ) in this critical point @xmath173 , , one performs the taylor expansion of @xmath175 in terms of @xmath84 and requires @xmath176 . then one can obtain constraints on the coefficients @xmath177 . the result is eq . ( [ solderivaaa ] ) ] one can obtain the critical operator of @xmath165 as @xmath178 the matrix @xmath179 is defined in eq . ( [ defmatr ] ) and the elements @xmath180 can depend on the index @xmath96 in principle . from eq . ( [ aoptilde ] ) , therefore , one can obtain @xmath181 here each term contains a phase factor @xmath182 . since @xmath183 is a multiple of @xmath2 ( say @xmath184 ) and @xmath185 , we get @xmath186 that is , the operator @xmath75 ( and similarly the operator @xmath76 ) is given as @xmath187 we then consider eq . ( [ twomata ] ) and eq . ( [ twomatc ] ) . the above operators @xmath188 and @xmath189 ( @xmath190 ) satisfy eq . ( [ twomata ] ) and eq . ( [ twomatc ] ) for each @xmath96 . by solving these equations , one can obtain the critical potentials and the solutions are realized when one has @xmath191 therefore , the critical behavior of matrix - valued operators @xmath165 and @xmath166 is finally given as @xmath192 several examples of critical potentials calculated by using the above procedure are listed in appendix [ listofcriticalpotentials ] and some critical potentials are drawn in fig . [ criticalpotential223 ] . note that all the parameters here , including @xmath90 , are fixed in this procedure . , but the critical points of eigenvalues are located at the origin , @xmath104 , we have chosen the scale , @xmath193 , and @xmath194 was fixed by choosing the convention of overall coefficient in the orthonormal polynomials ( [ onpmethod ] ) . in this sense , there is no free parameter here . one can also say that one can freely tune the parameter @xmath90 by turning on @xmath195 . this procedure is a redundant deformation which does not change the result , and we will use this fact in appendix [ listofcriticalpotentials ] . ] the result ( [ rareequal ] ) implies that all the @xmath2 coefficient sequences @xmath170 ( @xmath190 ) have the same asymptotic behavior at @xmath196 limit , i.e. at critical point , the coefficients satisfy @xmath197 even though @xmath198 can be approximated by @xmath2 different scaling functions ( e.g. @xmath199 ) in the sub - leading terms of eq . ( [ scalingfunctionsac ] ) . this was just an assumption in the previous discussion of the two - cut two - matrix models @xcite but here we have checked that this is generally correct in the multi - cut two - matrix models . next consider the critical points of @xmath200 realized in the potentials of degree @xmath174 ( see ( [ twomatpotentialorder ] ) ) . this contains @xmath0 symmetry breaking critical points @xmath18 . the critical operators of @xmath165 and @xmath166 are given as @xmath201 with @xmath202 ( see eq . ( [ barabcd ] ) ) . in particular , from eq . ( [ coptilde ] ) , one can see the following relations : @xmath203 therefore , the operators @xmath75 and @xmath76 are also obtained in the similar manner as eq . ( [ critiacwithoutt ] ) , and given as @xmath204 the critical points and potentials in this cases are also obtained with these operators and eq s ( [ twomata ] ) and ( [ twomatc ] ) , which are also listed in appendix [ listofcriticalpotentials ] . note that the critical potentials now depend on the parameters @xmath205 and @xmath206 . note that the difference between the @xmath34-rotated potentials ( [ rotatedpotentials ] ) and the real potentials ( [ realpotentials ] ) is almost the complex phase of @xmath207 @xmath208 in eq . ( [ accritibreaking1 ] ) . since we obtain the critical potentials by solving eq s ( [ twomata ] ) and ( [ twomatc ] ) by using eq . ( [ accritibreaking1 ] ) , the map from real critical potentials to @xmath34-rotated ones is also given by just analytically continuing the parameters @xmath209 in the real potentials to the complex values @xmath210 . this means that all the @xmath0 symmetric @xmath18 critical points and @xmath0-symmetry breaking @xmath18 critical points can be realized within both real potentials and @xmath34-rotated potentials of @xmath2-cut two - matrix models , and also that these real/@xmath34-rotated critical potentials are related to each other by just the above analytic continuation . , @xmath211 ; ( b ) @xmath212 , @xmath213 ; ( c ) @xmath214 , @xmath215 . ( a1 ) , ( b1 ) and ( c1 ) are plots in the real axes . the figures ( a2 ) , ( b2 ) and ( c2 ) are plots of the real part of the potentials @xmath216 on the complex sheet of @xmath12 . the straight lines with arrows on them correspond to the contour @xmath217 of integration . darker color means lower height . ] one can then realize the canonical operators in terms of the lax pair , and they are given by @xmath218 with @xmath219 { \partial}^{\hat q } + \sum_{n=0}^{\hat q-1}\widetilde h_n(t)\ , { \partial}^{\hat q- n},\label{tolax2}\end{aligned}\ ] ] and the canonical commutation relation ( [ twomatccr ] ) becomes the douglas equation , @xmath220=g_{str}i_k$ ] . the functions @xmath221 and @xmath222 are some combination of scaling functions in ( [ scalingfunctionsac ] ) . the parameter @xmath82 is a lattice spacing of @xmath223 . here we define @xmath224 here @xmath221 and @xmath225 are all real functions . since we can erase some parts of sub - leading scaling functions @xmath221 and @xmath222 by changing normalization of orthonormal polynomials ( some similarity transformation ) , we can always put @xmath226_{l+i , i } , \label{honeconst}\end{aligned}\ ] ] for @xmath227 . here we extend the meaning of indices as @xmath228_{i+lk , j+mk } \equiv \bigl[h_1(t)\bigr]_{i , j}$ ] it is also convenient to use the diagonalized basis of @xmath229 which is useful for analyzing multi - component kp hierarchy @xcite : @xmath230 and the matrix elements of @xmath231 are given by @xmath232 in this basis , the relation ( [ honeconst ] ) is simply given as @xmath233_{i , i } = 0\qquad ( i=1,2,\cdots , k).\end{aligned}\ ] ] however , from the analysis we have shown in section [ secrotatedvsreal ] , the component functions in the operators @xmath234 @xmath235 are complex - valued functions now . here we comment on the `` unitary '' series which is defined by the condition of @xmath236 . symmetric '' case in literature , in the sense of exchanging the matrices @xmath237 . since the terminology of `` @xmath123 symmetric '' is dedicated to @xmath0 symmetry in this paper , we will use the term `` unitary '' referring to the @xmath236 case . of course , the terminology of `` unitary '' should come from the worldsheet cft of dual string theory , and therefore our terminology `` unitary '' should be understood as a formal notion . in the case of one- and two - cut matrix models , this situation always gives unitary sequences although obtaining unitary sequences does not necessarily require this condition . ] we should note that the unitary series of the @xmath0 symmetric background is given by @xmath238 instead of @xmath239 . this can be understood as follows : in general , the `` unitary '' condition gives @xmath240 in the @xmath0 symmetric background . here @xmath241 . in the superstring ( @xmath132 ) cases , since the matrix @xmath158 satisfies @xmath242 the following combination of the operators @xmath165 and @xmath172 @xmath243 gives the @xmath244 system . the same argument also applies to the bosonic ( @xmath87 ) cases . for the @xmath245-cut @xmath0 symmetric background case , the matrices @xmath158 and @xmath246 are essentially different matrices ; hence , the previous argument fails . this means that the unitary series should be @xmath247 . on the other hand , if we consider the @xmath0 symmetry breaking critical points which correspond to minimal fractional superstring theory @xcite , @xmath248 & = a^{\hat q}\times \sqrt{r_*}\,k^{\hat q } \,\gamma\ , { \partial}^{\hat q } + \cdots,\end{aligned}\ ] ] the matrices of the leading terms in @xmath165 and @xmath172 are both given by @xmath158 . that is , the situation is similar to the super(bosonic)string cases . in fact , unitary minimal fractional superstring theory corresponds to the @xmath244 critical points , instead of @xmath247 . in this section , we consider off - critical behavior of the critical points we identified in the multi - cut matrix models . the prescription we adopt for studying off - critical amplitudes was developed in the one - cut two - matrix models @xcite , which we refer to as the dkk prescription . in the following , we first review their procedure then consider the extension to the multi - cut cases . at the @xmath13 one - cut critical points , the canonical pair of @xmath75 and @xmath76 operators was given as @xmath249 here we choose @xmath250 . considering the dependence on @xmath251 in the operators @xmath75 and @xmath76 is equivalent to turning on the coupling of the most relevant operator , @xmath121 . since we have dimensionful parameter @xmath121 , we can make a dimensionless combination with @xmath252 as @xmath253 = n \kappa t^n , \label{defofsmallz}\end{aligned}\ ] ] with @xmath254 . then the off - critical amplitudes of the canonical operators are generally expressed as @xmath255 note that the leading behavior of the polynomials @xmath256 and @xmath257 are @xmath258 so the commutation relation becomes @xmath259 = n^{-1 } \ , \frac{q\,\pi'_p(z)\ , \xi_q(z)-p\,\xi'_q(z)\,\pi_p(z)}{p+q-1},\end{aligned}\ ] ] and @xmath260 this was found in @xcite and is one of the main equation for off - critical amplitudes in the two - matrix models . it is known that this system has a simple solution for conformal background which is given by the chebyshev polynomials of the first kind , @xmath261 @xcite . in the case of unitary series @xmath262 , the equation ( [ dkkeq ] ) becomes the addition formula for the hyperbolic sine and cosine functions : @xmath263 consequently , the solution for @xmath264 is @xmath265 with the normalization factor @xmath266 which satisfies @xmath267 and @xmath268 . for the general @xmath13 cases , the chebyshev - polynomial is not a solution to eq . ( [ dkkeq ] ) . however , aside from the unitary series , the most relevant operator and cosmological constant are different in general ; considering this fact here , it means that there can be a jacobian factor in the eq . ( [ dkkeq ] ) in general . therefore , we will utilize this degree of freedom to deform the eq . ( [ dkkeq ] ) ; it was proposed in @xcite that the simple solutions with the chebyshev polynomials can be derived from @xmath269 here @xmath270 with @xmath271 . this procedure corresponds to utilizing the canonical pair @xmath272 of the cosmological constant @xmath14 @xcite , @xmath273 = \kappa \ , n\ , \mu^n,\end{aligned}\ ] ] instead of the canonical coordinates @xmath274 of the most relevant operator @xmath121 , @xmath275 = \kappa\ , n\ , t^n.\end{aligned}\ ] ] so in general the equation we have is @xmath276 in this procedure , the requirement of matrix model is that the functions of @xmath277 in the operators @xmath75 and @xmath278 are all polynomials in @xmath277 . one can expect that @xmath279 should be polynomials in general background . the solution for conformal background in the superstring cases was first obtained in @xcite . there are two classes of the solutions . the first class is called the one - cut - phase solution : @xmath280 the second is called the two - cut - phase solution : @xmath281 here we use @xmath282 and @xmath283 is the fzzt disk amplitude . realization of this solution in the pure - supergravity critical point was shown in @xcite . realization of @xmath18 solutions in the two - cut two - matrix models was shown in @xcite , with the sft formulation @xcite@xcite and its prescription utilizing the @xmath284 constraints @xcite . in the dkk prescription , what we need to consider is the douglas equation for the matrix - valued lax operators @xmath285 and @xmath286 of eq s ( [ aoptilde ] ) and ( [ coptilde ] ) . since @xmath285 and @xmath286 can generally be expanded in terms of pauli matrices @xmath287 , one can decompose the string equation in the following form : @xmath288\ , a^{-\frac{\hat p+\hat q}{2}}\lambda^{-(\hat p+\hat q ) } -\kappa i_2 = \text{(ccr0)}\ , \kappa^0 + \text{(ccr1 ) } \,\kappa i_2 + \text{(ccr2)}\ , \kappa \sigma_3+o(\kappa^2 ) . \label{doug2cut}\end{aligned}\ ] ] the one - cut solution is trivially the solution found in the bosonic case . the two - cut solution is less trivial but can be realized as @xmath289 with @xmath290 . one can easily show that this satisfies the douglas equations ( [ doug2cut ] ) . the curly bracket @xmath291 denotes anti - commutator which is necessary to satisfy ( ccr2 ) . the implications of this system are the following : * the zero - th order equation of the douglas equation , ( ccr0 ) , indicates that @xmath165 commutes with @xmath172 . that is , @xmath165 and @xmath172 can be diagonalized simultaneously . , corresponds to the fzzt brane amplitude with each different r - r charge . ] * eigenvalues of the matrices @xmath165 and @xmath172 satisfy the same equation ( [ dkkeq ] ) as in the bosonic case . the important difference is now the eigenvalues can include non - polynomial parts ( e.g. the square root in eq . ( [ 2cutsolsesh ] ) ) , as long as all the matrix elements remain to be polynomials in the matrix - model realization ( [ 2cutsoldkk ] ) . * usually , some naive expression of the matrix - model realization ( e.g. eq . ( [ 2cutsoldkk ] ) without the anti - commutator ) can not satisfy ( ccr2 ) . in this case , we need to add proper first order corrections to the operators @xmath165 and @xmath172 . and @xmath277 . but in the higher - cut cases , we need to add purely first order corrections to the scaling functions in the operators @xmath165 and @xmath172 . see section [ directanalysis11cases ] . ] now we can extend the previous discussion to the @xmath2-cut matrix models . here we will focus on the general considerations ; more detailed investigation of the three - cut string equation can be found in appendix [ directevaluationthreecutstringeq ] . for the sake of simplicity , we consider the models with the @xmath0 symmetric background . in this case , the operators @xmath285 and @xmath292 are restricted to @xmath293 the matrix - model requires that all the functions @xmath294 and @xmath295 are polynomials in @xmath277 . thus the diagonalization is just given by @xmath296 with @xmath297 and @xmath298 and @xmath299 . here @xmath300 and @xmath301 are all polynomials and the equality `` @xmath302 '' indicates diagonalization . then eq . ( [ dkkeq ] ) is expressed as @xmath303 since the left - hand side of the equation is a polynomial , we need to resolve the @xmath2-th root @xmath304 in the right - hand side . a natural ansatz is @xmath305 which is characterized by the two indices @xmath306 . then the equation turns out to be @xmath307 + { \nonumber}\\ & + \bigl[(\hat q-\hat p ) z + \hat p \gamma_{\hat p } - \hat q \gamma_{\hat q}\bigr ] \rho_{\hat p}(z)\ , \eta_{\hat q}(z)=\hat p+\hat q-1 , \label{cisyeq}\end{aligned}\ ] ] with @xmath308 this solution can be realized in the matrix model as @xmath309,\qquad \mathcal c^{\rm t}(n , z ) = -a^{\hat q/2 } \lambda^{\hat q } \eta_{\hat q } ( z ) \bigl [ \gamma^{-1 } z - \bar m\bigr],\end{aligned}\ ] ] with @xmath310 note that there is a sequence of @xmath97 @xmath266 s in @xmath311 and a sequence of @xmath97 @xmath312 s in @xmath313 . these matrices satisfy @xmath314 and @xmath315 = [ \gamma^{-1},\bar m]$ ] which result in ( ccr0 ) . of course , there is some ambiguity of taking permutation among the positions of @xmath312 and @xmath266 . where we explicitly study all possible solutions of the string equations in the three - cut cases . ] also we have the constraint of eq . ( [ honeconst ] ) , which means that @xmath316 with the above setting , we can solve equation ( [ cisyeq ] ) . from straightforward evaluation of eq . ( [ cisyeq ] ) , one observes that the parameters @xmath312 and @xmath266 satisfy @xmath317 except for @xmath318 case ; it can also be seen that @xmath319 is related to @xmath320 as @xmath321 for unitary series @xmath322 , one can see that they satisfy @xmath323 which leads to simple solutions of eq . ( [ cisyeq ] ) ; for instance , we list first five solutions of @xmath324 in the following : ) can generate polynomials ( i.e. solutions of eq . ( [ cisyeq ] ) ) in all the @xmath18 critical points not only in the unitary cases @xmath322 listed below . ] @xmath325 our claim is that the solutions for the unitary cases @xmath326 can be expressed by the jacobi polynomials , @xmath327 . in the diagonalized form ( [ diagac ] ) , the solution is given as @xmath328{\bigl(z - c\bigr)^l\bigl(z+c\bigr)^{k - l } } \times \omega { \nonumber}\\ \mathcal c^{\rm t}(n , z ) & \simeq a^{\hat q/2}\ , \frac{\gamma(\hat p)^2}{2c \gamma(2\hat p-1 ) } \bigl(2c k\lambda\bigr)^{\hat p } p^{(-\frac{2l - k}{k},\frac{2l - k}{k})}_{\hat p-1}(z / c ) \ , \sqrt[k]{\bigl(z - c\bigr)^{k - l}\bigl(z+c\bigr)^{l}}\times \omega^{-1 } , \label{cisysol}\end{aligned}\ ] ] with @xmath329 here @xmath330 is the beta function and also @xmath331 is given in ( [ defomega ] ) . note that the relation ( [ exchangeabxp ] ) is essentially the identity ( [ exchangeab ] ) of the jacobi polynomials , @xmath332 . we show that the above expressions of @xmath165 and @xmath166 solve eq . ( [ cisyeq ] ) for general indices @xmath306 in appendix [ jacobipproof ] . for the three - cut critical points of @xmath333 , some of the solutions can be written in terms of the chebyshev polynomials . since this solution is very similar to the solutions of one - cut and two - cut cases , it provides simple examples to see what happens in multi - cut cases . in the case of @xmath318 , the constraints ( [ honeconst ] ) turns out to be @xmath334 when we have @xmath324 , then @xmath335 . a solution which satisfies this constraint is given by @xmath336{\bigl(z - c\bigr)^2\bigl(z+2c\bigr)}\times \omega,{\nonumber}\\ \mathcal c^{\rm t}(n , z ) & \simeq a^{\hat q/2}\,c^{-1}\bigl(\frac{3c \lambda}{2}\bigr)^{\hat q } \bigl(2\cosh\bigl(\frac{3}{2}\tau\bigr)\bigr)^{\frac{1}{3}}\ , \cosh \bigl(\frac{2\hat q-1}{2}\tau\bigr)\times \omega^{-1}{\nonumber}\\ & = a^{\hat q/2}\,c^{-1}\bigl(\frac{3c \lambda}{2}\bigr)^{\hat q}\ , w_{\hat q-1}\bigl(z / c\bigr ) \sqrt[3]{\bigl(z - c\bigr)\bigl(z+2c\bigr)^2}\times \omega^{-1 } , \label{cheb3cutsol}\end{aligned}\ ] ] with @xmath337 . here the polynomial @xmath338 is the chebyshev polynomial of the fourth kind : @xmath339 this satisfies the deformed equation ( [ deformeddkkeq ] ) with the canonical pair @xmath340 , critical point in the @xmath341-symmetric 3-cut cases . ] @xmath342 = n \kappa \tilde\mu^n,\qquad w(z)\sim \cosh ( \hat q-2)\tau \sim t_{|\hat q-2|}(z / c ) , \ ] ] with some proper normalization . here we consider geometry of the multi - cut solutions we have obtained . the purpose is two - folded ; first , it serves as a check of our ansatz ( [ cisyansatz ] ) ; second , it provides some physical intuition to the multi - cut system . the algebraic equation of the @xmath318 solution ( [ cheb3cutsol ] ) is easily written down as @xmath343 where we define @xmath344 and @xmath345 are the eigenvalues of the operators @xmath346 and @xmath347 ( see eq . ( [ tolax1 ] ) ) . one can easily see that this equation is an algebraic equation in @xmath128 and @xmath11 . note that there are three branches , @xmath348 @xmath349 , of the algebraic equation , @xmath350 and this defines the three sheets of the riemann surface . these branches are related by the following shift of the parameter @xmath351 , @xmath352 in the expression of eq . ( [ cheb3cutsol ] ) , since this shift does not change the coordinate @xmath128 . this algebraic equation has singularities @xmath353 , @xmath354 at the origin @xmath355 and at @xmath356 of @xmath357 with @xmath358 and @xmath359 . the geometrical meaning of singular points is that different branches of riemann surface intersect at the points . since the singular point at the origin does not change under the shift of @xmath351 of eq . ( [ shiftbranches ] ) , this singular point joins the three sheets at the same time . on the other hand , the singular points given in ( [ singularpointspeq1 ] ) only join two sheets at the same time . the reason is following : the invariance of the singular points ( [ singularpointspeq1 ] ) under the shift ( [ shiftbranches ] ) is essentially given by the invariance of the factor @xmath360 in @xmath361 , that is , @xmath362 \qquad ( m=0,1,2).\end{aligned}\ ] ] since the @xmath341 symmetric @xmath318 critical points satisfy @xmath363 , the above condition is only satisfied by two different values of @xmath364 . this means that the singular points only join two sheets at the same time . the branch points @xmath365 of the function @xmath366 , @xmath367 appear when @xmath368 one can see that the singular points ( [ singularpointspeq1 ] ) appear between the origin and the branch points in the @xmath341 symmetric manner ( i.e. @xmath369 ) . the simplest curve ( the @xmath370 cases ) , @xmath371 is drawn in fig . [ 3cutgeometry ] . the topology of the curves is also shown in fig . [ 3cuttopology ] . note that there is a non - trivial handle in the algebraic curve . this comes from the cubic root in the curve . the appearance of non - trivial handle is explained in appendix [ appcubicroots ] , where some basic properties of cubic roots are summarized . case and @xmath372 , which is given by @xmath373 . the branch points are at @xmath374 @xmath375 . the singularity is at the origin @xmath376 . ] and ( b ) @xmath377 . the junction of three sheets is the singular point at the origin . the torus comes from the cubic - root cut . two - sheet junctions in ( b ) are the singular points of ( [ singularpointspeq1 ] ) , the number of which is nine . ] we also write the algebraic curve of the @xmath378 cases as @xmath379 with @xmath380 and @xmath381{\bigl(z-\frac{3f}{2}\bigr)^2\bigl(z+\frac{3f}{2}\bigr ) } , \quad q=\mu^{2/3 } \bigl(z-\frac{f}{2}\bigr)\,\sqrt[3]{\bigl(z-\frac{3f}{2}\bigr)\bigl(z+\frac{3f}{2}\bigr)^2}. \end{aligned}\ ] ] one can see that this solution also has three branch points . next we consider the counterpart solution of `` one - cut phase '' in the two - cut matrix models , i.e. @xmath382 . we here adopt the expression ( [ onecutthreecut ] ) derived from direct analysis of the string equation , and leave the details in appendix [ directevaluationthreecutstringeq ] . the diagonalization of the matrix operators gives @xmath383 here we put @xmath384 and @xmath385 ( see eq . ( [ defofsmallz ] ) ) . they give three algebraic equations : @xmath386 since the phase @xmath44 labels branches of the solution , the complete algebraic equation should be given by multiplication of these equations , @xmath387 this curve has no branch point but has three singular points which are at @xmath388 @xmath375 . one can also see that these singular points only join two sheets at the same time . the curve is drawn in fig . [ 3cutonecutgeometry ] . the topology of the curve is also shown in fig . [ 3cutonecuttopology ] . case and @xmath382 , which is given by @xmath389 . the singularities are at @xmath390 @xmath375 . there is no branch point , but we explicitly show the cut which indicates the eigenvalue condensation . this cut has a junction of three cuts , which is allowed in the cubic root . since there is no branch point attached by this cut , one can also push this cut away to infinity . ] . ( a ) is the case of finite @xmath14 and ( b ) is the case of @xmath391 . ] here in fig . [ 3cutonecutgeometry ] , we draw a single connected cut which does not attach any branch points . this cut has a junction of three cuts . this kind of junction is always allowed in general @xmath2-th root ( see appendix [ appcubicroots ] ) . since there is no branch point here , one can also push this cut away to infinity . although this cut can be pushed away , this has a physical meaning as the eigenvalue condensation . there is a similar phenomenon in the two - cut matrix models @xcite : in the one - cut phase of two - cut matrix models , the `` two cuts '' are completely connected with each other and form `` a single cut '' , which run along the whole real axes . since the branch points disappear in this procedure , one can also push this cut away to infinity . in our three - cut case , since the cubic - root system admits a @xmath341 symmetric cut which has a three - cut junction without any branch points , it is natural to interpret that this system still have this kind of cut which indicates the eigenvalue density . in this paper , we have provided quantitative analysis of @xmath18 critical points in the multi - cut two - matrix models based on an extension of the prescription developed by daul , kazakov and kostov in the one - cut two - matrix models @xcite . right at the critical points , we identified the minimal construction of critical resolvents and potentials which can give @xmath18 critical points of the multi - cut two - matrix models . the hermiticity of the lax operators realized in the multi - cut critical points is identified . we have also studied off - critical behavior of macroscopic loop amplitudes . in this study , we identified an ansatz for non - polynomial parts of amplitudes in the @xmath0 symmetric background . this ansatz enables us to generate several formulas for off - critical macroscopic loop amplitudes : the intriguing one is the formula written with the jacobi polynomials in the unitary cases @xmath322 . several future issues about our results are in order : * in identification of the critical points , we only consider the system of equations ( [ primitiveccr ] ) and ( [ steqmatrix ] ) . in this investigation , there is no essential difference between the real potential models and the @xmath34-rotated models , in the sense of existence of @xmath0 symmetry breaking critical points . as an independent check , one can perform direct evaluations using the monte carlo approach @xcite , since our real - potential solutions can avoid the difficulty of complex numbers in numerical analysis . * although we have found the formula for the unitary cases ( [ cisysol ] ) , the expression in terms of algebraic equation is missing . since underlying structure is multi - component kp hierarchy @xcite , they should be expressed by the algebraic equation of @xmath392 order , @xmath393 @xcite . * from our analysis , the system of douglas equation in the @xmath0 symmetric background admits one discrete parameter @xmath97 in the expression ( [ cisysol ] ) . the off - critical system is , however , controlled only by a dimension - full perturbation @xmath14 . this parameter @xmath97 seems mysterious because one may expect that there are at most two meaningful geometries which respect the sign of @xmath14 . the physical meaning of the parameter @xmath97 , therefore , should be an interesting question to be investigated . * since our analysis only concentrates on the possible asymptotic ( weak coupling ) geometries of macroscopic loop amplitudes , the consideration about interpolation between different asymptotic geometries ( which might be expressed by @xmath97 ) is necessary . this should be accomplished by non - perturbative analysis of the string equation as is in the two - cut one - matrix models @xcite . * since we have known about macroscopic loop amplitudes in the @xmath0 symmetric critical points , it should be interesting to identify the liouville continuum formulation corresponding to the @xmath0 symmetric critical points . * to see the relation to minimal fractional superstring theory , we need to calculate the amplitudes in the @xmath0 symmetry breaking critical points . since minimal fractional superstring theory is a natural generalization of bosonic and super string theories , the formula could be probably written by the jacobi polynomials of the ultraspherical sequence , @xmath394 . * one of the important features of the multi - cut matrix models is the @xmath0 charge of d - branes . in this sense , it should be important to see the annulus amplitude of macroscopic loop amplitudes @xcite . to see the correlations among @xmath0 charges through annulus amplitudes of the matrix models is also an interesting problem to be studied . * acknowledgment * the authors would like to thank shoichi kawamoto for the useful discussion and comments on this work . we also would like to thank chien - ho chen , wei - ming chen , kazuyuki furuuchi , pei - ming ho , hiroshi isono , shoichi kawamoto , jen - chi lee , feng - li lin , tomohisa takimi , dan tomino and wen - yu wen for attending minimal string study meeting and sharing valuable discussions . shih would like to show his gratitude to people in taiwan string focus group where most of the work has been done . ct is supported by national science council of taiwan under the contract no . 96 - 2112-m-021 - 002-my3 . the authors are also supported by national center for theoretical science under nsc no . 98 - 2119-m-002 - 001 . an intuitive understanding of the relation between the resolvents and orthonormal polynomials is following : there is a useful expression for the general orthonormal polynomials @xcite : @xmath395 here the expectation value is defined by path integral over the @xmath396 truncated matrices @xmath397 and @xmath398 as @xmath399 with @xmath400 . this implies the following powerful relation in the large @xmath401 limit , @xmath402 this relation has been argued and proved from several viewpoints @xcite @xcite @xcite . with this relation , solving the eigenvalue problems ( [ orthrecab ] ) and ( [ orthreccd ] ) in the large @xmath1 limit is translated into obtaining the resolvents of the matrix model . to compare our calculations with previous results in @xcite , we make a list of the critical potentials ( of real coefficients ) @xmath403 .. the gaussian integration over @xmath404 for the two - matrix model partition function gives back to the one - matrix models . however note that the potential @xmath403 here is defined as the usual one - matrix potential @xmath405 . ] note that our ( real - coefficient ) critical potentials are related to the ( complex - coefficient ) critical potentials in @xcite by replacing @xmath406 and @xmath407 . has a different meaning in the main text . ] the parameter @xmath312 corresponds to the critical point of eigenvalue ( @xmath408 ) , which one can always put zero , @xmath409 . the parameter @xmath410 is an irrelevant perturbation and essentially does not change the system but is necessary to obtain the critical points of odd @xmath411 ( one can always choose any finite non - zero number ) . @xmath412 the critical resolvents are @xmath413 defined by the operator @xmath414 as @xmath415 and @xmath416 here we turn on the coefficient @xmath418 in the potential @xmath419 and put @xmath250 . with suitable choices of the value of @xmath195 or @xmath90 ( one of them is a free parameter ) , we can obtain critical potentials with smaller coefficients . for example , two - cut @xmath420 unitary critical potential with @xmath38 is given as @xmath421 and one can see that the coefficients are larger than those of the @xmath250 case ( [ twocut(4,5)critiwtheta ] ) . also the critical point in eigenvalue space @xmath103 are chosen to be the origin @xmath422 . by making a shift of variable @xmath423 , we can eliminate the parameter @xmath312 like in appendix [ onematcritipotentials ] . [ [ the - unitary - cases ] ] the unitary cases : + + + + + + + + + + + + + + + + + + + the critical potential of the unitary cases @xmath424 ( i.e. @xmath239 unitary minimal superstrings ) are @xmath425 [ [ the - hat - phat - q - cases ] ] the @xmath18 cases : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the potentials @xmath426 of general @xmath427 cases are @xmath428 [ [ mathbb - z_2-breaking - cases ] ] @xmath123 breaking cases : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the @xmath123 breaking critical potentials ( @xmath410 is a breaking parameter ) are @xmath429 note that the critical potentials of @xmath430-rotated models are obtained by the analytic continuation , @xmath407 . [ [ the - unitary - cases - hat - phat - q ] ] the unitary cases ( @xmath322 ) : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the potentials of `` unitary '' cases @xmath424 are @xmath431 note that , for the case of @xmath318 , we consider the potential of @xmath432 . then the critical operator of @xmath433 is given as @xmath434 from the construction , the vicinity of @xmath435 is the @xmath436 critical point . in this class of solutions , we need to be careful about critical potentials with odd degrees . in general , negative roots for the critical potentials @xmath437 with @xmath438 , and the fermi sea is filled from origin to some point around the first negative root . however for @xmath318 , the fermi sea extends to negative infinity ( see eq . ( [ peq1cripotex ] ) with @xmath439 ) . hence we introduce a regularization parameter @xmath440 by eq . ( [ defepsilon ] ) . this will make the matrix model well - defined and we will also adjust the normalization factor @xmath195 accordingly . [ [ the - hat - phat - q1hat - q - cases ] ] the @xmath441 cases : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the potentials @xmath442 and @xmath443 of the @xmath333 case are @xmath444 since there is no @xmath445 critical point in three - cut cases , the above @xmath445 critical potential should result in the @xmath446 critical point . [ [ the - hat - phat - q - cases-1 ] ] the @xmath18 cases : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the potentials @xmath426 of general @xmath427 cases are @xmath447 [ [ the - unitary - cases - hat - phat - q-1 ] ] the unitary cases ( @xmath322 ) : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the potentials of `` unitary '' cases @xmath424 are @xmath448 [ [ the - hat - phat - q1hat - q - cases-1 ] ] the @xmath441 cases : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the potentials @xmath442 and @xmath443 of the @xmath333 case are @xmath449 since there is no @xmath446 critical point in the four - cut cases , the above @xmath446 critical potential should result in the @xmath450 critical point . [ [ the - hat - phat - q - cases-2 ] ] the @xmath18 cases : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the potentials @xmath426 of general @xmath427 cases are @xmath451 [ [ the - unitary - cases - hat - phat - q-2 ] ] the unitary cases ( @xmath322 ) : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the potentials of `` unitary '' cases @xmath424 are @xmath452 [ [ the - hat - phat - q1hat - q - cases-2 ] ] the @xmath441 cases : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the potentials @xmath442 and @xmath443 of the @xmath333 case are @xmath453 [ [ the - hat - phat - q - cases-3 ] ] the @xmath18 cases : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the potentials @xmath426 of general @xmath427 cases are @xmath454 we concentrate on the @xmath0 breaking critical points of the multi - cut matrix models which is characterized by the operators @xmath165 and @xmath172 of @xmath455 since they correspond to minimal fractional superstring theory @xcite . the parameter @xmath410 is the breaking parameter . the critical potentials of the @xmath34-rotated models are obtained by the analytic continuation , @xmath456 . we put @xmath250 . [ [ the - unitary - models - hat - qhat - p1 ] ] the unitary models @xmath457 : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the potentials @xmath458 of the unitary @xmath459 cases ( unitary minimal fractional superstring theory ) are @xmath460 & \hspace{8cm}\bigl(\theta= 5 + 21 \epsilon,\quad \epsilon \neq 1\bigr ) , \\[2pt ] & \underline{(\hat p,\hat q)=(2,3)}\quad \left\ { \begin{array}{l } { \displaystyle}v_1^{(3)}(x)= 3 g x^2 + \frac{13 x^3}{3 } -\frac{12 g x^5 } { 5 } -\frac{13 x^6}{6 } + \frac { g x^8}{8 } + \frac{x^9}{9 } , \cr { \displaystyle}v_2^{(2)}(y)= g y + \frac{5 g^2 y^2 } { 2 } -\frac{13 y^3}{3 } -\frac{5 g y^4 } { 4 } + \frac{y^6}{6 } , \end{array } \right . { \nonumber}\\ & \hspace{8cm}(\theta = 26 ) . \end{aligned}\ ] ] [ [ the - hat - phat - q1hat - q - cases-3 ] ] the @xmath441 cases : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the critical potentials @xmath461 and @xmath443 are @xmath462 [ [ the - unitary - models - hat - qhat - p1 - 1 ] ] the unitary models @xmath457 : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the potentials @xmath458 of the unitary @xmath459 cases ( unitary minimal fractional superstring theory ) are @xmath463 & \hspace{8cm}\bigl(\theta= 2 ( 4 + 15 \epsilon ) , \quad \epsilon \neq 1\bigr ) , \\[2pt ] & \underline{(\hat p,\hat q)=(2,3)}{\nonumber}\\ & \left\ { \begin{array}{l } { \displaystyle}v_1^{(3)}(x)= -\frac{25 g}{3 } x^3 -13 x^4 + \frac{125 g}{7 } x^7 + \frac{143 x^8}{8 } -\frac{25 g}{11 } x^{11 } -\frac{13 x^{12}}{6 } + \frac{g}{15 } x^{15 } + \frac{x^{16}}{16},\cr { \displaystyle}v_2^{(2)}(y)= g y + \frac{7 g^2}{2 } y^2 -\frac{13 y^4}{2 } -\frac{7 g}{5 } y^5 + \frac{y^8}{8 } , \end{array } \right . { \nonumber}\\ & \hspace{8 cm } ( \theta=65).\end{aligned}\ ] ] [ [ the - hat - phat - q1hat - q - cases-4 ] ] the @xmath441 cases : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the critical potentials @xmath461 and @xmath443 are @xmath464 [ [ the - unitary - models - hat - qhat - p1 - 2 ] ] the unitary models @xmath457 : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the potentials @xmath458 of the unitary @xmath459 cases ( unitary minimal fractional superstring theory ) are @xmath465 [ [ the - hat - phat - q1hat - q - cases-5 ] ] the @xmath441 cases : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the critical potentials @xmath461 and @xmath443 are @xmath466 in this appendix , we make a detailed analysis of the string equation in the three - cut matrix models . ] based on the formalism of @xmath2-component kp hierarchy @xcite@xcite . the motivation behind this analysis is to check that our ansatz ( [ cisyansatz ] ) corresponds to a special solution of lax pairs . on the other hand , the complete analysis also provides useful hints for other solutions in the multi - cut matrix models . the realization of @xmath0 symmetric @xmath18 critical points have been shown in eq s ( [ tolax1 ] ) and ( [ tolax2 ] ) . here we drop some irrelevant factors and the following simplified notation is used : @xmath467 the coefficient matrices @xmath468 and @xmath469 are all real functions . then the lax operators @xmath470 and @xmath471 are defined as @xmath472 with @xmath473 = [ { \boldsymbol{p}},{\boldsymbol{l}}]=0 $ ] and @xmath474 , and @xmath475 in the following discussion , we put @xmath476 for simplicity . the relation @xmath477 can be easily solved if derivative terms are neglected : @xmath478 with @xmath479 . the bracket @xmath480 is a totally symmetric product which is defined by @xmath481 here we define @xmath482\bigr],\qquad f = f^{(\rm diag)}+f^{(\rm off)}. \end{aligned}\ ] ] in the kp hierarchy basis @xmath483 , they are truly a diagonal part and an off - diagonal part of the matrix @xmath484 . since we impose the @xmath0 symmetry here , the kp operator @xmath19 corresponding to the matrix operator @xmath75 is given in eq s ( [ tolax1 ] ) and ( [ tolax2 ] ) as @xmath485 with @xmath486 then the equation @xmath473=0 $ ] gives @xmath487\bigr ] + ( \text{derivative terms } ) . \end{aligned}\ ] ] the operator @xmath20 for the @xmath333 case can be written @xcite as @xmath488 with @xmath489 . then the string equation can be easily derived from @xmath490.\end{aligned}\ ] ] one can easily see that @xmath491 ( mod . @xmath492 ) is not allowed as critical points . for example , in the first simplest case @xmath370 , the string equation is written as @xmath493+t \bigl[h,\gamma^2\bigr]\end{aligned}\ ] ] with @xmath494 . below , by imposing the information of the matrix model ( [ finhpeq1 ] ) , we solve this string equation at the first order of string coupling . there are four asymptotic solutions : the first one is generalization of `` one - cut solution '' in two - cut matrix models : @xmath495 with a shift of @xmath277 , this corresponds to our solution ( [ cisysol ] ) with the choice of @xmath382 . the others @xmath496 are essentially expressed as @xmath497 with @xmath498 they all satisfy @xmath499 . note that the forms of @xmath500 and @xmath501 @xmath496 are different from each other but eigenvalues are the same . with a shift of @xmath277 , this is equivalent to our ansatz ( [ cisyansatz ] ) and the solution ( [ cisysol ] ) of @xmath372 . we also note that the douglas equation , @xmath220=g_{str}i_3 $ ] , requires the following first order correction of string coupling : @xmath502 with @xmath503 this kind of first order correction can not be realized by the ordering problem of @xmath277 and @xmath121 like in the two - cut cases ( [ 2cutsoldkk ] ) . we consider the background @xmath504 . there are five asymptotic solutions : the first one is a generalization of `` one - cut solution '' in two - cut matrix models : @xmath505 other three solutions @xmath496 are essentially expressed as @xmath506 with @xmath507 these solutions precisely correspond to the ansatz ( [ cisyansatz ] ) . the last solution is different from the ansatz ( [ cisyansatz ] ) , and can be expressed as @xmath508 the functions @xmath509 and @xmath510 in @xmath511 ( see ( [ finhpeq1 ] ) ) are related as @xmath512 it would be of interest to make further study of these solutions . the kp operator @xmath19 which corresponds to the matrix operator @xmath165 is given in eq s ( [ tolax1 ] ) and ( [ tolax2 ] ) as @xmath513 with @xmath514 then the equation @xmath473=0 $ ] gives @xmath515+[h_2,s_{n-2 } ] \bigr ) \bigr ] + ( \text{derivative terms } ) . \end{aligned}\ ] ] the operator @xmath20 and the string equation for the @xmath378 case can be written as @xmath516 = g_{str } i_3.\end{aligned}\ ] ] there are also several equivalent solutions like the 2nd , 3rd and 4th solutions of the @xmath370 and @xmath446 cases . here we only list representative solutions . the first solution is the counterpart of the one - cut solution : there are three equivalent expressions . one of them is given as @xmath517 the parameter @xmath518 is fixed by the douglas equation @xmath220=g_{str}i_3 $ ] . this case is also related to our solution ( [ cisysol ] ) of @xmath382 . the second solution corresponds to our ansatz ( [ cisyansatz ] ) : there are also three expressions and one of them is given as @xmath519 the parameter @xmath518 is fixed by the douglas equation @xmath220=g_{str}i_3 $ ] . this case is also related to our solution ( [ cisysol ] ) of @xmath372 . here we write some basic facts about the jacobi polynomials , which we have used in this paper . some standard reference is @xcite , for example . the jacobi polynomial @xmath327 is a solution of the hypergeometric equation , @xmath520p_n^{(\alpha,\beta)}(z)=0,\label{jacobieq}\end{aligned}\ ] ] and is defined by @xmath521.\end{aligned}\ ] ] they are orthogonal polynomials with respect to the following inner product : @xmath522 the jacobi polynomials are related to other famous polynomials . the gegenbauer polynomials , @xmath523 , and the legendre polynomials , @xmath524 , are @xmath525 the chebyshev polynomials of the first kind , @xmath526 , and the second kind , @xmath527 , are @xmath528 ^ 2}\ , u_n(z).\end{aligned}\ ] ] they are also called the ultraspherical polynomials @xmath529 . the twisted type of ultraspherical polynomials , @xmath530 @xcite , can include the chebyshev polynomials of the third kind , @xmath531 , and the forth kind , @xmath338 : @xmath532 so we can see that @xmath533 . some useful relations are in order : + @xmath534 @xmath535 @xmath536 @xmath537 here we check that our formula ( [ cisysol ] ) @xmath538{\bigl(z-1\bigr)^l\bigl(z+1\bigr)^{k - l } } , { \nonumber}\\ \xi_{\hat q}(z ) & = p^{(-\frac{2l - k}{k},\frac{2l - k}{k})}_{\hat q-1}(z ) \ , \sqrt[k]{\bigl(z-1\bigr)^{k - l}\bigl(z+1\bigr)^{l } } , \end{aligned}\ ] ] generally satisfies eq . ( [ dkkeq ] ) @xmath539 in the case of @xmath322 . for sake of simplicity , we neglect normalization factors . what we need to show is that the derivative of eq . ( [ dkkeq2 ] ) vanishes , @xmath540 we first calculate @xmath541 as follows : @xmath542{\bigl(z-1\bigr)^l\bigl(z+1\bigr)^{k - l}}\,+{\nonumber}\\ & \qquad\quad\quad + 2\ , { \partial}_z p^{(\frac{2l - k}{k},-\frac{2l - k}{k})}_{\hat p-1}(z ) \times { \partial}_z \bigl(\sqrt[k]{\bigl(z-1\bigr)^l\bigl(z+1\bigr)^{k - l}}\bigr)+{\nonumber}\\[3pt ] & \qquad\quad\quad + p^{(\frac{2l - k}{k},-\frac{2l - k}{k})}_{\hat p-1}(z ) \times { \partial}_z^2\bigl(\sqrt[k]{\bigl(z-1\bigr)^l\bigl(z+1\bigr)^{k - l}}\bigr ) \biggr ] { \nonumber}\\ & = p^{(-\frac{2l - k}{k},\frac{2l - k}{k})}_{\hat p-1}(z ) \biggl [ ( z^2 - 1)\ , { \partial}_z^2 p^{(\frac{2l - k}{k},-\frac{2l - k}{k})}_{\hat p-1}(z ) + { \nonumber}\\ & \qquad+ 2\bigl(z+\dfrac{l}{k}-\dfrac{k - l}{k}\bigr)\ , { \partial}_z p^{(\frac{2l - k}{k},-\frac{2l - k}{k})}_{\hat p-1}(z ) - \dfrac{4\dfrac{l}{k}\bigl(\dfrac{k - l}{k}\bigr)}{z^2 - 1 } p^{(\frac{2l - k}{k},-\frac{2l - k}{k})}_{\hat p-1}(z ) \biggr ] { \nonumber}\\[3pt ] & = \biggl [ -\dfrac{4\dfrac{l}{k}\bigl(\dfrac{k - l}{k}\bigr)}{z^2 - 1 } + \hat p(\hat p-1 ) \biggr ] p^{(-\frac{2l - k}{k},\frac{2l - k}{k})}_{\hat p-1}(z)\times p^{(\frac{2l - k}{k},-\frac{2l - k}{k})}_{\hat p-1}(z)\end{aligned}\ ] ] here we use the following relations , @xmath543{\bigl(z-1\bigr)^{k - l}\bigl(z+1\bigr)^{l}}\times \sqrt[k]{\bigl(z-1\bigr)^l\bigl(z+1\bigr)^{k - l } } & = z^2 - 1 , \\[5pt ] \sqrt[k]{\bigl(z-1\bigr)^{k - l}\bigl(z+1\bigr)^{l}}\ , \frac{{\partial}}{{\partial}z } \bigl(\sqrt[k]{\bigl(z-1\bigr)^l\bigl(z+1\bigr)^{k - l}}\bigr ) & = z+\dfrac{l}{k}-\dfrac{k - l}{k } , \\ \sqrt[k]{\bigl(z-1\bigr)^{k - l}\bigl(z+1\bigr)^{l}}\ , \frac{{\partial}^2}{{\partial}z^2 } \bigl(\sqrt[k]{\bigl(z-1\bigr)^l\bigl(z+1\bigr)^{k - l}}\bigr ) & = -\dfrac{4\dfrac{l}{k}\bigl(\dfrac{k - l}{k}\bigr)}{z^2 - 1},\end{aligned}\ ] ] and the differential equation ( [ jacobieq ] ) , @xmath544p^{(\frac{2l - k}{k},-\frac{2l - k}{k})}_{\hat p-1}(z ) = \hat p(\hat p-1)\,p^{(\frac{2l - k}{k},-\frac{2l - k}{k})}_{\hat p-1}(z ) . \end{aligned}\ ] ] since the calculation of @xmath545 is similar to that of @xmath541 by exchanging the two indices : @xmath546 and one concludes @xmath547 this is the equation ( [ derdkkeq ] ) . one can calculate the constant in eq . ( [ dkkeq2 ] ) and fix the normalization of eq . ( [ cisysol ] ) by using eq s ( [ jacobinormalization1 ] ) and ( [ jacobinormalization2 ] ) . here we summarize basic properties of cubic roots of analytic functions and the topology of associated riemann surface , which serves as the simplest prototype of general @xmath2-th root cases . in the case of cubic root of a single complex variable , @xmath548{z^l}\ , \to\ , \sqrt[3]{(e^{2\pi i}z)^l } = \omega^{l } \sqrt[3]{z^l } \qquad ( l=0,1,2,\cdots),\end{aligned}\ ] ] we can define the charge @xmath549 of the branch point @xmath550 , according to the power of the fundamental phase @xmath97 , as @xmath551 . one can have branch cuts in the complex plane which connect branch points with vanishing total charges , @xmath552 . for cubic roots of an analytic function , there are two types of branch cuts : \(i ) type i branch cut connects two branch points of opposite charges . one concrete example is given by @xmath553{(z - a)(z - b)^2}.\end{aligned}\ ] ] here the branch point @xmath554 carries charge @xmath555 , and the branch point @xmath556 carries charge @xmath557 . the geometry of the associated riemann surface is given in fig . [ cubicroot1](a ) . by cutting the three complex planes along the branch cuts and gluing the six edges according to the identification rule , we can see clearly that the topology of the type i branch cut is given by a simple junction of the three sheets , as shown in fig . [ cubicroot1](b ) . \(ii ) type ii branch cut connects three branch points of identical charges . one concrete example is given by @xmath558{(z - a)(z - a\omega)(z - a\omega^2 ) } = \sqrt[3]{z^3-a^3}\end{aligned}\ ] ] here all three branch points @xmath559 carry charge @xmath555 . the geometry of the associated riemann surface is given in fig . [ cubicroot2](a ) . by cutting the three complex planes along the branch cuts and flipping the resulting complex plane into a hexagon ( including the point at infinity ) , one can transform the identification rule in fig . [ cubicroot2](a ) for three complex planes into a gluing rule for three hexagons . hence , the topology of the type ii branch cut is the same as that of a torus ( fig . [ cubicroot2](b ) ) . indicate the way of connection around the branch point @xmath560 . b ) cut the cuts out and open them up . after gluing the sheets , one can observe that the topology is torus . that is , the topological role of the second type cut is a junction of the three sheets with a torus . ] a. m. polyakov , `` quantum geometry of bosonic strings , '' phys . b * 103 * ( 1981 ) 207 ; `` quantum geometry of fermionic strings , '' phys . b * 103 * ( 1981 ) 211 . v. g. knizhnik , a. m. polyakov and a. b. zamolodchikov , `` fractal structure of 2d - quantum gravity , '' mod . lett . a * 3 * ( 1988 ) 819 . f. david , `` conformal field theories coupled to 2-d gravity in the conformal gauge , '' mod . phys . a * 3 * ( 1988 ) 1651 ; + j. distler and h. kawai , `` conformal field theory and 2-d quantum gravity , or who s afraid of joseph liouville ? , '' nucl . b * 321 * ( 1989 ) 509 . e. brezin and v. a. kazakov , `` exactly solvable field theories of closed strings , '' phys . lett . b * 236 * ( 1990 ) 144 ; + m. r. douglas and s. h. shenker , nucl . b * 335 * ( 1990 ) 635 ; + d. j. gross and a. a. migdal , `` nonperturbative two - dimensional quantum gravity , '' phys . * 64 * ( 1990 ) 127 . e. brezin , m. r. douglas , v. kazakov and s. h. shenker , `` the ising model coupled to 2-d gravity : a nonperturbative analysis , '' phys . b * 237 * ( 1990 ) 43 ; + d. j. gross and a. a. migdal , `` nonperturbative solution of the ising model on a random surface , '' phys . * 64 * ( 1990 ) 717 . d. j. gross and a. a. migdal , `` a nonperturbative treatment of two - dimensional quantum gravity , '' nucl . b * 340 * ( 1990 ) 333 . j. mcgreevy and h. l. verlinde , `` strings from tachyons : the @xmath561 matrix reloaded , '' jhep * 0312 * ( 2003 ) 054 [ arxiv : hep - th/0304224 ] . d. gaiotto and l. rastelli , `` a paradigm of open / closed duality : liouville d - branes and the kontsevich model , '' jhep * 0507 * ( 2005 ) 053 [ arxiv : hep - th/0312196 ] . d. j. gross and e. witten , `` possible third order phase transition in the large n lattice gauge theory , '' phys . d * 21 * ( 1980 ) 446 . v. periwal and d. shevitz , `` unitary matrix models as exactly solvable string theories , '' phys . lett . * 64 * ( 1990 ) 1326 ; `` exactly solvable unitary matrix models : multicritical potentials and correlations , '' nucl . b * 344 * ( 1990 ) 731 . m. r. douglas , n. seiberg and s. h. shenker , `` flow and instability in quantum gravity , '' phys . b * 244 * ( 1990 ) 381 . c. r. nappi , `` painleve - ii and odd polynomials , '' mod . lett . a * 5 * ( 1990 ) 2773 . c. crnkovic , m. r. douglas and g. w. moore , `` loop equations and the topological phase of multi - cut matrix models , '' int . j. mod . phys . a * 7 * ( 1992 ) 7693 [ arxiv : hep - th/9108014 ] . t. j. hollowood , l. miramontes , a. pasquinucci and c. nappi , `` hermitian versus anti - hermitian one matrix models and their hierarchies , '' nucl . b * 373 * ( 1992 ) 247 [ arxiv : hep - th/9109046 ] . t. takayanagi and n. toumbas , `` a matrix model dual of type 0b string theory in two dimensions , '' jhep * 0307 * ( 2003 ) 064 [ arxiv : hep - th/0307083 ] . m. r. douglas , i. r. klebanov , d. kutasov , j. m. maldacena , e. j. martinec and n. seiberg , `` a new hat for the @xmath561 matrix model , '' arxiv : hep - th/0307195 . i. r. klebanov , j. m. maldacena and n. seiberg , `` unitary and complex matrix models as 1-d type 0 strings , '' commun . * 252 * ( 2004 ) 275 [ arxiv : hep - th/0309168 ] . c. crnkovic and g. w. moore , `` multicritical multicut matrix models , '' phys . b * 257 * ( 1991 ) 322 . h. irie , `` fractional supersymmetric liouville theory and the multi - cut matrix models , '' nucl . b * 819 * ( 2009 ) 351 [ arxiv:0902.1676 [ hep - th ] ] . e. brezin , c. itzykson , g. parisi and j. b. zuber , `` planar diagrams , '' commun . * 59 * ( 1978 ) 35 . v. a. kazakov , `` the appearance of matter fields from quantum fluctuations of 2d gravity , '' mod . a * 4 * ( 1989 ) 2125 . i. k. kostov , `` strings embedded in dynkin diagrams , '' cargese 1990 , proceedings , random surfaces and quantum gravity , pp.135 - 149 . i. k. kostov , `` loop amplitudes for nonrational string theories , '' phys . b * 266 * ( 1991 ) 317 . i. k. kostov , `` strings with discrete target space , '' nucl . phys . b * 376 * ( 1992 ) 539 [ arxiv : hep - th/9112059 ] . t. banks , m. r. douglas , n. seiberg and s. h. shenker , `` microscopic and macroscopic loops in nonperturbative two - dimensional gravity , '' phys . b * 238 * ( 1990 ) 279 . g. w. moore , n. seiberg and m. staudacher , `` from loops to states in 2-d quantum gravity , '' nucl . b * 362 * ( 1991 ) 665 . j. m. daul , v. a. kazakov and i. k. kostov , `` rational theories of 2-d gravity from the two matrix model , '' nucl . b * 409 * ( 1993 ) 311 [ arxiv : hep - th/9303093 ] . m. fukuma and s. yahikozawa , `` comments on d - instantons in c @xmath562 1 strings , '' phys . b * 460 * ( 1999 ) 71 [ arxiv : hep - th/9902169 ] . j. m. maldacena , g. w. moore , n. seiberg and d. shih , `` exact vs. semiclassical target space of the minimal string , '' jhep * 0410 * ( 2004 ) 020 [ arxiv : hep - th/0408039 ] . m. r. douglas , `` strings in less than one - dimension and the generalized kdv hierarchies , '' phys . b * 238 * ( 1990 ) 176 . t. tada and m. yamaguchi , `` @xmath563 and @xmath11 operator analysis for two matrix model , '' phys . b * 250 * ( 1990 ) 38 ; + m. r. douglas , _ in * cargese 1990 , proceedings , random surfaces and quantum gravity * 77 - 83 . ( see high energy physics index 30 ( 1992 ) no . 17911 ) _ ; + t. tada , `` @xmath564 critical point from two matrix models , '' phys . b * 259 * ( 1991 ) 442 . m. fukuma , h. irie and y. matsuo , `` notes on the algebraic curves in @xmath13 minimal string theory , '' jhep * 0609 * ( 2006 ) 075 [ arxiv : hep - th/0602274 ] . m. fukuma and h. irie , `` a string field theoretical description of @xmath13 minimal superstrings , '' jhep * 0701 * ( 2007 ) 037 [ arxiv : hep - th/0611045 ] . n. seiberg and d. shih , `` branes , rings and matrix models in minimal ( super)string theory , '' jhep * 0402 * ( 2004 ) 021 [ arxiv : hep - th/0312170 ] . v. a. kazakov and i. k. kostov , `` instantons in non - critical strings from the two - matrix model , '' arxiv : hep - th/0403152 . a. sato and a. tsuchiya , `` zz brane amplitudes from matrix models , '' jhep * 0502 * ( 2005 ) 032 [ arxiv : hep - th/0412201 ] . n. seiberg and d. shih , `` flux vacua and branes of the minimal superstring , '' jhep * 0501 * ( 2005 ) 055 [ arxiv : hep - th/0412315 ] . m. fukuma , h. irie and s. seki , `` comments on the d - instanton calculus in @xmath565 minimal string theory , '' nucl . b * 728 * ( 2005 ) 67 [ arxiv : hep - th/0505253 ] . n. ishibashi , t. kuroki and a. yamaguchi , `` universality of nonperturbative effects in @xmath566 noncritical string theory , '' jhep * 0509 * ( 2005 ) 043 [ arxiv : hep - th/0507263 ] . f. david , `` phases of the large n matrix model and nonperturbative effects in 2-d gravity , '' nucl . b * 348 * ( 1991 ) 507 ; nonperturbative effects in matrix models and vacua of two - dimensional phys . b * 302 * ( 1993 ) 403 [ arxiv : hep - th/9212106 ] . p. h. ginsparg and j. zinn - justin , `` action principle and large order behavior of nonperturbative gravity , '' b. eynard and j. zinn - justin , `` large order behavior of 2-d gravity coupled to @xmath567 matter , '' phys . lett . b * 302 * ( 1993 ) 396 [ arxiv : hep - th/9301004 ] . m. fukuma and s. yahikozawa , `` nonperturbative effects in noncritical strings with soliton backgrounds , '' phys . b * 396 * ( 1997 ) 97 [ arxiv : hep - th/9609210 ] ; `` combinatorics of solitons in noncritical string theory , '' phys . b * 393 * ( 1997 ) 316 [ arxiv : hep - th/9610199 ] ; v. fateev , a. b. zamolodchikov and al . b. zamolodchikov , `` boundary liouville field theory . i : boundary state and boundary two - point function , '' arxiv : hep - th/0001012 ; + j. teschner , `` remarks on liouville theory with boundary , '' arxiv : hep - th/0009138 . a. b. zamolodchikov and al . b. zamolodchikov , `` liouville field theory on a pseudosphere , '' arxiv : hep - th/0101152 . m. l. mehta , `` a method of integration over matrix variables , '' commun . phys . * 79 * , 327 ( 1981 ) . e. j. martinec , `` on the origin of integrability in matrix models , '' commun . * 138 * ( 1991 ) 437 . m. fukuma , h. kawai and r. nakayama , `` continuum schwinger - dyson equations and universal structures in two - dimensional quantum gravity , '' int . j. mod . phys . a * 6 * ( 1991 ) 1385 ; `` infinite dimensional grassmannian structure of two - dimensional quantum gravity , '' commun . * 143 * ( 1992 ) 371 ; `` explicit solution for @xmath568@xmath569 duality in two - dimensional quantum gravity , '' commun . * 148 * ( 1992 ) 101 . n. kawahara , j. nishimura and a. yamaguchi , `` monte carlo approach to nonperturbative strings demonstration in noncritical string theory , '' jhep * 0706 * ( 2007 ) 076 [ arxiv : hep - th/0703209 ] . k. okuyama , `` annulus amplitudes in the minimal superstring , '' jhep * 0504 * ( 2005 ) 002 [ arxiv : hep - th/0503082 ] . h. irie , `` notes on d - branes and dualities in @xmath13 minimal superstring theory , '' nucl . phys . b * 794 * [ pm ] ( 2008 ) 402 [ arxiv:0706.4471 hep - th ] . g. w. moore , `` geometry of the string equations , '' commun . math . * 133 * ( 1990 ) 261 ; `` matrix models of 2-d gravity and isomonodromic deformation , '' prog . * 102 * ( 1990 ) 255 . m. sato , rims kokyuroku * 439 * ( 1981 ) 30 ; + e. date , m. jimbo , m. kashiwara and t. miwa , `` transformation groups for soliton equations . operator approach to the kadomtsev - petviashvili equation , '' rims-358 ; + m. jimbo and t. miwa , `` solitons and infinite dimensional lie algebras , '' publ . kyoto * 19 * ( 1983 ) 943 ; + v. g. kac and j. w. van de leur , `` the @xmath56-component kp hierarchy and representation theory , '' j. math . * 44 * ( 2003 ) 3245 [ arxiv : hep - th/9308137 ] . i. krichever , `` the dispersionless lax equations and topological minimal models , '' commun . * 143 * ( 1992 ) 415 . g. e. andrews , r. askey and r. roy , `` special functions , encyclopedia of mathematics and its applications 71 , '' cambridge university press ( 1999 ) 684 p.

according to the instanton liquid model ( ilm ) instantons are largely responsible for the spontaneous breaking of chiral symmetry and the low energy properties of light hadrons @xcite . this has been demonstrated using phenomenological instanton liquid models . the two main tasks involved in the construction are * producing an equilibrium instanton ensemble assuming some interaction between the pseudoparticles . this is typically done with a variational calculation @xcite . * computing hadronic observables using the above instanton ensemble as the gauge background . this is a formidable task and the computation necessarily involves several approximations . it would thus be very desirable to check the physical picture emerging from the ilm , starting from first principles . in the last few years there has been a considerable effort in this direction , using lattice simulations of qcd as a tool . the properties of the instanton ensemble have been extracted from monte carlo generated lattice configurations using several different `` smoothing '' techniques . the most important parameters , the topological susceptibility , the instanton density , and the average instanton size were found to be in qualitative agreement with the ilm . on the other hand although there is a considerable amount of indirect evidence @xcite so far the lattice has not given any direct results concerning the second part of the above programme . in ref . @xcite cooled lattice configurations were used to show that low eigenmodes of the dirac operator saturate the point - to - point hadronic correlators . more recently a strong correlation was found between the chiral density of low modes of the staggered dirac operator and the topological charge density @xcite . these results suggest that instantons probably really play an important role in qcd . all this work however has involved monte carlo generated or cooled lattices that contained other gauge field fluctuations in addition to the instantons . therefore , based solely on these studies it is impossible to separate the part of the effects due to the instantons and the part ( if any ) that is due to the rest of the gauge field fluctuations . to my knowledge , the only lattice study so far that makes such a separation possible has appeared in ref . @xcite . there , the instanton content of smoothed gauge configurations was reconstructed . hadron spectroscopy done on these artificial instanton backgrounds was compared to the spectroscopy on the corresponding smoothed configurations . while the smoothed ensemble yielded physically reasonable results for the pion and the rho mass , the instantons failed to reproduce this . in fact wilson spectroscopy on the instanton ensemble was hardly different from free field theory ( trivial gauge field background ) . in view of the positive results cited above , this was highly unexpected , especially that at first sight the only difference between the compared two ensembles appeared to be the presence versus the lack of confinement on the smoothed and the artificial instanton ensembles respectively . according to the ilm , confinement is not essential for the low energy properties of light hadrons @xcite . the findings of ref . @xcite indeed suggest that the physics in the chiral limit is rather insensitive to the string tension . with less than half of the original string tension , the low fermion modes on the cooled configurations still produced physically sensible hadron correlators . in the present paper i address the question of how to reconcile the insensitivity of light hadrons to the string tension ( also expected from the ilm ) with the failure of the instantons alone to reproduce the correct physics ( found in @xcite ) . i will exclusively consider the su(2 ) gauge group . along the way i will also obtain some useful insight into the structure of the low end of the spectrum of lattice dirac operators and in particular i will show the importance of using a ( close to ) chiral lattice dirac operator . this is essential for a faithful reproduction of continuum - like quark ( quasi)zero modes . the plan of the paper is as follows . in section [ se : zmz ] i briefly discuss the instanton liquid picture of physics in the chiral limit , and in particular the role of the zero modes and the so called `` zero mode zone '' . in section [ se : cs ] i point out the role of the explicit chiral symmetry breaking of the lattice dirac operator and the importance of minimising it . i show that the optimised clover dirac operator introduced in @xcite performs much better in this respect than the wilson operator . i explicitly study the zero mode zone in the simplest nontrivial case , that of an instanton antiinstanton pair and find that the mixing of zero modes closely follows the pattern expected in the continuum . in section [ se : spectrum ] i find that the optimised clover operator produces spectroscopy results markedly different from the wilson operator but it still falls short of reproducing the correct physics on the instanton backgrounds . instead , it yields a peculiar mixture of qcd and free field theory . i demonstrate that this is due to the presence of a large number of low lying free field modes in the spectrum of the dirac operator with the instanton backgrounds that produce artificially light hadrons . the free modes are completely absent from the smoothed configurations . they can be also eliminated from the instanton configurations by superimposing an ensemble of zero momentum gauge field fluctuations ( torons ) over the instantons . these gauge field fluctuations survive even deep in the perturbative regime . i argue that they are probably needed to compensate for the anomalously large small - volume effects present in the instanton ensemble due to the lack of confinement . by comparing the density of eigenvalues around zero on the physical ( smoothed ) , the toron , and toron plus instanton ensembles , i obtain direct evidence that the instantons are responsible for chiral symmetry breaking . after the elimination of the free quark modes , the instanton backgrounds are shown to reproduce the correct physical pion and rho correlators . section [ se : con ] contains my conclusions along with some speculations on still unanswered questions . finally , in the appendix , for reference i collected some general properties of wilson type lattice dirac operators that i used in the main text . some of these can be also found in the literature . in this section i describe how instantons can produce low lying modes of the dirac operator and why this is important for the physics in the chiral limit . for simplicity i start with the continuum theory and in the second part of this section i discuss the complications that appear on the lattice . let @xmath0 be the continuum massless dirac operator which implicitly depends on a gauge field background . the basic building block of the physics in the chiral limit is the @xmath1 limit of the massive quark propagator , @xmath2 . from the spectral decomposition in terms of eigenmodes of @xmath0 @xmath3 it is easily seen that in the chiral limit the most important modes are the ones with eigenvalues @xmath4 close to 0 . the crucial assumption of the ilm is that it is the instantons that are responsible for generating the bulk of these lowest eigenmodes . this can be qualitatively understood as follows . according to the atiyah - singer index theorem , in the background of an instanton ( antiinstanton ) @xmath0 has at least one negative ( positive ) chirality zero mode . in the presence of an infinitely separated instanton antiinstanton pair we still expect to find opposite chirality zero modes localised on the two objects . if the members of the pair are brought closer to one another , the two degenerate zero eigenvalues will in general split into two complex ones still close to the origin . a very simplistic but qualitatively correct description of this can be given as follows . let @xmath5 and @xmath6 be the zero modes of the dirac operators @xmath7 and @xmath8 respectively . since we consider @xmath0 in different gauge backgrounds , we explicitly indicate this ; @xmath9 and @xmath10 refer to an instanton and an antiinstanton . we now want to describe the spectrum of @xmath11 where @xmath12 is some superposition of the gauge fields @xmath9 and @xmath10 . since @xmath5 and @xmath6 are of opposite chirality , @xmath13 . let us arbitrarily complete the set of these two vectors into an orthonormal basis and construct the matrix of @xmath11 in this basis . the elements of this matrix in the subspace spanned by @xmath5 and @xmath6 are @xmath14 where @xmath15 the diagonal matrix elements vanish because @xmath0 maps left handed vectors into right handed ones and vice versa . by a suitable choice of the phases , the off - diagonal elements can always be made real and due to the anti - hermiticity of @xmath0 they are of equal magnitude and have opposite signs . if we now assume that @xmath16 and @xmath17 lie approximately in this two - dimensional subspace , i.e. the matrix of @xmath11 contains the above @xmath18 block - diagonal part , then @xmath11 can be easily seen to have two complex conjugate eigenvalues @xmath19 with the corresponding ( unnormalised ) eigenvectors being @xmath20 . this mechanism can be generalised to gauge fields which are superpositions of several instantons and antiinstantons . the basis in this case consists of the zero modes of the individual ( anti)instantons . the only additional approximation involved is the assumption that the zero modes corresponding to different ( anti)instantons are orthogonal . if the pseudoparticles are well separated , this is approximately true . one can thus consider the matrix of @xmath0 restricted to the subspace of zero modes called the `` zero mode zone '' in a general instanton background . if the off diagonal elements of this matrix are small then it will have complex eigenvalues close to the origin , in addition to a number of exact zero eigenvalues corresponding to the total topological charge . moreover , the complex eigenmodes will be approximately linear combinations of several instanton and antiinstanton zero modes and thus become highly delocalised , making it possible to propagate quarks to large distances . the most important quantities in this construction are the off diagonal instanton antiinstanton matrix elements , the @xmath21 s . in the simplest case of only one pair , @xmath21 is known to depend on the distance and the relative orientation of the members of the pair as @xmath22 where @xmath23 describes the relative orientation in group space , @xmath24 and @xmath25 is the relative position of the instanton and the antiinstanton @xcite . @xmath26 is a function of the instanton and antiinstanton scale parameters and the distance between them . at small distances it depends on the ansatz used to combine the field of the instanton and the antiinstanton , at large separation it is expected to fall off as @xmath27 . we have seen how instanton zero modes can give rise to the low - lying eigenmodes that are the most important ones in the chiral limit . close to the chiral limit the number of modes giving a substantial contribution to the quark propagator becomes smaller and smaller . it is thus crucial to reproduce these low eigenmodes and eigenvalues as faithfully as possible on the lattice if we want to test the ilm . the most important obstruction to this is that due to the nielsen - ninomiya theorem it is impossible to construct a local lattice dirac operator describing one fermion species with exact chiral symmetry @xcite . in the present paper i shall limit the discussion to wilson type lattice dirac operators that describe one fermion species but explicitly break chiral symmetry . in the absence of chiral symmetry the fermion zero modes corresponding to the topology of the gauge field are not protected ; any small fluctuation of the gauge field can shift the zero eigenvalues . in order to understand how this happens , a useful starting point is the following property shared by the lattice and the continuum dirac operator . let @xmath28 be an eigenvector of @xmath0 corresponding to the eigenvalue @xmath29 . if @xmath0 has no degeneracies then @xmath30 if and only if @xmath29 is real . this is a simple consequence of the so - called `` @xmath31 hermiticity '' , @xmath32 , a common property of the continuum and the lattice dirac operator . a proof of the above statement is given in the appendix . in the continuum , @xmath0 is anti - hermitian and has eigenvalues on the imaginary axis . thus the only real eigenvalue it can have is 0 . chirality makes a sharp distinction between zero and non - zero modes in the continuum . zero modes have a chirality of @xmath33 while all the non - zero eigenmodes have zero chirality . since the eigenmodes and their chiralities depend continuously on the gauge field , zero modes are protected , they can not be shifted away from zero by continuous deformations of the gauge field . on the other hand , the lattice dirac operator is not anti - hermitian , and its eigenvalues are not constrained to be on the imaginary axis . while lattice zero modes are still protected from being shifted off the real axis by smooth deformations of the gauge field ( this would cause their chirality to jump to zero ) , there is nothing preventing them from moving continuously along the real axis . indeed , a lattice dirac operator will in general not have any exact zero modes unless the gauge field is fine tuned . it is now clear that the lattice analogues of the continuum zero modes are real modes of the dirac operator . these have non - vanishing chiralities but in general their magnitude is smaller than 1 . the crucial role chiral symmetry plays in this discussion can be understood by noting that chiral symmetry of @xmath0 ( i.e. @xmath34 ) together with @xmath31 hermiticity would imply that @xmath0 is anti - hermitian and has protected zero modes . the smaller the explicit chiral symmetry breaking of @xmath0 is , the more its low lying spectrum will resemble that of an anti - hermitian operator . recently it has been shown that the closest a lattice dirac operator can come to being chirally symmetric is by obeying the ginsparg - wilson relation and thus having an ultralocal chirality breaking in the quark propagator @xcite . in this case the spectrum of @xmath0 lies on a circle in the complex plane passing through the origin . indeed the low end of this spectrum is almost on the imaginary axis and there are protected zero modes . i shall now discuss how the continuum description of the zero mode zone presented in the previous subsection has to be modified on the lattice . assume that there is an instanton on the lattice . in the limit when the instanton size goes to infinity ( in units of the lattice spacing ) , the instanton `` does not see '' the lattice at all , and the continuum description applies . the lattice dirac operator , @xmath7 has an exact zero eigenvalue with a corresponding @xmath35 chirality eigenmode @xmath5 . if the instanton is not infinitely large , the real eigenvalue gets shifted away from zero , its magnitude depends on the instanton size and the chirality breaking of @xmath7 @xcite . the corresponding eigenmode also becomes only an approximate eigenvector of @xmath31 and thus @xmath36 . let us now consider the dirac operator in the presence of an instanton antiinstanton pair . following the continuum discussion in the previous subsection , its matrix elements between the ( lattice approximate ) `` zero modes '' @xmath5 and @xmath6 can be parametrised as @xmath37 where @xmath38 is real , @xmath39 and @xmath40 are non - negative real numbers . in the su(2 ) case the diagonal elements are automatically real and the off - diagonal ones can be also made real by a suitable choice of the phases of the eigenvectors . this will be proved in the appendix . due to the @xmath31 hermiticity of @xmath0 , since @xmath5 and @xmath6 are approximate eigenvectors of @xmath31 , the off - diagonal terms have opposite signs and about the same magnitude . the eigenvalues of this matrix are @xmath41 with the corresponding ( unnormalised ) eigenvectors being @xmath42 the most important feature of these eigenvectors is that as in the continuum they are mixtures of the instanton and antiinstanton zero modes with roughly the same magnitude ( the ratio being @xmath43 ) and with a relative phase @xmath44 , up to a correction proportional to @xmath38 . this mechanism produces highly delocalised eigenmodes in the zero mode zone and thus facilitates the propagation of quarks by jumping from instanton to antiinstanton . maintaining this feature on the lattice is thus crucial for the description of the zero mode zone . the requirement for the proper mixing of zero modes is the inequality @xmath45 for producing qualitatively continuum - like eigenmodes , the difference of the diagonal elements has to be much smaller than the magnitude of the off - diagonal mixing matrix elements of @xmath11 . otherwise the two eigenvalues will be real and the corresponding eigenmodes rapidly become localised on the instanton and the antiinstanton respectively . recall that in the continuum , the mixing matrix elements @xmath46 are proportional to @xmath47 , where @xmath48 is the invariant angle of the instanton antiinstanton relative orientation ( see eq . ( [ eq : relor ] ) ) . @xmath38 ( and @xmath49 ) is expected to be of the order of the real eigenvalues of @xmath7 and @xmath8 , both small but nonzero numbers . this means that inequality ( [ eq : ineq ] ) will be violated for @xmath48 sufficiently close to @xmath50 . we can minimise the range in relative orientation , where this happens by making @xmath38 small , i.e. having the would - be zero modes close to zero . in this section i study how the general features of the zero mode zone described in the previous section are realised by two commonly used lattice dirac operators , the wilson and the clover operator . the gauge background i use for this test contains an instanton of size @xmath51 and an antiinstanton of size @xmath52 , separated by a distance @xmath53 in the time direction . the lattice size is @xmath54 and i always use antiperiodic boundary conditions in the time direction , periodic in all other directions . these parameters are typical in the spectroscopy calculations that i shall present in the next section . let me start with the wilson operator . in the presence of only the ( anti)instanton with the above parameters , @xmath55 has a real eigenvalue at ( 0.17 ) 0.51 with chirality ( 0.39 ) -0.53 . it is clear that for these small instantons the wilson operator does not even come close to satisfying the conditions of the previous section : the `` zero modes '' are far away from zero with chiralities of magnitude much smaller than 1 . the simplified description of the instanton - antiinstanton configuration in terms of the @xmath18 matrix is then inadequate . if the instanton and antiinstanton are close to being parallel @xmath56 then the real part of the ( anti)instanton related eigenvalues is around 0.3 . moreover already at @xmath57 the two complex eigenvalues become real and the corresponding modes are rather localised at the instanton and the antiinstanton respectively . a much better description of the zero mode zone is provided by the optimised clover action proposed in ref . ( see also @xcite for more discussion and a recent scaling test of this action . ) the clover coefficient was chosen to minimise the range in which the real eigenvalues occur in the physical branch of the spectrum on locally smooth gauge backgrounds . as explained in the previous section , this is exactly what is needed for a good description of the mixing of zero modes . the optimal value for @xmath58 was found to be around 1.2 on a set of ape smeared @xmath59 quenched su(3 ) configurations . physically these are quite similar to the su(2 ) configurations appearing in the present study , therefore in what follows i shall always choose @xmath60 . the real modes occurring in the single ( anti)instanton background already reveal that the clover action is much closer to the continuum than the wilson action . the eigenvalues are located at -0.014 and -0.002 and the chiralities of the corresponding eigenmodes are 0.993 and -0.995 respectively . in the instanton plus antiinstanton background the lowest two modes are always complex for almost any relative orientation of the pair , even at @xmath61 . we can hope that the simple description in terms of the mixing matrix works well in this case . indeed , the magnitude of the diagonal elements is small , neither @xmath38 nor @xmath29 goes above 0.02 in magnitude for any relative orientation . in fig . [ fig : t12 ] i plotted the two off - diagonal matrix elements , @xmath62 versus @xmath48 along with the one - parameter fits of the form const.@xmath63 . as expected , the mixing matrix elements are very well described by the continuum ansatz of eq.([eq : relor ] ) . a comparison of the eigenvalues of the @xmath18 mixing matrix and the explicitly computed eigenvalues of the lattice dirac operator in the corresponding gauge backgrounds shows agreement to within @xmath64 in the whole range of @xmath65 . we can conclude that the @xmath60 clover operator should give a satisfactory treatment of the zero mode zone as long as the instanton size does not drop below around @xmath51 . this is of course true only if the gauge configuration is locally smooth . otherwise the fermion mass acquires an additive renormalisation which completely destroys the above features important for the description of the zero mode zone . having a fermion action that is expected to describe the zero modes and their mixing reasonably well , we can discuss the main topic of the paper , namely , what part of qcd is reproduced by instantons alone . in this section i present quenched hadron spectroscopy results obtained with the @xmath60 clover fermion action . the gauge field backgrounds on which the spectroscopy is done are the ones used in ref . the starting point is a set of 28 @xmath54 su(2 ) configurations generated with a fixed point action . the lattice spacing is @xmath66fm , as fixed by the sommer parameter of the heavy quark potential . these monte carlo generated configurations are not yet suitable for our purposes since they are not locally `` smooth '' . both the instantons and the details of the zero mode zone are obscured by short distance fluctuations . cycling , a smoothing technique based on the renormalisation group , has been shown to preserve the main physical features of the gauge configurations . after 9 cycling steps the string tension @xcite and @xmath67 versus @xmath68 are essentially unchanged . moreover , these lattices are smooth enough that their instanton content can be unambiguously identified . in fact , about @xmath69 of the action is accounted for by instantons ( assuming there is no interaction between them ) . we can create artificial lattice configurations that have the same instanton content as the smoothed ones . the instanton sizes and locations are reproduced but the relative orientation in group space is distributed according to the su(2 ) haar measure . recently a study of the relative orientation appeared in @xcite , however reproducing the relative orientations would make our procedure much more cumbersome . i compare the spectroscopy done on the smoothed ( cycled ) lattices and the corresponding artificial instanton configurations with the same instanton content . these ensembles are both locally smooth enough that the optimised clover action should give a satisfactory description of their zero mode zone . in fig . [ fig : pirho ] the vector meson mass is plotted as a function of the pseudoscalar mass for both ensembles . the correlators were always fitted with the assumption that there is only one lightest particle dominating both the pseudoscalar and the vector channels . i also show the @xmath70 free field line . in fact , the wilson fermion action produces exactly degenerate free field like pions and rhos on the instanton backgrounds . compared to the wilson action , the clover action shows a marked improvement , but it still fails to reproduce the @xmath67 vs. @xmath68 observed on the smoothed lattices . [ fig : pi ] shows the pion mass sqared vs. the bare quark mass for both ensembles . on the smoothed ensemble @xmath71 , as expected from pcac . this is clearly not the case on the instanton ensemble , where a best fit to the form @xmath72 gives @xmath73 which is between @xmath74 and the free field value , @xmath75 . we also note that since both ensembles are locally very smooth , the additive mass renormalisations are very close to zero . therefore , it is also meaningful to compare hadron masses obtained at the same bare quark mass on the two gauge ensembles . this comparison reveals that the instanton ensemble typically produces much lighter hadrons than the smoothed ensemble . in spite of its substantially improved chiral properties , even the clover dirac operator fails to reproduce the correct physics of qcd on the instanton backgrounds . instead , it yields a peculiar mixture of qcd and free field theory . is there a problem with the instanton liquid model or the physics of these instanton lattices is still not close enough to the continuum ? to answer this question we look at the low lying eigenmodes of the dirac operator in more detail . in fig . [ fig : dspec ] i plot the lowest 30 eigenvalues in the complex plane for the instanton and the smoothed ensemble . the eigenvalues of all 28 configurations are superimposed . the boundary condition is antiperiodic in the time direction , periodic in all other directions . in both cases most of the eigenvalues lie very close to a circle passing through the origin . this is a sign of the approximate chiral symmetry of the action . the most striking difference between the two ensembles is the appearance of a gap at around im@xmath29=0.5 on the instanton ensemble . it is also instructive to compare the density of eigenvalues . since the low eigenvalues are almost imaginary i project them on the imaginary axis and plot the densities as a function of the imaginary part of the eigenvalues . this corresponds to the density of eigenvalues around zero in the continuum . in fig . [ fig : ev_hist1 ] we can compare the densities corresponding to the two ensembles . the real modes have been removed , these would show up as spikes at the origin . the densities at zero seem to agree quite well . according to the banks - casher relation the chiral condensate is proportional to the density of modes around zero @xcite ( excluding exact zero modes , which do not contribute in the infinite volume limit ) . this shows that on both the instanton and the smoothed ensemble chiral symmetry is broken and the value of the chiral condensate is approximately the same . apart from the vicinity of the origin , however , the two distributions are quite different . on the instanton ensemble there is a substantial `` piling up '' of modes above im(@xmath76 and then a `` thinning '' of modes at 0.5 . what is the reason of this huge difference between the two distributions ? the answer can be easily given by looking at the eigenvectors of the dirac operator . it turns out that the quark density of the eigenvectors on the instanton ensemble corresponding to the peak of the distribution are much more delocalised than all the other eigenvectors occurring in either ensemble . in fact , these modes are essentially free quark modes . this can be seen as follows . on a trivial gauge field configuration ( all links are 1 ) with periodic bondary conditions in all directions , there are @xmath77 trivial ( constant ) zero modes , this is the number of c - number degrees of freedom corresponding to one fermion flavour . in the presence of antiperiodic boundary condition in the time direction , the lowest eigenmodes are shifted away from zero , to @xmath78 since in our case @xmath79 . both eigenvalues are @xmath77-fold degenerate . the mixing of the free field modes into a particular eigenmode @xmath28 can be characterised by @xmath80 , where @xmath81 is the projection of the normalised eigenmode @xmath28 onto the eigenspace corresponding to the eigenvalue @xmath82 . this quantity is not gauge invariant therefore i work in landau gauge . any generic fermion mode on a non - trivial but locally smooth configuration will have a nonzero projection on the @xmath82 eigenspaces . the question is whether this is just an accidental mixing or the given ensemble really contains close to free field modes . to decide this , a good quantity to look at is @xmath83 . if the mixing is accidental , we expect @xmath84 to fluctuate around 1 , independently of the corresponding eigenvalue @xmath29 . on the other hand , if there are free field like modes on a given configuration then @xmath84 will increase substantially when @xmath29 approaches the free field eigenvalue @xmath85 . in fig . [ fig : proj ] i plotted @xmath84 versus the distance of the given eigenvalue from the free field mode @xmath85 . the figure shows a random selection of eigenvalues with imaginary parts between 0.0 and 0.2 . in the smoothed ensemble @xmath84 fluctuates around 1 everywhere , there is no trace of the free field modes . on the other hand , in the instanton ensemble , the modes close to @xmath85 have a substantially larger projection on the @xmath85 eigenspace than on the @xmath86 subspace . in fact , these modes , close to @xmath85 have @xmath87 , so they are essentially free field modes . a detailed study of the chiral density @xmath88 reveals that all the modes with @xmath89 look like mixtures of instanton zero modes with the density concentrated in several lumps . this is to be contrasted with the chiral density of the modes with @xmath90 that spreads almost homogenously over the whole lattice , as expected of the lowest free field eigenmodes . now we can understand why the instanton ensemble gave substantially smaller hadron masses than the smoothed ensemble . the reason is that the former had a large number of free quark modes . even if the zero mode zones of the two ensembles are similar , the free field modes provide a very efficient way to propagate quarks to large distances . this is why the instanton ensemble shows a peculiar mixture of qcd and free field characteristics . in the smoothed ensemble confinement completely eliminates the freely propagating quarks . according to the instanton liquid model , instantons alone , without confinement , can reproduce most of the properties of the light hadrons . our result shows that in the presence of instantons only , without confinement , free quark modes provide a very effective way to propagate quarks and they contaminate the hadron correlators producing unusually light hadrons . one possible way to suppress the free quark modes is to add to the instantons some locally very smooth background that contains only long wavelength fluctuations . there has to be enough long distance structure to suppress the free propagation of quarks but on the other hand they have to be locally smooth enough not to distort the instantons . cycling is a very efficient way to create such backgrounds . to this end i started with an ensemble of wilson @xmath91 ( @xmath92fm ) @xmath93 su(2 ) gauge configurations , performed 4 cycling steps and finally inverse blocked them to size @xmath54 . this produced a set of lattices extremely smooth on the few lattice spacing scale but containing the longest wavelength fluctuations characteristic to a @xmath94@xmath95 lattice . their spatial size was a bit smaller , their temporal size a bit larger than the confinement scale . moreover these configurations had a small enough physical size that they almost never contained instantons . after superimposing these lattices on the instanton configurations by simply multiplying the corresponding links ( in landau gauge ) , i checked that the original topological charge of the instanton configurations changed very little . to show how the addition of the smooth background affects the hadron correlators , in fig . [ fig : pirho_b2.4 ] i plotted typical pseudoscalar and vector correlators ( at @xmath96 ) . the change is dramatic . while on the instanton configurations the pion and the rho are very light ( crosses ) , the addition of the smooth background brings the correlators ( bursts ) very close to the physical ones obtained on the smoothed lattices ( boxes ) . the good agreement of the physical correlators and the instanton plus smooth background correlators persists in the whole range of masses where i tested it ( @xmath97 ) . it would be tempting to interpret this as a consequence of confinement ; the superimposed smooth backgrounds still contained the longest scale fluctuations of the confining lattices and these removed the free field modes that contaminated the spectroscopy before . but is confinement really needed for that ? this can be easily tested by replacing the superimposed backgrounds with ones of much smaller physical size . the lattices that i used for this purpose were generated in exactly the same way as before , except that the starting @xmath98 lattices were produced at wilson @xmath99 . these lattices have a tiny physical size , they are expected to be completely perturbative . in fig . [ fig : pirho_b3.0 ] i plotted again the pion and rho correlators at @xmath96 . quite surprisingly , the correlators have absolutely no dependence on the @xmath100 at which the background was created , they are the same on the instantons plus confining and instantons plus perturbative configurations . apparently , there is still something nontrivial on these physically very tiny perturbative lattices that has the same effect on the correlators as a confining background . what could this be ? the answer can be given by noting that the @xmath101 limit of yang - mills theory on the torus is not completely trivial because there is not only one gauge field configuration with zero action . the set of flat ( minimal action ) gauge field configurations can be characterised by the su(2 ) conjugacy classes represented by four mutually commuting gauge group elements corresponding to the ( constant ) polyakov loops in the four perpendicular directions . these are essentially zero momentum gauge field configurations , with a constant gauge potential ( link ) , sometimes referred to as torons . their contribution to physical quantities is a finite size effect , and has been estimated in ref . @xcite . to explicitly check that it is really the torons that affect the correlators so strongly , i extracted the `` toron content '' of the perturbative background fields . this was done in the following way . after landau gauge fixing i averaged the polyakov loops in all four directions . since the polyakov loops along any given direction were almost constant ( up to small perturbative fluctuations ) , the averages were close to being su(2 ) elements . i projected the averages back onto su(2 ) and then distributed the average polyakov loop evenly among the links in the given direction . this resulted in configurations with constant gauge fields ( links in any direction ) that carried the average polyakov loops of the original perturbative configurations . i shall refer to these lattices as `` toron '' configurations . we can now compare the correlators in the instantons plus toron background and the instantons plus the full perturbative background . the result again in the typical case of @xmath96 is shown in fig . [ fig : toron ] . the instantons and the torons together fully reproduce the correlators , demonstrating that the most important fluctuations on the perturbative lattices are indeed the torons . it is also quite instructive to compare the distribution of the lowest eigenvalues of the dirac operator on the different types of gauge configurations . [ fig : bc_hist ] shows the densities on the smoothed lattices ( solid lines ) , the perturbative ones ( dotted line ) and the lattices containing both the instantons and the perturbative backgrounds ( dashed line ) . the three distributions agree quite well away from zero . however around zero , the perturbative ensemble produces only a very low density ( compatible with zero ) , while the two other ensembles yield qualitatively similar densities , both nonzero . this demonstrates very clearly that it is the instantons that are responsible for creating a nonzero eigenvalue density around zero which through the banks - casher relation @xcite implies the spontaneous breaking of chiral symmetry . we also note that the peak corresponding to the free field modes ( see fig . [ fig : ev_hist1 ] ) has been completely eliminated by the perturbative backgrounds . this again can be attributed to the torons , as can be checked by looking at the eigenvalue distribution on the toron lattices ( not shown ) which is very similar to the dotted line in fig . [ fig : bc_hist ] . in the present paper i addressed the question of how much of qcd in the chiral limit is reproduced by restricting the fluctuations of the gauge field to instantons only . i reconstructed the instanton content of smoothed monte carlo lattice configurations and compared hadron spectroscopy on this instanton ensemble to the spectroscopy on the original `` physical '' smoothed configurations . the fermion action used for the spectroscopy was a `` chirally optimised '' clover action . i explicitly studied the fermion zero modes in the presence of an instanton and also the mixing of the zero modes corresponding to an instanton and an antiinstanton . i found that in these simple cases the optimised action described the continuum features of the zero mode zone rather well . in spite of this , the optimised action still failed to reproduce the physical hadron spectrum . it yielded anomalously light hadrons with a mixture of qcd and free - field like features . a closer look at the spectrum of the dirac operator revealed that on the instanton ensemble it had a large number of free quark modes . these were absent on the physical smoothed configurations . they provided a very efficient way of propagating quarks to large distances thereby substantially reduced the hadron masses and contaminated the spectroscopy . superimposing an ensemble of very smooth , essentially perturbative gauge field fluctuations on the instantons , was enough to eliminate the free quark modes and to restore the physical hadron correlators . by comparing the density of low eigenvalues of the dirac operator i also obtained direct evidence that it is the instantons that are responsible for creating a nonzero density of modes around zero and by the banks - casher relation also for chiral symmetry breaking . the important fluctuations on the superimposed perturbative lattices , needed to recover the physical hadron correlators , turned out to be zero momentum ( constant link or gauge potential ) configurations , torons . i would like to emphasise here that this toron ensemble was a perturbative one , produced at large @xmath100 i.e. small physical lattice size . the meson correlators appeared to be largely independent of the @xmath100 at which the superimposed background was created , as long as it was above the finite temperature phase transition on the given lattice size . it seems rather surprising that such `` mild '' gauge field fluctuations as the torons , can have so profound an impact on physical quantities . after all , torons exist only on the torus , they are absent in an infinite space - time and thus constitute only finite size effects . on the other hand we know that the instantons by themselves do not confine @xcite . this means that the instanton ensemble does not produce a mass gap and it is expected to exhibit much stronger finite size effects than qcd , from which the instantons have been extracted . in fact , the free field modes that are eliminated by the torons are also absent in the infinite volume limit , since they are not normalisable . in free field theory the density of modes around zero is proportional to the linear size of the box @xmath102 , whereas the density of instanton related modes grows proportionally to the volume . it is thus not inconceivable that in increasingly bigger volumes , although the torons have less and less influence , at the same time they become less and less needed since the relative importance of the free field modes also dies out . the torons might be needed only for compensating the unnaturally large finite size effects due to the absence of confinement in the instanton ensemble . what is the picture emerging from this lattice study concerning the instanton liquid model ? the qcd vacuum contains an ensemble of instantons that is generated by the non - perturbative gauge field dynamics . the instantons by themselves break chiral symmetry but hadron correlators in the instanton backgrounds are strongly contaminated by freely propagating quarks . this yields anomalously light mesons and a small splitting between the pseudoscalar and the vector channel . the free quarks can be eliminated and the physical hadron correlators can be restored by superimposing a perturbative ensemble of zero momentum gauge configurations ( torons ) on the instantons . this suggests that the contamination by free quarks is most likely a finite volume effect which , in qcd , is completely suppressed by confinement . more evidence in favour of this scenario could be gathered by studying the volume dependence of these effects on larger volumes . another important question , not answered by the present study is why it is exactly the perturbative toron ensemble that is needed to restore the physical meson correlators . it would be also very desirable to extend this investigation to a larger class of observables , in particular baryon correlators , for which the su(3 ) case need be considered . finally , for more precise tests a better chiral action , such as the overlap @xcite could be used . in the appendix , for reference , i collected some properties of wilson type lattice dirac operators that were used in the main body of the paper . some of these results can be found in the literature , in particular in refs . @xcite . i assume that the dirac operator satisfies @xmath31 hermiticity , i.e. @xmath103 let @xmath0 be a @xmath31-hermitian dirac operator with no degeneracy in its spectrum , i.e. having as many different eigenvalues as the dimension of the space it acts on . let @xmath29 , and @xmath49 be two eigenvalues of @xmath0 with the corresponding eigenvectors being @xmath28 and @xmath104 . then @xmath105 if and only if @xmath106 . _ proof : _ a simple consequence of @xmath31 hermiticity is that @xmath107 which implies @xmath108 it immediately follows that if @xmath109 then @xmath106 . to prove the converse we first note that since @xmath0 has as many different eigenvalues as the dimension of the space it acts on , its eigenvectors form a basis . we have already seen that the only eigenvector on which @xmath110 can have a nonzero projection , is @xmath111 . if this projection were also zero , then @xmath110 would be orthogonal to all the vectors in a complete set and consequently it would be zero . this is impossible since @xmath31 has a trivial kernel . for proving that @xmath112 it was essential that the eigenvalues of @xmath0 be non - degenerate and its eigenvectors form a complete set . generally this is expected to be the case unless the gauge field is fine tuned . there is however an important exception . when topology as seen by the fermions changes through a smooth deformation of the gauge fields , two opposite chirality real eigenvalues collide and leave the real axis as a complex conjugate pair . it can be shown that when the two eigenvalues coincide , the corresponding eigenspace is one - dimensional , the eigenvector has zero chirality , @xmath0 does not possess a complete set of eigenvectors and therefore has no spectral decomposition . a simple consequence of statement 1 is that if @xmath0 has a complete non - degenerate spectrum , then the chirality of an eigenvector , @xmath88 is nonzero if and only if the corresponding eigenvalue , @xmath29 , is real . in the remainder of the appendix i shall prove the following property of wilson type su(2 ) lattice dirac operators . let @xmath113 and @xmath114 be real eigenvectors of two ( not necessarily the same ) su(2 ) lattice dirac operators such that the corresponding eigenvalues and their complex conjugates are non - degenerate . it is then possible to choose the phases of @xmath113 and @xmath114 such @xmath115 , @xmath116 , @xmath117 , and @xmath118 are all real for any ( third ) lattice dirac operator @xmath0 . _ proof : _ besides @xmath31 hermiticity , this property depends on an additional symmetry of the dirac operator relating @xmath0 to its complex conjugate @xcite . this is specific to the su(2 ) case . let @xmath119 be the charge conjugation operator and @xmath120 , where @xmath121 is a pauli matrix acting in colour space . using that @xmath122 ( where the last equality follows because we work in a representation with all @xmath123 s hermitian ) , and that for any @xmath124 @xmath125 , it is not hard to prove that @xmath126 now if @xmath28 is an eigenvector of @xmath0 with eigenvalue @xmath29 then @xmath127 which means that @xmath128 is an eigenvector of @xmath0 with eigenvalue @xmath129 . due to the non - degeneracy of the eigenvalues and that @xmath130 , it follows that @xmath131 in the special case when @xmath29 is real , the phase choice @xmath132 eliminates the extra phase factor on the r.h.s . ( [ eq : phase ] ) . now writing @xmath133 , and making use of eqs . ( [ eq : dstar ] ) and ( [ eq : phase ] ) with the above phase choice , the non - diagonal matrix elements can be easily seen to be real . the same argument also shows that the diagonal matrix elements are automatically real , regardless of the phase choice . this completes the proof . we would like to note that while any wilson type dirac operator has a spectrum symmetric with respect to the real axis ( this follows from @xmath31 hermiticity ) , in general there is no simple relation between the eigenvectors corresponding to complex conjugate eigenvalues . the only exception is the su(2 ) case , when they are related by eq . ( [ eq : phase ] ) . i thank pierre van baal for discussions , tom degrand and anna hasenfratz for correspondence . the computations were performed using a computer code based on the milc @xcite code and the arpack @xcite package . i thank the milc collaboration and the authors of the arpack package for making their code publicly available . this work has been supported by fom . de forcrand , m. garca prez , and i .- o . stamatescu , nucl . phys . * b499 * ( 1997 ) 409 ; c. michael , and p.s . spencer , phys . * d52 * ( 1995 ) 4691 ; d.a . smith , and m.j . teper , phys . * d58 * ( 1998 ) 014505 ; b. alles , m. campostrini , a. di giacomo , y. gunduc , e. vicari , phys . * d48 * ( 1993 ) 2284 ; a. hasenfratz , and c. nieter , phys . * b439 * ( 1998 ) 366 . edwards , u.m . heller , j. kiskis , and r. narayanan , phys . * 82 * ( 1999 ) 4188 ; r.g . edwards , u.m . heller , j. kiskis , and r. narayanan , chiral condensate in the deconfined phase of quenched gauge theories , hep - lat/9919941 .

a double exchange model was introduced in @xcite to describe the motion of conduction electrons in the background of magnetic ions . in its simplest form the hamiltonian for the model can be written as : @xmath0 where @xmath1 is the electron annihilation operator on the site @xmath2 of the lattice , @xmath3 is a spin @xmath4 of an electron with implicit summation over repeated spin indices @xmath5 . ] at site @xmath2 , @xmath6 is a core spin . the first term of ( [ eq : de ] ) describes the hopping of conduction electrons between nearest neighbor sites @xmath2 and @xmath7 of the lattice . the hopping amplitude is @xmath8 . the second term is a ferromagnetic hund s interaction between the core spin and the spin of the conduction electron at the same lattice site . to specify the model completely we have to fix the total number of conduction electrons in the sample . we denote filling factor @xmath9 , where @xmath10 is a total number of sites of the lattice . in this paper we are interested @xcite in the limit of a very strong hund s interaction @xmath11 and of very large core spins @xmath12 . in ( [ eq : de ] ) we did not include any ( bare ) core spin interactions , e.g. , due to superexchange . the only interaction between core spins in the model ( [ eq : de ] ) occurs through the exchange by conduction electrons . because of an infinite hund s coupling @xmath13 an electron spin is always aligned with a core spin on any given lattice site . therefore , the effective hopping amplitude of an electron between two nearest neighbors is the largest when corresponding core spins are parallel . in this case an electron can hop between lattice sites without changing its spin orientation . one concludes that conduction electrons induce a ferromagnetic interaction between core spins @xcite . in the limit of large core spins @xmath12 we consider one can use an adiabatic approximation and think of core spins as of slow variables so that the conduction electrons ( fast variables ) are always in the ground state in the background of slowly moving core spins . it is tempting to average over the motion of conduction electrons ( to integrate conduction electrons out ) and obtain an effective spin model describing the dynamics of core spins ( and electron spins ) induced by charge degrees of freedom . in an almost ferromagnetic state the average spin per lattice site is @xmath14 corresponding to the sum of a core spin and an average electron spin . therefore , heuristically , one arrives to the model of ferromagnetically interacting _ fractional _ spins . however , this immediately raises the question of what one means by a fractional spin . although the average spin per lattice site is fractional , the total spin operator on a given site may take only values @xmath15 ( no electron on the site ) or @xmath16 ( there is an electron on the site ) . moreover , a fractional value of @xmath17 is incompatible with quantum mechanics of spins ( see section [ sec : singlespin ] ) . resolving this apparent controversy is a main goal of this paper . the paper is organized in the following way . the aim of the section [ sec : singlespin ] is mainly pedagogical . we establish notations and remind the reader how to write the path integral for a single spin . we show how the ambiguity of the berry phase for a single spin requires @xmath17 be an integer . for completeness we derive the berry phase for a single spin in appendix [ app : pathint ] using a fermionic determinant representation for the action . the reader familiar with the subject can start from the section [ sec : deinfj ] where we consider the double exchange model ( [ eq : de ] ) in the limit @xmath11 and @xmath12 . in section [ sec : de2 ] we formulate a double exchange model on two core spins and obtain our main result an effective action for core spins induced by a single conduction electron . this action ( namely , its kinetic part ) answers the question of how to write berry phase for a system of ( two ) fractional spins . appendix [ app : diagrams ] contains more accurate calculation of the effective action which uses the method of gradient expansion . we discuss the topology behind berry phase for fractional spins in section [ sec : tabp ] and conclude in section [ sec : concl ] . in this section we remind the reader how to write down the path integral for a single quantum spin and discuss topological reasons for spin quantization . quantum spin @xmath18 is defined as a representation of dimension @xmath19 of an @xmath20 lie algebra with commutation relations @xmath21 & = & 2s^{z } , \nonumber \\ \left[s^{z},s^{\pm}\right ] & = & \pm s^{\pm } , \label{eq : su2}\end{aligned}\ ] ] where as usual @xmath22 . an @xmath20 lie algebra has irreducible finite dimensional representations of dimension @xmath19 labelled by spin @xmath15 taking integer or half - integer values so that @xmath23 . we would like to construct the classical action for the time dependent unit vector @xmath24 which upon quantization by path integral reproduces ( [ eq : su2 ] ) . one could start directly from ( [ eq : su2 ] ) and construct the path integral using coherent states corresponding to an @xmath20 group ( see e.g. , ref.@xcite ) . in appendix [ app : pathint ] we give an alternative derivation of this path integral using the fermionic determinant representation . the result is given by partition function @xmath25 } , \label{eq : partition}\end{aligned}\ ] ] where @xmath26 & = & i ( 2s ) 2\pi \int_{\cal b } d^{2}x\ , \frac{1}{8\pi}\epsilon^{abc}\epsilon_{\mu\nu } n^{a}\partial_{\mu}n^{b}\partial_{\nu}n^{c}. \label{eq : wznw}\end{aligned}\ ] ] let us explain what ( [ eq : partition],[eq : wznw ] ) mean . first of all we used an euclidean time representation replacing real physical time @xmath8 by an imaginary time @xmath27 . the change to an imaginary time changes quantum amplitude @xmath28 used in path integral into a `` boltzmann '' factor @xmath29 . we impose periodic boundary conditions ( in imaginary time ) on @xmath30 so that @xmath31 and @xmath32 is a maximal span of an imaginary time . with these boundary conditions one can think of time as of a one - dimensional circle with identification of point @xmath33 with @xmath34 . physical results will be obtained in the limit @xmath35 after an analytical continuation back to the real time @xmath36 . imaginary time representation is slightly more convenient for calculations and for a discussion of topological aspects of spin quantization . secondly , the two - dimensional integration in ( [ eq : wznw ] ) is performed over some closed domain @xmath37 of two dimensional plane such that the boundary of this domain is identified with an imaginary time . the values of @xmath38 field in the area @xmath37 are defined as an extension from the physical trajectory @xmath30 so that @xmath39 is a smooth function of @xmath40 $ ] and auxiliary variable @xmath41 $ ] such that @xmath42 and @xmath43 . we shall refer to the action ( [ eq : wznw ] ) as to a berry phase @xcite for a spin ( see appendix [ app : pathint ] ) . although the action ( [ eq : wznw ] ) is given in terms of two - dimensional integral over auxiliary space , the variation of the integrand is a full derivative . therefore , the variation of berry phase ( [ eq : wznw ] ) is given by a one - dimensional integral over the `` physical '' ( imaginary ) time as @xmath44 we see that the variation of ( [ eq : wznw ] ) does not depend on the choice of an extension of @xmath30 in the unphysical domain @xmath37 . using ( [ eq : variation ] ) one can show that the equation @xmath45 , where @xmath46 is a coupling to an external magnetic field gives a classical equation of motion @xmath47 $ ] . the latter is the equation of precession of spin in magnetic field @xmath48 . partition function ( [ eq : partition ] ) with the action ( [ eq : wznw ] ) solve the problem of path integral representation for the algebra ( [ eq : su2 ] ) . namely , the commutation relations ( [ eq : su2 ] ) can be obtained as a result of quantization @xcite of `` classical action '' ( [ eq : wznw ] ) . we have already checked that an infinitesimal variation of berry phase @xmath49 $ ] does not depend on the extension of @xmath30 into the auxiliary domain @xmath37 . how about @xmath50 $ ] itself ? one can easily show ( see appendix [ app : pathint ] ) that the action ( [ eq : wznw ] ) is equal to @xmath51 where @xmath52 is a solid angle swept by a unit vector @xmath38 during its motion ( see fig.[fig : wznw ] ) . this angle is not uniquely defined : one can always add or subtract any angle which is a multiple of a full solid angle @xmath53 . the unit vector @xmath30 draws a closed line on the surface of a sphere with unit radius during its motion in imaginary time . berry phase is proportional to the solid angle ( shaded region ) swept by the vector @xmath30 . one can calculate this solid angle by extending @xmath38 into the two - dimensional domain @xmath37 as @xmath54 and calculating ( [ eq : wznw ] ) . ] therefore , @xmath50 $ ] is not unambiguously defined and is a multi - valued functional of @xmath30 . multi - valued functionals of this type were studied in both mathematics and quantum field theory ( see e.g. , @xcite ) and are often referred to as wznw term wess - zumino - novikov - witten term . although @xmath55 is a multiple - valued functional , in partition function ( [ eq : partition ] ) one needs @xmath56}$ ] which is a single - valued functional of @xmath30 under the condition that the coefficient @xmath17 is an integer . then @xmath57 and the boltzmann weights @xmath58 is defined unambiguously by physical configurations @xmath30 . one can think of this constraint on the value of @xmath17 as of the _ reason _ for spin to be an integer or a half - integer . having non - integer @xmath17 in ( [ eq : wznw ] ) amounts to having _ inconsistent _ theory . there is a one more representation of berry phase which is particularly convenient for the purposes of this paper . it is a so - called @xmath59 representation . we write @xmath60 , where @xmath61 is a complex two - component vector satisfying the constraint @xmath62 . the latter reproduces @xmath63 . in terms of @xmath64-variable we have for berry phase @xmath26 & = & -i ( 2s ) \int_{0}^{\beta } d\tau \ , z^{\dagger}i\partial_{\tau}z \label{eq : wznwcp1}\end{aligned}\ ] ] which is apparently independent from the auxiliary coordinate @xmath65 . this `` simplification '' is , however , deceptive because the equation @xmath60 does not define @xmath64 unambiguously . one can always change @xmath66 with any function @xmath67 satisfying @xmath68 . the latter is necessary to conserve periodic boundary conditions @xmath69 . this `` gauge '' transformation obviously does not change the vector @xmath38 . the change of wznw action ( [ eq : wznwcp1 ] ) under this transformation is @xmath70= i ( 2s)2\pi k$ ] where @xmath71 is any integer . we arrive to the same result : the action @xmath72 itself is multivalued while @xmath29 is single - valued under _ the condition _ that @xmath17 is an _ integer number _ ! let us now briefly discuss topological aspects of berry phase ( wznw term ) for a single spin ( [ eq : wznw ] ) . the classical unit vector @xmath38 from ( [ eq : wznw ] ) takes values on a unit two - dimensional sphere @xmath73 the target space of the problem . there exists nontrivial expression ( the integrand of ( [ eq : wznw ] ) ) depending on vector @xmath38 and its derivatives which is ( i ) local ( ii ) being integrated over two - dimensional sphere gives integer values for all possible smooth configurations of @xmath38 defined on @xmath73 . mathematically we say that there exists a closed but non - exact 2-form on @xmath73 ( target space ) or even more formally that the second cohomology group of @xmath73 is @xmath74 . this allows one to write down the topological term ( [ eq : wznw ] ) where an integration of the mentioned 2-form is performed over a domain @xmath37 with physical time as the boundary of this disk . the special properties mentioned above guarantee that ( [ eq : wznw ] ) unambiguously defines the path integral ( [ eq : partition ] ) . let us now consider the double exchange model ( [ eq : de ] ) and take a limit of infinite hund s constant @xmath75 . we will do it in the lagrangian formulation . first we use the wznw action ( [ eq : wznwcp1 ] ) to write the lagrangian ( in this section we use real time representation ) of the double exchange model corresponding to the hamiltonian ( [ eq : de ] ) @xmath76 - h. \label{eq : delagr}\ ] ] here @xmath77 is a chemical potential which is chosen to guarantee the correct filling factor of electrons and @xmath78 is a complex vector corresponding to the core spin @xmath79 at the site @xmath2 of the lattice . we will make the following change of variables in the lagrangian @xmath80 . here @xmath81 are spinless fermions , @xmath82 is a core spin variable , and @xmath83 is a complex vector orthogonal to @xmath82 , namely , @xmath84 . physically , @xmath85 is a component of an electron on the site @xmath2 with a spin parallel to the core spin at the same site and @xmath86 is the one with a spin antiparallel to the core spin . taking limit @xmath87 corresponds to neglecting @xmath88 component of an electron . we substitute @xmath89 into ( [ eq : delagr ] ) and omitting constant terms obtain @xmath90 \nonumber \\ & + & \sum_{i}(2s+\psi_{i}^{\dagger}\psi_{i } ) z_{i}^{\dagger}i\partial_{t}z_{i}. \label{eq : delagrproj}\end{aligned}\ ] ] the first term of ( [ eq : delagrproj ] ) is a kinetic term for projected electrons . let us take a close look at the second and the third terms . we assume that the motion of spins is much slower then the motion of electrons . this is justified in the limit of large core spins @xmath12 . then one can think of electrons as moving in the background of almost static @xmath64 s . the second term in ( [ eq : delagrproj ] ) describes the hopping of these electrons with the effective hopping amplitude @xmath91 . this amplitude is complex number and has both the magnitude and the phase . the magnitude of the effective hopping amplitude is given by @xmath92 or @xmath93 where @xmath94 is the angle between two core spins . in this derivation we used a relation @xmath95 and the known identity on pauli matrices @xmath96 . the absolute value of the effective hopping is proportional to the cosine of the half of the angle between core spins @xcite . it is maximal and equal @xmath8 for parallel core spins and vanishes for antiparallel spins . the phase of the effective amplitude @xmath97 is not defined unambiguously because of the gauge freedom ( see the previous section ) . consider , however , an electron moving along some closed trajectory in a background of a smooth configuration of core spins . then along the path we can write @xmath98 , where @xmath99 is small , and @xmath100 . the product of all the effective amplitudes along the path is @xmath101 , where we use parameterization @xmath102 of @xmath82 in terms of euler angles @xmath103 , and @xmath85 . along the closed path @xmath104 , so only the second term contributes to the phase picked up by an electron . it is easy to see that this second term is equal to the half of the solid angle enclosed by vectors @xmath38 along the trajectory of an electron . we see that ( projected ) electrons moving along some closed trajectory acquire the phase equal to the half of solid angle formed by spins along this trajectory . one can say , therefore , that they `` feel '' the solid angle formed by core spins as a flux of magnetic field . this flux is ( modulo @xmath105 ) a gauge invariant and well defined quantity . finally , the last term in ( [ eq : delagrproj ] ) is a `` modified '' kinetic term for core spins . one might expect that after the averaging over electrons this term will describe the dynamics of effective spins of magnitude @xmath106 . however , having non - integer coefficient in front of berry phase results in an inconsistent theory . to make this seeming paradox more precise we average over electrons accurately by integrating fermions @xcite out of ( [ eq : delagrproj ] ) . we obtain an effective action for core spins in terms of @xmath64-variables @xmath107 \nonumber \\ & + & \int dt\ , \sum_{i } 2s z_{i}^{\dagger}i\partial_{t}z_{i } . \label{eq : deeff}\end{aligned}\ ] ] here the expression in the determinant is both an operator in time and @xmath108 matrix in lattice sites . the matrix element @xmath109 is present only if @xmath110 are nearest neighbors . in the adiabatic limit there are two main effects produced by an integration over fermionic degrees of freedom . first , the energy of the filled ( to the filling factor @xmath111 ) fermi sea is added to the action . this leads to an effective ferromagnetic interaction between core spins . second , the correction to the berry phase of core spins must reflect the fact that spins of conduction electrons are effectively added to core spins . the latter correction is the matter of consideration of this paper . the variation of ( [ eq : deeff ] ) over @xmath112 immediately gives @xmath113 , where @xmath114 are some fractional numbers which in general depend on the configuration of core spins ( @xmath82 s ) . this implies the presence of the term @xmath115 in the action ( [ eq : deeff ] ) and confirms the result we expected from the heuristic `` averaging '' over electrons . however , the presence of this term alone renders theory inconsistent . in the next section we calculate the fermionic determinant of the type ( [ eq : deeff ] ) albeit for a simplified double exchange model on two lattice sites . we show that there is an additional term present in the effective action for core spins . this additional term will make the effective theory consistent as it is shown in section [ sec : tabp ] . the paradox of having fractional spin in the effective action for a double exchange model is present already for a much simpler double exchange model on two lattice sites with a single conduction electron . in addition the electron spectrum in such model is discrete which makes integration over fermions in adiabatic approximation to be a well - defined procedure . therefore , in this section we consider the double exchange model on two lattice sites with one conduction electron and show how to resolve the issue of fractional spin on this simple example . the projected ( @xmath87 ) double exchange model ( [ eq : delagrproj ] ) on two lattice sites ( in imaginary time ) @xmath116 \nonumber \\ & - & t\left [ \psi_{1}^{\dagger}(z_{1}^{\dagger}z_{2})\psi_{2 } + h.c.\right ] , \label{eq : delagrproj2}\end{aligned}\ ] ] where @xmath117 is a euclidean action . chemical potential @xmath77 must be chosen to guarantee that there is exactly one conduction electron in the system , i.e. , @xmath118 . one can repeat the heuristic argument of the previous section replacing @xmath119 in the second term of ( [ eq : delagrproj2 ] ) by its average value @xmath4 . one obtains two effective spins of magnitude @xmath120 instead of bare core spins . again , we have the problem of having the fractional coefficient in the berry phase term . we rewrite the lagrangian ( [ eq : delagrproj2 ] ) as @xmath121 + \psi^{\dagger}(\partial_{\tau}+i\mu+\hat{a})\psi . \label{eq : proj2}\ ] ] here we defined @xmath122 and @xmath123 matrix @xmath124 @xmath125 where we introduced explicitly the phase and the magnitude of an effective hopping amplitude @xmath126 the absolute value of an effective hopping amplitude @xmath127 , where @xmath128 is an angle between core spins @xmath129 and @xmath130 . it is equal to zero only for strictly antiparallel core spins @xmath131 . for fixed directions of core spins the matrix @xmath124 becomes constant matrix with eigenvalues @xmath132 corresponding to the two - level spectrum of an electron hopping between two core sites in double exchange model with infinite @xmath13 . two - level spectrum is symmetric and one can choose the chemical potential @xmath133 so that there is only one filled energy level at any given moment of time . then the spinless fermion @xmath134 ( electron with projected spin ) occupies the lowest energy level @xmath135 which is minimal ( @xmath136 ) when core spins are parallel . changing the angle between core spins requires the energy of the order of @xmath8 while we are interested in low energy dynamics with typical values of energy @xmath137 . therefore , we assume that @xmath138 does not vanish and is of the order of @xmath8 . the grassman variable @xmath139 corresponding to the conduction electron must satisfy antiperiodic boundary conditions in imaginary time @xmath140 : @xmath141 . we represent @xmath142 as a product of spinless fermion @xmath143 and spin @xmath4 boson @xmath144 , which after projection @xmath87 becomes @xmath64-boson corresponding to core spins . the boundary conditions for @xmath145 and @xmath146 are antiperiodic and periodic respectively . then we have for @xmath147 and @xmath148 ( see ( [ eq : proj2],[eq : notation ] ) ) @xmath149 where @xmath71 is an integer the winding number of the phase @xmath150 in imaginary time @xmath151 . the effective action of core spins is given by @xmath152 \,d\tau + w_{e}^{ind } , \label{eq : ea}\ ] ] where the effective action of core spins induced by fermions @xmath153 here the operator @xmath154 is linear differential operator with explicit dependence on time @xmath140 through the @xmath123 matrix @xmath155 and we put @xmath133 corresponding to having only one conduction electron in the system . to calculate the functional trace ( [ eq : ft2 ] ) we find the eigenvalues of this operator by solving differential equation @xmath156 with boundary conditions ( [ eq : boundarypsi],[eq : boundarya ] ) and calculate the sum of logarithms of all eigenvalues @xmath157 . matrix ( [ eq : a2 ] ) can be rewritten as @xmath158 where @xmath159 we would like to solve ( [ eq : equation1 ] ) with the matrix @xmath155 from ( [ eq : a20 ] ) in an adiabatic approximation , e.g. , assuming that all time derivatives are much smaller than the energy scale @xmath160 . first we `` unwind '' wavefunction @xmath161 by a unitary transformation @xmath162 where @xmath163 means the derivative of @xmath150 over @xmath164 . then , instead of ( [ eq : equation1 ] ) the function @xmath165 satisfies @xmath166 where we defined `` zero order '' operator @xmath167 and perturbation @xmath168 which is proportional to time derivatives of core spins as @xmath169 now we can solve ( [ eq : ptequation ] ) using the perturbation theory in @xmath168 . the boundary conditions of @xmath165 are given by ( [ eq : boundarypsi ] ) and ( [ eq : transformation ] ) as @xmath170 where berry phase @xmath171 where we used that @xmath172 ( @xmath71 is an integer , see ( [ eq : boundarya ] ) ) and that @xmath173 . we are going to find the eigenvalues of ( [ eq : ptequation ] ) using the perturbation theory in @xmath168 . this is justified in an adiabatic limit when @xmath174 proportional to the rate of change of core spins is much smaller than the typical fermionic energy @xmath138 or in other words when the conduction electron is the fast degree of freedom compared to core spins . the solutions of the zero order equation @xmath175 are labelled by an integer number @xmath176 and by plus or minus sign @xmath177 with eigenvalues @xmath178 where @xmath179 there are no corrections to these eigenvalues in the first order of perturbation theory in @xmath168 . therefore , up to the second order of perturbation theory we obtain @xmath180 where we have restored the chemical potential @xmath77 which should be @xmath133 for the problem with a single conduction electron . the effective action ( [ eq : ft2 ] ) is given by @xmath181 . \label{eq : ft3}\ ] ] to calculate the obviously divergent sum in ( [ eq : ft3 ] ) we regularize it by subtracting the effective action for the system without conduction electrons . that is we calculate the difference between ( [ eq : ft3 ] ) calculated at @xmath133 and the one calculated at @xmath182 . @xmath183 . \label{eq : ft4}\end{aligned}\ ] ] this sum is still logarithmically divergent but one can show that the divergent part is a constant not depending on physical parameters @xmath184 and @xmath185 . calculating the sum in ( [ eq : ft4 ] ) we obtain @xmath186 in the limit @xmath182 we obtain @xmath187 this is the final result for the effective action calculated at finite `` temperature '' @xmath32 . notice that the only imaginary part of a euclidean effective action is the berry phase @xmath185 . we have obtained this phase in the first order of perturbation theory . however , the same expression is also valid in all orders of perturbation theory . it is easy to show for our simple model because the higher orders of perturbation theory can only give corrections to the real part of ( [ eq : firstorder ] ) which does not contribute to the imaginary part of an euclidean action . we consider the higher orders of perturbation theory using the regular gradient expansion in appendix [ app : diagrams ] . in fact , the imaginary part of a euclidean action has a topological nature and , therefore , can not be changed by small perturbations ( see the discussion in section [ sec : tabp ] ) . in the `` zero temperature '' limit @xmath188 we have @xmath189 or explicitly @xmath190 the latter result can of course be obtained just by replacing the discrete sum over @xmath176 in ( [ eq : ft4 ] ) by an integral . for the full effective action ( [ eq : ea ] ) we obtain @xmath191 , \label{eq : kin } \\ w_{e}^{dyn } & = & -\int_{0}^{\beta}\delta\,d\tau + \ldots . \label{eq : dyn}\end{aligned}\ ] ] the first term of the dynamic part of an effective action @xmath192 is just the time integral of the energy of the ground state of the conduction electron in a fixed instantaneous configuration of core spins . the dots denote the higher order corrections to ( [ eq : dyn ] ) . these corrections are small if the background core spins are moving adiabatically and they are diverging if derivatives @xmath193 are bigger than @xmath138 . for such processes an adiabatic approximation fails . in particular , if core spins have opposite directions @xmath194 we have @xmath195 and an adiabatic approximation is not applicable at any rate of the change of core spins . the berry phase term ( [ eq : kin ] ) is the main result of this paper . we see that the the correction to the berry phase of bare core spins induced by the conduction electron is twofold . first , as we expected averaging over fermions in ( [ eq : delagrproj2 ] ) there is a change of the coefficient in berry phases of core spins from @xmath17 to @xmath196 . in addition , there is a correction @xmath197 , where the integer number @xmath71 is a winding number ( [ eq : boundarya ] ) of the phase @xmath150 in imaginary time . the latter term is a full time derivative and , therefore , does not change the classical dynamics of spins . classically , the effective `` core '' spins behave as if they had the value of @xmath120 ! let us show now that the full berry phase ( [ eq : kin ] ) is consistent with quantum mechanics of spins and , therefore , answers the question of how to write down the berry phase for two fractional spins @xmath120 . as we have seen at the end of the section [ sec : singlespin ] the inconsistency occurs if the partition function is not invariant under gauge transformations of @xmath64 variables . we perform the gauge transformation @xmath198 independently for first and second core spins . in order not to change the periodic boundary conditions of @xmath199 we require @xmath200 , where @xmath201 are arbitrary integer numbers . under this transformation the three terms of ( [ eq : kin ] ) depending on @xmath202 , @xmath203 , and @xmath150 change by @xmath204 , @xmath205 , and @xmath206 respectively . it is easy to see that neither of these three terms gives an invariant contribution to a partition function . e.g. , for @xmath207 the change of the first term is @xmath208 which gives @xmath209 being exponentiated . however , the total kinetic term ( [ eq : kin ] ) is equal to the sum of all these terms and changes under the gauge transformation by @xmath210 $ ] which is a multiple of @xmath105 and , therefore , does not change the weight @xmath211 . this means that neither of three terms considered makes any sense separately from the others . only their gauge invariant ( modulo @xmath105 ) sum has a physical meaning of the combined berry phase of the system of two core spins plus one conduction electron . we have seen in section [ sec : singlespin ] that the existence of wznw term for a single unit vector ( single spin ) is due to the fact that the target space @xmath73 has a nontrivial second cohomology group @xmath74 . the addition of wznw term to the action with the coefficient @xmath17 is consistent only if this coefficient is quantized ( is an integer number ) . for two independent spins the target space becomes @xmath212 and one obviously can write two independent wznw terms corresponding to @xmath213 . this is apparently true for two independent spins . however , the result is different for the conduction electron induced dynamics of core spins in the double exchange model . let us discuss why this is so . to obtain the induced dynamics of core spins we have excluded the charge degree of freedom of conduction electron integrating it out in an adiabatic approximation . this approximation , however , breaks when the the rate of the change of core spins becomes comparable to the energy scale @xmath214 given by the levels of conduction electron in the background of time independent core spins . there is a special point though when core spins are opposite to each other . at this point @xmath215 and an adiabatic approximation breaks no matter how slow the motion of core spins is . two degenerate states with antiparallel core spins . an electron with spin aligned along the core spin can not hop from one lattice site to another and can be either on the right or on the left lattice site . ] this is quite obvious from the physical point of view . consider two time independent core spins opposite to each other . the effective hopping amplitude of a conduction electron is @xmath216 and one has two different configurations . the electron can be either on the left lattice site or on the right one ( see figure [ fig : oppspins ] ) and the matrix element of the hopping term is zero between those states . this means that the ground state of the system is doubly degenerate . the charge degree of freedom ( left or right position of the electron ) is essential for this particular configuration of core spins and can not be integrated out . in our derivation we used an adiabatic approximation and neglected such configurations . this is justified in the limit @xmath12 for ferromagnetically interacting core spins . however , this approximation drastically changes the topology of the system . essentially , we prohibited the configurations @xmath194 . the relevant target space changes from @xmath212 for two independent spins to @xmath217 . the latter expression denotes the direct product @xmath212 with points @xmath194 ( topologically equivalent to one more @xmath73 ) removed . the resulting target space @xmath218 is topologically equivalent to @xmath73 . to prove it one can show is , in fact , homotopic to a tangent bundle to @xmath73 which is in turn can be contracted onto @xmath73 . ] that it can be contracted along itself onto @xmath73 . we illustrate the similar phenomenon on the corresponding one - dimensional example in figure [ fig : cutting ] . the contraction @xmath219 is shown as a one - dimensional illustration of @xmath220 . a ) the torus space @xmath221 is represented as a square with the opposite sides identified according to the direction of arrows . the diagonal shows the points given by @xmath194 . b ) the space @xmath219 is obtained by the removal of the diagonal . c ) one pair of opposite sides is glued . d ) the second pair is glued making a cylinder . e ) the cylinder is contracted onto the unit circle . ] therefore , the relevant target space @xmath222 . to check for allowed wznw terms we calculate @xmath223 . there is only one wznw term allowed for the chosen target space ! the existence of only one allowed topological term explains why only the full berry phase ( [ eq : kin ] ) is well defined . the most general form ( in @xmath59 representation ) of the topological term for two spins is given by @xmath224 . \label{eq : tterm}\end{aligned}\ ] ] coefficients @xmath225 satisfy quantization conditions ( or gauge invariance conditions ) . namely , @xmath226 and @xmath227 must be integers . we see that for the double exchange model of two core spins and a single conduction electron we have ( [ eq : kin ] ) which corresponds to particular integer values @xmath228 and @xmath229 . if by some reason the conduction electron spends unequal time at two sites ( e.g. , the local magnetic field acting on electron spin on one of sites is added ) we have an asymmetric situation @xmath230 . however , the combination @xmath226 and @xmath227 must stay integer to guarantee the consistency of the theory . in this paper we considered the infinite hund s coupling limit of the double exchange model with a given concentration @xmath111 of conduction electrons . in the limit of large core spins @xmath12 one can think of electrons and core spins as of fast and slow degrees of freedom respectively . the superficial averaging over the fast motion of electrons leads to an inconsistent theory with effective fractional spins . using the example of the double exchange model on two lattice sites we showed that in addition to the change of the value of the effective core spin on site from @xmath15 to @xmath231 a new topological term arises in the effective action . with this term the exponent of the effective action becomes single - valued and well defined . we showed that the full berry phase ( [ eq : kin ] ) reflects the change of topology of the system which occurs due to the adiabaticity condition the condition justifying the averaging over the charge degrees of freedom . the generalization of our results to the double exchange model for many spins is relatively straightforward . the only major difference with the case of two spins is that the average occupation number of a given lattice site by conduction electrons depends on the configuration of core spins . this makes the magnitude of a `` fractional '' spin on the given site a fluctuating quantity with correspondent modifications of berry phase terms . the latter effect is not expected to be strong in the ferromagnetically ordered state . we are grateful to boris spivak who stimulated us to think about fractional spins in ferromagnetic metals . the discussion of topological aspects of berry phase with dmitri ivanov is deeply appreciated . a. g. a. is alfred p. sloan research fellow . the work of ar . a. is supported by dr project 200153 at los alamos national laboratory . in this section for the sake of completeness we rederive the well - known path integral representation for the quantum mechanics of a single spin . we reduce the problem to a calculation of a simple fermionic determinant . the method we use differs from the commonly used coherent state representation ( see e.g. , @xcite ) for the path integral . it is conceptually simpler but involves the gradient expansion calculation of a fermionic determinant . it turns out that the similar determinant is needed for our analysis of a double exchange model for two core spins . it is easy to check that the representation @xmath232 with constraint @xmath233 is a faithful representation of ( [ eq : su2 ] ) with @xmath234 . here @xmath235 is a set of pauli matrices and @xmath236 with @xmath5 are annihilation ( creation ) fermi operators respectively which satisfy the following ( anti)commutation relations @xmath237 and @xmath238 . in fact , this representation reflects how spin operators appear in nature with @xmath134 being an annihilation operator of the physical electron . we consider the following hamiltonian @xmath239 , where @xmath240 is a spin of a fermion and @xmath138 is a strength of the coupling of this spin with the unit vector @xmath38 . the correspondent action is @xmath241\psi , \label{eq : actionspin}\ ] ] where we introduced a notation @xmath242 . one can think of @xmath243 as of the magnetic field acting on the fermionic spin @xmath18 . in the limit of an infinite coupling @xmath244 ( infinite magnetic field ) the fermionic spin @xmath18 will be `` frozen '' along the classical vector @xmath38 so that @xmath38 in the limit of infinite @xmath138 becomes a direction of quantization axis of a quantum spin of a fermion . if now we change the direction of the classical vector @xmath38 with time , the fermionic spin will follow the direction of @xmath38 . therefore , we expect that in the limit of an infinite @xmath138 an effective dynamics of classical vector @xmath38 induced by a fermion will be that of the quantum spin @xmath4 . the path integral representation for spin @xmath4 is then given by @xmath245 } , \label{eq : pathint}\end{aligned}\ ] ] where @xmath246=\lim_{\delta\to\infty}w^{eff}[\vec{n}]$ ] is a classical action corresponding to a quantum spin @xmath4 . this classical action is a functional of the trajectory of the time dependent unit vector @xmath24 . the effective action @xmath247 $ ] is defined as @xmath248 } & = & \int d\psi d\psi^{\dagger } \ ; e^{i\int dt\ , \psi^{\dagger } \left[i\partial_{t}+\delta\hat{n}\right]\psi}. \label{eq : fermint}\end{aligned}\ ] ] the functional integration in ( [ eq : fermint ] ) is performed over grassman variables@xcite @xmath249 classical analogs of fermionic annihilation ( creation ) operators . integrating over fermions we obtain @xmath250 & = & -i \ln \,{\rm det}\ , \left[i\partial_{t}+\delta\hat{n}\right ] . \label{eq : effdet}\end{aligned}\ ] ] to generalize the result to an arbitrary spin @xmath15 we can consider the same fermionic theory ( [ eq : fermint ] ) but with @xmath17 independent species of fermions coupled to the same vector @xmath38 . in this case all spins @xmath4 of fermions will be aligned along @xmath38 so that ( [ eq : pathint ] ) will correspond to quantum spin @xmath15 . integration over all fermionic species is independent and we have for an effective action @xmath250 & = & -i 2s \ln\,{\rm det}\ , \left[i\partial_{t}+\delta\hat{n}\right ] . \label{eq : effaction}\end{aligned}\ ] ] the only problem which is left is technical . it is to calculate the fermionic determinant ( [ eq : effaction ] ) in the limit of an infinite @xmath138 . one can obtain the result without ( almost ) any calculations . let us introduce the complex , two - component , normalized vector @xmath251 , @xmath252 so that it is an eigenvector of @xmath253 with the eigenvalue @xmath254 : @xmath255 . for a time independent @xmath253 this vector is a ground state of our hamiltonian @xmath256 with the ground state energy @xmath257 . the other eigenstate has an energy @xmath258 and in the limit of an infinite @xmath138 is never excited by a motion of the unit vector @xmath38 . therefore , in the limit of an infinite @xmath138 we can use @xmath259 with @xmath165 spinless ( projected ) fermion . substituting this into the action ( [ eq : actionspin ] ) we obtain @xmath260.\ ] ] the variable @xmath165 does not have any dynamics in the limit of an infinite @xmath138 . we average over @xmath165 using @xmath261 and obtain @xmath262 for spin @xmath4 . in case of spin @xmath15 one has @xmath17 species of fermions , @xmath263 and @xmath264 . this is the desired result for the action for a single spin @xmath15 . however , the main point of this paper is to show that the `` naive '' averaging over fermions instead of the careful determinant calculation can result in error . in this case one can show that due to topological reasons the calculation we performed is a legitimate one . to add even more weight to this statement we proceed with a formal calculation of the fermionic determinant ( [ eq : effaction ] ) and show that it gives the same result for a single spin action . before going to the calculation of fermionic determinant ( [ eq : effaction ] ) we will perform wick rotation of time and go to an imaginary time representation . imaginary time representation is slightly more convenient for calculations for reasons that will be clear later . the change to an imaginary time changes quantum amplitude @xmath28 into a `` boltzmann '' factor @xmath29 , where subscript `` e '' stands for `` euclidean '' . in imaginary time @xmath265 we have instead of ( [ eq : effaction ] ) @xmath266 & = & - 2s \ln\,{\rm det}\ , \left[\partial_{\tau}-\delta\hat{n}\right ] \label{eq : effactione}\end{aligned}\ ] ] with the partition function for a single spin @xmath25 } , \label{eq : pathinteapp}\end{aligned}\ ] ] where @xmath50=\lim_{\delta\to\infty}w_{e}^{eff}[\vec{n}]$ ] . we impose periodic boundary conditions ( in imaginary time ) on @xmath30 so that @xmath31 and @xmath32 is a maximal span of an imaginary time . with these boundary conditions we can think of time as of the one - dimensional circle with identification of points @xmath33 with @xmath34 . physical results will be obtained in the limit @xmath35 after analytical continuation back to real time @xmath36 . using the method presented in appendix [ app : diagrams ] one can calculate the effective action ( [ eq : effactione ] ) in adiabatic approximation with an arbitrary accuracy . however , we are interested in the limit @xmath244 . in this limit all terms of gradient expansion in @xmath267 except for the first ( topological ) one will vanish . here we show how to obtain the most important ( topological ) @xmath184-independent part of the effective action ( [ eq : effactione ] ) which is the exact result for the action for a single spin . determinant of an operator does not change if one applies a unitary transformation to it . this is , however , a potentially dangerous procedure in the context of gradient expansion as one has to make sure the boundary conditions for electrons do not change in the process of such transformation . it turns out that this is not an issue for the single spin problem . however , e.g. , for the double exchange problem the boundary condition of fermions do change in a non - trivial way under the unitary rotation ( see section [ sec : de2 ] ) . let us use the transformation @xmath268 such that @xmath269 . this transformation @xmath268 is easily found using the so called @xmath59 representation . in this representation we introduce a two - component complex vector @xmath64 with the constraint @xmath270 such that @xmath271 . then the matrix @xmath268 can be written as @xmath272 we write the equation ( [ eq : effactione ] ) in the form @xmath273 & = & - 2s \,{\rm{tr}\,}\ln \left[\partial_{\tau}+\hat{a}-\delta\sigma ^{3}\right ] , \label{eq : effactione2}\end{aligned}\ ] ] where @xmath274 and @xmath275 means both functional trace and the trace over @xmath276-matrices . the matrix elements of @xmath155 are proportional to the first time derivative of spin variables and are small in comparison to @xmath138 . we can expand the logarithm in ( [ eq : effactione2 ] ) in powers of @xmath155 . we keep only the first @xmath155 dependent term as the other terms will have additional powers of @xmath138 in denominators . ( one can obtain those terms in a regular way explained in appendix [ app : diagrams ] ) @xmath277 & = & 2s \,{\rm{tr}\,}\frac{\partial _ { \tau } + \delta \sigma ^{3 } } { -\partial ^{2}_{\tau } + \delta ^{2}}\hat{a } \nonumber \\ & = & i\int \frac{d\omega } { 2\pi } \,\mbox{tr}\,\frac{-\omega + i\delta \sigma ^{3 } } { \omega ^{2}+\delta ^{2}}\hat{a}(\omega = 0 ) \nonumber \\ & = & 2s\frac{1}{2}\int_{0}^{\beta } d\tau \,\mbox{tr}\ , \sigma ^{3}\hat{a } \nonumber \\ & = & - i 2s\int_{0}^{\beta } d\tau\ , z^{\dagger } i\partial _ { \tau } z. \label{eq : effactione3}\end{aligned}\ ] ] it is instructive to rewrite ( [ eq : effactione3 ] ) in terms of the original variables @xmath38 . we parameterize complex vector @xmath278 . it is easy to see that the angles @xmath128 and @xmath279 are the polar and azimuthal angles corresponding to the direction of the unit vector @xmath38 . the phase @xmath134 can be chosen arbitrarily as it does not affect @xmath38 . up to the full time derivative of @xmath134 we obtain @xmath280 , where @xmath52 is a solid angle enclosed by the unit vector @xmath38 during its time evolution . one can check that this solid angle can be rewritten in the form ( [ eq : wznw ] ) which is explicitly rotationally invariant . in this appendix we calculate the effective action ( [ eq : ft2 ] ) which can be written as @xmath281 where the differential operator @xmath282 is defined as @xmath283 with matrix @xmath155 given by ( [ eq : a2],[eq : a20 ] ) . we rewrite @xmath284 where @xmath285 , @xmath286 , and @xmath287 . we use an adiabatic approximation using the ( gradient ) expansion in @xmath267 . the operator @xmath282 contains the magnitude of the effective hopping @xmath138 which is slowly varying but it is not small itself . to avoid this difficulty we calculate instead of effective action its variation and then restore the result from this variation @xmath288 , \nonumber\ ] ] where @xmath289 let us now calculate the functional trace in the basis of plane waves @xmath290 here the integral over time @xmath140 is the calculation of the diagonal matrix element @xmath291 and integration over @xmath292 is the calculation of the functional trace . the residual trace @xmath293 is the trace in the space of @xmath123 matrices . `` pulling '' @xmath294 from right to the left we have @xmath295 in the latter expression the partial derivative @xmath296 is assumed to act `` to the right '' . this means that when we expand @xmath297 \nonumber \\ & = & \int dt\left[\delta l_{0}+\delta l_{1}+\delta l_{2}+\dots \right ] , \label{eq : expansion}\end{aligned}\ ] ] all derivatives are assumed to act on all terms to the right . e.g. , @xmath298 . we introduced @xmath299 and @xmath300 we notice that @xmath301 because @xmath302 by adiabatic approximation ( it contains time derivative of core spins ) . therefore , the denominator of ( [ green ] ) is never zero and integrals ( [ deltal ] ) are not singular . the expansion ( [ eq : expansion ] ) is essentially the gradient expansion , i.e. , the expansion in the number of time derivatives . one can see that @xmath303 contains at least one time derivative , @xmath304 at least two etc . the calculation of @xmath305 is straightforward and gives @xmath306 \label{deltal0}\ ] ] or removing variation @xmath307 to obtain the latter expression we expanded up to the second order in time derivatives using @xmath308 . let us now prove that the whole dependence of the full result ( [ eq : expansion ] ) on @xmath309 comes from @xmath310 . varying the action ( [ eq : proj2 ] ) over @xmath309 we obtain @xmath311 . this statement is exact because the total number of electrons in the system is fixed @xmath312 . for the effective action we have , therefore , @xmath313 . however , from ( [ deltal0 ] ) we have @xmath314 which gives @xmath315 and all higher order terms @xmath303 , @xmath304 etc . do not depend on @xmath309 . this simplifies further calculations allowing to use @xmath316 for all higher order terms . as a result of calculations one obtains from ( [ deltal ] ) @xmath317 this is identical in form to a berry phase of a single spin @xmath4 given by ( [ eq : variation ] ) with identification @xmath318 . therefore , @xmath303 is given by formula analogous to the berry phase for a single spin ( [ eq : wznw ] ) and is to be interpreted as a half of solid angle swept by vector @xmath319 during its time evolution . assume for a moment now that @xmath320 . then the vector @xmath321 moves along the equator of unit sphere and covers the solid angle @xmath322 during its motion . in this approximation we obtain @xmath323 which gives half of the solid angle after time integration . is known as the method of embedding in quantum field theory @xcite ] calculating corrections due to nonzero @xmath324 from ( [ eq : l1 ] ) we obtain @xmath325 where we expanded the latter expression up to the terms of the second order in time derivatives . finally , combining terms ( [ fulll0],[fulll1],[fulll2 ] ) we obtain an effective action up to the second order terms @xmath327 it is interesting to notice that although neither of contributions ( [ fulll0],[fulll1],[fulll2 ] ) is gauge invariant , the full action ( [ fullle ] ) is gauge invariant . this means that the final result ( [ fullle ] ) can be rewritten in terms of two unit vectors @xmath328 and @xmath329 which represent physical degrees of freedom core spins of double exchange model ( [ eq : de ] ) .

modern atomic spectroscopy studies the structure and properties of practically any atom of the periodic table as well as of ions of any ionization degree . particular attention is paid to their energy spectra . for the investigations of many - electron atoms and ions , it is of great importance to combine experimental and theoretical methods . nowadays the possibilities of theoretical spectroscopy are much enlarged thanks to the wide use of powerful computers . theoretical methods utilized must be fairly universal and must ensure reasonably accurate values of physical quantities studied . many - electron atom usually is considered as many - body problem and is described by the wave function constructed from the wave functions of one electron , moving in the central nuclear charge field and in the screening field of the remaining electrons . then the wave function of this electron may be represented as a product of radial and spin - angular parts . the radial part is usually found by solving various modifications of the hartree - fock equations and can be represented in a numerical or analytical forms ( froese fischer @xcite ) whereas the angular part is expressed in terms of spherical functions . then the wave function of the whole atom can be constructed in some standard way ( cowan @xcite , jucys and savukynas @xcite , nikitin and rudzikas @xcite ) starting with these one - electron functions and may be used further on for the calculations of any matrix elements representing physical quantities . for obtaining the values of atomic quantities it is necessary to solve so - called eigenvalue problem @xmath1 where @xmath2 is the wave function of the system under investigation and @xmath3 is its hamiltonian . in order to obtain accurate values of atomic quantities it is necessary to account for relativistic and correlation effects . it turned out that for very large variety of atoms and their ionization degrees the relativistic effects may be taken into account fairly accurately as breit - pauli corrections ( nikitin and rudzikas @xcite , rudzikas @xcite ) . it is convenient to present the hamiltonian as consisting of two parts in breit - pauli approach , namely , @xmath4 where @xmath5 is the non - relativistic hamiltonian and @xmath6 stands for the relativistic contibution . non - relativistic hamiltonian is the sum of the kinetic energy of electrons @xmath7 , their potential energy @xmath8 , and the electrostatic electron interaction @xmath9 @xmath10 the first two terms , @xmath7 and @xmath8 are the one - particle operators , whereas the third term , @xmath9 , is two - particle operator . the @xmath6 may also be subdivided into non - fine structure and fine structure contributions @xmath11 the non - fine structure contributions @xmath12 shift non - relativistic energy levels without any splitting of them . the mass - velocity term @xmath13 describes the variation of mass with velocity . the one- and two - body darwin ( contact ) terms @xmath14 and @xmath15 are the corrections of the one - electron dirac equation due to the retardation of the electromagnetic field produced by an electron ( contact interaction ) . the spin - spin - contact term @xmath16 accounts for the interaction of the spin magnetic moments of two electrons occupying the same space . the orbit - orbit interaction @xmath17 accounts for the interaction of two orbital moments . the fine - structure contributions @xmath18 split the non - relativistic energy levels ( terms ) into a series of closely - spaced levels . the most important of these is the nuclear spin - orbit interaction @xmath19 ( spin - own - orbit ) representing the interaction of the spin and angular magnetic moments of an electron in the field of the nucleus . the spin - other - orbit @xmath20 and spin - spin @xmath21 contributions in a rough sense may be viewed as corrections to the nuclear spin - orbit interaction due to the presence of the other electrons in the system . unfortunately , practical calculations show that all realistic atomic hamiltonians do not lead straightforwardly to eigenvalue problem ( [ eq : sch ] ) . actually we have to calculate all non - zero matrix elements of the hamiltonian considered including those non - diagonal with respect to electronic configurations , then to form energy matrix , to diagonalize it , obtaining in this way the values of the energy levels as well as the eigenfunctions ( the wave functions in the intermediate coupling scheme ) . the latter may be used then to calculate electronic transitions as well as the other properties and processes . such a necessity raises special requirements for the theory . the total matrix element of each term of the energy operator in the case of complex electronic configuration will consist of matrix elements , describing the interaction inside each shell of equivalent electrons as well as between these shells . going beyond the single - configuration approximation we have to be able to take into account in the same way non - diagonal , with respect to configurations , matrix elements . to find the expressions for the matrix elements of all terms of the hamiltonian considered for complex electronic configurations , having several open shells , is a task very far from the trivial one . a considerable part of the effort must be devoted to coping with integrations over spin - angular variables , occurring in the matrix elements of the operators under consideration . this paper presents the general methodology , leading to optimal expressions for operators and matrix elements . a number of methodologies to calculate the angular parts of matrix elements exists in literature . many existing codes for integrating the spin - angular parts of matrix elements ( glass @xcite , glass and hibbert @xcite , grant @xcite , burke _ @xcite ) are based on the computational scheme proposed by fano @xcite . this methodology is based on having the total wavefunction of an atom built from the antisymmetrized wavefunctions of separate shells , and this antisymmetrization is done via coefficients of fractional parentage . the shells are coupled one to another via their angular momenta . so , the finding of matrix elements amounts to finding the recoupling matrices and the coefficients of fractional parentage . suppose that we have a bra function with @xmath22 shells in @xmath0 coupling : @xmath23{c } \psi _ u^{bra}\left ( lsm_lm_s\right ) \equiv \\ \equiv ( n_1l_1n_2l_2 ... n_ul_u\alpha _ 1l_1s_1\alpha _ 2l_2s_2 ... \alpha _ ul_us_u{\cal a}lsm_lm_s| \end{array}\ ] ] and a ket function : @xmath24{c } \psi _ u^{ket}\left ( l^{\prime } s^{\prime } m^{\prime}_lm^{\prime}_s\right ) \equiv \\ n_ul_u\alpha _ 1^{\prime } l_1^{\prime } s_1^{\prime } \alpha _ 2^{\prime } l_2^{\prime } s_2^{\prime } ... \alpha _ u^{\prime } l_u^{\prime } s_u^{\prime } { \cal a}^{\prime } l^{\prime } s^{\prime}m^{\prime}_lm^{\prime}_s ) , \end{array}\ ] ] where @xmath25 stands for all intermediate quantum numbers , depending on the order of coupling of momenta @xmath26 . using the wigner - eckart theorem in @xmath0 space we shift from the matrix element of any two - particle operator @xmath27 between functions ( [ eq : ma - a ] ) and ( [ eq : ma - b ] ) to the submatrix element ( reduced matrix ) @xmath28 of this operator . when the two - particle operator acts upon four distinct shells , then the finding general expressions of matrix elements , according to methodology by fano @xcite , is based upon the formula @xmath29{c } ( \psi _ u^{bra}\left ( ls\right ) || \widehat{g}^{\left ( \kappa _ 1\kappa _ 2k,\sigma_1\sigma _ 2k\right ) } \sim \displaystyle { \sum_{n_{i } \lambda _ { i } , n_{j } \lambda _ { j } , n_{i^{\prime } } \lambda _ { i^{\prime } } , n_{j^{\prime } } \lambda _ { j^{\prime } } } } \left ( -1\right ) ^\delta \left [ n_{i } \left ( n_{j}-\delta(i , j ) \right ) n_{i^{\prime}}^{\prime } \left ( n_{j^{\prime}}^{\prime}-\delta(i^{\prime},j^{\prime } ) \right ) \right]^{1/2 } \times \\ \times \left ( l_{i}^{n_{i}}\;\alpha_{i } l_{i}s_{i}||l_{i}^{n_{i}-1}\;\left ( \alpha_{i}^{\prime } l^{\prime}_{i } s^{\prime } _ { i } \right ) , l_{i } \right ) \left ( l_{j}^{n_{j}}\;\alpha_{j } l_{j}s_{j}||l_{j}^{n_{j}-1}\;\left ( \alpha_{j}^{\prime } l^{\prime}_{j } s^{\prime } _ { j } \right ) , l_{j } \right ) \times \\ \times \left ( l_{i^{\prime } } ^{n_{i^{\prime } } } \;\alpha_{i^{\prime } } l_{i^{\prime } } s_{i^{\prime } } ||l_{i^{\prime } } ^{n_{i^{\prime } } -1}\;\left ( \alpha_{i^{\prime } } ^{\prime } l^{\prime}_{i^{\prime } } s^{\prime } _ { i^{\prime } } \right ) , l_{i^{\prime } } \right ) \left ( l_{j^{\prime } } ^{n_{j^{\prime } } } \;\alpha_{j^{\prime } } l_{j^{\prime } } s_{j^{\prime } } ||l_{j^{\prime } } ^{n_{j^{\prime } } -1}\;\left ( \alpha_{j^{\prime } } ^{\prime } l^{\prime}_{j^{\prime } } s^{\prime } _ { j^{\prime } } \right ) , l_{j^{\prime } } \right ) \times \\ \times \left\ { r_{d } \left ( \lambda _ i,\lambda _ j,\lambda _ i^{\prime } , \lambda _ j^{\prime } , \kappa_1 , \kappa_2 , \kappa , \sigma_1 , \sigma_2 , \sigma , \lambda ^{bra } , \lambda ^{ket } \right ) \right . \\ \times \left ( 1 + \delta(i , j ) \delta(i^{\prime } , j^{\prime } ) \right)^{-1 } \left ( n_i\lambda _ in_j\lambda _ j|| g^{\left ( \kappa _ 1\kappa _ 2k,\sigma_1\sigma _ 2k\right ) } - r_{e } \left ( \lambda _ i,\lambda _ j,\lambda _ i^{\prime } , \lambda _ j^{\prime } , \kappa_1 , \kappa_2 , \kappa , \sigma_1 , \sigma_2 , \sigma , \lambda ^{bra } , \lambda ^{ket } \right ) \\ \times \left . \left ( 1 + \delta(i , j ) ) ( 1 - \delta(i^{\prime } , j^{\prime } ) \right ) \left ( n_i\lambda _ in_j\lambda _ j|| g^{\left ( \kappa _ 1\kappa _ 2k,\sigma_1\sigma _ 2k\right ) } \right\ } , \end{array}\ ] ] where @xmath30 is a phase factor , ( see for example in @xcite ) , @xmath31 is the array for the bra function shells terms , and similarly for @xmath32 . the coefficient @xmath33 is a fractional parentage coefficient , and coefficients @xmath34 and @xmath35 are the recoupling matrices in @xmath36 and @xmath37 - spaces of direct and exchange terms , respectively . for more detailes on recoupling matrices see in grant @xcite , burke _ et al _ @xcite . the summation in expression ( [ eq : ma - c ] ) implies that the quantum numbers @xmath38 , @xmath39 of all participating shells are included . there are four such pairs of @xmath38 , @xmath39 in the sum . in essence , the fano calculation scheme consists of evaluating recoupling matrices . although such an approach uses classical racah algebra @xcite on the level of coefficients of fractional parentage , it may be necessary to carry out multiple summations over intermediate terms . due to these summations and the complexity of the recoupling matrix itself , the associated computer codes become rather time consuming . jucys and vizbarait . e @xcite proposed to use the two - electron coefficients of fractioanl parentage instead of ordinary ones , in matrix elements calculations , but even that did not solve the abovementioned problems . a solution to this problem was found by burke _ they tabulated separate standard parts of recoupling matrices along with coefficients of fractional parentage at the beginning of a calculation and used them further on to calculate the needed coefficients . computer codes by glass @xcite , glass and hibbert @xcite , grant @xcite , burke _ @xcite utilize the program njsym ( burke @xcite ) or njgraf ( bar - shalom and klapisch @xcite ) for the calculation of recoupling matrices . both are rather time consuming when calculating matrix elements of complex operators or electronic configurations with many open shells . in order to simplify the calculations , cowan @xcite suggested that matrix elements be grouped into classes ( see cowan @xcite figure 13 - 5 ) . unfortunately , this approach was not generalized to all two - electron operators . perhaps for this reason cowan s approach is not very popular although the program itself , based on this approach , is widely used . many approaches for the calculation of spin - angular coefficients ( glass @xcite , glass and hibbert @xcite , grant @xcite , burke _ et al _ @xcite ) are based on the usage of racah algebra only on the level of coefficients of fractional parentage . a few authors ( jucys and savukynas @xcite , cowan @xcite ) utilize the unit tensors , simplifying the calculations in this way , because use can be made of the tables of unit tensors and selection rules can be used prior to computation to check whether the spin - angular coefficients are zero or not . moreover , the recoupling matrices themselves have a simpler form . unfortunately , these ideas were applied only to diagonal matrix elements with respect to configurations , though cowan @xcite suggested the usage of unit tensors for non - diagonal ones as well . all the above mentioned approaches were applied in the coordinate representation . the second quantization formalism ( judd @xcite , rudzikas and kaniauskas @xcite and rudzikas @xcite ) has a number of advantages compared to coordinate representation . first of all , it is much easier to find algebraic expressions for complex operators and their matrix elements , when relying on second quantization formalism . it has contributed significantly to the successful development of perturbation theory ( see lindgren and morrison @xcite , merkelis _ et al _ @xcite , and orthogonal operators ( uylings @xcite ) , where three - particle operators already occur . uylings @xcite suggested a fairly simple approach for dealing with separate cases of three - particle operators . moreover , in the second quantization approach the quasispin formalism was efficiently developed by innes @xcite , pakauskas _ et al _ @xcite , rudzikas and kaniauskas @xcite , fano and rau @xcite . the main advantage of this approach is that applying the quasispin method for calculating the matrix elements of any operator , we can use all advantages of the new version of racah algebra ( see rudzikas @xcite ) for integration of spin - angular part of any one- and two - particle operator . for example , the reduced coefficients of fractional parentage are independent of the occupation number of the shell ( see rudzikas and kaniauskas @xcite , gaigalas _ et al _ all this enabled merkelis and gaigalas @xcite to work out a general perturbation theory approach for complex cases of several open shells . in the paper by merkelis @xcite a detailed review of a version of graphical methodology is presented that allows one to represent the operators graphically and to find the matrix elements of these operators using diagrammatic technique . the majority of methods and computer codes of finding angular coefficients discussed above were faced with a number of problems , the main of these being : * the high demand of cpu time for calculating the angular parts of matrix elements even on the modern computer . therefore the high accuracy of characteristics of atomic quantities somtimes is even unattainable . * the methods are applicable in practice only for the comparatively simple systems , because in the programs based on classical racah algebra the treatment of recoupling matrices is rather complicated , especialy when finding the matrix elements of breit - pauli operators between complex configurations . * the configurations with open @xmath40-shells must often be included in theoretical calculations . this causes problems in a number of methodologies , because the complete account of @xmath40- shells implies using a large number of coefficients of fractional parentage . gaigalas and rudzikas @xcite , gaigalas _ @xcite , gaigalas _ @xcite , gaigalas _ et al _ @xcite and gaigalas and rudzikas @xcite suggested an efficient and general approach for finding the spin - angular parts of matrix elements of atomic interactions , relying on the combination of the second - quantization approach in the coupled tensorial form , the generalized graphical technique and angular momentum theory in orbital , spin and quasispin spaces as well as on the symetry properties of the quantities considered . this approach is free of previous shortcomings . it is the main goal of this work to present all this methodology consistently and in a unifying manner , paying the special attention to its main ideas . also , we aim at a detailed discussion of obtaining the efficient tensorial expressions of a two - particle operator , as well as the analytical expressions for recoupling matrices . in addition to that , in this work we aim at comparing this methodology to other calculation of angular coefficients schemes . from what is said above we see that the treatment of angular parts of matrix elements is a many - sided question . the classical scheme by fano was developed in various aspects . the methods were developed that used racah algebra @xcite on a higher level , or used the new version of the racah algebra ( angular momentum theory in orbital , spin and quasispin spaces ) ( see rudzikas @xcite ) . the ways were searched for , to obtain the expressions that had simpler recoupling matrices . in order to discuss all that in more detail , to compare the existing methods of obtaining the angular parts and to mark the advantages of one method or another , the finding angular parts must be looked at from different angles . thus , the first thing to discuss is the expression for any physical operator ( see sections 2,3,4 ) , because already the form of it determines the level of application of tensor algebra in the calculation of matrix elements of this operator . also , it is very important to make clear to what extent the racah algebra is exploited in one or another methodology of angular parts treatment . this helps to mark the advantages of one method against another ( see section 5 ) . in addition , it is of importance to compare the ways of calculating the recoupling matrix in various methods ( see section 6 ) . it is well known in the literature that a scalar two - particle operator may be presented the following tensorial form ( see jucys and savukynas @xcite , glass @xcite ) : @xmath41_{p - p}^{(k~k ) } , \ ] ] where @xmath42 is the radial part of operator , @xmath43 is a tensor acting upon the orbital and spin variables of the @xmath44-th function , @xmath45 , @xmath46 are the ranks of operator acting in orbital space , and @xmath47 , @xmath48 are the ranks of operator acting in spin space . all the above mentioned approaches were usually applied in the coordinate representation . now we will investigate into the second quantization formalism , which is broadly applied in atomic physics as well . a two - particle operator in second quantization method is written as follows : @xmath49 where @xmath50 , @xmath51 is the two - electron matrix element of operator @xmath52 , and @xmath53 is the electron creation and @xmath54 electron annihilation operators . meanwhile two tensorial forms are well known in second quantization . in the first form the operators of second quantization follow in the normal order : @xmath55{c } g_i^{\left ( \kappa_1 \kappa_2 k , \sigma_1 \sigma_2 k \right)}= \displaystyle { \sum_{n_il_i , n_jl_j , n_i^{\prime } l_i^{\prime } , n_j^{\prime } l_j^{\prime } } } \widehat{g}_i(ij , i^{\prime } j^{\prime } ) = \\ = -\frac 12 \displaystyle { \sum_{n_il_i , n_jl_j , n_i^{\prime } l_i^{\prime } , n_j^{\prime } l_j^{\prime } } } \displaystyle { \sum_{\kappa _ { 12}\kappa _ { 12}^{^{\prime } } \sigma _ { 12}\sigma _ { 12}^{^{\prime } } } } \displaystyle { \sum_p}\left ( -1\right ) ^{k - p}\left [ \kappa _ { 12},\kappa _ { 12}^{\prime } , \sigma _ { 12},\sigma _ { 12}^{\prime } \right ] ^{1/2}\times \\ \times \left ( n_i\lambda _ in_j\lambda _ j||g^{\left ( \kappa _ 1\kappa _ 2k,\sigma _ 1\sigma _ 2k\right ) } ||n_i^{\prime } \lambda _ i^{\prime } n_j^{\prime } \lambda _ j^{\prime } \right ) \times \\ \times \left\ { \begin{array}{ccc } l_i^{\prime } & l_j^{\prime } & \kappa _ { 12}^{\prime } \\ \kappa _ 1 & \kappa _ 2 & k \\ l_i & l_j & \kappa _ { 12 } \end{array } \right\ } \left\ { \begin{array}{ccc } s & s & \sigma _ { 12}^{\prime } \\ \sigma _ 1 & \sigma _ 2 & k \\ s & s & \sigma _ { 12 } \end{array } \right\ } \times \\ \times \left [ \left [ a^{\left ( \lambda _ i\right ) } \times a^{\left ( \lambda _ j\right ) } \right ] ^{\left ( \kappa _ { 12}\sigma _ { 12}\right ) } \times \left [ \stackrel{\sim } { a}^{\left ( \lambda _ i^{\prime } \right ) } \times \stackrel{% \sim } { a}^{\left ( \lambda _ j^{\prime } \right ) } \right ] ^{\left ( \kappa _ { 12}^{\prime } \sigma _ { 12}^{\prime } \right ) } \right ] _ { p - p}^{\left ( kk\right ) } , \end{array}\ ] ] where @xmath56 = \left ( 2a+1\right ) \left ( 2b+1\right ) $ ] , @xmath57 , @xmath58 is the two - electron submatrix ( reduced matrix ) element of operator @xmath59 and tensor @xmath60 is defined as ( see for example rudzikas @xcite ) @xmath61 the product of tensors @xmath62 ^{\left ( \kappa _ { 12}\sigma _ { 12}\right ) } \times \left [ \stackrel{\sim } { a}^{\left ( \lambda _ i^{\prime } \right ) } \times \stackrel{% \sim } { a}^{\left ( \lambda _ j^{\prime } \right ) } \right ] ^{\left ( \kappa _ { 12}^{\prime } \sigma _ { 12}^{\prime } \right ) } \right ] _ { p - p}^{\left ( kk\right ) } $ ] denotes tensorial part of operator @xmath63 . in another form the second quantization operators are coupled by pairs consisting of electron creation and annihilation operators . in coupled tensorial form : @xmath64{c } g_{ii}^{\left ( \kappa_1 \kappa_2 \kappa , \sigma_1 \sigma_2 \sigma \right)}= \displaystyle { \sum_{n_il_i , n_jl_j , n_i^{\prime } l_i^{\prime } , n_j^{\prime } l_j^{\prime } } } \widehat{g}_{ii}(ij , i^{\prime } j^{\prime } ) = \\ = \frac 12 \displaystyle { \sum_{n_il_i , n_jl_j , n_i^{\prime } l_i^{\prime } , n_j^{\prime } l_j^{\prime } } } \displaystyle { \sum_p}\left ( -1\right ) ^{k - p}\left ( n_i\lambda _ in_j\lambda _ j||g^{\left ( \kappa _ 1\kappa _ 2k,\sigma _ 1\sigma _ 2k\right ) } ||n_i^{\prime } \lambda _ i^{\prime } n_j^{\prime } \lambda _ j^{\prime } \right ) \times \\ \times \left\ { \left [ \kappa _ 1,\kappa _ 2,\sigma _ 1,\sigma _ 2\right ] ^{-1/2}\times \left [ \left [ a^{\left ( \lambda _ i\right ) } \times \stackrel{\sim } { a}^{\left ( \lambda _ i^{\prime } \right ) } \right ] ^{\left ( \kappa _ 1\sigma _ 1\right ) } \times \left [ a^{\left ( \lambda _ j\right ) } \times \stackrel{\sim } { a}^{\left ( \lambda _ j^{\prime } \right ) } \right ] ^{\left ( \kappa _ 2\sigma _ 2\right ) } \right ] _ { p - p}^{\left ( kk\right ) } - \right . -\left ( -1\right ) ^{l_i+l_j^{\prime } } \left\ { \begin{array}{ccc } \kappa _ 1 & \kappa _ 2 & k \\ l_j^{\prime } & l_i & l_j \end{array } \right\ } \left\ { \begin{array}{ccc } \sigma _ 1 & \sigma _ 2 & k \\ s & s & s \end{array } \right\ } \left [ a^{\left ( \lambda _ i\right ) } \times \stackrel{\sim } { a}% ^{\left ( \lambda _ j^{\prime } \right ) } \right ] _ { p - p}^{\left ( kk\right ) } \delta \left ( n_jl_j , n_i^{\prime } l_i^{\prime } \right ) \right\}. \end{array}\ ] ] the expression ( [ eq : op - c ] ) consists of only one tensorial product whereas ( [ eq : op - e ] ) has two , but the summation in the first formula is also over intermediate ranks @xmath65 , @xmath66 , @xmath67 and @xmath68 , complicating in this way the calculations . the advantages or disadvantages of these alternative forms of arbitrary two - electron operator may be revealed in practical applications . in these forms the product of second quantization operators denotes tensorial part of operator @xmath52 . for instance , the tensorial structure of electrostatic ( coulomb ) electron interaction operator is @xmath69 ( jucys and savukynas @xcite ) , and only the two - electron submatrix elements @xmath70 of these operators are different . in the case of electrostatic interaction : @xmath71{c } \left ( n_i\lambda _ in_j\lambda _ j||g_{coulomb}^{\left ( kk0,000\right ) } ||n_i^{\prime } \lambda _ i^{\prime } n_j^{\prime } \lambda _ j^{\prime } \right ) = \\ = 2\left [ k\right ] ^{1/2}\left ( l_i||c^{\left ( k\right ) } ||l_i^{\prime } \right ) \left ( l_j||c^{\left ( k\right ) } ||l_j^{\prime } \right ) r_k\left ( n_il_in_i^{\prime } l_i^{\prime } , n_jl_jn_j^{\prime } l_j^{\prime } \right ) . \end{array}\ ] ] from ( [ eq : op - f ] ) , by ( [ eq : op - c ] ) and ( [ eq : op - e ] ) , we finally obtain the following two secondly quantized expressions for coulomb operator ( see merkelis _ @xcite ) : @xmath72{c } \widehat{v}_i=-\frac 12 \displaystyle { \sum_{n_il_in_jl_jn_i^{\prime } l_i^{\prime } n_j^{\prime } l_j^{\prime } } } \displaystyle { \sum_{\kappa _ { 12}\sigma _ { 12}k}}\left ( -1\right ) ^{l_j+l_i^{\prime } + k+\kappa _ { 12}}\left [ \kappa _ { 12},\sigma _ { 12}\right ] ^{1/2}\left ( l_i||c^{\left ( k\right ) } ||l_i^{\prime } \right ) \times \\ \times \left ( l_j||c^{\left ( k\right ) } ||l_j^{\prime } \right ) r_k\left ( n_il_in_i^{\prime } l_i^{\prime } , n_jl_jn_j^{\prime } l_j^{\prime } \right ) \left\ { \begin{array}{ccc } l_i & l_i^{\prime } & k \\ l_j^{\prime } & l_i & \kappa _ { 12 } \end{array } \right\ } \times \\ \times \left [ \left [ a^{\left ( \lambda _ i\right ) } \times a^{\left ( \lambda _ j\right ) } \right ] ^{\left ( \kappa _ { 12}\sigma _ { 12}\right ) } \times \left [ \stackrel{\sim } { a}^{\left ( \lambda _ i^{\prime } \right ) } \times \stackrel{% \sim } { a}^{\left ( \lambda _ j^{\prime } \right ) } \right ] ^{\left ( \kappa _ { 12}\sigma _ { 12}\right ) } \right ] ^{\left ( 00\right ) } , \end{array}\ ] ] @xmath73{c } \widehat{v}_{ii}= \displaystyle { \sum_{n_il_in_jl_jn_i^{\prime } l_i^{\prime } n_j^{\prime } l_j^{\prime } } } \displaystyle { \sum_k}\left ( l_i||c^{\left ( k\right ) } ||l_i^{\prime } \right ) \left ( l_j||c^{\left ( k\right ) } ||l_j^{\prime } \right ) r_k\left ( n_il_in_i^{\prime } l_i^{\prime } , n_jl_jn_j^{\prime } l_j^{\prime } \right ) \times \\ \times \left\ { \left [ k\right ] ^{-1/2}\left [ \left [ a^{\left ( \lambda _ i\right ) } \times \stackrel{\sim } { a}^{\left ( \lambda _ i^{\prime } \right ) } \right ] ^{\left ( k0\right ) } \times \left [ a^{\left ( \lambda _ j\right ) } \times \stackrel{\sim } { a}^{\left ( \lambda _ j^{\prime } \right ) } \right ] ^{\left ( k0\right ) } \right ] ^{\left ( 00\right ) } + \right . \\ \left . + \left ( 2\left [ l_i\right ] \right ) ^{-1/2}\left [ a^{\left ( \lambda _ i\right ) } \times \stackrel{\sim } { a}% ^{\left ( \lambda _ j^{\prime } \right ) } \right ] ^{\left ( 00\right ) } \delta \left ( n_jl_j , n_i^{\prime } l_i^{\prime } \right ) \right\ } , \end{array}\ ] ] the tensorial expressions for orbit - orbit and other physical operators in second quantization form may be obtained in the same manner . it is worth mentioning that the expressions ( [ eq : op - g ] ) and ( [ eq : op - h ] ) embrace , already in an operator form , both the diagonal interaction terms , relative to configurations , and the non - diagonal ones . non - diagonal terms define the interaction between all the possible electron distributions over the configurations considered , differing by quantum numbers of not more than two electrons . the merits of representing operators in one form or another ( [ eq : op - g ] ) or ( [ eq : op - h ] ) are mostly determined by the technique used to find their matrix elements and quantities in terms of which they are expressed . in the paper gaigalas and rudzikas @xcite it was shown that the tensorial forms ( [ eq : op - a ] ) , ( [ eq : op - c ] ) , ( [ eq : op - e ] ) of two - particle operator do not take full advantage of tensor algebra . the most characteristic examples are when configurations considered have many open shells , or when the non - diagonal matrix elements are seeked . in the paper gaigalas _ et al _ @xcite the following optimal expression of two - particle operator is proposed , which allows one to make the most of the advantages of racah algebra ( see racah @xcite ) . @xmath74{c } \widehat{g}^{\left ( \kappa_1 \kappa_2 k , \sigma_1 \sigma_2 k \right)}\sim \displaystyle { \sum_{\alpha } } \displaystyle { \sum_{\kappa _ { 12},\sigma _ { 12},\kappa _ { 12}^{\prime } , \sigma _ { 12}^{\prime } } } \theta \left ( \xi \right ) \left\ { a_{p ,- p}^{\left ( kk\right ) } \left ( n_\alpha \lambda _ \alpha , \xi \right ) \delta \left ( u,1\right ) \right . \\ + \displaystyle { \sum_{\beta } } \left [ b^{\left ( \kappa _ { 12}\sigma _ { 12}\right ) } \left ( n_\alpha \lambda _ \alpha , \xi \right ) \times c^{\left ( \kappa _ { 12}^{\prime } \sigma _ { 12}^{\prime } \right ) } \left ( n_\beta \lambda _ \beta , \xi \right ) \right ] _ { p ,- p}^{\left ( kk\right ) } \delta \left ( u,2\right ) \\ + \displaystyle { \sum_{\beta \gamma } } \left [ \left [ d^{\left ( l_\alpha s\right ) } \times d^{\left ( l_\beta s\right ) } \right ] ^{\left ( \kappa _ { 12}\sigma _ { 12}\right ) } \times e^{\left ( \kappa _ { 12}^{\prime } \sigma _ { 12}^{\prime } \right ) } \left ( n_\gamma \lambda _ \gamma , \xi \right ) \right ] _ { p ,- p}^{\left ( kk\right ) } \delta \left ( u,3\right ) \\ \left . + \displaystyle { \sum_{\beta \gamma \delta } } \left [ \left [ d^{\left ( l_\alpha s\right ) } \times d^{\left ( l_\beta s\right ) } \right ] ^{\left ( \kappa _ { 12}\sigma _ { 12}\right ) } \times \left [ d^{\left ( l_\gamma s\right ) } \times d^{\left ( l_\delta s\right ) } \right ] ^{\left ( \kappa _ { 12}^{\prime } \sigma _ { 12}^{\prime } \right ) } \right ] _ { p ,- p}^{\left ( kk\right ) } \delta \left ( u,4\right ) \right\ } . \end{array}\ ] ] whereas in traditional expressions , e. g. ( [ eq : op - c ] ) , the summation runs over the principle and the orbital quantum numbers of open shells without detailing these , in the expression written above the first term represents the case of a two - particle operator acting upon the same shell @xmath75 , the second term corresponds to operator @xmath76 acting upon two different shells @xmath75 , @xmath77 . when operator @xmath76 acts upon three shells the third term in ( [ eq : op - i ] ) must be considered and when it acts upon four - the fourth one . we define in this expression the shells @xmath75 , @xmath77 , @xmath78 , @xmath79 to be different . the tensorial part of a two - particle operator is expressed in terms of operators of the type @xmath80 , @xmath81 , @xmath82 , @xmath83 , @xmath84 . they correspond to one of the forms : @xmath85 @xmath86 ^{\left ( \kappa _ 1\sigma _ 1\right ) } , \ ] ] @xmath87 ^{\left ( \kappa _ 1\sigma _ 1\right ) } \right ] ^{\left ( \kappa _ 2\sigma _ 2\right ) } , \ ] ] @xmath88 ^{\left ( \kappa _ 1\sigma _ 1\right ) } \times a_{m_{q3}}^{\left ( q\lambda \right ) } \right ] ^{\left ( \kappa _ 2\sigma _ 2\right ) } , \ ] ] @xmath89 ^{\left ( \kappa _ 1\sigma _ 1\right ) } \times \left [ a_{m_{q3}}^{\left ( q\lambda \right ) } \times a_{m_{q4}}^{\left ( q\lambda \right ) } \right ] ^{\left ( \kappa _ 2\sigma _ 2\right ) } \right ] ^{\left ( kk\right ) } , \ ] ] for example , if we take a two - particle operator acting upon two shells , then we see from expression ( [ eq : op - i ] ) that the angular part of two - particle operator is expressed via operators @xmath90 and @xmath91 . in a case when the operator @xmath76 acts in such a manner that two operators of second quantization act upon one shell and two act upon another , the @xmath90 and @xmath91 are expressed as ( [ eq : op - k ] ) . but in a case when three operators of second quantization act upon one shell and one acts upon another , then @xmath90 and @xmath91 are expressed either as ( [ eq : op - j ] ) and ( [ eq : op - l ] ) or ( [ eq : op - j ] ) and ( [ eq : op - m ] ) . in writing down the expressions ( [ eq : op - j ] ) - ( [ eq : op - n ] ) the quasispin formalism was used , where @xmath92 and @xmath93 are components of the tensor @xmath94 , having in additional quasispin space the rank @xmath95 and projections @xmath96 , i.e. @xmath97 and @xmath98 in the expression ( [ eq : op - i ] ) @xmath22 is the overall number of shells acted upon by a given tensorial product of creation / annihilation operators . parameter @xmath99 implies the whole array of parameters ( and sometimes an internal summation over some of these is implied , as well ) that connect the amplitudes @xmath100 of tensorial products of creation / annihilation operators in the expression ( [ eq : op - i ] ) to these tensorial products ( see gaigalas _ et al _ also , attention must be paid to the fact that the ranks @xmath101 , @xmath102 , @xmath103 , @xmath104 , @xmath105 and @xmath106 are also included into the parameter @xmath99 . the amplitudes @xmath107 are all proportional to the submatrix element of a two - particle operator @xmath108 , @xmath109 to obtain the expression of a concrete physical operator , analogous to expression ( [ eq : op - i ] ) , the tensorial structure of the operator and the two - particle matrix elements ( [ eq : op - r ] ) must be known . we shall investigate this now . the _ electrostatic ( coulomb ) electron interaction _ operator @xmath110 itself contains the tensorial structure @xmath111{c } h^{coulomb } \equiv \displaystyle { \sum_{k } } h_{coulomb}^{(kk0,000 ) } \end{array}\ ] ] and its submatrix element is @xmath112{c } \left ( n_i\lambda _ in_j\lambda _ j\left\| h_{coulomb}^{(kk0,000 ) } \right\| n_{i^{\prime } } \lambda _ { i^{\prime } } n_{j^{\prime } } \lambda _ { j^{\prime } } \right ) = \\ 2 [ k]^{1/2 } \left ( l_i\left\| c^{\left ( k \right ) } \right\| l_{i^{\prime } } \right ) \left ( l_j\left\| c^{\left ( k \right ) } \right\| l_{j^{\prime } } \right ) r_{k}\left ( n_il_i n_{i^{\prime } } l_{i^{\prime } } , n_jl_j n_{j^{\prime } } l_{j^{\prime } } \right ) . \end{array}\ ] ] the _ spin - spin _ operator @xmath113 itself contains tensorial structure of two different types , summed over @xmath114 ( gaigalas and rudzikas @xcite ) , @xmath115{c } h^{ss } \equiv \displaystyle { \sum_{k } } \left [ h_{ss}^{(k+1 k-1 2,112 ) } + h_{ss}^{(k-1 k+1 2,112 ) } \right ] . \end{array}\ ] ] their submatrix elements are ( jucys and savukynas @xcite ) @xmath116{c } \left ( n_i\lambda _ in_j\lambda _ j\left\| h_{ss}^{\left ( k+1k-12,112\right ) } \right\| n_{i^{\prime } } \lambda _ { i^{\prime } } n_{j^{\prime } } \lambda _ { j^{\prime } } \right ) = \frac 3 { \sqrt{5}}\sqrt{\left ( 2k+3\right ) ^{\left ( 5\right ) } } \times \\ \times \left ( l_i\left\| c^{\left ( k+1\right ) } \right\| l_{i^{\prime } } \right ) \left ( l_j\left\| c^{\left ( k-1\right ) } \right\| l_{j^{\prime } } \right ) n^{k-1}\left ( n_il_in_jl_j , n_{i^{\prime } } l_{i^{\prime } } n_{j^{\prime } } l_{j^{\prime } } \right ) \end{array}\ ] ] @xmath117{c } \left ( n_i\lambda _ in_j\lambda _ j\left\| h_{ss}^{\left ( k-1k+12,112\right ) } \right\| n_{i^{\prime } } \lambda _ { i^{\prime } } n_{j^{\prime } } \lambda _ { j^{\prime } } \right ) = \frac 3 { \sqrt{5}}\sqrt{\left ( 2k+3\right ) ^{\left ( 5\right ) } } \times \\ \times \left ( l_i\left\| c^{\left ( k-1\right ) } \right\| l_{i^{\prime } } \right ) \left ( l_j\left\| c^{\left ( k+1\right ) } \right\| l_{j^{\prime } } \right ) n^{k-1}\left ( n_jl_jn_il_i , n_{j^{\prime } } l_{j^{\prime } } n_{i^{\prime } } l_{i^{\prime } } \right ) , \end{array}\ ] ] where we use a shorthand notation @xmath118 and radial integral ( [ eq : ss - b ] ) , ( [ eq : ss - c ] ) is defined as in glass and hibbert @xcite : @xmath119{c } n^k\left ( n_il_in_jl_j , n_{i^{\prime } } l_{i^{\prime } } n_{j^{\prime } } l_{j^{\prime } } \right ) \\ = \frac{\alpha ^2}4\int_0^\infty \int_0^\infty p_i\left ( r_1\right ) p_j\left ( r_2\right ) \frac{r_2^k}{r_1^{k+3}}\epsilon ( r_1-r_2)p_{i^{\prime } } \left ( r_1\right ) p_{j^{\prime } } \left ( r_2\right ) dr_1dr_{2 , } \end{array}\ ] ] where @xmath120 is a heaviside step - function , @xmath121 the _ spin - other - orbit _ operator @xmath122 itself contains tensorial structure of six different types , summed over @xmath114 ( see gaigalas _ et al _ @xcite ) : @xmath123{c } h^{sso } \equiv \displaystyle { \sum_{k } } \left [ h_{sso}^{(k-1 k 1,101 ) } + h_{sso}^{(k-1 k 1,011)}+ \right . \\ h_{sso}^{(k k 1,101 ) } + h_{sso}^{(k k 1,011)}+ h_{sso}^{(k+1 k 1,101 ) } + h_{sso}^{(k+1 k 1,011 ) } \right ] . \end{array}\ ] ] their submatrix elements are : @xmath124{c } \left ( n_i\lambda _ in_j\lambda _ j\left\| h_{soo}^{\left ( k-1k1,\sigma _ 1\sigma _ 21\right ) } \right\| n_{i^{\prime } } \lambda _ { i^{\prime } } n_{j^{\prime } } \lambda _ { j^{\prime } } \right ) = 2\cdot 2^{\sigma _ 2}\left\ { \left ( 2k-1\right ) \left ( 2k+1\right ) \right . \\ \times \left . \left ( l_i+l_{i^{\prime } } -k+1\right ) \left ( k - l_i+l_{i^{\prime } } \right ) \left ( k+l_i - l_{i^{\prime } } \right ) \left ( k+l_i+l_{i^{\prime } } + 1\right ) \right\ } ^{1/2 } \\ \times \left ( k\right ) ^{-1/2}\left ( l_i\left\| c^{\left ( k\right ) } \right\| l_{i^{\prime } } \right ) \left ( l_j\left\| c^{\left ( k\right ) } \right\| l_{j^{\prime } } \right ) n^{k-2}\left ( n_jl_jn_il_i , n_{j^{\prime } } l_{j^{\prime } } n_{i^{\prime } } l_{i^{\prime } } \right ) , \end{array}\ ] ] @xmath125{c } \left ( n_i\lambda _ in_j\lambda _ j\left\| h_{soo}^{\left ( kk1,\sigma _ 1\sigma _ 21\right ) } \right\| n_{i^{\prime } } \lambda _ { i^{\prime } } n_{j^{\prime } } \lambda _ { j^{\prime } } \right ) = -2\cdot 2^{\sigma _ 2}\left ( 2k+1\right ) ^{1/2}\left ( l_i\left\| c^{\left ( k\right ) } \right\| l_{i^{\prime } } \right ) \\ \times \left ( l_j\left\| c^{\left ( k\right ) } \right\| l_{j^{\prime } } \right ) \left\ { \left ( k\left ( k+1\right ) \right ) ^{-1/2}\left ( l_i\left ( l_i+1\right ) -k\left ( k+1\right ) -l_{i^{\prime } } \left ( l_{i^{\prime } } + 1\right ) \right ) \right . \\ \times \left\ { \left ( k+1\right ) n^{k-2}\left ( n_jl_jn_il_i , n_{j^{\prime } } l_{j^{\prime } } n_{i^{\prime } } l_{i^{\prime } } \right ) -kn^k\left ( n_il_in_jl_j , n_{i^{\prime } } l_{i^{\prime } } n_{j^{\prime } } l_{j^{\prime } } \right ) \right\ } \\ \left . -2\left ( k\left ( k+1\right ) \right ) ^{1/2}v^{k-1}\left ( n_il_in_jl_j , n_{i^{\prime } } l_{i^{\prime } } n_{j^{\prime } } l_{j^{\prime } } \right ) \right\ } , \end{array}\ ] ] @xmath126{c } \left ( n_i\lambda _ in_j\lambda _ j\left\| h_{soo}^{\left ( k+1k1,\sigma _ 1\sigma _ 21\right ) } \right\| n_{i^{\prime } } \lambda _ { i^{\prime } } n_{j^{\prime } } \lambda _ { j^{\prime } } \right ) = 2\cdot 2^{\sigma _ 2}\left\ { \left ( 2k+1\right ) \left ( 2k+3\right ) \right . \\ \times \left . \left ( l_i+l_{i^{\prime } } -k\right ) \left ( k - l_i+l_{i^{\prime } } + 1\right ) \left ( k+l_i - l_{i^{\prime } } + 1\right ) \left ( k+l_i+l_{i^{\prime } } + 2\right ) \right\ } ^{1/2 } \\ \times \left ( k+1\right ) ^{-1/2}\left ( l_i\left\| c^{\left ( k\right ) } \right\| l_{i^{\prime } } \right ) \left ( l_j\left\| c^{\left ( k\right ) } \right\| l_{j^{\prime } } \right ) n^k\left ( n_il_in_jl_j , n_{i^{\prime } } l_{i^{\prime } } n_{j^{\prime } } l_{j^{\prime } } \right ) . \end{array}\ ] ] the radial integrals in ( [ eq : soo - b ] ) - ( [ eq : soo - d ] ) are ( see glass and hibbert @xcite ) : @xmath127{c } v^k\left ( n_il_in_jl_j , n_{i^{\prime } } l_{i^{\prime } } n_{j^{\prime } } l_{j^{\prime } } \right ) \\ = \frac{\alpha ^2}4\int_0^\infty \int_0^\infty p_i\left ( r_1\right ) p_j\left ( r_2\right ) \frac{r_{<}^{k-1}}{r_{>}^{k+2}}r_2\frac \partial { \partial r_1}p_{i^{\prime } } \left ( r_1\right ) p_{j^{\prime } } \left ( r_2\right ) dr_1dr_2 . \end{array}\ ] ] the tensorial form of _ orbit - orbit _ operator is ( see eissner _ @xcite ) @xmath128 the sum of submatrix elements of three terms @xmath129 , @xmath130 and @xmath131 is equal to ( see badnell @xcite ) : @xmath132{c } \left ( n_i\lambda _ in_j\lambda _ j\left\| h_{oo1}^{(kk0,000 ) } + h_{oo2}^{(kk0,000 ) } + h_{oo4}^{(kk0,000 ) } \right\| n_{i^{\prime } } \lambda _ { i^{\prime } } n_{j^{\prime } } \lambda _ { j^{\prime } } \right ) = \\ = \left\ { -2k\left ( k+1\right ) \left ( t^{k+1}\left ( n_il_in_jl_j , n_{i^{\prime } } l_{i^{\prime } } n_{j^{\prime } } l_{j^{\prime } } \right ) -t^{k-1}\left ( n_il_in_jl_j , n_{i^{\prime } } l_{i^{\prime } } n_{j^{\prime } } l_{j^{\prime } } \right ) \right ) \right . \\ + \left ( l_i\left ( l_i+1\right ) -k\left ( k+1\right ) -l_{i^{\prime } } \left ( l_{i^{\prime } } + 1\right ) \right ) \\ \times \left ( u^{k+1}\left ( n_il_in_jl_j , n_{i^{\prime } } l_{i^{\prime } } n_{j^{\prime } } l_{j^{\prime } } \right ) -u^{k-1}\left ( n_il_in_jl_j , n_{i^{\prime } } l_{i^{\prime } } n_{j^{\prime } } l_{j^{\prime } } \right ) \right ) \\ + \left ( l_j\left ( l_j+1\right ) -k\left ( k+1\right ) -l_{j^{\prime } } \left ( l_{j^{\prime } } + 1\right ) \right ) \\ \times \left ( u^{k+1}\left ( n_jl_jn_il_i , n_{j^{\prime } } l_{j^{\prime } } n_{i^{\prime } } l_{i^{\prime } } \right ) -u^{k-1}\left ( n_jl_jn_il_i , n_{j^{\prime } } l_{j^{\prime } } n_{i^{\prime } } l_{i^{\prime } } \right ) \right ) \\ + \frac 12\left ( l_i\left ( l_i+1\right ) -k\left ( k+1\right ) -l_{i^{\prime } } \left ( l_{i^{\prime } } + 1\right ) \right ) \left ( l_j\left ( l_j+1\right ) -k\left ( k+1\right ) -l_{j^{\prime } } \left ( l_{j^{\prime } } + 1\right ) \right ) \\ \times \left [ \frac{k-2}{k\left ( 2k-1\right ) } \left ( n^{k-2}\left ( n_il_in_jl_j , n_{i^{\prime } } l_{i^{\prime } } n_{j^{\prime } } l_{j^{\prime } } \right ) + n^{k-2}\left ( n_jl_jn_il_i , n_{j^{\prime } } l_{j^{\prime } } n_{i^{\prime } } l_{i^{\prime } } \right ) \right ) \right . \\ -\frac{k+3}{\left ( k+1\right ) \left ( 2k+3\right ) } \left ( n^k\left ( n_il_in_jl_j , n_{i^{\prime } } l_{i^{\prime } } n_{j^{\prime } } l_{j^{\prime } } \right ) + n^k\left ( n_jl_jn_il_i , n_{j^{\prime } } l_{j^{\prime } } n_{i^{\prime } } l_{i^{\prime } } \right ) \right ) \right ] \right\ } \\ \times \left ( 1-\delta \left ( k,0\right ) \right ) , \end{array}\ ] ] where @xmath133{c } t^k\left ( n_il_in_jl_j , n_{i^{\prime } } l_{i^{\prime } } n_{j^{\prime } } l_{j^{\prime } } \right ) = \frac{\alpha ^2}{4\left ( 2k+1\right ) } \\ \times \int_0^\infty \int_0^\infty p_i\left ( r_1\right ) p_j\left ( r_2\right ) \frac{r_<^k}{r_>^{k+1 } } \left(\frac{\partial}{\partial r_1}+\frac{1}{r_1}\right ) p_{i^{\prime } } \left ( r_1\right ) \left(\frac{\partial}{\partial r_2}+\frac{1}{r_2}\right ) p_{j^{\prime } } \left ( r_2\right ) dr_1dr_{2 } , \end{array}\ ] ] @xmath134{c } u^k\left ( n_il_in_jl_j , n_{i^{\prime } } l_{i^{\prime } } n_{j^{\prime } } l_{j^{\prime } } \right ) = \frac{\alpha ^2}{4\left ( 2k+1\right ) } \\ \times \int_0^\infty \int_0^\infty p_i\left ( r_1\right ) p_j\left ( r_2\right)\left ( ( k-1)\frac{r_2^k}{r_1^{k+2}}\epsilon ( r_1-r_2 ) -(k+2)\frac{r_1^{k-1}}{r_2^{k+1}}\epsilon ( r_2-r_1)\right ) \\ \times p_{i^{\prime } } \left ( r_1\right ) \left(\frac{\partial}{\partial r_2}+\frac{1}{r_2}\right ) p_{j^{\prime } } \left ( r_2\right ) dr_1dr_{2}. \end{array}\ ] ] the submatrix element of remaining term @xmath135 is : @xmath136{c } \left ( n_i\lambda _ in_j\lambda _ j\left\|h_{oo3}^{(kk0,000 ) } \right\| n_{i^{\prime } } \lambda _ { i^{\prime } } n_{j^{\prime } } \lambda _ { j^{\prime } } \right ) = 2 \sqrt{2k+1}\frac 1{k(k+1 ) } \\ \times \left ( \left ( l_i+l_{i^{\prime } } + k+2\right ) \left ( l_i+l_{i^{\prime } } -k\right ) \left ( l_i - l_{i^{\prime } } + k+1\right ) \left ( l_{i^{\prime } } -l_i+k+1\right ) \right . \\ \times \left . \left ( l_j+l_{j^{\prime } } + k+2\right ) \times \left ( l_j+l_{j^{\prime } } -k\right ) \left ( l_j - l_{j^{\prime } } + k+1\right ) \left ( l_{j^{\prime } } -l_j+k+1\right ) \right ) ^{1/2 } \\ \times \left ( l_i||c^{\left ( k+1\right ) } ||l_{i^{\prime } } \right ) \left ( l_j||c^{\left ( k+1\right ) } ||l_{j^{\prime } } \right ) \\ \times \left ( n^{k-1}\left ( n_il_in_jl_j , n_{i^{\prime } } l_{i^{\prime } } n_{j^{\prime } } l_{j^{\prime } } \right ) + n^{k-1}\left ( n_jl_jn_il_i , n_{j^{\prime } } l_{j^{\prime } } n_{i^{\prime } } l_{i^{\prime } } \right ) \right ) . \end{array}\ ] ] the rest of two - particle breit - pauli operators that we did not investigate so far are the two - body darwin and spin - spin - contact terms . they do not bring any additional difficulties into the investigation of hamiltonian , but for the sake of completeness of presentation we will discuss them briefly the _ two - body darwin _ operator @xmath15 ( see for more detail nikitin and rudzikas @xcite ) , as well as the _ spin - spin - contact _ operator @xmath16 ( see shalit and talmi @xcite and feneuille @xcite ) , both have the following tensorial structure : @xmath137{c } h \equiv \displaystyle { \sum_{k } } h^{(kk0,000)}. \end{array}\ ] ] these two terms are included into calculation by adding to the radial integral @xmath138 a term @xmath139 where @xmath140{c } x\left ( n_il_in_jl_j , n_{i^{\prime } } l_{i^{\prime } } n_{j^{\prime } } l_{j^{\prime } } \right ) \\ = \frac{\alpha ^2}4\int_0^\infty \int_0^\infty p_i\left ( r_1\right ) p_j\left ( r_2\right ) \frac{1}{r_1^{2 } } \delta \left ( r_1-r_2 \right ) p_{i^{\prime } } \left ( r_1\right ) p_{j^{\prime } } \left ( r_2\right ) dr_1dr_{2}. \end{array}\ ] ] the expression ( [ eq : op - i ] ) has a series of terms , and thus at a first glance seems to be difficult to apply . for this purpose in the next sections we shall discuss in more detail : * obtaining the expression by wick s theorem . * the compact written form of all terms , using the extended graphical technique . * obtaining the values of recoupling matrix and of the standard quantities . we shall also compare the existing methodologies of finding angular parts , showing the advantages and shortcomings of one methodology or another . this methodology is fully applicable also for the one - particle operator ( see gaigalas and rudzikas @xcite , gaigalas _ et al _ as this operator does not cause big problems in the atomic physics , we will not stop for its details . wick s theorem in the second quantization formalism is formulated as follows ( see wick @xcite ; bogoliubov and shirkov @xcite ) : if @xmath141 is a product of creation and annihilation operators , then @xmath142 where @xmath143 represents the normal form of @xmath141 and @xmath144 represents the sum of the normal - ordered terms obtained by making all possible single , double , ... contractions within @xmath141 . based on the identification in bogoliubov and shirkov @xcite , the operator is presented in normal form when all of the operators of annihilation included in it are transferred to the right of the creation operators . usually , wick s theorem is applied when treating complex operators that are represented by a large number of second quantization operators in a non - normal product form . in atomic physics such operators are used in perturbation theory ( see fetter and waleka @xcite ; lindgren and morrison @xcite ; merkelis _ et al _ @xcite ) and in the orthogonal operator method ( see uylings @xcite ) . most often the wick s theorem is applied to the products of second quantization operators that are not tensorially coupled ( see lindgren and morrison @xcite ) . while applying the perturbation theory in an extended model space , two different groups of second quantization operators are defined ( see for details in lindgren and morrison @xcite ) . the second quantization operators acting upon core states belong to one group , whereas the operators acting upon open and excited shells belong to another one . these two groups are very different in applying wick s theorem to them . in the first group , the creation operators are re - named to annihilation operators and are called the hole absorption operators , while the annihilation operators are re - named to creation operators and are called the hole creation operators . the creation operators of the second group are called the particle creation operators , while annihilation ones are called the particle absorption operators . such a division of second quantization operators into two groups is called the particle - hole formalism . merkelis _ et al _ @xcite have proposed to use the so - called graphical analogue of wick s theorem in perturbation theory ( see gaigalas _ et al _ @xcite , gaigalas @xcite ) . it is an efficient tool for obtaining the normal products of second quantization operators in a coupled tensorial form . before applying this theorem , particular second quantization terms are in a normal product in coupled form . in addition , this theorem is applied in the particle - hole formalism , too ( see merkelis _ @xcite , merkelis _ et al _ @xcite ) . in all the cases mentioned above , the wick s theorem is applied for the most general case of operators , i.e. the shells that are acted upon are not detailed . but , in the case of the extended model space , the group that the second quantization operators belong to , depending on the electronic structure of atom or ion under investigation , is defined . wick s theorem is not applied in investigations of ordinary physical operators . @xcite proposed a new version of the wick s theorem application , where the optimal tensorial expression of any two - particle operator is easily obtained . the specificity of wick s theorem application in this case lies in applying it only when the shells that are acted upon by the secondary quantization operators are known , i.e. it is applied for each term @xmath145 separately . such an interpretation of wick s theorem bears similarity with the particle - hole formalism . the only difference is that in this case the second quantization operators are differentiated formally not on the basis of structure of atom under investigation , but on the basis of shells acted upon . this is done in the following way . the second quantization operators acting upon a shell with a lowest index are attributed to the first group . those acting upon a shell with a next - lowest index are attributed to the second group , etc . in the most general case we have four distinct groups . for example , suppose we have a two - particle operator @xmath146 , @xmath147 ( where @xmath148 , @xmath149 , @xmath150 ) , the matrix element of which between the functions @xmath151 and @xmath152 we must obtain . then the operators acting upon the second shell are attributed to the first group , the ones acting upon the third shell - to the second , and the ones acting upon the fifth - to the third group . assuming that all the operators from first group are the creation ones , and the rest are annihilation operators , we apply the wick s theorem . thus we obtain that the operators from the first group are one beside another , and all are positioned after the operators from the first group . thus we obtain : @xmath153 after that , we apply wick s theorem assuming that the operators from the first and the second groups are creation ones , and the rest of them are annihilation operators . thus we obtain that the operators from the second group are one beside another , and all are positioned after the first group operators : @xmath154 if in the product that we investigate there are operators of second quantization acting upon four distinct shells , then we apply wick s theorem once again , assuming that operators from the first , second and third groups are creation ones , and from the fourth group - annihilation ones . in this case the wick s theorem is applied to the second quantization operators in uncoupled form . from ( [ eq : op - b ] ) we see that in second quantization a two - particle operator is written as a sum , where parameters @xmath44 , @xmath155 , @xmath156 , @xmath157 run over all possible arrays of quantum numbers . so , the greater the number of open shells in bra and ket functions , the greater the number of terms @xmath145 in the expression of two - particle operator . it is obvious that all these terms must be systematized in order to obtain in general case the most efficient tensorial expression of a two - particle operator , in the way described above . .distributions of shells , upon which the second quantization operators are acting , that appear in the submatrix elements of any two - particle operator , when bra and ket functions have @xmath22 open shells ( see gaigalas _ @xcite ) [ cols=">,^,^,^,^,^ " , ] in the work by gaigalas _ et al _ @xcite it is chosen an optimal number of distributions , which is necessary to obtain the matrix elements of any two - particle operator , when the bra and ket functions consist of arbitrary number of shells . this is presented in table 1 . we point out that for distributions 2 - 5 and 19 - 42 the shells sequence numbers @xmath158 , @xmath159 , @xmath160 , @xmath161 ( in bra and ket functions of a submatrix element ) satisfy the condition @xmath162 , while for distributions 6 - 18 no conditions upon @xmath158 , @xmath159 , @xmath160 , @xmath161 are imposed ( this permits to reduce the number of distributions ) . for distributions 19 - 42 this condition is imposed only for obtaining simple analytical expressions for the recoupling matrices @xmath163 . this will be discussed in more detail in the next section . so , in the way that is described earlier , the wick s theorem is applied , assuming that the second quantization operators acting upon shells @xmath164 , @xmath165 , @xmath160 and @xmath161 belong to different groups . in the next section we discuss the way to obtain irreducible tensorial form of these distributions . in addition , the arguments will be given in evidence of superiority of the obtained tensorial expressions against other expressions known in the literature . the methodology presented in this section demonstrates the way to obtain optimal arrangement of the second quantization operators , for any two - particle operator . it can be applied without restrictions for obtaining the optimal tensorial form of two - particle terms of orthogonal operators and of perturbation theory operators , too . the graphical technique of angular momentum is widely used in the atomic physics : see yutsis _ @xcite , jucys and bandzaitis @xcite , brink and satcher @xcite , el - baz @xcite . it is applied efficiently both in the coordinate representation ( see for example jucys and bandzaitis @xcite ) , and in the second quantization formalism ( gaigalas _ et al _ the use of it allows one to obtain the analytical expressions for the recoupling matrices conveniently ( see for example kaniauskas and rudzikas @xcite ) , to investigate the tensorial products of operators ( see for example jucys _ @xcite ) , and to seek for the matrix elements of operators ( see for example huang and starace @xcite ) . gaigalas and merkelis @xcite have proposed a graphical way to obtain the values of matrix elements when the operator is a many - particle ( one- , two- , three- , etc . ) one and has irreducible tensorial form . for example , when a two - particle operator is considered , it has the form ( [ eq : op - c ] ) or ( [ eq : op - e ] ) . the matrix elements in this methodology are expressed not only in the terms of coefficients of fractional parentage or reduced coefficients of fractional parentage , but in terms of standard quantities @xmath166 and @xmath167 , too . @xcite have proposed to calculate the matrix elements by using the tensorial expressions for such two - particle operator , that take full advantage of racah algebra . in this case the tensorial form of operator depends on the shells that the operator acts upon ( distributions 1 - 42 from table 1 ) . this is the difference of this methodology from other . it is most convenient to obtain the tensorial expressions for 42 distributions graphically , using the generalized graphical technique by gaigalas _ et al _ @xcite . in such a case not only the similarities between different distributions are easily seen , but also the compact graphical representation of the obtained expressions is possible . we will stop for details on this in the present section . a two - particle operator may be represented graphically by a feynman - goldstone diagram @xmath168 from figure [ fig : dd - a ] ( lindgren and morrison @xcite ) . as it is shown in the paper bolotin _ @xcite , the feynman - goldstone diagrams are topologically equivalent to the angular momentum graphs . due to that , an irreducible tensorial form for every feynman - goldstone diagram may be obtained ( see merkelis _ the graph @xmath169 is the angular momentum graph corresponding to the diagram @xmath168 . so the two - particle operator will be written down as follows : @xmath170{c } \widehat{g } ( ij , i^{\prime } j^{\prime } ) = d_{1}= - \frac 12 \displaystyle { \sum_{m_{\lambda _ i}m_{\lambda _ j } m_{\lambda ^{\prime } _ i}m_{\lambda ^{\prime } _ j } } } \displaystyle { \sum_p } \left [ \kappa _ { 1},\kappa _ { 2},\sigma _ { 1},\sigma _ { 2}\right ] ^{-1/2 } \times \\ \times \left ( n_i\lambda _ in_j\lambda _ j||g^{\left ( \kappa _ 1\kappa _ 2k,\sigma _ 1\sigma _ 2k\right ) } ||n_i^{\prime } \lambda _ i^{\prime } n_j^{\prime } \lambda _ j^{\prime } \right ) ~d_{2}~~ a^{\left ( \lambda _ i\right ) } _ { m_{\lambda _ i } } a^{\left ( \lambda _ j\right ) } _ { m_{\lambda _ j } } \stackrel{\sim } { a}^{\left ( \lambda _ i^{\prime } \right ) } _ { m_{\lambda ^{\prime } _ i } } \stackrel{\sim } { a}^{\left ( \lambda _ j^{\prime } \right ) } _ { m_{\lambda ^{\prime } _ j } } , \end{array}\ ] ] ( 148,180 ) ( 10,140)(0,1)10 ( 10,150)(0,1)2 ( 5,148)(0,0)[t]@xmath171 ( 20,140)(0,1)10 ( 20,150)(0,1)2 ( 25,148)(0,0)[t]@xmath172 ( 11,140)(2,0)5(2,1)[t ] ( 15,143)(0,0)@xmath173 ( 10,130)(0,1)10 ( 10,130)(0,1)3 ( 5,133)(0,0)@xmath174 ( 10,128)(0,1)3 ( 20,130)(0,1)10 ( 20,130)(0,1)3 ( 25,133)(0,0)@xmath175 ( 20,128)(0,1)3 ( 15,120)(0,0)@xmath176 ( 45,140)(0,1)12 ( 45,140 ) ( 40,148)(0,0)[t]@xmath177 ( 75,140)(0,12)12 ( 75,152)(0,-1)3 ( 75,140 ) ( 81,148)(0,0)[t]@xmath178 ( 45,140.6)(6,0)6 ( 45,140.3)(6,0)6 ( 45,140)(20,0)30 ( 45,139.7)(6,0)6 ( 45,139.4)(6,0)6 ( 52,143)(0,0)@xmath179 ( 69,140.6)(6,0)6 ( 69,140.3)(6,0)6 ( 69,139.7)(6,0)6 ( 69,139.4)(6,0)6 ( 68,143)(0,0)@xmath180 ( 60,140 ) ( 60,145)(0,0)@xmath181 ( 60.6,135)(0,1)5 ( 60.3,135)(0,1)5 ( 60,135)(0,1)5 ( 59.7,135)(0,1)5 ( 59.4,135)(0,1)5 ( 60,132)(0,0)@xmath182 ( 45,130)(0,1)10 ( 39,134)(0,0)@xmath183 ( 40,140)(0,0)@xmath184 ( 45,130)(0,-1)3 ( 75,130)(0,1)10 ( 81,134)(0,0)@xmath185 ( 80,140)(0,0)@xmath181 ( 75,127)(0,1)3 ( 60,120)(0,0)@xmath186 ( 100,136)(0,1)12 ( 100,147)(5,5)[b ] ( 94,144)(0,0)[t]@xmath171 ( 100,120)(0,0)@xmath187 ( 120,126)(0,1)10 ( 114,130)(0,0)@xmath172 ( 120,123)(0,1)3 ( 120,125)(5,5)[t ] ( 120,120)(0,0)@xmath188 ( 10,90)(0,1)10 ( 10,100)(0,1)2 ( 5,98)(0,0)[t]@xmath171 ( 20,90)(0,1)10 ( 20,100)(0,1)2 ( 25,98)(0,0)[t]@xmath172 ( 10,90)(1,0)10 ( 15,93)(0,0)@xmath173 ( 10,80)(0,1)10 ( 10,80)(0,1)3 ( 5,83)(0,0)@xmath174 ( 10,78)(0,1)3 ( 20,80)(0,1)10 ( 20,80)(0,1)3 ( 25,83)(0,0)@xmath175 ( 20,78)(0,1)3 ( 15,70)(0,0)@xmath189 ( 80,90)(0,1)12 ( 80,101)(5,5)[b ] ( 80,90 ) ( 75,98)(0,0)[t]@xmath171 ( 110,90)(0,12)12 ( 110,102)(0,-1)3 ( 110,101)(5,5)[b ] ( 110,90 ) ( 115,98)(0,0)[t]@xmath174 ( 80,90.6)(6,0)6 ( 80,90.3)(6,0)6 ( 80,90)(20,0)30 ( 80,89.7)(6,0)6 ( 80,89.4)(6,0)6 ( 87,93)(0,0)@xmath190 ( 104,90.6)(6,0)6 ( 104,90.3)(6,0)6 ( 104,89.7)(6,0)6 ( 104,89.4)(6,0)6 ( 103,93)(0,0)@xmath191 ( 95,90 ) ( 95,95)(0,0)@xmath181 ( 95.6,85)(0,1)5 ( 95.3,85)(0,1)5 ( 95,85)(0,1)5 ( 94.7,85)(0,1)5 ( 94.4,85)(0,1)5 ( 95,82)(0,0)@xmath182 ( 80,80)(0,1)10 ( 74,84)(0,0)@xmath172 ( 75,90)(0,0)@xmath184 ( 80,80)(0,-1)3 ( 80,79)(5,5)[t ] ( 110,80)(0,1)10 ( 116,84)(0,0)@xmath175 ( 115,90)(0,0)@xmath181 ( 110,77)(0,1)3 ( 110,79)(5,5)[t ] ( 95,70)(0,0)@xmath192 ( 10,40)(0,1)10 ( 10,50)(0,1)2 ( 5,48)(0,0)[t]@xmath171 ( 10,40)(1,0)10 ( 15,37)(0,0)@xmath173 ( 16,48)(0,0)@xmath193 ( 16,45)(-1,0)2 ( 20,30)(0,1)10 ( 15,40)(10,10)[t ] ( 20,30)(0,1)3 ( 26,33)(0,0)@xmath175 ( 20,28)(0,1)3 ( 15,20)(0,0)@xmath194 ( 45,40)(0,1)12 ( 45,51)(5,5)[b ] ( 45,40 ) ( 40,48)(0,0)[t]@xmath171 ( 75,40)(0,1)12 ( 75,51)(5,5)[b ] ( 75,40 ) ( 80,48)(0,0)[t]@xmath172 ( 45,40.6)(6,0)6 ( 45,40.3)(6,0)6 ( 45,40)(20,0)30 ( 45,39.7)(6,0)6 ( 45,39.4)(6,0)6 ( 52,43)(0,0)@xmath179 ( 69,40.6)(6,0)6 ( 69,40.3)(6,0)6 ( 69,39.7)(6,0)6 ( 69,39.4)(6,0)6 ( 68,43)(0,0)@xmath195 ( 60,40 ) ( 60,45)(0,0)@xmath181 ( 60.6,35)(0,1)5 ( 60.3,35)(0,1)5 ( 60,35)(0,1)5 ( 59.7,35)(0,1)5 ( 59.4,35)(0,1)5 ( 60,32)(0,0)@xmath182 ( 45,30)(0,1)10 ( 39,34)(0,0)@xmath174 ( 40,40)(0,0)@xmath184 ( 45,27)(0,1)3 ( 45,29)(5,5)[t ] ( 75,30)(0,1)10 ( 81,34)(0,0)@xmath175 ( 80,40)(0,0)@xmath181 ( 75,27)(0,1)3 ( 75,29)(5,5)[t ] ( 60,20)(0,0)@xmath196 ( 110,42 ) ( 110,37)(0,0)@xmath184 ( 110.6,42)(0,1)5 ( 110.3,42)(0,1)5 ( 110,42)(0,1)5 ( 109.7,42)(0,1)5 ( 109.4,42)(0,1)5 ( 115,45)(0,0)@xmath182 ( 100,32)(1,1)10 ( 100,27)(0,1)5 ( 94,34)(0,0)@xmath171 ( 100,30)(0,-1)3 ( 100,29)(5,5)[t ] ( 120,32)(-1,1)10 ( 120,27)(0,1)5 ( 126,34)(0,0)@xmath175 ( 120,27)(0,1)3 ( 120,29)(5,5)[t ] ( 110,20)(0,0)@xmath197 it must be noted that in expression ( [ eq : pas - a ] ) the projection @xmath198 of @xmath199 , as well as that of the momentum line @xmath200 in graph @xmath169 are the same . this is also to be said about the remaining operators of second quantization and the three open lines af graph @xmath169 . as it has been mentiond in section 2 , this operator has two tensorial forms , ( [ eq : op - c ] ) and ( [ eq : op - d ] ) . these may be represented graphically , since the creation operator @xmath201 , as well as operator @xmath202 respectivelly , are graphically denoted by diagrams @xmath203 and @xmath204 . the first form ( [ eq : op - c ] ) of two - particle operator @xmath205 is represented as : @xmath206{c } g_i(ij , i^{\prime } j^{\prime } ) = d_1=-\frac 12 \displaystyle { \sum_{\kappa _ { 12}\kappa _ { 12}^{^{\prime } } \sigma _ { 12}\sigma _ { 12}^{^{\prime } } } } \displaystyle { \sum_p}\left ( -1\right ) ^{k - p}\left [ \kappa _ 1,\kappa _ 2,\sigma _ 1,\sigma _ 2\right ] ^{1/2}\times \\ \times \left ( n_i\lambda _ in_j\lambda _ j||g^{\left ( \kappa _ 1\kappa _ 2k,\sigma _ 1\sigma _ 2k\right ) } ||n_i^{\prime } \lambda _ i^{\prime } n_j^{\prime } \lambda _ j^{\prime } \right ) \left\ { \begin{array}{ccc } l_i^{\prime } & l_j^{\prime } & \kappa _ { 12}^{\prime } \\ \kappa _ 1 & \kappa _ 2 & k \\ l_i & l_j & \kappa _ { 12 } \end{array } \right\ } \times \\ \times \left\ { \begin{array}{ccc } s & s & \sigma _ { 12}^{\prime } \\ \sigma _ 1 & \sigma _ 2 & k \\ s & s & \sigma _ { 12 } \end{array } \right\ } d_6 \end{array}\ ] ] whereas the second ( [ eq : op - d ] ) : @xmath207{c } g_{ii}(ij , i^{\prime } j^{\prime } ) = d_5+d_7= \\ = \frac 12 \displaystyle { \sum_p}\left ( -1\right ) ^{k - p}\left ( n_i\lambda _ in_j\lambda _ j||g^{\left ( \kappa _ 1\kappa _ 2k,\sigma _ 1\sigma _ 2k\right ) } ||n_i^{\prime } \lambda _ i^{\prime } n_j^{\prime } \lambda _ j^{\prime } \right ) \times \\ \times \{\left [ \kappa _ 1,\kappa _ 2,\sigma _ 1,\sigma _ 2\right ] ^{-1/2}d_8-\left ( -1\right ) ^{l_i+l_j^{\prime } } \left\ { \begin{array}{ccc } \kappa _ 1 & \kappa _ 2 & k \\ l_j^{\prime } & l_i & l_j \end{array } \right\ } \times \\ \times \left\ { \begin{array}{ccc } \sigma _ 1 & \sigma _ 2 & k \\ s & s & s \end{array } \right\ } \delta \left ( n_jl_j , n_i^{\prime } l_i^{\prime } \right ) d_9\}. \end{array}\ ] ] we emphasize here that the winding line of interaction in the feynman - goldstone diagram corresponds to the operators of second quantization in the normal order ( figure [ fig : dd - a ] , @xmath208 ) . whereas the dotted interaction line indicates that the second quantization operators are ordered in pairs of creation - annihilation operators . in the latter case first comes the pair on the left side of a feynman - goldstone diagram ( figure [ fig : dd - a ] , @xmath209 ) . such a notation of two kinds for an interaction line is meaningful only for two - particle ( or more ) operators , since for any one - particle operator both the winding and dotted lines correspond to the same order of creation and annihilation operators . from expressions ( [ eq : pas - b ] ) , ( [ eq : pas - c ] ) we see that the two - particle operator in the first form is represented by one feynman - goldstone diagram @xmath208 , whereas in the second - by two diagrams @xmath209 and @xmath210 . the diagrams , corresponding to tensorial product , have the following algebraic expressions : @xmath211 ^{\left ( \kappa _ { 12}\sigma _ { 12}\right ) } \times \left [ \stackrel{\sim } { a}^{\left ( \lambda _ i^{\prime } \right ) } \times \stackrel{\sim } { a}^{\left ( \lambda _ j^{\prime } \right ) } \right ] ^{\left ( \kappa _ { 12}^{\prime } \sigma _ { 12}^{\prime } \right ) } \right ] _ { p - p}^{\left ( kk\right ) } , \ ] ] @xmath212 ^{\left ( \kappa _ 1\sigma _ 1\right ) } \times \left [ a^{\left ( \lambda _ j\right ) } \times \stackrel{\sim } { a}^{\left ( \lambda _ j^{\prime } \right ) } \right ] ^{\left ( \kappa _ 2\sigma _ 2\right ) } \right ] _ { p - p}^{\left ( kk\right ) } \ ] ] @xmath213 _ { p - p}^{\left ( kk\right ) } \ ] ] the positions of the second quantization operators in the diagrams @xmath214 , @xmath215 and @xmath216 define their order in tensorial products : the first place in tensorial product occupies the upper right second quantization operator , the second - lower right , after them the upper left and lower left operators follow . the angular momenta diagram defines their coupling scheme into tensorial product . for more detail see gaigalas and rudzikas @xcite . as it has been mentioned earlier , these two forms do not always take full advantage of the racah algebra ( see gaigalas and rudzikas @xcite ) . the expression ( [ eq : op - i ] ) has no such shortcomings . now we will demonstrate the way to obtain graphically a tensorial expressions for particular distributions 1 - 42 from table 1 . ( 148,120 ) ( 10,105)(0,1)10 ( 10,115)(0,1)2 ( 5,113)(0,0)[t]@xmath217 ( 20,105)(0,1)10 ( 20,115)(0,1)2 ( 25,113)(0,0)[t]@xmath218 ( 11,105)(2,0)5(2,1)[t ] ( 15,110)(0,0)@xmath173 ( 10,95)(0,1)10 ( 10,95)(0,1)3 ( 5,98)(0,0)@xmath219 ( 10,93)(0,1)3 ( 20,95)(0,1)10 ( 20,95)(0,1)3 ( 25,98)(0,0)@xmath217 ( 20,93)(0,1)3 ( 15,85)(0,0)@xmath220 ( 70,105)(0,1)12 ( 70,105 ) ( 65,113)(0,0)[t]@xmath221 ( 100,105)(0,12)12 ( 100,117)(0,-1)3 ( 100,105 ) ( 106,113)(0,0)[t]@xmath222 ( 70,105.6)(6,0)6 ( 70,105.3)(6,0)6 ( 70,105)(20,0)30 ( 70,104.7)(6,0)6 ( 70,104.4)(6,0)6 ( 77,108)(0,0)@xmath179 ( 94,105.6)(6,0)6 ( 94,105.3)(6,0)6 ( 94,104.7)(6,0)6 ( 69,104.4)(6,0)6 ( 93,108)(0,0)@xmath180 ( 85,105 ) ( 85,110)(0,0)@xmath181 ( 85.6,100)(0,1)5 ( 85.3,100)(0,1)5 ( 85,100)(0,1)5 ( 84.7,100)(0,1)5 ( 84.4,100)(0,1)5 ( 85,97)(0,0)@xmath182 ( 70,95)(0,1)10 ( 65,99)(0,0)@xmath223 ( 65,105)(0,0)@xmath184 ( 70,95)(0,-1)3 ( 100,95)(0,1)10 ( 106,99)(0,0)@xmath224 ( 105,105)(0,0)@xmath181 ( 100,92)(0,1)3 ( 85,85)(0,0)@xmath225 ( 10,50)(0,1)12 ( 10,61)(5,5)[b ] ( 10,50 ) ( 5,58)(0,0)[t]@xmath226 ( 40,50)(0,12)12 ( 40,62)(0,-1)3 ( 40,61)(5,5)[b ] ( 40,50 ) ( 46,58)(0,0)[t]@xmath227 ( 10,50.6)(6,0)6 ( 10,50.3)(6,0)6 ( 10,50)(20,0)30 ( 10,49.7)(6,0)6 ( 10,49.4)(6,0)6 ( 17,53)(0,0)@xmath190 ( 34,50.6)(6,0)6 ( 34,50.3)(6,0)6 ( 34,49.7)(6,0)6 ( 34,49.4)(6,0)6 ( 33,53)(0,0)@xmath191 ( 25,50 ) ( 25,55)(0,0)@xmath181 ( 25.6,45)(0,1)5 ( 25.3,45)(0,1)5 ( 25,45)(0,1)5 ( 24.7,45)(0,1)5 ( 24.4,45)(0,1)5 ( 25,42)(0,0)@xmath182 ( 10,40)(0,1)10 ( 4,44)(0,0)@xmath226 ( 5,50)(0,0)@xmath184 ( 10,40)(0,-1)3 ( 10,39)(5,5)[t ] ( 40,40)(0,1)10 ( 46,44)(0,0)@xmath228 ( 45,50)(0,0)@xmath181 ( 40,37)(0,1)3 ( 40,39)(5,5)[t ] ( 25,20)(0,0)@xmath229 ( 70,63 ) ( 100,63 ) ( 70,63)(1,-1)30 ( 100,63)(-1,-1)30 ( 70,63.6)(6,0)6 ( 70,63.3)(6,0)6 ( 70,63)(20,0)30 ( 70,62.7)(6,0)6 ( 70,62.4)(6,0)6 ( 77,66)(0,0)@xmath179 ( 94,63.6)(6,0)6 ( 94,63.3)(6,0)6 ( 94,62.7)(6,0)6 ( 94,62.4)(6,0)6 ( 93,66)(0,0)@xmath180 ( 85,63 ) ( 85,68)(0,0)@xmath184 ( 85.6,63)(0,-1)30 ( 85.3,63)(0,-1)30 ( 85,63)(0,-1)30 ( 84.7,63)(0,-1)30 ( 84.4,63)(0,-1)30 ( 81,59)(0,0)@xmath182 ( 77,55)(0,0)[t]@xmath230 ( 65,48)(0,0)[t]@xmath221 ( 70,68)(0,0)@xmath181 ( 70,63)(0,-1)30 ( 95,55)(0,0)[t]@xmath221 ( 106,48)(0,0)[t]@xmath222 ( 100,68)(0,0)@xmath181 ( 100,63)(0,-1)30 ( 85,33 ) ( 85,28)(0,0)@xmath184 ( 70,28)(0,0)@xmath181 ( 70,33 ) ( 70,33.6)(6,0)6 ( 70,33.3)(6,0)6 ( 70,33)(20,0)30 ( 70,32.7)(6,0)6 ( 70,32.4)(6,0)6 ( 77,30)(0,0)@xmath190 ( 100,28)(0,0)@xmath184 ( 100,33 ) ( 94,33.6)(6,0)6 ( 94,33.3)(6,0)6 ( 94,32.7)(6,0)6 ( 94,32.4)(6,0)6 ( 93,30)(0,0)@xmath191 ( 85,20)(0,0)@xmath231 we take the distribution @xmath160 , @xmath159 , @xmath158 , @xmath160 ( 13 form table 1 ) as an example for investigation . then the feynman - goldstone diagram of operator @xmath232 is @xmath233 , the angular momentum graph is @xmath234 , and the second quantization operators are in the following order : @xmath235 applying the wick s theorem as described in section 3 , and assuming that the operators acting upon shell @xmath160 belong to the first group , the once acting upon @xmath159 belong to the second , and the ones acting upon @xmath158 belong to the third group , we obtain the following order of operators : @xmath236 using the generalized graphical technique of angular momentum by gaigalas _ @xcite , we couple the operators of second quantization into tensorial product @xmath237 ( see figure [ fig : dd - c ] ) : @xmath238 ^{\left ( \kappa _ { 12}\sigma _ { 12}\right ) } \times \left [ \stackrel{\sim } { a}^{\left ( \lambda _ { \beta } \right ) } \times \stackrel{\sim } { a}^{\left ( \lambda _ { \alpha}\right ) } \right ] ^{\left ( \kappa _ { 12}^{\prime } \sigma _ { 12}^{\prime } \right ) } \right ] _ { p - p}^{\left ( kk\right ) } .\ ] ] in the course of obtaining @xmath237 graphically , a recoupling matrix @xmath239 appears , whose analytical expression is readily obtained by using the graphical technique of jucys and bandzaitis @xcite . all the needed expressions are obtained in the same way . ( 148,105 ) ( 20,90)(0,1)10 ( 20,100)(0,1)2 ( 15,98)(0,0)[t]@xmath240 ( 30,90)(0,1)10 ( 30,100)(0,1)2 ( 35,98)(0,0)[t]@xmath241 ( 20,90)(1,0)10 ( 25,93)(0,0)@xmath173 ( 20,80)(0,1)10 ( 20,80)(0,1)3 ( 15,83)(0,0)@xmath242 ( 20,78)(0,1)3 ( 30,80)(0,1)10 ( 30,80)(0,1)3 ( 35,83)(0,0)@xmath243 ( 30,78)(0,1)3 ( 25,70)(0,0)@xmath244 ( 80,90)(0,1)12 ( 80,101)(5,5)[b ] ( 80,90 ) ( 74,98)(0,0)[t]@xmath240 ( 110,90)(0,1)12 ( 110,101)(5,5)[b ] ( 110,90 ) ( 115,98)(0,0)[t]@xmath241 ( 80,90.6)(6,0)6 ( 80,90.3)(6,0)6 ( 80,90)(20,0)30 ( 80,89.7)(6,0)6 ( 80,89.4)(6,0)6 ( 87,93)(0,0)@xmath179 ( 104,90.6)(6,0)6 ( 104,90.3)(6,0)6 ( 104,89.7)(6,0)6 ( 104,89.4)(6,0)6 ( 103,93)(0,0)@xmath195 ( 95,90 ) ( 95,95)(0,0)@xmath181 ( 95.6,85)(0,1)5 ( 95.3,85)(0,1)5 ( 95,85)(0,1)5 ( 94.7,85)(0,1)5 ( 94.4,85)(0,1)5 ( 95,82)(0,0)@xmath182 ( 80,80)(0,1)10 ( 74,84)(0,0)@xmath242 ( 75,90)(0,0)@xmath184 ( 80,77)(0,1)3 ( 80,79)(5,5)[t ] ( 110,80)(0,1)10 ( 116,84)(0,0)@xmath243 ( 115,90)(0,0)@xmath181 ( 110,77)(0,1)3 ( 110,79)(5,5)[t ] ( 95,70)(0,0)@xmath245 ( 10,40)(0,1)12 ( 10,51)(5,5)[b ] ( 10,40 ) ( 4,48)(0,0)[t]@xmath241 ( 40,40)(0,1)12 ( 40,51)(5,5)[b ] ( 40,40 ) ( 45,48)(0,0)[t]@xmath240 ( 10,40.6)(6,0)6 ( 10,40.3)(6,0)6 ( 10,40)(20,0)30 ( 10,39.7)(6,0)6 ( 10,39.4)(6,0)6 ( 17,43)(0,0)@xmath179 ( 34,40.6)(6,0)6 ( 34,40.3)(6,0)6 ( 34,39.7)(6,0)6 ( 34,39.4)(6,0)6 ( 33,43)(0,0)@xmath195 ( 25,40 ) ( 25,45)(0,0)@xmath184 ( 28,37)(0,0)@xmath246 ( 25.6,35)(0,1)5 ( 25.3,35)(0,1)5 ( 25,35)(0,1)5 ( 24.7,35)(0,1)5 ( 24.4,35)(0,1)5 ( 25,32)(0,0)@xmath182 ( 10,30)(0,1)10 ( 4,34)(0,0)@xmath243 ( 5,40)(0,0)@xmath184 ( 10,27)(0,1)3 ( 10,29)(5,5)[t ] ( 40,30)(0,1)10 ( 46,34)(0,0)@xmath242 ( 45,40)(0,0)@xmath181 ( 40,27)(0,1)3 ( 40,29)(5,5)[t ] ( 25,20)(0,0)@xmath247 ( 80,40)(0,1)12 ( 80,51)(5,5)[b ] ( 80,40 ) ( 74,48)(0,0)[t]@xmath241 ( 110,40)(0,1)12 ( 110,51)(5,5)[b ] ( 110,40 ) ( 115,48)(0,0)[t]@xmath240 ( 80,40.6)(6,0)6 ( 80,40.3)(6,0)6 ( 80,40)(20,0)30 ( 80,39.7)(6,0)6 ( 80,39.4)(6,0)6 ( 87,43)(0,0)@xmath179 ( 104,40.6)(6,0)6 ( 104,40.3)(6,0)6 ( 104,39.7)(6,0)6 ( 104,39.4)(6,0)6 ( 103,43)(0,0)@xmath195 ( 95,40 ) ( 95,45)(0,0)@xmath181 ( 98,37)(0,0)@xmath246 ( 95.6,35)(0,1)5 ( 95.3,35)(0,1)5 ( 95,35)(0,1)5 ( 94.7,35)(0,1)5 ( 94.4,35)(0,1)5 ( 95,32)(0,0)@xmath182 ( 80,30)(0,1)10 ( 74,34)(0,0)@xmath243 ( 75,40)(0,0)@xmath184 ( 80,27)(0,1)3 ( 80,29)(5,5)[t ] ( 110,30)(0,1)10 ( 116,34)(0,0)@xmath242 ( 115,40)(0,0)@xmath181 ( 110,27)(0,1)3 ( 110,29)(5,5)[t ] ( 95,20)(0,0)@xmath248 in obtaining these expressions , as well as in representing them graphically , it is very convenient to use the rule of changing the sign of a node , existing in the graphical technique of angular momentum ( see jucys and bandzaitis @xcite ) . therefore now we will treat an example of using such a rule . suppose , we have the following correspondence between diagrams ( figure [ fig : dd - c ] ) : @xmath249 in which the second quantization operators are in the order @xmath250 @xmath251 @xmath252 @xmath253 . our goal is to obtain the diagram corresponding to the order @xmath254 @xmath253 @xmath255 @xmath251 . bearing in mind that the second quantization operators anticommute with each other and they all act on different electronic shells and we are not changing the order of their coupling into tensorial product , we arrive at @xmath256 let us also discuss another situation : we have defined the disposition of the operators and we want to couple them into certain tensorial product . suppose that we want to represent graphically the following tensorial product : @xmath257 ^{\left ( \kappa _ 1\sigma _ 1\right ) } \times \left [ a^{(\lambda _ 3)}\times \tilde a^{(\lambda _ 4)}\right ] ^{\left ( \kappa _ 2\sigma _ 2\right ) } \right ] ^{\left ( \kappa \sigma \right ) } .\ ] ] for this purpose we have to rearrange the generalized clebsch - gordan coefficient , which is defining the scheme of coupling of the operators into the tensorial product . it is easy to notice that this coefficient will properly define the tensorial product , if we change the sign of the vertex @xmath258 in diagram @xmath259 : @xmath260 the procedures described are fairly simple , however , they are sufficient for the majority of cases . the more complete description of this generalized graphical approach may be found in gaigalas _ @xcite , gaigalas @xcite , gaigalas and merkelis @xcite . all the analytical expressions for distributions 1 - 42 from table 1 are presented in the paper gaigalas _ et al _ they are written down using the generalized graphical methodology of angular momentum , and the vortex sign change rule , which was discussed in this section . as a consequence of that , the analytical expressions for 42 terms may be written down via 6 different expressions . this , undoubtedly , facilitates a lot the implementation of methodology proposed in gaigalas _ et al _ @xcite . in this section we will discuss several ways to obtain matrix elements of a two - particle operator . as it was mentioned earlier , up to now the fano calculation scheme @xcite is the most popular one . its general expression when a two - particle operator acts upon different shells is presented in ( [ eq : ma - c ] ) . the general expression for a matrix element in other cases is similar . for example , when the operator acts only upon one shell , we have @xmath261{c } ( \psi _ u^{bra}\left ( ls\right ) || \widehat{g}^{\left ( \kappa _ 1\kappa _ 2k,\sigma_1\sigma _ 2k\right ) } \sim \displaystyle { \sum_{n_{i } \lambda _ { i } } } \left ( -1\right ) ^\delta n_{i } \left ( n_{i}-1 \right ) \times \\ \times \displaystyle { \sum _ { \ { t \ } } } \left ( l_{i}^{n_{i}}\;\alpha_{i } l_{i}s_{i}||l_{i}^{n_{i}-1}\;\left ( \alpha_{i}^{\prime \prime } l^{\prime \prime}_{i } s^{\prime \prime}_{i } \right ) , l_{i } \right ) \left ( l_{i}^{n_{i}-1}\;\alpha_{i}^{\prime \prime } l_{i}^{\prime \prime}s_{i}^{\prime \prime } \alpha_{i}^{\prime \prime \prime } l_{i}^{\prime \prime \prime } s^{\prime \prime \prime}_{i } \right ) , l_{i } \right ) \times \\ \times \left ( l_{i}^{n_{i}^{\prime } } \ ; \alpha_{i}^{\prime \prime \prime } l_{i}^{\prime \prime \prime}s_{i}^{\prime \prime \prime } \alpha_{i}^{\prime v } l^{\prime v}_{i } s^{\prime v}_{i } \right ) , l_{i } \right ) \left ( l_{i}^{n_{i}^{\prime } -1}\;\alpha_{i}^{\prime v } l_{i}^{\prime v}s_{i}^{\prime v}||l_{i}^{n_{i}^{\prime } -2}\;\left ( \alpha_{i}^{\prime } l^{\prime}_{i } s^{\prime } _ { i } \right ) , l_{i } \right ) \times \\ \times r_{d } \left ( \lambda _ i,\lambda _ i,\lambda _ i,\lambda _ i , \kappa_1 , \kappa_2 , \kappa , \sigma_1 , \sigma_2 , \sigma , \lambda ^{bra } , \lambda ^{ket } \right ) \times \\ \times \left ( n_i\lambda _ in_i\lambda _ i|| g^{\left ( \kappa _ 1\kappa _ 2k,\sigma_1\sigma _ 2k\right ) } \end{array}\ ] ] as we see here , in contrast to ( [ eq : ma - c ] ) , the summation is performed only over one array @xmath38 , @xmath39 of quantum numbers , because the operator acts only upon one shell . but here a summation over @xmath262 occurs , though , which indicates the summation over arrays of intermediate terms @xmath263 , @xmath264 , @xmath265 . remembering the relationship between a coefficient of fractional parentage and a reduced matrix element of a second quantization operator ( see pakauskas _ et al _ @xcite , rudzikas and kaniauskas @xcite ) : @xmath266{c } \left ( l^{n}\;\alpha ls||a^{(ls)}|| l^{n-1}\ ; \alpha^{\prime } l^{\prime } s^{\prime } \right ) = \\ = ( -1)^{n+(n+1)\varphi ( n ) } \sqrt{n[l , s ] } \left ( l^{n}\;\alpha ls||l^{n-1}\;\left ( \alpha^{\prime } l^{\prime } s^{\prime } \right ) , l \right ) \end{array}\ ] ] we see that the racah algebra in expressions ( [ eq : ma - c ] ) , ( [ eq : ma - d ] ) is used only on the level of coefficients of fractional parentage . in separate cases , e.g. when the two - particle operator acts upon one or two shells , it is possible to use expressions which exploit the racah algebra at a higher level , i.e. to take more advantage of the tensor algebra elements ( see judd @xcite , jucys and savukynas @xcite ) . for example , let us investigate the case when a matrix element is calculated for bra and ket functions having one shell only . the tensorial forms ( [ eq : op - c ] ) and ( [ eq : op - e ] ) are of value here . taking the second one of these , we have @xmath267{c } ( nl^{n}\alpha ls|\widehat{g}_{ii}|nl^{n}\alpha ^{\prime}l^{\prime}s^{\prime})= \\ = \frac 12 \left ( n\lambda n\lambda n\lambda \right ) \left\ { \left [ \kappa _ 1,\kappa _ 2,\sigma_1,\sigma _ 2\right ] ^{-1/2 } \right . \times \\ \times ( nl^{n}\alpha ls| \left [ \left [ a^{\left ( \lambda \right ) } \times \stackrel{\sim } { a}^{\left ( \lambda \right ) } \right ] ^{\left ( \kappa _ 1\sigma _ 1\right ) } \times \left [ a^{\left ( \lambda \right ) } \times \stackrel{\sim } { a}^{\left ( \lambda \right ) } \right ] ^{\left ( \kappa _ 2\sigma _ 2\right ) } \right ] ^{\left ( kk\right ) } l^{\prime } s^{\prime } ) - \\ -\left\ { \begin{array}{ccc } \kappa _ 1 & \kappa _ 2 & k \\ l & l & l \end{array } \right\ } \left\ { \begin{array}{ccc } \sigma _ 1 & \sigma _ 2 & k \\ s & s & s \end{array } \right\ } \times \\ \left . \times ( nl^{n}\alpha ls| \left [ a^{\left ( \lambda \right ) } \times \stackrel{\sim } { a}^{\left ( \lambda \right ) } \right ] ^{\left ( kk\right ) } |nl^{n}\alpha l^{\prime } s^{\prime } ) \right\}. \end{array}\ ] ] using the relationships between the tensorial product of creation and annihilation operators and the tensorial quantities @xmath166 and @xmath167 ( see rudzikas and kaniauskas @xcite ) , the expression ( [ eq : ma - f ] ) for matrix elements can be writen down in terms of @xmath166 and @xmath167 . in comparing ( [ eq : ma - d ] ) to ( [ eq : ma - f ] ) we see that the summation over intermediate terms @xmath263 , @xmath264 , @xmath265 is already performed in expression ( [ eq : ma - f ] ) . so , in this case the racah algebra is exploited at the level of standard quantities @xmath166 ir @xmath167 . this simplifies calculations a lot : * for zero matrix elements are easily tracked down from triangular conditions even before the actual calculation is performed . in case ( [ eq : ma - f ] ) only the triangular conditions @xmath268 and @xmath269 are present , but their number may be greater in other cases . ( in the above , the notation @xmath268 means the triangular condition @xmath270 . ) * the tables of standard quantities ( see nielson and koster @xcite , karazija _ et al _ @xcite , cowan @xcite ) may be used . * the recoupling matrix is simpler in this case , and it has an analytical expression . so , the expressions exploiting the racah algebra at the level of @xmath166 and @xmath167 are much more advantageous than ( [ eq : ma - c ] ) . such expressions are obtained for all physical operators . for example , the expressions for spin - other - orbit operator are presented in papers horie @xcite , karazija _ et al _ @xcite and vizbarait . e _ et al _ @xcite , the ones for spin - spin operator - in papers horie @xcite and karazija _ et al _ @xcite , and the ones for orbit - orbit operator in the monograph jucys and savukynas @xcite . the shortcoming of the expressions of this type is that the racah algebra is exploited to its full extent in separate cases only . this is discussed in detail in paper by gaigalas _ _ @xcite . gaigalas _ et al _ @xcite have proposed a methodology which allows one to take all the advantages of the racah algebra in the most general case . according to the approach by gaigalas _ et al _ @xcite , a general expression of submatrix element for any two - particle operator between functions with @xmath22 open shells can be written down as follows : @xmath271{c } ( \psi _ u^{bra}\left ( ls\right ) || \widehat{g}^{\left ( \kappa _ 1\kappa _ 2k,\sigma_1\sigma _ 2k\right ) } } \right ) ) = \\ = \displaystyle { \sum_{n_il_i , n_jl_j , n_i^{\prime } l_i^{\prime } , n_j^{\prime } l_j^{\prime } } } ( \psi _ u^{bra}\left ( ls\right ) ||\widehat{g } \left ( n_il_i , n_jl_j , n_i^{\prime } l_i^{\prime } , n_j^{\prime } l_j^{\prime } \right ) } \right ) ) = \\ = \displaystyle { \sum_{n_il_i , n_jl_j , n_i^{\prime } l_i^{\prime } , n_j^{\prime } l_j^{\prime } } } \displaystyle { \sum_{\kappa _ { 12},\sigma _ { 12},\kappa _ { 12}^{\prime } , \sigma _ { 12}^{\prime } } } \sum \left ( -1\right ) ^\delta \theta ^{\prime } \left ( n_i\lambda _ i , n_j\lambda _ j , n_i^{\prime } \lambda _ i^{\prime } , n_j^{\prime } \lambda _ j^{\prime } , \xi \right ) \times \\ \times t\left ( n_i\lambda _ i , n_j\lambda _ j , n_i^{\prime } \lambda _ i^{\prime } , n_j^{\prime } \lambda _ j^{\prime } , \lambda ^{bra},\lambda ^{ket},\xi , \gamma \right ) \times \\ \times r\left ( \lambda _ i,\lambda _ j,\lambda _ i^{\prime } , \lambda _ j^{\prime } , \lambda ^{bra},\lambda ^{ket},\gamma \right ) . \end{array}\ ] ] in calculating the spin - angular part of a submatrix element using ( [ eq : ma - g ] ) , one has to compute the following quantities ( for more detail see gaigalas @xcite ) : 1 . the recoupling matrix @xmath272 . this recoupling matrix accounts for the change in going from matrix element + @xmath273 , which has @xmath22 open shells in the bra and ket functions , to the submatrix element + @xmath274 , which has only the shells being acted upon by the two - particle operator in its bra and ket functions . 2 . the submatrix element @xmath275 , which denotes the submatrix elements of operators ( [ eq : op - j ] ) - ( [ eq : op - n ] ) . here @xmath276 refers to the array of coupling parameters connecting the recoupling matrix @xmath272 to the submatrix element . phase factor @xmath277 ( for more detail see gaigalas @xcite ) . 4 . @xmath278 , which is proportional to the radial part and corresponds to one of @xmath279, ... ,@xmath280 . it consists of a submatrix element @xmath281 , and in some cases of simple factors and 3@xmath282-coefficients ( for more detail see gaigalas @xcite ) . in the next section we shall discuss finding of the recoupling matrix . now we shall analyse the submatrix element @xmath275 . as the angular part of expression ( [ eq : ma - g ] ) contains the tensors ( [ eq : op - j ] ) - ( [ eq : op - n ] ) , so we will discuss the derivation of submatrix elements of these operators , and present the expressions for these quantities . it is worth noting that these tensorial quantities all act upon the _ same _ shell . so , all the advantages of tensor algebra and the quasispin formalism may be exploited efficiently . we obtain the submatrix elements of operator ( [ eq : op - j ] ) by straightforwardly using the wigner - eckart theorem in quasispin space ( see rudzikas @xcite ) : @xmath283{c } \left ( l^n\;\alpha qls||a_{m_q}^{\left ( qls\right ) } ||l^{n^{\prime } } \;\alpha ^{\prime } q^{\prime } l^{\prime } s^{\prime } \right ) = -\left [ q\right ] ^{-1/2}\left [ \begin{array}{ccc } q^{\prime } & 1/2 & q \\ m_q^{\prime } & m_q & m_q \end{array } \right ] \times \\ \times \left ( l\;\alpha qls|||a^{\left ( qls\right ) } |||l\;\alpha ^{\prime } q^{\prime } l^{\prime } s^{\prime } \right ) , \end{array}\ ] ] where the last multiplier in ( [ eq : ma - h ] ) is the so - called completely reduced ( reduced in the quasispin , orbital and spin spaces ) matrix element . the value of the submatrix element of operator ( [ eq : op - k ] ) is obtained by @xmath284{c } \left ( nl^n\;\alpha qls||\left [ a_{m_{q1}}^{\left ( q\lambda \right ) } \times a_{m_{q2}}^{\left ( q\lambda \right ) } \right ] ^{\left ( k_1k_2\right ) } ||nl^{n^{\prime } } \;\alpha ^{\prime } q^{\prime } l^{\prime } s^{\prime } \right ) = \\ = \displaystyle { \sum_{\epsilon , m_\epsilon } } \left [ q\right ] ^{-1/2}\left [ \begin{array}{ccc } q & q & \epsilon \\ m_{q1 } & m_{q2 } & m_\epsilon \end{array } \right ] \left [ \begin{array}{ccc } q^{\prime } & \epsilon & q \\ m_q^{\prime } & m_\epsilon & m_q \end{array } \right ] \times \\ \times \left ( nl\;\alpha qls|||w^{\left ( \epsilon k_1k_2\right ) } |||nl\;\alpha ^{\prime } q^{\prime } l^{\prime } s^{\prime } \right ) . \end{array}\ ] ] on the right - hand side of equations ( [ eq : ma - h ] ) and ( [ eq : ma - i ] ) only the clebsch - gordan coefficient @xmath285 $ ] depends on the number @xmath286 of equivalent electrons . @xmath287 denotes reduced in quasispin space submatrix element ( completely reduced matrix element ) of the triple tensor @xmath288 ^{\left ( \epsilon k_1k_2\right ) } $ ] . it is related to the completely reduced coefficients ( subcoefficients ) of fractional parentage in a following way : @xmath289{c } \left ( nl\;\alpha qls|||w^{\left ( \epsilon k_1k_2\right ) } |||nl\;\alpha ^{\prime } q^{\prime } l^{\prime } s^{\prime } \right ) = \\ = \left ( -1\right ) ^{q+l+s+q^{\prime } + l^{\prime } + s^{\prime } + \epsilon + k_1+k_2}\left [ \epsilon , k_1,k_2\right ] ^{1/2}\times \\ \times \displaystyle { \sum_{\alpha ^{\prime \prime } q^{\prime \prime } l^{\prime \prime } s^{\prime \prime } } } \left ( l\;\alpha qls|||a^{\left ( qls\right ) } |||l\;\alpha ^{\prime \prime } q^{\prime \prime } l^{\prime \prime } s^{\prime \prime } \right ) \times \\ \times \left ( l\;\alpha ^{\prime \prime } q^{\prime \prime } l^{\prime \prime } s^{\prime \prime } |||a^{\left ( qls\right ) } |||l\;\alpha ^{\prime } q^{\prime } l^{\prime } s^{\prime } \right ) \times \\ \times \left\ { \begin{array}{ccc } q & q & \epsilon \\ q^{\prime } & q & q^{\prime \prime } \end{array } \right\ } \left\ { \begin{array}{ccc } l & l & k_1 \\ l^{\prime } & l & l^{\prime \prime } \end{array } \right\ } \left\ { \begin{array}{ccc } s & s & k_2 \\ s^{\prime } & s & s^{\prime \prime } \end{array } \right\ } . \end{array}\ ] ] in the other three cases ( [ eq : op - l ] ) , ( [ eq : op - m ] ) , ( [ eq : op - n ] ) we obtain the submatrix elements of these operators by using ( 2.28 ) of jucys and savukynas @xcite : @xmath290{c } ( nl^n\;\alpha qls||\left [ f^{\left ( \kappa _ 1\sigma _ 1\right ) } \left ( n\lambda \right ) \times g^{(\kappa _ 2\sigma _ 2)}\left ( n\lambda \right ) \right ] ^{\left ( kk\right ) } ||nl^{n^{\prime } } \;\alpha ^{\prime } q^{\prime } l^{\prime } s^{\prime } ) = \\ = \left ( -1\right ) ^{l+s+l^{\prime } + s^{\prime } + 2k}\left [ k\right ] \times \\ \times \displaystyle { \sum_{\alpha ^{\prime \prime } q^{\prime \prime } l^{\prime \prime } s^{\prime \prime } } } ( nl^n\;\alpha qls||f^{\left ( \kappa _ 1\sigma _ 1\right ) } \left ( n\lambda \right ) ||nl^{n^{\prime \prime } } \;\alpha ^{\prime \prime } q^{\prime \prime } l^{\prime \prime } s^{\prime \prime } ) \times \\ \times ( nl^{n^{\prime \prime } } \;\alpha ^{\prime \prime } q^{\prime \prime } l^{\prime \prime } s^{\prime \prime } ||g^{(\kappa _ 2\sigma _ 2)}\left ( n\lambda \right ) ||nl^{n^{\prime } } \;\alpha ^{\prime } q^{\prime } l^{\prime } s^{\prime } ) \times \\ \times \left\ { \begin{array}{ccc } \kappa _ 1 & \kappa _ 2 & k \\ l^{\prime } & l & l^{\prime \prime } \end{array } \right\ } \left\ { \begin{array}{ccc } \sigma _ 1 & \sigma _ 2 & k \\ s^{\prime } & s & s^{\prime \prime } \end{array } \right\ } , \end{array}\ ] ] where @xmath291 , @xmath292 are one of ( [ eq : op - j ] ) or ( [ eq : op - k ] ) and the submatrix elements correspondingly are defined by ( [ eq : ma - h ] ) , ( [ eq : ma - i ] ) and ( [ eq : ma - j ] ) . @xmath293 is defined by the second quantization operators occurring in @xmath294 and @xmath295 . as is seen , by using this approach gaigalas _ et al _ @xcite , the calculation of the angular parts of matrix elements between functions with @xmath22 open shells is reduced to requiring the reduced coefficients of fractional parentage or the tensors ( for example @xmath296 ) , which are independent of the occupation number of the shell and are acting on one shell of equivalent electrons . the main advantage of this approach is that the standard data tables in such a case will be much smaller in comparison with tables of the usual coefficients @xmath297 @xmath298 ( see jucys and savukynas @xcite ) and , therefore , many summations will be less time - consuming . also one can see that in such an approach the submatrix elements of standard tensors and subcoefficients of fractional parentage actually can be treated in a uniform way as they all are the completely reduced matrix elements of the second quantization operators . hence , all methodology of calculation of matrix elements will be much more universal in comparison with the traditional one ( see cowan @xcite , jucys and savukynas @xcite , wybourne @xcite ) . while seeking the matrix elements of one- or two- particle operators , it is necessary to obtain the values of a recoupling matrix @xmath272 , if we use the methodology by gaigalas _ et al _ @xcite ( see expresion ( [ eq : ma - g ] ) ) , or the recoupling matrices @xmath34 and @xmath35 , if we use the methodology by fano @xcite ( see expresion ( [ eq : ma - c ] ) ) . in the case of several open shells the expressions for matrix elements of every physical operator are published , where the recoupling matrices are in the form of simple factors . usually all these are written in the coordinate representation . they can be found in karazija _ @xcite , jucys and savukynas @xcite , karazija @xcite , rudzikas @xcite . meanwhile , for the more complex configurations , i.e. the ones having many shells , the recoupling matrices are much more complicated . beside that , the complexity of a two - particle operator adds to this . when the tensorial structure of an operator is complex , the recoupling matrix is rather complex , too , e.g. the spin - other - orbit operator ( see gaigalas @xcite ) . while attempting to calculate the angular part of matrix elements in all the mentioned cases , a general methodology for calculating the recoupling matrices is necessary . it has to be efficient , too , because the speed of calculation of angular parts of matrix elements depends on that . the majority of methodologies to obtain angular parts are based on the fano @xcite calculation scheme ( see for example glass @xcite , glass and hibbert @xcite , grant @xcite ) . in finding the matrix elements using this , one of the tasks is to obtain the recoupling matrices for direct and exchange terms . let us treat a matrix element of a two - particle operator acting upon four distinct shells @xmath299 for such an operator , the recoupling matrix in @xmath300 - space , using graphical technique of jucys and bandzaitis @xcite , is represented in figure [ fig : a ] . as we speak of calculations in @xmath0 - coupling , the analogous recoupling matrix in @xmath301 - space has to be calculated . in the figure [ fig : a ] the tree containing * b * nodes represents the bra function , and the one with * k * nodes represents the ket function . bra , as well as ket function , both contain @xmath22 open shells in this case . now let us treat a case when three distinct shells are acted upon by the operator . for example , the matrix element of operator @xmath302 is sought . then in the recoupling matrix the node iii will be on @xmath303 line ( see figure [ fig : a ] ) . this means that there will be two nodes on the resulting orbital moment @xmath303 line of shell @xmath44 . these will be nodes i and iii . the resulting orbital moment @xmath304 of the shell @xmath44 of bra function is attributed to the line connecting nodes @xmath305 and i. the resulting orbital moment @xmath306 of the shell @xmath44 of ket function is attributed to the line between nodes iii and @xmath307 . finally , characteristics @xmath308 of all possible intermediate momenta for that particular shell , which are summed over , are attributed to the line between nodes i and iii . so , in treating the operator acting upon three distinct shells , an intermediate sum in the recoupling matrix appears . similarly , two intermediate sums occur when two distinct shells are acted upon by the operator . and when it acts upon one shell only , three intermediate sums are present . the lines @xmath114 , @xmath309 and @xmath310 in the recoupling matrix represent the structure of two- particle operator in @xmath36 - space . see tutlys @xcite , grant @xcite for details . this is also valid for the exchange term , only the line @xmath311 must be connected to the node * d * , and line @xmath312 - to the node * c*. the first program to calculate the recoupling matrices of this type , njsym , was written by burke @xcite . it performs the calculations in two stages : 1 ) the recoupling matrix is expressed as a sum of products of the @xmath313 - coefficients ; 2 ) this expression is used in calculation . tutlys @xcite wrote a program to calculate angular parts of matrix elements , angula , which expressed the recoupling matrix in terms of clebsch - gordan coefficients before the actual calculations . while finding the recoupling matrix by the clebsch - gordan coefficient summation , this program eliminates trivial coefficients from the expression . bar - shalom and klapisch @xcite developed a new program njgraf . this program calculates the recoupling matrix in several stages . on the basis of graphical methodology by yutsis , levinson and vanagas @xcite , the recoupling matrix is analysed graphically and an optimal expression is found . afterwards , the value of recoupling matrix itself is calculated . an analogous program recoup was written by lima @xcite , and a program newgraph was written by fack _ et al _ @xcite . all these ( njgraf , recoup and newgraph ) are based on the same principle . an optimal analytical expression for the recoupling matrix is obtained by graphical method , and then the calculations are carried out according to it . but the optimal expressions they find are different quite often , and are not really optimal . as it was mentioned above , the methodology of angular calculation based on the fano calculation scheme has a shortcoming that the intermediate sums appear in complex recoupling matrices . due to these summations and the complexity of the recoupling matrix itself , the associated computer codes become rather time consuming . a solution to this problem was found by burke _ they tabulated separate standard parts of recoupling matrices along with coefficients of fractional parentage at the beginning of a calculation and then used them later to calculate the coefficients needed . computer codes by glass @xcite , glass and hibbert @xcite , burke _ @xcite , fischer @xcite , fischer @xcite and dyall _ et al _ @xcite utilize the program njsym ( burke @xcite ) or njgraf ( bar - shalom and klapisch @xcite ) for the calculation of recoupling matrices . both are rather time consuming when calculating matrix elements of complex operators or electronic configurations when calculating matrix elements of complex operators or electronic configurations with many open shells . in order to simplify the calculations , cowan @xcite suggested grouping matrix elements into classes ( see cowan @xcite , figure 13 - 15 ) . unfortunately this approach was not generalized to all two - electron operators . perhaps this is the reason why cowan s approach is not widely used although the program itself , based on this approach is widely used . gaigalas _ et al _ @xcite proposed a methodology where the analytical expressions for recoupling matrices are obtained for the most general case . in this methodology , analogically as in cowan @xcite , the matrix elements are attributed to four different groups . the operators acting upon only one shell belong to the first group ( distribution 1 from table 1 ) , the ones acting upon two - to the second ( distributions 2 - 10 from table 1 ) , upon three - to the third ( distributions 11 - 18 from table 1 ) , and upon four - to the fourth group ( distributions 19 - 42 from table 1 ) respectively . each group has a different recoupling matrix , @xmath314 - @xmath315 . they all are shown in figures [ fig : b ] - [ fig : e ] . representing the recoupling matrix @xmath272 when a two - particle operator acts upon a single shell.,height=510 ] representing the recoupling matrix @xmath272 when a two - particle operator acts upon a two shell.,height=510 ] representing the recoupling matrix @xmath272 when a two - particle operator acts upon a three shell.,height=510 ] representing the recoupling matrix @xmath272 when a two - particle operator acts upon a four shells.,height=510 ] let us treat the recoupling matrix @xmath316 first ( see figure [ fig : b ] ) . it represents the case when a physical two - particle operator acts upon a single shell . here , similarly as in figure [ fig : a ] , the graphical technique of jucys and bandzaitis @xcite is used . the bra and ket functions are represented as in figure [ fig : a ] . as the two - particle operator acts upon only one shell in this case , this operator may be treated as a single - particle one from the recoupling matrix point of view . using a graphical rule allowing one to cut two lines and connect the loose ends , we disconnect the nodes @xmath317 and @xmath318 from the general recoupling matrix . the part separated from recoupling corresponds to delta - functions only , @xmath319 , @xmath320 . in addition , using graphical technique it is possible to cut all the nodes until @xmath305 and @xmath321 out of the general recoupling matrix . all the cut nodes contribute the same delta functions which may be written as @xmath322 , @xmath323 . in the next stage , it remains to treat the part of recoupling matrix from the nodes @xmath305 and @xmath324 up to @xmath325 and @xmath326 . beside that , the lines @xmath327 and @xmath328 are connected in this remaining diagram . before obtaining the analytical expression for the remaining recoupling matrix we introduce a notation @xmath329 . it will help us to describe the procedure of obtaining analytical expression for the remaining recoupling matrix in a simpler way . so , the diagram in figure [ fig : f ] is denoted as @xmath329 . it equals to .,height=529 ] @xmath330 ^{1/2}\left\ { \begin{array}{ccc } j_{1 } & j_{2 } & j_{12 } \\ j_{3 } & j & j_{32 } \end{array } \right\ } .\ ] ] now , using the graphical rule of cutting three lines , we cut out the nodes @xmath331 , @xmath332 and @xmath333 from the recoupling matrix . thus we obtain a diagram @xmath334 . this coefficient in the paper by gaigalas _ @xcite is denoted as @xmath335 , and equals to @xmath336 ^{1/2}\left\ { \begin{array}{ccc } k & l_a^{\prime } & l_a \\ j & t & t^{\prime } \end{array } \right\ } .\ ] ] the diagram is cut from the nodes @xmath337 and @xmath338 up to @xmath339 and @xmath340 in the same way . it is expressed in terms of diagrams @xmath341{c } c_2\left ( k , k_{\min } , k_{\max } \right ) = \displaystyle { \stackrel{k_{\max } } { \prod_{i = k_{\min } } } } a_{1}(+ - + - ; k l'_{12 ... i } l_{12 ... i } ; l_{i } l_{12 ... i-1 } l'_{12 ... i-1 } ) . \end{array}\ ] ] it is easy to notice that @xmath342 corresponds to the coefficient @xmath342 defined in gaigalas _ @xcite . before going into investigation of the recoupling matrix pictured in figure [ fig : c ] , we denote the diagram @xmath343 ( see figure [ fig : g ] ) as @xmath344 . its analytical expression is .,height=453 ] @xmath345 ^{1/2}\left\ { \begin{array}{ccc } j_{1 } & j_{2 } & j_{12 } \\ j_{3 } & j_{4 } & j_{34 } \\ j_{13 } & j_{24 } & j \end{array } \right\ } .\ ] ] in the @xmath346 diagram , first the signs of nodes 1 - 6 , and then the moments are presented . while investigating the recoupling matrix ( see figure [ fig : c ] ) it is easy to notice that there is an additional graphic element . after cutting this out , we get a diagram the further investigation of which is very similar to the investigation of @xmath314 . that additional matrix element is @xmath347 . it is obtained by cutting the nodes @xmath337 , @xmath348 , @xmath349 , @xmath338 out of the diagram @xmath350 . after doing this , the diagram under investigation is split into two diagrams closely resembling the recoupling matrix @xmath316 . as they are cut in the very same way as @xmath316 is , we will not stop at recoupling matrix @xmath350 for details . in the methodology presented in gaigalas _ @xcite , the recoupling matrix is represented by a diagram @xmath351 ( see figure [ fig : d ] ) when the two - particle operator is acting upon three shells . similarly as for the recoupling matrix @xmath350 , the cutting out of nodes @xmath337 , @xmath348 , @xmath349 and @xmath352 leads to a diagram @xmath353 , and the cutting out of nodes @xmath354 , @xmath355 , @xmath356 ir @xmath357 leads to a diagram @xmath358 . in that case the diagram splits up into three simple diagrams , the further investigation of which is analogous to that of @xmath316 recoupling matrix . the most complex recoupling matrix @xmath359 ( see figure [ fig : e ] ) comes into view when all the second quantization operators act upon different shells . the part of this recoupling matrix from nodes @xmath317 and @xmath318 to nodes @xmath360 and @xmath361 , and also from nodes @xmath362 and @xmath363 to nodes @xmath325 and @xmath326 is very similar to the recoupling matrices described above , therefore their investigation is the same as that of the diagrams described earlier . we will investigate in more detail the part of diagram from nodes @xmath360 and @xmath361 to nodes @xmath362 and @xmath363 . that is pictured in figure [ fig : e - b ] . after cutting the diagram across four lines @xmath364 , @xmath365 , @xmath366 and @xmath367 and connecting these according the graphical technique of jucys and bandzaitis @xcite with a generalized clebsch - gordan coefficient , then connecting the lines @xmath368 , @xmath366 and @xmath369 with an ordinary clebsch - gordan coefficient , we obtain a diagram which is expressed via two @xmath313-coefficients . the exact analytical expression of this diagram is presented in the paper by gaigalas _ @xcite ( see expresion ( 29 ) ) . .,height=510 ] after cutting the nodes @xmath362 , @xmath356 , @xmath370 , @xmath371 and @xmath363 out of the remaining part and investigating the obtained diagram according to the jucys and bandzaitis @xcite technique , we obtain a diagram which corresponds to a @xmath372- coefficient . the analytical expression of that is presented in paper by gaigalas _ @xcite , expr . ( 31 ) . if a shell @xmath373 is beside shell @xmath374 , i.e. there are no additional nodes between nodes @xmath360 , @xmath361 and @xmath362 , @xmath363 , then the additional graphical elements vanish . so , we obtained the full analytical expression of recoupling matrix @xmath375 . if between the abovementioned shells a single additional shell is present , then we obtain an additional diagram which is expressed as a product of two @xmath313-coefficients ( see ( 30 ) from gaigalas _ et al _ if the number of shells is greater , then the obatined diagram in the most general case is written via ( 32 ) from gaigalas _ et al _ @xcite . the methodology that is based on the second quantization in coupled tensorial form , on the angular momentum theory in three spaces ( orbital , spin and quasispin ) , on the wick s theorem and on the generalized graphical technique of angular momentum , give the possibility to efficiently calculate the matrix elements of energy operators in general case ( gaigalas and rudzikas @xcite , gaigalas _ @xcite , this paper ) . the main ideas of this methodology are : 1 . a number of theoretical methods is known in atomic physics that facilitate a lot the treatment of angular part of matrix elements . these are the theory of angular momentum , its graphical representation , the quasispin , the second quantization and its coupled tensorial form . but while treating the matrix elements of physical operators in general , the methods mentiond above are applied only some part , or very inefficiently . + an idea is presented and carried out in the work gaigalas and rudzikas @xcite , gaigalas _ et al _ @xcite and this paper , of unifying all these methods in order to optimize the way to treat matrix elements of physical operators , to investigate efficiently even the most complex cases of atoms and ions . the wick s theorem is widely known in the theory of atom . up to now it was applied to the most general products of the operators of second quantization , i.e. when the particular arrays of quantum numbers for each operator of second quantization were not yet defined . in the methodology described there is proposed to apply the wick s theorem for the products of second quantization operators where they have the values of quantum numbers already defined . this allows one , using the methodology of second quantization in coupled tensorial form , to abtain immediately the optimal tensorial expressions for any operator . then , in treting the matrix elements of physical operators , the advantages of a new modification of the racah algebra are exploited to their full extent . 3 . while analysing the physical operators presented as products of tensors @xmath376 , @xmath377 , @xmath378^{(k_q k_l k_s)}$ ] , @xmath379^{(k_q k_l k_s)}$ ] , + @xmath380^{(k_q k_l k_s)}$ ] , it is possible to obtain convenient analytical expressions for recoupling matrices that must be taken into account in finding the matrix elements ( non - diagonal with respect to configuration included ) of any physical operator between complex configurations ( with any number of open shells ) . an idea is proposed and carried out , concerning the most efficient way to apply the tensor algebra and the quasispin formalism in the most general case , for the diagonal and the non - diagonal matrix elements as well , when the bra- and ket- functions have any number of open shells . the combination of all these improvements allows to efficiently account for correlation efects practically for any atom or ion of periodical table . the author is grateful to professor z. rudzikas for encouraging and valuable remarks . , `` on the method of calculation of matrix elements of the operators of atomic quantities in the case of complex configurations '' , proceedings of the academy of sciences of lithuanian ssr , b series v. 4 ( 27 ) , p. 45 - 57 ( 1961 ) ( in russian ) . , `` irreducible tensorial form of the effective hamiltonian of an atom and the diagrammatic representation in the first two orders of the stationary perturbation theory '' , liet . fiz . rink . ( english translation sov . phys . coll . ) , v. 25 , no . 6 , p. 14 - 31 ( 1985 ) . , `` energies of @xmath286 equivalent electrons expressed in terms of two - electron energies and independent three - electron parameters : a new complete set of orthogonal operators : i. theory '' , j. phys . b : at . mol . phys . , v. 17 , p. 2375 - 2392 ( 1984 ) . , `` reduced in quasi - spin space coefficients of fractional parentage and matrix elements of tensorial operators '' , liet . fiz . rink . , v. 18 , no . 293 ( 1978 ) . ( english translation sov . phys . coll . * 18*(3 ) , 1 ( 1978 ) from allerton press , new york ) . , `` expressions for reduced matrix elements of non - relativistic effective hamiltonian of atom in the first orders of stationary perturbation theory for configurations with two unifilled shells @xmath383 '' , in : _ spectroscopy of autoionized states of atoms and ions _ ( moscow , scientific council of spectroscopy , 1985 ) , p. 20 - 42 . , `` diagramatic approach to irreducible tensorial operators and its application to perturbation theory '' , in : _ spectroscopy of autoionized states of atoms and ions _ ( moscow , scientific council of spectroscopy , 1985 ) , p. 43 - 61 . , `` application of the graphical methods of angular momentum theory to the investigation of the expansion of stationary perturbation theory '' , izv . uchebn . zaved . 50 , no . 7 , p. 1403 - 1410 ( 1986 ) . , _ tables for the calculation of matrix elements of atomic quantities _ , ( moscow , 1967 ) ; english translation by e.k . wilip , anl - trans-563 ( national technical information service , springfield , va . , 1968 ) . , `` a program to calculate the matrix elements in multi - configuration approximation '' , in : _ sbornik programm po matematicheskomy obespecheniju atomnyh raschotov_. vipusk 4 . ( vilnius , 1980 ) ( in russian ) .

in this paper we study the effects of rotation on the radiation driven winds of early - type supergiants . we will focus on the explanation for the occurrence of disks around fast rotating b[e ] supergiants . b[e ] supergiants , also designated sgb[e ] stars ( lamers et al . @xcite ) , are b type supergiants that exhibit forbidden emission lines in their optical spectra . the observations of hybrid spectra , i.e. spectra with _ broad _ uv p cygni features and _ narrow _ emission lines and dust emission of b[e ] supergiants in the magellanic clouds led to the proposal of a disk wind model by zickgraf et al.(@xcite ) . this model postulates a dense disk of outflowing material in a fast line driven wind to explain the observed characteristics of the spectra of these stars ( see zickgraf @xcite ) . the stellar wind at the equator is about ten times slower than that at the pole . also , the wind at the equator is about a hundred times denser than at the pole . additional evidence for a two - component outflow has been obtained from polarimetric measurements , e.g. by zickgraf & schulte - ladbeck ( @xcite ) . the precise mechanism behind the formation of these disks is still a mystery ( cassinelli 1998 ) . it is however clear that the origin of an axisymmetric wind structure such as a disk may well be connected to the fast rotation of a star . two theories using rotation in a different way have been considered for the formation of these disk winds : ( 1 ) the wind compressed disk ( wcd ) model of bjorkman & cassinelli ( @xcite ) , that invokes the kinematics of the winds from rotating stars . the streamlines of the gas in the wind from both hemispheres of a rapidly rotating star cross in the equatorial plane . the concentration of the gas and the shock in the equatorial plane produce an outflowing equatorial disk , with a thickness on the order of a few degrees . owocki & cranmer ( 1994 ) have argued that the motion to the equatorial plane may be counteracted by the radiation force perpendicular from the plane due to lines . however , this `` wind - inhibition effect '' may not be effective in winds with a strong density gradient in the equatorial direction . we return to this in sect . ( 2 ) the rotationally induced bi - stability model ( rib ) of lamers & pauldrach ( @xcite ) invokes an increase in the mass flux from the equator and a decrease in the equatorial wind velocity compared to the poles from the bi - stability jump . this jump in mass flux and in wind velocity is due to the temperature difference between the pole and the equator of a fast rotating b[e ] supergiant with gravity darkening . the jump will occur for stars with effective temperatures between 20 000 and 30 000 k. in reality , both effects , i.e. the wind compression and the rotation induced bi - stability , may be operating together and amplifying one another in the line driven winds of rapidly rotating early - b stars . ( for a detailed explanation of both models , see lamers & cassinelli ( 1999 ) ( hereafter @xmath5 ) , chapter 11 . ) the theory of radiation driven winds for non - rotating stars was developed by castor et al . ( @xcite ) ( hereafter cak ) and predicts the mass loss rate and @xmath6 for spherical winds . the influence of rotation on line driven winds was investigated by friend & abbott ( @xcite ) . they found an increase in the mass - loss rate at the equator and a corresponding decrease in @xmath6 , but these authors did not consider the effects of gravity darkening and the oblate shape of the star . cranmer & owocki ( @xcite ) described the finite disk correction factor in case of an oblate star . in this paper we will modify the line driven wind theory , including these effects of oblateness and gravity darkening as well as the rotational terms in the equation of motion . this will then be applied to rapidly rotating b[e ] supergiants to investigate whether these effects can explain the occurrence of disks . we will also apply recent bi - stability calculations to study the effect of the rib - model to explain the occurrence of outflowing disks around rapidly rotating b[e ] supergiants . in sect . 2 we describe the theoretical background of the radiation driven wind theory and the bi - stability effect . in sect . 3 we derive the equations for the winds from oblate rotating stars with a temperature gradient between poles and equator due to gravity darkening . in sect . 4 we describe the method for solving the equations and for calculating the winds from rotating stars . these calculations will be applied to b[e ] supergiants in sect . [ s_b[e ] ] . in sect . [ s_bs ] we will investigate the effect of bi - stability on rotating stars and sect . [ s_concl ] concludes with a summary of this work and a discussion on the formation of disks of b[e ] supergiants . the computation of the dynamics of line driven stellar winds is a complicated problem in radiation hydrodynamics . it involves the simultaneous solution of the equations of motion , the rate equations and the radiative transfer equations in order to calculate the radiative acceleration . for hot luminous stars the line forces are the most important driving forces in the wind . cak have shown that the line acceleration can be parameterized in terms of the optical depth parameter @xmath7 as @xmath8 , where @xmath0 and @xmath1 are parameters that depend on composition and temperature of the wind . in this expression @xmath0 is a measure of the number of lines and @xmath1 is a measure of the distribution of the line strengths with @xmath9 or 1 for a pure mix of optically thin or thick lines respectively . in this paper we adopt the cak formalism and simplify the equations of motion by assuming a stationary , radial flow of a viscousless fluid . the possible influence of magnetic fields will be ignored . these simplifications are subject to the following restrictions ( see e.g. abbott 1980 ) : * the cak theory treats the line acceleration in a simplified way in terms of the force multipliers , described above . detailed calculations of the line driving , including millions of lines , show that the total line acceleration can be well described by this simple representation ( pauldrach et al . both observations and calculations show that line driven stellar winds are not stationary . even a wind in a stationary solution will develop shocks ( see e.g. lucy @xcite ) . however , the time - averaged structure follows the stationary state quite well ( see owocki et al . @xcite , and feldmeier 1999 ) . therefore we will restrict ourselves to stationary models . * a sectorial model will be adopted . this means that for every latitude a one dimensional problem will be solved . wind compression as in the wcd model will thus be neglected , but the main effect of wind compression is expected to _ redistribute _ the mass loss and not to change the _ total _ mass that is lost from the rotating star . * the absence of viscosity is a good approximation at the high density of line driven winds ( see cak ) . the representation of the line force by a simple power law may seem to be a gross simplification of the underlying physics of myriad line absorption processes , but it can be shown to hold for a homogeneously distributed mix of optically thick and thin lines ( abbott @xcite , gayley @xcite ) . calculations of cak and more recently in nlte by vink et al . ( @xcite ) of realistic model atmospheres , based on calculations of unified atmosphere / wind models with the isa - wind code ( de koter et al . 1997 ) , confirm this to be generally true to good accuracy for the important part of the wind ( from about the sonic point up to a few stellar radii ) . therefore , the complex physics of ion populations can be ignored in the investigation of the various effects of rotation on the stellar wind . however , there is one notable exception which is called the bi - stability jump . this refers to a jump in @xmath6 around the temperature of 21000 k where @xmath10 climbs from 1.3 ( lower temperature ) to 2.6 ( higher temperature ) . this jump was observed by lamers et al . ( @xcite ) in b supergiants . it is linked to a shift in ionization states of fe in the lower part of the wind . the line driving in the lower part of the wind is dominated by iron . below about @xmath11 k , fe iv recombines to fe iii and since fe iii is a more efficient line driver than fe iv , the wind structure changes dramatically ( vink et al . @xcite ) . the bi - stability jump may also occur in the temperature difference between the pole and the equator of a fast rotating star with gravity darkening . therefore two sets of force multiplier parameters for the wind will be adopted to reflect the sudden change in ionization states ( a high - temperature set for the pole and a low - temperature set for the equator ) . intuitively it is clear that rotation has an effect on the shape of a star and on the motion of the gas in the wind . an additional effect that the rotation of a star can have is the darkening of the equatorial regions of a star via the von zeipel effect . these effects will modify the wind of a star considerably and we will incorporate them in the line driven wind theory for rotating stars . the shape of a uniformly rotating star with all its mass concentrated in the core is determined by the equipotential surfaces of the potential in a rotating frame ( roche model ) : @xmath12 where @xmath13 , @xmath14 is the co - latitude ( @xmath14 = 0 at the pole ) and @xmath15 . @xmath16 is of course independent of the longitude @xmath17 . note that @xmath18 . so the maximum oblateness is @xmath19 . the critical ( break - up ) velocity is defined in this paper as : @xmath20 where @xmath21 is the effective mass of the star , @xmath22 which is the reduced mass due to radiation pressure by electron scattering ( see below ) . the surface of a rotating star is implicitly given by eq . [ eq_firot ] . solving this expression for @xmath23 is equivalent to the solution of a 3rd degree polynomial , which results in @xmath24 because of the difference in radius @xmath25 and the difference in rotational velocity between pole and equator , the radial escape velocity at the equatorial region is smaller than at the pole @xmath26 we can then roughly estimate the effect of rotation on @xmath6 as this value has an almost linear dependence on @xmath27 ( see later in [ s_ssol ] ) . the von zeipel theorem ( @xcite ) for distorted stars states that the radiative flux from a point on the star is proportional to the local effective gravity : @xmath28 with @xmath29 the boltzmann constant . as @xmath30 , we can write down the flux of a rotating star as a function of the co - latitude @xmath14 . this ( lengthy ) expression for the flux as a function of @xmath14 is plotted in fig . [ fg_f ] for a few different values of @xmath31 . the rotation of the star can induce a @xmath14 component of the flow of the matter in the wind . in this study a possible @xmath14 component is neglected as was discussed in [ s_physics ] , but the motion in the longitudinal @xmath17 direction must still be taken into account . in the absence of forces that can exert a torque on the wind , the equation of motion for @xmath32 is given by the conservation of angular momentum . note however that there could be a torque from the line forces themselves , which is dependent on the velocity gradient of the gas . we assume that there are no external torques , so conservation of angular momentum gives : @xmath33 the equation of motion for the radial direction becomes : @xmath34 where @xmath35 is the radial velocity and @xmath36 is the stellar mass . the conservation of mass for a non - spherical sectorial wind can be written as @xmath37 where @xmath38 is the local mass loss rate , i.e. the total mass loss rate if the solution for this latitude where valid for a spherical star . the total mass loss rate from the star is @xmath39 the equation of motion together with the conservation of mass governs the dynamics of the wind given the equation of state , @xmath40 ( @xmath41 is the isothermal sound speed ) , and the radiative acceleration @xmath42 . using the conservation of mass , the pressure term can be rewritten as : @xmath43 as in the cak - theory the temperature structure is specified a priori as a function of @xmath44 . ( the results depend only very weakly on the chosen temperature structure ; see pauldrach et al . @xcite ) . the radiative acceleration consists of two components . the continuum component due to the electron scattering , @xmath45 , and a second component , @xmath46 , due to line scattering and absorption processes . the continuum acceleration is given in terms of @xmath47 : @xmath48 where @xmath49 is the electron opacity , @xmath50 is the stellar luminosity and @xmath51 is the radiation flux . the contributions by other continuum opacities are small . nevertheless , they were included in the calculation of the force multipliers by vink et al . ( 1999 ) used in our bi - stable wind models ( sect . for a homogeneous spherical star , @xmath47 is given by as the ratio between the continuum force and gravity and not as the ratio between continuum force and the critical radiation force for a rotating star ] : @xmath52 in the more general case of a non - homogeneous , non - spherical star the continuum acceleration can be defined as a correction to the classical continuum acceleration of eq . [ eq_gam ] : @xmath53 where @xmath54 is the continuum correction factor which is given by : @xmath55 the line acceleration @xmath56 has a more complicated form , since it is a function of distance @xmath44 as well as the velocity @xmath35 and velocity gradient @xmath57 . combining the equation of motion ( eq . [ eq_bewvgl ] ) with the rewritten pressure and continuum acceleration terms and multiplying by @xmath58 , one finds : @xmath59 this equation is valid for each co - latitude @xmath14 . following cak , the critical point of this equation is found by imposing the singularity condition ( where the subscript @xmath60 indicates values at the critical point ) : @xmath61 and the regularity condition : @xmath62 note that the critical point is not the sonic point @xmath63 ( @xmath64 ) due to the fact that there is an additional dependence on @xmath65 in the line acceleration @xmath46 ( abbott 1980 ; @xmath5 chapt 3.3 ) . the line forces are described within the framework of cak in the sobolev approximation . this means that the intristic line absorption profiles are considered to be infinitely sharp . then the line force becomes a function of local properties of the wind only . the acceleration due to an ensemble of lines is given by the summation of the contributions of all individual lines with rest frequencies @xmath66 . @xmath46 is given by ( castor @xcite ) : @xmath67 where @xmath68 is the absorption coefficient per gram for the @xmath69th line , @xmath70 , @xmath71 the intensity at the rest frequency @xmath72 and @xmath73 is the sobolev optical depth of @xmath69th line , defined as @xmath74 . the integration is performed over the complete visible disk of the star . following cak , the line acceleration can be rewritten in terms of the force multiplier as a function of the optical depth parameter @xmath75 . where @xmath75 is defined as : @xmath76 with @xmath77 being the thermal velocity of the protons and @xmath78 is some reference value for the electron scattering @xmath79 @xmath80 g@xmath81 ( see _ isw _ , chapt . the line acceleration in terms of the continuum acceleration is give by : @xmath82 if the star is assumed to be a point source , @xmath83 can be approximated by a simple power law parameterization ( cak , abbott 1982 ) : @xmath84 where the @xmath85 accounts for the effect of the electron density on the ionization balance in the wind . here @xmath86 is the electron density in units of @xmath87 @xmath88 and @xmath89 is the geometric dilution factor . friend & abbott ( 1986 ) and pauldrach et al . ( 1986 ) showed the importance of the finite disk correction on the line force in case of an extended source . this finite disk correction factor @xmath90 is given by ( cak ) : @xmath91 where @xmath92 this expression is completely general : the actual shape of the star and its intensity distribution enter through the integration domain and the @xmath93 dependence . however , here we assume the intensity @xmath94 to be locally isotropic , i.e. we neglect limb darkening . we have tested the effect of limbdarkening . the inclusion of limbdarkening results in a small @xmath14-independent correction on @xmath90 . the line acceleration @xmath46 including all the neccesary correction factors is now given by : @xmath95 combining eqs . [ eq_bewvglf ] and [ eq_gl ] gives the full equation of motion for the sectorial wind of a rotating star @xmath96 for line driven winds of non - rotating stars solutions of the equation of motion have been retrieved by e.g. cak and pauldrach et al . ( 1986 ) . these will be presented to serve as an illustration of the solution of the equation of motion and as a basis to interpret the more complicated results from the full equation of motion ( [ eq_bewvglrot ] ) . if the star is considered to be a point soure the integral of the line acceleration in eq . [ eq_gmeerlijn ] collapses to one point . in this case @xmath90 disappears from eq . [ eq_gl ] and the line acceleration becomes a simple function @xmath97 . the solution of eq . [ eq_bewvglf ] was found by cak and can easily be explained by neglecting the gas pressure terms ( @xmath98 ) . in this case [ eq_bewvglf ] reduces to @xmath99 where @xmath100 is a constant containing the mass loss rate . the equation is solved by imposing uniqueness of the solution ( kudritzki et al . 1989 ) which then fixes the value of the constant and thus @xmath101 . the solution for the mass loss and velocity law is given by @xmath102 @xmath103 with @xmath104 this simplified solution is equal to the full cak solution , in the limit of small sound speed , @xmath105 ( see also @xmath5 chapt 8) . in case of a homogeneous spherical star the finite disk correction factor @xmath90 ( eq . [ eq_corr ] ) can be calculated analytically ( cak ) : @xmath106 including the finite disk correction results in an increase in @xmath6 , a decrease in @xmath107 and a modification of the simple scaling laws that were found in the original cak approach ( see friend & abbott @xcite ; pauldrach et al . the decrease of the mass loss rate compared to the point source case is due to the decrease in @xmath46 close to the star where @xmath101 is determined . close to the stellar photosphere the finite disk correction is smaller than one , viz . @xmath108 . the accompanying increase in @xmath6 of typically a factor of two is due to two effects : ( 1 ) a reduction of @xmath107 results in a smaller amount of material to be accelerated and ( 2 ) far from the photosphere , the correction factor @xmath90 becomes larger than 1 . the resulting dependence between @xmath6 and @xmath27 is approximated by friend & abbott ( 1986 ) : @xmath109 they also found a relation for the modified mass loss rate : @xmath110 the solution of eq . [ eq_bewvglrot ] gives the mass loss rate , or rather the local mass loss rate @xmath38 , and velocity structure of the wind . the solution is complicated however by the presence of @xmath90 . this is an integral of a velocity - dependent function times the surface intensity over the visible section of the star . the appearance of the star varies throughout the wind in intensity distribution as well as in shape , as the star is no longer spherical and @xmath111 is a function of latitude . since the analytic solution would be cumbersome , a numerical approach of the solution is chosen . this is done with the analytic solutions of simple , i.e. the non - rotating , models in mind for comparison . the numerical solution of eq . [ eq_bewvglrot ] is relatively straightforward in case the function @xmath90 is a given function of @xmath44 . therefore we solve eq . [ eq_bewvglrot ] with @xmath112 in an iterative way as follows : 1 . a @xmath113 law for @xmath114 is assumed , viz . @xmath112 is calculated using this velocity law . note that @xmath90 is not dependent on the actual value of @xmath114 but on the velocity gradient . 3 . eq . [ eq_bewvglrot ] is solved using this correction factor @xmath112 , obtaining a new velocity law @xmath114 . step 2 - 3 are repeated until convergence is reached . typically three iterations are sufficient , since further iterations changed the obtained values by less than 0.5 % . we checked that the solutions converged to the same value , independent of the starting value . the calculation of the correction factor @xmath90 from the velocity law ( step 1 . in the scheme described above ) is performed by numerical evaluation of eq . [ eq_corr ] . this is a non - trivial task , since both the integrand ( eq . [ eq_zeipel ] ) and the shape of the visible `` disk '' ( eq . [ eq_rtheta ] ) are dependent on @xmath44 . the main parameters needed to obtain @xmath90 for a fixed co - latitude @xmath116 are : the stellar surface @xmath117 and the surface temperature @xmath118 , both given as a function of the co - latitude @xmath14 , as well as the velocity law @xmath114 . it is convenient to change the integration variables in eq . [ eq_corr ] from the wind centered coordinates @xmath14 and @xmath17 to the star centered coordinates @xmath119 and @xmath17 . for the definition of these coordinates , see fig . [ fg_ofr ] . note that here @xmath17 is not the stellar longitude anymore , but the angle of rotation around the line from a point @xmath120 in the wind to the stellar center . @xmath121 where @xmath94 is the frequency - integrated intensity @xmath122 the angles are related to one another via @xmath123 with the stellar surface given by @xmath124 , and @xmath125 equation [ eq_coint ] can be integrated , considering that the limb angle @xmath126 can be determined numerically , yielding the finite disk correction for the line acceleration of an arbitrary shaped star . the correction factor @xmath54 for the electron acceleration ( defined in eq . [ eq_dc ] ) similarly becomes : @xmath127 this correction is also included in the solution of the equation of motion . [ tb_res1 ] [ cols="^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^ " , ] ( 1 ) : the adopted stellar parameters are : @xmath128 ; @xmath129 ; @xmath130 k. to solve the equation of motion it is neccesary to determine the conditions at the critical point : @xmath131 , @xmath132 and @xmath133 , or , equivalently , @xmath131 , @xmath132 and @xmath134 , as well as the local mass loss rate @xmath135 . apart from eqs . [ eq_sing ] and [ eq_regu ] and the equation of motion ( [ eq_bewvglrot ] ) itself the additional constraint for the continuum optical depth @xmath136 at the stellar radius is neccesary ( pauldrach et al . 1986 ) to uniquely determine these quantities : @xmath137 if the condition for @xmath138 is used one may adopt any small value , say @xmath139 km / s , since even a large error in @xmath138 will only yield a relatively small error in @xmath6 and @xmath135 . this is due to the fact that at the surface of the star @xmath114 and thus also @xmath140 vary exponentially with scale height @xmath141 . this means that the radius where eq . [ eq_tau2/3 ] is fulfilled is very close to @xmath142 . this has been checked a posteriori . in our calculations we have adopted this boundary condition for @xmath138 in the solution of the full momentum equation . a first guess for the local mass loss rate obtained from the cak solution is used to integrate the equation of motion from the stellar surface outward . this gives an approximate value for the critical point where @xmath143 if @xmath144 . this approximate value for the critical point can then be used to consecutively improve the guessed value of @xmath135 using eqs . [ eq_sing ] and [ eq_regu ] . this iterative procedure is terminated when the local mass loss rate converges . the calculation is then completed by integrating past the critical point on the solution which extends to infinity ( see cak or abbott 1980 for a description of the topology of the solutions ) . to deal with a numerical singularity at the sonic point the equations are solved for @xmath145 rather than for @xmath114 . we have calculated models for rotating b[e ] stars using the above described method . for the first models we adopted the following stellar parameters , which are typical for b[e ] stars : @xmath146 , @xmath147 , @xmath148 , @xmath149 and solar abundances . the effective temperature is defined as @xmath150 where @xmath151 is the total surface of the distorted star . the following force multiplier parameters from pauldrach et al . ( @xcite ) were adopted to describe the line acceleration : @xmath152 , @xmath153 , @xmath154 . an overview of the results for this generic b[e ] star rotating at @xmath31 = 0 , 0.3 , 0.4 , 0.5 and 0.6 times the critical rotation speed is given in table 1 . the calculation of models with higher values of @xmath31 runs into numerical problems , because in the iteration , the approximate critical point can be below the sonic point . the table shows for different co - latitudes @xmath14 and different rotation velocities @xmath31 the values of @xmath6 , the local mass loss rate @xmath38 and the value of the velocity law parameter @xmath113 , obtained by fitting the calculated velocity structure from @xmath155 to @xmath156 to a @xmath113-law . we see that the value of @xmath113 does not change much . [ fg_vr ] shows the velocity structure of the wind and the location of the critical point for @xmath31 = 0 , 0.3 and 0.6 . note that the critical point is very close the star . this is true for all finite disk corrected solutions . figure [ fg_vmp ] shows the behaviour of @xmath6 versus @xmath135 for various rotational speeds and latitudes . as expected , @xmath6 at the equator is smaller than at the pole because @xmath6 scales roughly with @xmath27 and due to the smaller value of @xmath27 at the equator . figure [ fg_mp ] shows the latitude dependence of the local mass - loss rate and of the terminal velocity from equator to pole , for different rotational speeds , indicated by @xmath31 . at the pole ( @xmath157 ) @xmath6 is about constant for various rotational speeds the mass - loss rate at the equator decreases slightly with increasing rotation . the increase of the mass loss rate at the equator as calculated by pauldrach et al . ( @xcite ) and friend & abbott ( @xcite ) is offset by a decrease in mass loss rate due to the smaller radiative flux at the equator due to the von zeipel effect . the mass loss rate at the pole increases strongly with increasing rotation rate due to the increase in luminosity at the pole . this is because the total luminosity of the star is fixed and the smaller luminosity at the equator must therefore be compensated by a higher luminosity at the pole . the overall effect of rotation is to increase the mass loss rate of a star . this can be seen in fig . [ fg_mp ] where the overall mass - loss rate ( i.e. @xmath135 integrated over the stellar surface ) is plotted vs. the rotational velocity . the mass - loss rate of the models with @xmath158 vary with @xmath31 as @xmath159 . figure [ fg_ldens ] shows the resulting density contrast ( far from the star ) between pole and equator ( @xmath160 ) . the densities at pole and equator are essentially the same : the smaller @xmath6 at the equator is offset by the larger mass loss rate @xmath135 at the pole . [ fg_dc ] we have also calculated models for different luminosities ( see table 2 ) , and different stellar masses ( see table 3 ) . the mass - loss rate for the different luminosities is also plotted in fig . [ fg_mp ] . the density contrast for various luminosities is plotted in fig . [ fg_ldens ] . we see that the density at the pole compared to the equator increases for increasing luminosity and the effect on the mass - loss rate is only visible for the highest luminosity . this is an effect of the continuum acceleration @xmath54 . 7 illustrates the effect of the continuum acceleration on the forces in the wind for a star with high luminosity ( @xmath162 ) . the net effect of the continuum radiation pressure and the gravity , i.e. @xmath163 , is an inward force in the equatorial region and an outward force in the polar region where the flux is higher . for small luminosity the variation in @xmath164 through @xmath54 is not important because @xmath47 itself is small . for high luminosity @xmath165 increases and thus the effect of @xmath166 on @xmath164 becomes noticeable ( this is with all other properties of the star fixed ) . in eq . [ eq_vinfsimple ] we see that a larger value of @xmath166 at the pole causes a relatively larger decrease ( through @xmath164 ) in @xmath27 and thus in @xmath6 , whereas the mass loss rate at the equator will increase less because of the smaller @xmath54 . this is only a qualitative explanation and the actual dependence of @xmath167 does not follow the relation suggested by eqs . 29 and 31 . we have found that the actual values of @xmath167 could be approximated quite well ( within 10 percent ) with the following relation : @xmath168 where the @xmath169 is the radius where @xmath54 reaches its minimum value . so we see that although rotation alone modifies the wind structure considerably , it can not be responsible for the density contrast between the equatorial and polar wind observed in b[e ] stars , in case only radial effects are considered . we conclude that at least one other physical effect must be responsible for the formation of disks of rotating b[e ] stars . the two most likely additional effects are the bi - stability and the flow of the wind material towards the equator . in the section above we have shown that rotation alone can not explain the observed large density contrast between the poles and the equator of b[e ] stars . one of the possible additional mechanisms to enhance this contrast is the bi - stability jump . to investigate whether the bi - stability jump observed in normal b - supergiants can explain the formation of disks around b[e ] stars , we have calculated bi - stable wind models for a fast rotating typical b[e ] supergiant . spherical , nlte wind models for normal b supergiants have been calculated by vink et al . ( 1999 ) . from these models cak - like force multiplier parameters have been derived using a monte - carlo method to simulate photon - gas interactions . these models show the occurence of a bi - stability jump around @xmath11 k. for two models on either side of the bi - stability jump ( @xmath170 = 17 500 and 30 000 k ) , the cak parameters were determined by fitting a power law from about the sonic point to about @xmath171 ( from @xmath172 to @xmath173 in optical depth , @xmath75 is defined in eq . [ eq_t ] ) to the calculated force multiplier values . the resulting values for @xmath0 , @xmath1 and @xmath2 are listed in table [ tb_ka ] . note that the @xmath2 parameter is taken equal to zero as it was not possible to extract it explicitly from the models . its effect is hidden in the constant @xmath0 . as expected , the values for the force multiplier parameters are quite different for the two cases , because the ionization has changed dramatically over the bi - stability jump . cccc & @xmath0 & @xmath1 & @xmath2 + 17500 k & 0.57 & 0.45 & 0.0 + 30000 k & 0.06 & 0.65 & 0.0 + [ tb_ka ] the predicted bi - stability jump occurs around @xmath174 kk . therefore we calculated a model for a rotating b[e ] star with the following properties : @xmath175 k , @xmath176 , @xmath177 and solar abundances . this implies that the pole of the rapidly rotating star will be hotter than 25 kk and the equator will be cooler than 25 kk . for the pole , the force multipliers @xmath0 and @xmath1 from the hot monte carlo model of 30 000 k were used , whereas for the equator , @xmath0 and @xmath1 from the cool model of 17 500 k were used . the resulting @xmath178 and mass loss rates @xmath38 are plotted in figs . [ fg_vtheta ] and [ fg_mdtheta ] for @xmath31 = 0.6 . clearly visible is the drastic decrease in @xmath6 and the drastic increase of @xmath135 towards the equator . this occurs around a co - latitude @xmath14 ( angle from the pole ) of about 60 degrees , where @xmath179 k. ( the precise location of the jump depends on the effective temperature of the star ) . since both @xmath135 and @xmath6 have a equator / pole contrast of about a factor of three , the resulting density contrast is about a factor of 10 in a disklike region with a half opening angle of 30 degrees . our calculations show that the density contrast in the wind between pole and equator due to the bi - stability jump is significantly larger than without the bi - stability jump . yet , the calculated value of about a factor 10 is not sufficient to explain the observed density contrast of a factor @xmath180 . so another mechanism is needed to enhance the density contrast even further . this is most likely the wind compression mechanism ( bjorkman & cassinelli 1993 ; bjorkman 1998 ) . we have modified the line driven wind theory for rotating stars by including the effects of oblateness and gravity darkening as well as rotational terms in the equation of motion . we considered a sectorial wind , i.e. we neglected the effect of the motion of the wind towards the equator or towards the pole . this assumption is justified close to the stellar surface below the critical point of the momentum equation . this implies that our method will predict about the correct distibution of mass - loss rates from the star as a function of stellar latitude , but it may not be accurate enough to predict the velocity and density distribution further away from the star , if motions in the @xmath14 direction become important . the equation of motion was solved using an iterative numerical scheme , that includes the conservation of angular momentum and the correction factors to the radiative acceleration by lines and by the continuum , due to the non - spherical shape of the star and due to the latitude dependence of the radiative flux . this method was applied to study its possible effects on the formation of disks around fast rotating b[e ] supergiants . the models with constant force multiplier parameters @xmath0 , @xmath1 and @xmath2 show a decrease of _ both _ the mass loss rate and the terminal velocity from pole to equator . this is mainly due to two effects : the reduction of the escape velocity from pole to equator , resulting in a higher terminal velocity at the pole , and the reduction of the radiative flux from the pole to the equator due to gravity darkening , which results in a decrease in mass loss rate at the equator . for a star with a fixed luminosity and a fixed polar radius , the temperature at the pole increases and the temperature at the equator decreases with increasing rotation rate . the terminal velocity of the wind from the poles is almost independent of @xmath181 but at the equator @xmath6 decreases with increasing rotation rate . the mass - loss rate at the pole increases with increasing @xmath181 , due to the increase in @xmath170 at the poles , but the mass loss rate from the equator is almost independent of the rotation . the combination of these effects alone produce a density contrast between the polar and the equatorial wind of a factor @xmath182 , except for fast rotating stars with high luminosity ( in which case @xmath183 ) . our results agree quantitatively with those obtained by maeder ( @xcite ) for the latitudinal dependence of @xmath135 . our predictions of the global mass loss rates are also in general agreement with those of maeder for models with luminosities far from the eddington limit . for high luminosities we find a strong polar outflow , whereas maeder s models predict an enhanced equatorial outflow on the basis of the original cak scaling laws . this difference is due to the fact that we have included an improved description of the radiation pressure in a rotating star . the difference between the polar wind and the equatorial wind is strongly enhanced when the bi - stability of radiation driven winds between @xmath149 and 30 000 k is taken into account . in this case the force multipliers @xmath0 and @xmath1 of a rapidly rotating star change drastically with stellar latitude if the pole is hotter than 25 000 k and the equator is cooler . in [ s_bs ] we have applied the newly calculated force multipliers from vink et al . ( 1999 ) above and below the bi - stability jump to the models of rotating b[e ] supergiants . one might argue that it is not allowed to apply the force multipliers of ( bi - stable ) spherical wind models to aspherical winds of rotating stars . however , we have shown that a disk formed through bi - stability will be sufficiently thick ( typically about 30 degrees for a star with an effective temperature equal to about 25 000 ) to make these spherical models reasonable approximations for the conditions in the wind close to the star , where the mass loss rate is determined . we find that rotationally induced bi - stability models of b[e ] stars reach a density contrast of about a factor 10 between the dense equatorial wind and the less dense polar wind . this is less than the factor @xmath184 that is observed ( see e.g. bjorkman @xcite ) . the extra density increase is most likely due to the wind compression . the gas that leaves the photosphere from a rotating wind , will follow an orbit in a tilted plane defined by the local rotation vector and the center of the star . if the rotational velocity is large or the wind velocity is small , this orbit will cross the equatorial plane where the streamlines of the wind from different stellar latitudes cross . the resulting shock will compress the gas into a thin outflowing disk with an opening angle of only a few degrees ( bjorkman & cassinelli 1993 ; @xmath5 chapt 11 ) . we have not considered this wind compression in our model . however , it is likely to occur in rotationally induced bi - stable winds , because the wind velocity in the equatorial plane is about a factor three smaller than from the poles , thus facillitating the wind compression . owocki & cranmer ( 1994 ) have argued that the flow of the wind towards the equatorial plane , predicted in the wcd - theory , may be offset by a @xmath14-component of the line acceleration towards the polar regions . this `` disk inhibition '' mechanism operates in wind models with constant force multipliers @xmath0 and @xmath1 . however , it is not clear that this mechanism is sufficiently strong to overcome the wind compression in the rib - model . this is because @xmath0 and @xmath1 change with latitude in the rib - model and there is a strong density gradient in the wind from the equator to the poles ( owocki et al . 1998 , puls et al . 1999 ) . the combination of the rib and the wcd mechanism may offer the best possibility for explaining the disks of b[e]-supergiants . whether the compression is strong enough to explain the observed high density contrast between the polar and the equatorial wind , remains to be seen . it can be calculated by combining the solutions of the wind momentum equation of rotating oblate stars with gravity darkening ( derived in this paper ) , and the calculation of the resulting trajectories of the wind . the combination of the rotation induced bi - stability model and the wind compressed disk model is promising for explaining the disks of b[e ] stars because the rib - mechanism explains the increased mass loss and the small velocity from the equatorial regions and the wcd - mechanism explains the strong compression of the disk . abbott d.c . , 1980 , apj 242 , 1183 abbott d.c . , 1982 , apj 259 , 282 bjorkman j.e . , 1998 , in _ b[e ] stars _ , eds a.m. hubert , c. jaschek kluwer : dordrecht , assl 233 , p 189 bjorkman j.e . , cassinelli j.p . , 1993 , apj 409 , 429 cassinelli 1998 in _ b[e ] stars _ , eds a.m. hubert , c. jaschek kluwer : dordrecht , assl 233 , p 177 castor j.i . , 1974 , mnras 169 , 279 castor j.i . , abbott d.c . , klein r.i . , 1975 , apj 195 , 157 cranmer s.r . , owocki s.p . , 1994 , apj 440 , 308 feldmeier a. , 1999 in _ variable and non - spherical stellar winds in luminous hot stars _ , lecture notes in physics , eds . b. wolf et al . springer : heidelberg , p 285 . friend d.b . , abbott d.c . , 1986 , apj 311 , 701 gayley k.g . , 1995 , apj 454 , 410 kudritzki r.p . , pauldrach a. , puls j. , abbott d.c . , 1989 , a&a 219 , 205 lamers h.j.g.l.m . , cassinelli j.p . , 1999 , introduction to stellar winds , cambridge univ . press . ( @xmath5 ) lamers h.j.g.l.m . , pauldrach a. , 1991 , a&a 244 , l5 lamers h.j.g.l.m . , , lindholm d.m . , 1995 , apj 455 , 269 lamers h.j.g.l.m . , zickgraf f .- j . , de winter d. , houziaux l. , zorec j. , 1998 , a&a 340 , 117 lucy l.b . , 1982 , apj 255 , 286 maeder a. , 1999 , a&a 347 , 185 owocki s.p . , castor j.i . , rybicki g.b . , 1988 , apj 335 , 914 owocki s.p . , cranmer r.c . , 1994 , apj 424 , 887 owocki s.p . , cranmer r.c . , gayley k.g . , 1998 , ap&ss 260 , 149 pauldrach a. , puls j. , kudritzki r.p . , 1986 , a&a 164 , 86 puls j. , petrenz p. , owocki s.p . , 1999 in _ variable and non - spherical stellar winds in luminous hot stars _ , lecture notes in physics , eds . b. wolf et al . springer : heidelberg , p 131 . , de koter a. , lamers h.j.g.l.m . , 1999 , a&a 350 , 181 von zeipel h. , 1924 , mnras 84 , 665 zickgraf f .- j . , 1992 , in _ astronomical ccd observing and reduction techniques _ howell s.b . , asp conference series vol . 22 , 75 zickgraf f .- j . , schulte - ladbeck r.e . , 1989 , a&a 214 , 274 zickgraf f .- j . , wolf b. , stahl o. , leitherer c. , klare g. , 1985 , a&a 143 , 421

the 40 transiting planet systems confirmed as of august 2008 show a remarkable diversity of properties , which is indicative of the complexity of planet formation and evolution processes . many different follow - up studies enabled by the special orientation of these systems ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * see charbonneau et al . 2007 for a review ) have brought about rapid improvements in evolutionary models of planet interiors and atmospheres @xcite . the increasing predictive power of these models is beginning to drive even more challenging observations . on the other hand , the precision and accuracy with which the most basic planet properties such as the mass and radius can be determined is currently limited by our knowledge of the properties of the host stars . significant uncertainties remain in the stellar mass and radius determinations of many systems . in some cases this is due to poorly determined photospheric properties ( mainly temperature and metallicity ) , and in others to a lack of an accurate luminosity estimate . additionally , the variety of methodologies used for these determinations and the different approaches towards systematic errors have resulted in a rather inhomogeneous set of planet properties , as discussed by @xcite . this complicates the interpretation of patterns and correlations that are being proposed ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? recent improvements in the analysis techniques have the potential to increase the accuracy of the stellar and planetary parameters significantly , especially for the large majority without a direct distance estimate . in particular , the application of the constraint on the stellar density that comes directly from the transit light curves has been shown to be superior to the use of other indicators of luminosity such as the surface gravity ( @xmath15 ) determined spectroscopically @xcite . @xcite have recently reanalyzed a large subset of the known transiting planets , incorporating these improvements and applying a uniform methodology to all systems . in the present work we focus on two of the recently discovered transiting systems , @xcite and @xcite , which lack accurate estimates for the photospheric properties of the parent stars and as a result have more uncertain stellar and planetary parameters . to improve upon these properties , we present new radial - velocity and photometric observations of with which we refine both the light curve solution and the spectroscopic orbit . we also carry out a reanalysis of existing photometry utilizing a technique which treats stellar limb darkening using adjustable parameters . we perform the first detailed spectroscopic determination of the photospheric properties of both stars , and we make use of the constraint on the stellar density mentioned above to infer more accurate values for the stellar and planetary masses and radii . our paper is organized as follows . in [ sec : obs ] we summarize the observations . in [ sec : atmpar ] we present our effective temperature and metallicity determinations , along with several consistency checks aimed at establishing the accuracy of the temperatures . in [ sec : orb ] we report an updated spectroscopic orbital solution for , and the light curve solutions for both systems are discussed in [ lcanalysis ] . section[sec : physics ] then describes our determination of the stellar masses and radii , which in turn lead to refined values for the planetary parameters over those reported in the discovery papers . we conclude in [ sec : summ ] by providing a summary of our results and by discussing whether the properties of the host stars give any useful clues on the origin of the strongly contrasting densities of their close - in gas giant planets , particularly in comparison with the other known transiting planet systems . high - resolution , high - snr ( signal - to - noise ratio ) spectroscopic observations of were obtained in july and september 2007 with the hires spectrograph on the keck i telescope @xcite , using essentially the same setup as in the discovery paper . the spectra cover the effective wavelength range @xmath1632008800 , and an iodine ( i@xmath17 ) gas absorption cell placed in front of the spectrograph slit was used to superimpose a dense set of narrow molecular lines between @xmath165000 and @xmath166000 . this iodine spectrum provides a stable wavelength reference and a means of monitoring the instrumental profile that are crucial for achieving precise radial velocity determinations . the observations were reduced and extracted using the makee software written by t. barlow . we obtained 4 spectra with typical exposure times of 15 min resulting in snrs of @xmath16100 pixel@xmath18 , which we combined with the discovery observations reported by @xcite for a total of 11 iodine spectra . precise radial velocities were measured following a procedure based on the methodology developed for the afoe spectrograph @xcite , and adapted for the processing of hires spectra . conceptually , the method models the observed star and i@xmath17 spectrum by using templates for the stellar and i@xmath17 spectra . this model includes physically motivated parameters that describe the spatial and temporal variations of the instrument , including the instrumental profile , as well as the sought - after relative doppler shift with respect to the stellar template . the model parameters are adjusted to minimize the difference between the model and the observations in a least - squares sense , down to the poisson noise . the stellar template is estimated by deconvolving an observation taken without superimposing i@xmath17 , whereas the template for the i@xmath17 is based on a high - resolution and high snr scan of the gas cell using the fourier transform spectrometer on the mcmath solar telescope at the kitt peak observatory ( see * ? ? ? each spectral order is modeled as a whole but analyzed independently . the final radial velocities are estimated from the mean of the doppler shifts computed for each order and their uncertainties from the standard deviation of that mean . the resulting velocities , expressed in the solar system barycentric frame , along with their associated formal errors are reported in table [ rvtres3 ] . these measurements include and supersede the velocities presented in the discovery paper , and they also correct a minor error in the previously published dates of observation . the revised orbital solution is presented and discussed in [ sec : orb ] . in addition to the photometric measurements presented in the discovery paper @xcite , we have collected six other high - cadence , high - precision transit light - curves of . one was obtained in the @xmath19 band on ut 2007 april 23 using the ccd camera of the iac80 telescope at the observatorio del teide , tenerife , spain , and the other five were gathered using keplercam ( see , e.g. , * ? ? ? * ) at the flwo 1.2-m telescope : two in the sloan @xmath20 and @xmath21 bands on ut 2007 april 25 and 2008 march 27 , and the remaining three in the @xmath22 band on ut 2008 march 9 , april 12 , and may 8 . on all nights conditions were quite good , except for the presence of high cirrus and highly variable seeing on ut 2008 april 12 . the transits were observed at airmasses ranging between @xmath161.00 and @xmath161.80 . all datasets were reduced using standard calibration techniques ( overscan correction , trimming , bias subtraction , flat fielding ) . we then performed aperture photometry of and between 10 and 30 comparison stars , depending on the filter and exposure time . we experimented with different choices for the aperture size , comparison star ensemble , and weighting of the comparison stars , aiming for the smallest out - of - transit ( oot ) rms . in practice , the best aperture size was approximately twice the fwhm of the stellar image on each night , and the best results were obtained using a straight average of the normalized light curves of the comparison stars . as for the formal errors on each photometric data point , we used the product of the oot rms of each light - curve and the factor @xmath23 , used to account for departures from gaussian ( `` white '' ) uncorrelated noise ( see , e.g. , * ? ? ? * ) . for consistency , we also re - processed in the same way the two discovery light - curves of presented in odonovan et al . ( 2007 ) , as well as the three photometric datasets utilized by mandushev et al . ( 2007 ) in their discovery announcement of . the main characteristics of the and light - curves are summarized in table [ tbl : lctres3 ] . the final set of photometric time - series of in all filters ( including the discovery data ) is available in machine - readable form in the electronic version of table [ tbl : phottres3 ] , and is plotted in figure [ tres3phot ] . table [ tbl : phottres4 ] reports all photometric data for , which are shown graphically in figure [ tres4phot ] . a detailed abundance analysis was carried out using the keck / hires template spectra of and . the stellar atmosphere parameters @xmath24 , @xmath15 , and [ fe / h ] were determined using standard methodology ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , which we summarize as follows . equivalent widths ( ews ) for a set of relatively weak and lines were measured manually in the keck spectra using the splot task in iraf ( see , e.g. , * ? ? ? * and references therein , for details on the specific choice of lines ) . these were then used together with a grid of kurucz atlas plane - parallel stellar model atmospheres @xcite as inputs to the 2002 version of the moog spectral synthesis code @xcite . atmospheric parameters were derived under the assumption of local thermodynamic equilibrium ( lte ) , imposing excitation and ionization balance . formal uncertainties on @xmath24 , @xmath15 , and microturbulent velocity @xmath25 were derived using the approach described in @xcite and @xcite , while the nominal uncertainty for [ fe / h ] corresponds to the scatter obtained from the lines rather than the formal error of the mean . the resulting set of parameters for is @xmath26 k , @xmath27 , @xmath28 @xmath29 , and [ fe / h ] = @xmath30 , and for we obtain @xmath31 k , @xmath32 , @xmath33 @xmath29 , and [ fe / h ] = @xmath34 . these values are collected in table [ star_tres3 ] and table [ star_tres4 ] , respectively . as a check , the manual ew measurements used above were compared against the results from the automated software aressousasag / ares . ] , made available to the community by @xcite . figure [ aresiraf ] shows the comparison between the ews measured for using ares and those from the manual approach with iraf . the top panel indicates excellent agreement between the two ; the linear fit has a slope of 1.011 . there are no significant correlations with the line strength ( see middle panel ) : the fractional difference between ews measured by the two methods has a probability of no correlation of 0.1 . finally , the histogram in the bottom panel shows there is no appreciable systematic difference between the methods . the mean difference and scatter are only 0.67 m and 2.6 m , respectively , which are smaller than found by @xcite from a similar comparison between feros , harps and uves spectra . consequently , the results of the abundance analysis with moog using ews measured with ares are virtually identical to those presented earlier . following @xcite we also synthesized a number of unblended lines in the template spectra of both and , and determined projected rotational velocities of @xmath35 @xmath29 and @xmath36 @xmath29 , respectively . our new determinations of the atmospheric parameters for the parent stars of and are generally in good agreement with the values reported by @xcite and @xcite , the main difference being that those authors assumed [ fe / h ] = 0.0 in their studies , whereas we find significant departures from solar metallicity in both stars . we describe here additional estimates of the effective temperature for and that serve to test the accuracy of our determinations above . both planet host stars were observed spectroscopically with the center for astrophysics ( cfa ) digital speedometer @xcite as part of the regular follow - up after the discovery of photometric signals suggesting transits . these observations cover 45 in a single echelle order centered at 5187 , and have @xmath38 . while this resolving power is moderately high , the snrs of the spectra are low , ranging from 7 to 13 per resolution element of 8.5 @xmath29 for the 13 exposures of , and snrs of 1113 for the 7 exposures of . nevertheless , useful information on the stellar properties can be extracted from these spectra as described by @xcite . briefly , the observed spectra are cross - correlated against a library of synthetic spectra based on kurucz model atmospheres ( see * ? ? ? * ; * ? ? ? * ) , calculated over a wide range of values of @xmath37 , @xmath15 , [ fe / h ] , and rotational velocity @xmath39 . the combination of parameters yielding the highest correlation averaged over all exposures is adopted as a representation of the properties of the star . due to the narrow wavelength range and limited snrs , the first three of the above properties are typically strongly correlated and are difficult to determine simultaneously . given these constraints , in the discovery papers we initially held the metallicity fixed at the solar value for both stars . for , which our keck spectroscopy indicates is slightly metal - poor , we repeated the above determination using a fixed metallicity of [ fe / h ] @xmath40 , and interpolated @xmath37 and @xmath15 to the precise composition from table [ star_tres3 ] . we obtained @xmath41 k and @xmath42 , where the uncertainties include contributions both from the error in the adopted [ fe / h ] and the internal errors . the temperature as well as the surface gravity are consistent with the keck determinations , within the uncertainties . similarly for we repeated the determinations for [ fe / h ] @xmath43 , given the metal - rich composition indicated by the keck spectroscopy , and interpolated to the intermediate value from table [ star_tres4 ] . the results are @xmath44 k and @xmath45 , which are once again in good agreement with the more reliable estimate from keck . while the temperature determinations discussed so far are implicitly based on the _ strength _ of the spectral lines , @xcite have demonstrated that highly precise information can also be extracted using the _ ratio _ of the depths of two spectral lines having different sensitivity to temperature . however , rather than yielding absolute temperatures , this technique in its original formulation only measures changes in temperature , albeit with extremely high precision often reaching a few kelvin ( see , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? absolute temperatures can still be obtained with recourse to external color - temperature calibrations , since line - depth ratios ( ldrs ) are usually strongly correlated with the color index of the star . the accuracy of such @xmath37 determinations is then limited by the calibrations themselves . @xcite have presented ldr-@xmath37 calibrations based on 26 carefully selected lines of fe , v , sc , si , and ni in the spectral interval 61906280 , grouped into 16 line pairs . these calibrations are valid for stars with temperatures between @xmath163800 k and @xmath166000 k , so unfortunately the technique is not applicable to . for we measured all 26 lines in our high - snr keck template spectrum , and adopted the @xcite calibrations appropriate for stars with rotational velocities of 10 @xmath29 ( very close to the @xmath46 value reported by @xcite and obtained also in this work ) . the temperature scale of these relations relies on a transformation between @xmath47 and @xmath37 by @xcite , which makes use of a mixture of dwarf and giant temperatures obtained by many different methods and does not account for differences in metallicity . for the present work we have preferred to use more sophisticated color - temperature relations such as those by @xcite and @xcite , which account not only for luminosity class but include also metallicity terms , and are based on effective temperatures derived homogeneously by the infrared flux method . the conversion from the ldr - based @xmath37 inferred from the @xcite relations back to an average color for the star using the @xcite prescription gives @xmath48 . our two preferred color - temperature relations mentioned above then yield a weighted average temperature of @xmath49 k for , in which the uncertainty includes observational errors propagated from the ldr measurements as well as the scatter of the calibrations . this result is consistent with our more direct estimate in [ sec : keck ] . as is well known , the wings of the @xmath51 line ( but not its core , formed higher up in the atmosphere under non - lte conditions ) are very sensitive to @xmath24 variations , but are relatively insensitive to changes in @xmath15 and [ fe / h ] ( see , e.g. , * ? ? ? * ; * ? ? ? * and references therein ) . this allows for a useful consistency check on our @xmath24 estimates above . we compared the observed h@xmath50 line profiles in our keck template spectra of and against synthetic profiles for solar - metallicity dwarfs ( [ fe / h ] = 0.0 , @xmath52 ) from the kurucz database . the results of this exercise are displayed in figure [ halpha ] , in which 10 regions centered on h@xmath50 are shown for each star together with four calculated profiles for different values of @xmath24 . in both cases the temperatures one would infer from these comparisons agree well with the estimates reported in table [ star_tres3 ] and table [ star_tres4 ] . an additional check on the effective temperatures is available from the multi - color photometry for and . measurements in johnson @xmath53 , cousins @xmath54 , and 2mass @xmath55 were used to derive seven color indices for , and nine for when considering also the @xmath56 and @xmath57 measurements from the tycho-2 catalog @xcite . the calibrations by @xcite and @xcite then yielded weighted average temperatures of @xmath58 k and @xmath59 k for and , respectively , ignoring extinction and adopting [ fe / h ] in each case as determined in [ sec : keck ] . the uncertainties include photometric errors and metallicity errors , as well as the scatter of the calibrations , but exclude unquantified systematic errors in the calibrations themselves . these temperatures are 260 k and 65 k cooler than our spectroscopic determinations above . they can be reconciled with the values in [ sec : keck ] if we assume the presence of reddening , and correct each of the indices . for the required value of @xmath60 is hardly significant ( @xmath61 mag ) , but for we obtain @xmath62 mag . , in which the ldr - based @xmath47 color is significantly bluer than the measured value of @xmath63 @xcite . ] for comparison , the reddening maps of @xcite and @xcite indicate a _ total _ reddening along the line of sight to each star of only @xmath64 mag , and similar results are obtained from the model of galactic dust distribution by @xcite . this is not inconsistent with the small value we infer for , but it is much smaller than our estimate for . possible explanations include patchy interstellar material combined with the relatively coarse resolution of these maps ( a few arc minutes for * ? ? ? * ) , or perhaps the presence of circumstellar material in . the reliability of age indicators for stars older than 12 gyr , such as chromospheric activity and lithium ( li ) abundance , as well as their interagreement , have been the subject of much debate in the literature ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the difficulties are due to many factors , including ( but not limited to ) non - trivial correlations between chromospheric activity , rotation , mass , and age , limited availability of activity estimates averaged over entire stellar activity cycles that range from @xmath65 to @xmath66 . these correspond to ages of @xmath67 gyr and @xmath16 2.5 gyr , respectively ( see , e.g. , * ? ? ? ] , insufficient understanding of the temporal evolution of li depletion due to the complex interplay between various processes ( e.g. , convection , mixing , diffusion , mass loss ) , and non - negligible differences in the observed behavior of chromospheric activity and li depletion as a function of mass , age , and chemical abundance between stars in young and old clusters , and in the field . we have nonetheless attempted to use the activity indicator and the lithium abundance as measured in our hires spectra of and to provide independent constraints for comparison with the formal age estimates determined below from evolutionary models ( see [ sec : physics ] ) , as well as to search for possible peculiarities of these two planet hosts compared to other samples of stars with and without planets . figure [ ca_tres3_4 ] shows a region of the hires template spectra for each star centered on the h line . clear emission is seen in the core of the line of , but not in . following the procedure outlined in @xcite we measured the chromospheric emission ratio @xmath68 , corrected for the photospheric contribution , from the h and k lines in each of our spectra . we obtained @xmath69 and @xmath70 for and , respectively . thus appears moderately active , while is quite the opposite . the resulting chromospheric age estimates , using the relations summarized in @xcite , are @xmath71 gyr for and @xmath72 gyr for . in figure [ licomp ] we show the results of the spectral synthesis of a 10 region centered on the li @xmath736707.8 line , using the atmospheric parameters derived from the fe - line analysis and the line list of @xcite . the two panels display the comparison between the observed spectra and three synthetic spectra , each differing only in the assumed li abundance . neither star shows a measurable li line , and we can only place upper limits of @xmath74 and @xmath75 for and , respectively . by comparison with average li abundance curves as a function of effective temperature for clusters of different ages @xcite , one would infer a rather old age for of @xmath76 gyr , nominally inconsistent with the estimate above from the activity index . based on its li the object appears decidedly older than hyades stars of the same @xmath24 , and this argument is corroborated ( e.g. , * ? ? ? * ) by the small @xmath46 we measure ( see table [ star_tres3 ] ) . for , the age inferred from the @xcite relations is @xmath77 gyr , which is in qualitative agreement with the absence of significant chromospheric activity as well as with the inferred value of the surface gravity ( see table [ star_tres4 ] ) , which indicates the star is somewhat evolved . the measured li abundance for is in line with values determined by @xcite for a sub - sample of nearby planet hosts with the same @xmath24 , while appears more depleted than other planet hosts of similar @xmath24 , @xmath78 , and [ fe / h ] . we discuss briefly in [ sec : summ ] some implications of these findings in the broader context of the existence of chemical peculiarities in planet hosts in comparison with stars without detected planets . from the measurements in this section we conclude that the above results are consistent with the notion that neither star is very young ( @xmath79 12 gyr ) , that is likely to be of intermediate age ( @xmath163 gyr ) , and that appears older and more evolved . using the radial velocities presented in [ sec : rvs ] we have updated the orbital solution given by @xcite , adopting the improved ephemeris described below in [ ttv ] . a keplerian circular orbit was adjusted to the data , and the results may be seen in table [ tab : par_tres3 ] and are shown graphically in figure [ specorb_tres3_4 ] . as found also by @xcite , the solution gives a scatter that is larger than expected from the internal velocity errors . this is most likely due to velocity `` jitter '' associated with chromospheric activity . if we model this as excess scatter to be added quadratically to the internal errors , we find that the amount of jitter required to produce a reduced @xmath80 value near unity is @xmath1618.5 @xmath81 . external estimates of the jitter for can be made on the basis of the spectral type , the measured value of @xmath82 @xmath29 , and the activity index @xmath83 , and vary considerably but generally range from about 5 to 20 @xmath81@xcite . these estimates are thus consistent with our findings . nevertheless , as a test we also modeled the data with an eccentric orbit but did not obtain much improvement , and the result for the eccentricity was not significantly different from zero ( @xmath84 ) . there are no indications of long - term variations in the observations at hand . for we adopt in the following the spectroscopic orbit by @xcite , since the radial - velocity material has not changed . for clarity , the orbital solution is reported again in table [ tab : par_tres4 ] . the analysis of the differential photometry for and was carried out in essentially the same manner as described in @xcite . we refer the reader to that paper for the details . given that we have multiple light curves for each system , we first determined the individual times of transit as described below in order to improve the ephemeris , and we subsequently refined the light - curve parameters . to determine the center of each measured transit event we initially adopted the light - curve parameters from the discovery papers . for , we adopted a quadratic limb darkening ( ld ) law with coefficients from @xcite , interpolated to the values @xmath85 k , @xmath86 dex , [ m / h ] @xmath87 dex , and @xmath88 @xmath29 . all ld coefficients for the various filters are reported in table [ tbl : ldcoeff ] . we fitted each light curve individually to solve only for the time of transit center @xmath89 and also a linear function of time ( two parameters ) to describe the oot flux ( the slope in the oot flux accounts for systematic errors , including differential extinction ) . the measured mid - transit times are presented in table [ midtrans ] . we then fitted a straight line to the central times of the eight transits of the form @xmath90 and derived the following new ephemeris ( which we report in table [ tab : par_tres3 ] ) : @xmath91 ( hjd ) , @xmath92 days . with a number of degrees of freedom @xmath93 , the resulting reduced @xmath94 indicates a rather poor fit . the transit timing residuals for are shown in figure [ trestiming ] . there are four outliers at the 23.5@xmath95 level . this could be seen as evidence suggesting that the period is not constant . alternatively , the errors might have been underestimated . the data available are not enough to draw any significant conclusion on the nature of these variations , but clearly additional observations are warranted . we leave for a future study the evaluation of the dynamical interpretation and significance of the transit times , as was done , for example , by @xcite for tres-1 and by @xcite for ogle - tr-111b . similarly for , we re - fitted the three discovery light - curves using the @xcite ld coefficients listed in table [ tbl : ldcoeff ] appropriate for @xmath96 k , @xmath97 dex , [ m / h]@xmath98 dex , and @xmath25 = 2.0 @xmath29 . the corresponding times of transit center , derived with the same procedure adopted for , are presented in table [ midtrans ] . given the short time baseline of these observations , we did not derive a new ephemeris for the system , but for the purpose of the analysis presented in the next section we simply adopted the mandushev et al . ( 2007 ) values of @xmath99 and @xmath89 . next , for both and we locked the transit times and oot baseline functions at the values indicated above and then , under the assumption of a circular orbit , we fitted all light curves simultaneously using the algorithm by @xcite to derive the relevant quantities radius ratio @xmath100 ( where @xmath101 and @xmath102 are the planetary and stellar radius , respectively ) , inclination @xmath22 , projected separation @xmath6 ( where @xmath103 is the semi - major axis ) , and impact parameter @xmath104 . for we found @xmath105 , @xmath106 , @xmath107 , and @xmath108 . for , we obtained @xmath109 , @xmath110 , @xmath111 , and @xmath112 . these estimates were derived assuming initial values for the stellar and planetary masses of @xmath113 @xmath114 and @xmath115 @xmath116 for , and @xmath117 @xmath114 and @xmath118 @xmath116 for , although the results are insensitive to these values . recent studies of transiting exoplanets light curves ( e.g. , * ? ? ? * and references therein ) have highlighted some issues related to how much effect different treatments of limb darkening can have on the light - curve solutions . in particular , these do not appear to be significantly affected by the specific choice of the ld law , while fixing the ld coefficients to their theoretical values seems to result in significantly smaller uncertainties in the fitted parameters with respect to the case in which the ld coefficients are also adjusted during the fitting procedure . the above results were obtained keeping all ld coefficients fixed . we decided to investigate the effect of fitting for the ld coefficients in the following way . given the quality of the data available to us , we can only fit for one coefficient ( or one combination of the two coefficients ) . we chose to fix the quadratic coefficient @xmath119 and solve for the linear coefficient @xmath120 . we found that allowing complete freedom in the linear coefficient resulted in `` unphysical '' solutions ( e.g. , largely negative coefficients ) . @xcite found that the fitted ld coefficients are usually within 0.10.2 of the theoretical claret values , and that is consistent with our experience ( see , e.g. , * ? ? ? * ) . we then decided to use an a priori constraint enforcing an agreement of @xmath121 , modifying the merit function as follows : @xmath122 ^ 2+\left(\frac{u - u_1}{0.2}\right)^2\ , , \ ] ] where @xmath123 is the stellar flux observed at time @xmath124 , @xmath125 its corresponding error , @xmath126 is the model value , @xmath127 is the adjustable linear ld coefficient , and @xmath120 its theoretical value ( appropriate for each band - pass ) . furthermore , for all three @xmath22-band light curves were required to agree on the ld parameter , and the same requirement was set on the two @xmath128-band light curves for . the results for the system parameters in this case were @xmath129 , @xmath130 , and @xmath131 . for , we obtained @xmath132 , @xmath133 , and @xmath134 . in both cases , the agreement between the system parameters derived keeping the ld coefficients fixed and those when the ld coefficients are part of the solution was excellent , with only a slight increase in the estimated uncertainties for the latter case . indeed , as already noted by @xcite , fixing the ld coefficients at their theoretically predicted values does not appear to significantly bias the results . however , in the interest of providing more conservative error estimates , we believe that , for the purpose of the analysis of high - quality light - curves such as the ones presented in this paper , a procedure that treats ld coefficients as adjustable parameters is preferable . based on the above considerations , for both and we elected to include in table [ tab : par_tres3 ] and table [ tab : par_tres4 ] the values of the systems parameters obtained from the light - curve analysis in the case in which the linear ld coefficient was allowed to float . we note that this approach provides a departure from the @xcite analysis . the revised spectroscopic determinations of @xmath37 and [ fe / h ] , along with the new spectroscopic orbital solution for and the light curve fits presented above , allow us to refine the determination of the stellar and planetary properties for both systems . to establish the properties of the parent stars we rely on stellar evolution models by @xcite and @xcite . we make explicit use of the constraint on the stellar density provided by the light - curve quantity @xmath135 , as described by @xcite . the procedure follows closely that given in the previous citation . briefly , we seek the best match ( in a @xmath80 sense ) between the measured \{@xmath37 , [ fe / h ] , @xmath135 } and points on a finely interpolated grid of isochrones spanning a wide range of metallicities and ages . two minor improvements in this procedure , described in more detail by @xcite , have to do with the weighting of each point sampled along the isochrones according to the distance in @xmath37-[fe / h]-@xmath135 space compared to the observed values , and an additional weighting according to the likelihood that the star is in a particular evolutionary state . the latter effect is accounted for by multiplying the first weight by the expected number density of stars at each location in the h - r diagram , according to an adopted initial mass function ( which in this case is simply a power law with a salpeter index ) . the impact of these weighting factors is generally minor . the results for the stellar properties of and are presented in table[star_tres3 ] and table[star_tres4 ] . the distance estimate to accounts for interstellar extinction ( @xmath136 mag ) , as described in [ sec : phot_teff ] . for both stars the surface gravity inferred from the models is typically much more accurate than the spectroscopic gravity determination , showing the power of the constraint on the luminosity and size of the star afforded by @xmath135 . the results for both and supersede those given recently by @xcite because of the new photometry and radial velocities contributed here in the former case , and the new light - curve solutions for both systems . the evolutionary ages for the two stars are qualitatively consistent with our conclusions from [ sec : age ] , but are somewhat more uncertain in the case of . finally , using the distance estimates inferred in this work , the ucac2 proper motion components @xcite reported by @xcite and @xcite , and the mean radial velocity values @xmath137 @xmath29 and @xmath138 @xmath29 as measured with the cfa digital speedometers , we obtain galactic space motion vectors [ @xmath139 , @xmath19 , @xmath140 = [ + 27.3 , + 6.7 , + 33.0 ] @xmath29 and [ @xmath139 , @xmath19 , @xmath140 = [ @xmath14143.9 , @xmath14139.1 , @xmath1416.9 ] @xmath29 for and , respectively ( where @xmath139 is taken to be positive toward the galactic anticenter ) . we collect these results along with the other properties derived previously in table [ star_tres3 ] and table [ star_tres4 ] . the planet parameters follow from the stellar properties and the results of the transit light curve and spectroscopic orbits , and are presented in table[tab : par_tres3 ] and table[tab : par_tres4 ] . for , our planetary mass and radius are significantly larger compared with the determinations in the discovery paper and in @xcite . this is due in part to the increased mass and radius for the parent star , but also to the larger radius ratio based on the new photometry . our detailed spectroscopic analyses of and have yielded accurate values of the atmospheric properties ( @xmath24 and [ fe / h ] ) , which are critical for establishing the fundamental properties of the hosts . the accuracy of the temperatures is supported by a number of independent checks ( low - resolution spectroscopy , line - depth ratios , h@xmath50 line profiles , color - temperature calibrations ) that gives us confidence that the inferred stellar properties are reliable . we find that is a main - sequence g dwarf with a metallicity about 1.5 times lower than the sun s , not a very common occurence among exoplanets hosts . is a somewhat evolved late f star that is nearing the end of its main - sequence phase , and is slightly enhanced in its iron content with respect to the solar abundance . the agreement we find between age indicators for and based on measurements of the activity levels , the lithium abundances , and rotation , and the evolutionary age inferred from the models is fair , although , as discussed in [ sec : age ] , the reliability of empirical age estimates for stars that are not young ( @xmath142 gyr ) is somewhat questionable . the model estimates themselves are not without their problems . nevertheless , we conclude the stars are 13 gyr old . neither star stands out as peculiar when compared with other planet hosts with similar physical properties . in the small @xmath46 value and the fact that we can only place an upper limit on the li abundance are consistent with the notion that planet hosts with @xmath24 similar to the sun appear to rotate more slowly and are more li - depleted than stars without detected planets . as pointed out recently ( * ? ? ? * and references therein ) , this evidence suggests that a planet - forming disk may induce additional rotational braking , leading to enhanced mixing in the stellar envelope , which in turn accelerates the destruction of lithium . , on the other hand , does not seem consistent with the claim by @xcite that hotter planet hosts with @xmath143 k have higher li abundances , possibly due to self - enrichment processes . it is worth keeping in mind , however , that these discrepancies may not be significant given the large spread in the li abundance for field stars with the temperature and mass of ( e.g. , * ? ? ? new investigations on these issues are clearly needed based on uniform analyses of large samples of planet hosts and statistically significant , well - defined control samples of stars without detected planets . new radial - velocity measurements for presented here have enabled us to revise the spectroscopic orbit for that system . we detect no indication of any longer - term variations in the radial velocities that might suggest the presence of another body in the system . however , the small number of observations and their limited time span of only 6 months emphasize the need for continued doppler monitoring of this and other transiting planet systems to investigate the possibility of additional companions . in stark contrast to the considerable ground - based and space - based efforts invested in studying in great detail the atmospheric properties of many of these objects , which have undoubtedly led to tremendous insights into their structure , formation , and evolution , the amount of radial - velocity data available for transiting planets is meager , and often does not go beyond the handful of observations published in the discovery papers . interest in the radial velocities seems to be quickly lost . it should be pointed out that the frequency of close - in giant planets ( @xmath144 day ) with additional massive planets in outer orbits ( up to the detection limit of today s doppler surveys , @xmath145 au ; see , e.g. , * ? ? ? * and references therein ) is about 12% ( 8 out of 70 systems discovered via rv methods ) , and it would therefore be wise to continue the velocity monitoring of some of the transiting systems . if additional planets in a transiting system were detected , they might also be found to undergo transits , and such a discovery would allow us to constrain structural models for gas giants akin to jupiter or saturn , and open exciting opportunities for additional investigations with present ( _ spitzer _ ) and upcoming ( jwst ) space - borne observatories . accurate stellar properties for and have been derived here following the approach described in @xcite and @xcite , comparing the spectroscopically determined @xmath24 and [ fe / h ] values along with the photometrically measured @xmath6 with current stellar evolution models . these properties have in turn allowed us to refine the determination of the mass and radius of the planets . in particular , we confirm that is the planet with the largest radius among the currently known transiting hot jupiters . recently , winn et al . ( 2008a ) derived upper limits on the albedo of tres-3 based on the nondetection of occultations observed at optical wavelengths . our findings are relevant to this study in two ways . firstly , they strengthen the case for a circular orbit , which is important because if the orbit is eccentric then it would be possible that winn et al . ( 2008a ) did not observe tres-3 at the actual times of occultations and that their data place no constraint on the albedo . secondly , our revised light - curve parameters are relevant because the upper limit on the geometric albedo ( @xmath146 ) was inferred from the measured upper limit on the planet - to - star flux ratio ( @xmath147 ) according to our revised value of @xmath149 therefore leads to revised upper limits on the geometric albedo of tres-3 . however , this revision turns out to be minor : the new value of @xmath149 is only 1.5% larger than the value used by winn et al . ( 2008a ) . the upper limits on the geometric albedo become weaker by about 3% . at 99% confidence , the revised upper limits are 0.31 , 0.64 , and 1.10 in @xmath150 , @xmath151 , and @xmath152band , respectively . there is a large spread in the observed radii and densities for transiting planets of comparable mass placed at similar orbital distance from stars of very similar properties ( @xmath24 , @xmath15 , [ fe / h ] , and age ) . for example , if we consider along with 5 other transiting planet hosts ( excluding ) with similar characteristics ( hd 149026 , hd 209458 , ogle - tr-56 , ogle - tr-132 , and wasp-1 ) , the nominal masses of the attending planets vary by a factor of @xmath153 , but reported densities vary by a factor of @xmath154 ( e.g. , * ? ? ? many theoretical mechanisms have been proposed to inflate the radius of a strongly irradiated planet , such as additional sources of internal heating due to stellar insolation @xcite , tidal heating due to non - zero eccentricity caused by gravitational interaction with an outer companion @xcite or by rotational obliquity @xcite , elevated interior opacity due to enhanced atmospheric metallicity @xcite , or varying core masses @xcite . none of these appear capable of explaining the observed spread in density in a natural way ( see also * ? ? ? * ) . of the 6 stars just mentioned , all are more metal - rich than the sun , except for hd 209458 , which has [ fe / h ] close to solar . some of the above models @xcite predict a positive correlation between the inferred planetary core mass and the host star s metallicity , as in the framework of the core - accretion model of giant planet formation ( e.g. , * ? ? ? this idea assumes [ fe / h ] closely tracks the metallicity of the protoplanetary disk , so that more metal - rich stars should be orbited by more metal - rich planets , with a larger heavy - element content . however , among these 6 systems only one planet ( hd 149026b ) has an inferred core mass significantly larger than zero , and four of the other planets have measured radii so large that the cores are likely to be insignificant . in fact , even in the absence of a core the observed radii can not be reproduced by the models , with being the extreme case . interestingly , over 40% of the transiting planets reported in table 5 of @xcite do not appear to require any core at all to explain their sizes ( according to the models of * ? ? ? * ) . taking all this into consideration , the claimed evidence for a core mass metallicity correlation could indeed be seen as supporting the more widely accepted scenario of formation by core accretion , but from the indications above it may also be that a significant fraction of these objects formed in a different way ( e.g. , * ? ? ? * and references therein ) . given the evidence collected so far , we suggest that simply connecting the host star s characteristics to the structural properties of transiting planets may in fact be an over - simplification . we conclude by stressing the importance of refining our understanding of the complex interplay between the disk environment and a forming giant planet , and its evolutionary history after envelope accretion , which might turn out to be more directly responsible for its final structure and composition than the metal content of the parent star . as gratefully acknowledges the kepler mission for partial support under nasa cooperative agreement ncc 2 - 1390 . gt acknowledges partial support for this work from nasa origins grant nng04lg89 g . dc is supported in part by nasa origins grant nng05gj29 g . ftod acknowledges partial support for this work provided through the nasa postdoctoral program at the goddard space flight center , administered by oak ridge associated universities through a contract with nasa . jbl gratefully acknowledges support from nsf grant ast-0307340 . some of the data presented herein were obtained at the w. m. keck observatory , which is operated as a scientific partnership among the california institute of technology , the university of california and the national aeronautics and space administration . the observatory was made possible by the generous financial support of the w. m. keck foundation . the authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of mauna kea has always had within the indigenous hawaiian community . we are most fortunate to have the opportunity to conduct observations from this mountain . this research has made use of nasa s astrophysics data system abstract service and of the simbad database , operated at cds , strasbourg , france . durisen , r. h. , boss , a. p. , mayer , l. , nelson , a. f. , quinn , t. , & rice , w. k. m. 2007 , in protostars and planets v , b. reipurth , d. jewitt , and k. keil ( eds . ) , university of arizona press ( tucson ) , 607 lcc 54187.04136 & 176.0 & 8.2 + 54187.13832 & 99.4 & 10.7 + 54188.02349 & 189.5 & 11.2 + 54188.12067 & 292.4 & 13.3 + 54189.01696 & @xmath141331.6 & 5.5 + 54189.09680 & @xmath141246.4 & 11.4 + 54189.14038 & @xmath141195.8 & 7.3 + 54288.85394 & 241.6 & 7.8 + 54288.98956 & 116.7 & 11.5 + 54289.82337 & @xmath1415.9 & 10.8 + 54372.86102 & @xmath141335.9 & 9.5 + + 2007 - 03 - 25 & @xmath128 & flwo & 1.73 & 0.0012 & 0.0011 & 1.17 + 2007 - 04 - 08 & @xmath155 & ftn & 1.15 & 0.0011 & 0.0014 & 1.46 + 2007 - 04 - 24 & @xmath19 & ot & 1.34 & 0.0016 & 0.0016 & 1.30 + 2007 - 04 - 25 & @xmath20 & flwo & 0.98 & 0.0014 & 0.0015 & 1.16 + 2008 - 03 - 09 & @xmath22 & flwo & 0.73 & 0.0018 & 0.0016 & 1.00 + 2008 - 03 - 27 & @xmath21 & flwo & 0.73 & 0.0014 & 0.0014 & 1.07 + 2008 - 04 - 12 & @xmath22 & flwo & 1.05 & 0.0015 & 0.0015 & 1.02 + 2008 - 05 - 08 & @xmath22 & flwo & 0.73 & 0.0021 & 0.0020 & 1.16 + + 2007 - 05 - 03 & @xmath128 & flwo & 0.73 & 0.0015 & 0.0016 & 1.26 + 2007 - 05 - 10 & @xmath128 & flwo & 0.73 & 0.0019 & 0.0017 & 1.23 + 2007 - 05 - 10 & @xmath155 & lowell & 1.54 & 0.0015 & 0.0015 & 1.22 + + 2454185.850884 & 0.99965 & 0.00115 + 2454185.853303 & 0.99948 & 0.00115 + 2454185.854495 & 0.99852 & 0.00115 + + 2454198.948010 & 0.99987 & 0.00109 + 2454198.948809 & 1.00054 & 0.00109 + 2454198.949606 & 0.99878 & 0.00109 + + 2454214.574418 & 1.00159 & 0.00157 + 2454214.575348 & 1.00247 & 0.00157 + 2454214.576280 & 0.99766 & 0.00157 + + 2454215.850393 & 0.99951 & 0.00150 + 2454215.851076 & 0.99912 & 0.00150 + 2454215.851771 & 0.99842 & 0.00150 + + 2454535.897759 & 0.99906 & 0.00163 + 2454535.898268 & 0.99996 & 0.00163 + 2454535.898766 & 0.99755 & 0.00163 + + 2454552.866642 & 0.99987 & 0.00135 + 2454552.867140 & 1.00181 & 0.00135 + 2454552.867638 & 1.00086 & 0.00135 + + 2454569.875280 & 1.00116 & 0.00149 + 2454569.875778 & 0.99975 & 0.00149 + 2454569.876287 & 1.00090 & 0.00149 + + 2454594.713269 & 1.00015 & 0.00204 + 2454594.713778 & 0.99628 & 0.00204 + 2454594.714276 & 1.00394 & 0.00204 + + 2454223.741853 & 0.99446 & 0.00160 + 2454223.745823 & 0.99218 & 0.00160 + 2454223.746321 & 0.99209 & 0.00160 + + 2454230.705029 & 0.99892 & 0.00170 + 2454230.705550 & 1.00057 & 0.00170 + 2454230.706059 & 0.99974 & 0.00170 + + 2454230.770620 & 0.99944 & 0.00150 + 2454230.773240 & 0.99791 & 0.00150 + 2454230.774280 & 0.99852 & 0.00150 + + @xmath155 & 0.6379 & 0.1792 + @xmath19 & 0.4378 & 0.2933 + @xmath20 & 0.5535 & 0.2351 + @xmath21 & 0.3643 & 0.3178 + @xmath22 & 0.2777 & 0.3191 + @xmath128 & 0.2179 & 0.3162 + + @xmath128 & 0.1483 & 0.3600 + @xmath155 & 0.5377 & 0.2579 + + 2454185.910430 & 0.000198 & 0 + 2454198.973147 & 0.000223 & 10 + 2454214.646298 & 0.000280 & 22 + 2454215.952080 & 0.000214 & 23 + 2454535.968246 & 0.000166 & 268 + 2454552.948971 & 0.000147 & 281 + 2454569.929089 & 0.000153 & 294 + 2454594.745943 & 0.000253 & 313 + + 2454223.797215 & 0.000847 & 0 + 2454230.904913 & 0.000656 & 7 + 2454230.905624 & 0.001106 & 7 + lc @xmath24 ( k ) & 5650 @xmath156 75 + @xmath15 & 4.4 @xmath156 0.1 + @xmath15 & @xmath157 + @xmath158 ( km s@xmath18 ) & 1.5 @xmath156 1.0 + @xmath25 ( km s@xmath18 ) & 0.85 @xmath156 0.1 + @xmath159fe / h@xmath160 $ ] & @xmath1410.19 @xmath156 0.08 + @xmath161 & @xmath162 + @xmath163 & @xmath164 + @xmath165 ( g @xmath166 ) & @xmath167 + @xmath168 ( @xmath169 ) & @xmath170 + @xmath102 ( @xmath171 ) & @xmath172 + age ( gyr ) & @xmath173 + @xmath174 ( @xmath175 ) & @xmath176 + @xmath177 ( mag ) & @xmath178 + distance ( pc ) & @xmath179 + @xmath139 , @xmath19 , @xmath180 ( km s@xmath18 ) & [ + 27.3 , + 6.7 , + 33.0 ] + lc @xmath24 ( k ) & 6200 @xmath156 75 + @xmath15 & 4.0 @xmath156 0.1 + @xmath15 & @xmath181 + @xmath158 ( km s@xmath18 ) & 8.5 @xmath156 0.5 + @xmath25 ( km s@xmath18 ) & 1.50 @xmath156 0.05 + @xmath159fe / h@xmath160 $ ] & + 0.14 @xmath156 0.09 + @xmath161 & @xmath182 + @xmath163 & @xmath183 + @xmath165 ( g @xmath166 ) & @xmath184 + @xmath168 ( @xmath169 ) & @xmath185 + @xmath102 ( @xmath171 ) & @xmath186 + age ( gyr ) & @xmath187 + @xmath174 ( @xmath175 ) & @xmath188 + @xmath177 ( mag ) & @xmath189 + distance ( pc ) & @xmath190 + @xmath139 , @xmath19 , @xmath180 ( km s@xmath18 ) & [ @xmath14143.9 , @xmath14139.1 , @xmath1416.9 ] + lc @xmath99 ( days ) & 1.30618581 ( fixed ) + @xmath89 ( hjd)&2,454,185.9104 ( fixed ) + @xmath191 & @xmath192 + @xmath193 & @xmath194 + @xmath195 & @xmath196 + @xmath22 ( deg ) & @xmath197 + @xmath198 ( @xmath81 ) & @xmath199 + @xmath200 ( @xmath81 ) & @xmath201 + @xmath202 & 0(fixed ) + @xmath203 ( @xmath114 ) & @xmath204 + rms ( @xmath81 ) & 22.0 + @xmath205 ( @xmath206 ) & @xmath207 + @xmath208 ( @xmath209 ) & @xmath210 + @xmath211 ( @xmath212 ) & @xmath213 + @xmath103 ( au ) & @xmath214 + @xmath215 ( cgs ) & @xmath216 + lc @xmath99 ( days ) & 3.553945 ( fixed ) + @xmath89 ( hjd ) & 2,454,230.9053 ( fixed ) + @xmath191 & @xmath217 + @xmath193 & @xmath218 + @xmath195 & @xmath219 + @xmath22 ( deg ) & @xmath220 + @xmath198 ( @xmath81 ) & @xmath221 + @xmath200 ( @xmath81 ) & @xmath222 + @xmath202 & 0 ( fixed ) + @xmath203 ( @xmath114 ) & @xmath223 + rms ( @xmath81 ) & 11.1 + @xmath205 ( @xmath206 ) & @xmath224 + @xmath208 ( @xmath209 ) & @xmath225 + @xmath211 ( @xmath212 ) & @xmath226 + @xmath103 ( au ) & @xmath227 + @xmath215 ( cgs ) & @xmath228 + . each of the light - curves is labeled with the telescope and filter employed . we have overplotted the simultaneous best - fit solution , adopting the appropriate quadratic limb - darkening parameters for each band pass ( see text for details).[tres3phot ] ] . each of the light - curves is labeled with the telescope and filter employed . as in figure [ tres3phot ] , we have overplotted the simultaneous best - fit model , adopting the appropriate quadratic limb - darkening parameters for each band pass ( see text for details).[tres4phot ] ]

since the seminal work of @xcite , it has been well - known that the motion of particles near a rotating black hole is strictly regular . this remarkable property is guaranteed by the full integrability of the system due to the existence of the fourth constant of motion , and it holds even in the case of an electrically - charged rotating black hole . on the other hand , as soon as the black hole is embedded in an external magnetic field , chaotic dynamics may appear @xcite . while the electric charge of astronomical black holes is negligible , the role of the magnetic field is important ( e.g. , * ? ? ? * ) . in this paper we study the role of large - scale ( organized ) magnetic fields on the properties of motion near a rotating black hole . recent general - relativistic magnetohydrodynamic ( grmhd ) simulations @xcite suggest that accretion flows onto black holes are complemented by outflows emerging from accretion disks and tori , and , in some regions of the flow , produce large - scale ( organized ) magnetic loops in a self - consistent manner . the mechanism leading to the development of ordered bundles of the magnetic lines of force with a significant degree of long - range coherence appears to be highly relevant in the context of formation of jets and outflows from accreting black holes . simulations indicate that the role of black hole rotation is important and this supports the idea of blandford znajek mechanism @xcite as the origin of acceleration and collimation in the vicinity of black holes @xcite . the formulation of the problem of particle acceleration in the context of oblique pulsar magnetospheres has been discussed by @xcite . we explore the effects of organized magnetic fields in a different context of particle motion near a weakly - magnetized black hole ( the magnetic field does not change the spacetime metric ) . we address the problem of regularity versus chaoticity of the resulting motion of electrically charged particles ( electrons and ions ) resulting from the mutual interplay of gravitational , electromagnetic , and gravito - magnetic effects of general relativity . as a matter of principle , we want to understand under what condition chaos emerges and drives the particle motion near magnetized black holes . to this end we assume that the magnetic field arises from currents flowing far out in the accretion disk . unlike most of previous analytical works , we do not impose axial symmetry ( the magnetic field can be inclined with respect to the rotation axis of the black hole ) . the particle approximation allows us to concentrate our attention on purely general - relativistic effects of the curved electro - vacuum spacetime while neglecting collective interactions and shocks in the plasma @xcite . the adopted approximation is thus relevant to situations when the grmhd conditions are not satisfied : in our case , particle mean free path @xmath0 is assumed to be very long ( exceeding the gravitational radius of the black hole , @xmath1 ) while , simultaneously , the gravitational field is very strong ( curvature radius evaluated from the riemann curvature tensor @xmath2 . explicit form of kretschmann scalar for kerr newman spacetime is given by @xcite . characteristic length scale of the spacetime time curvature ( curvature radius ) is then expressed as @xmath3 . ] is smaller than @xmath4 ) ; see @xcite and references therein . this work represents a natural extension of our previous studies @xcite in which we considered dynamic properties of charged matter orbiting in the vicinity of massive objects . our primary concern was to identify and investigate regions of a stable , off - equatorial motion . trajectories of ionized particles confined to such regions constitute a basic _ non - interacting test particle _ approximation of astrophysical coronae formed by diluted plasma above and below the accretion disk of accreting systems . several models have been considered so far . as a first step , the existence of off - equatorial orbits was investigated in the spacetime described by the exact kerr newman solution , i.e. , the case of a rotating electrically charged black hole . considering only the astrophysically - relevant region above the outer horizon of the black hole , it was found that there are no stable circular orbits ( _ halo orbits _ ) outside the equatorial plane @xcite . therefore , we altered our model , namely , we employed test field solutions describing large - scale , ordered magnetic fields ( asymptotically uniform or dipole type ) in which the compact object is immersed . in particular , we investigated a system consisting of schwarzschild black hole with a rotating magnetic dipole field , and the case of the kerr black hole in an asymptotically - uniform magnetic field , which both appeared to host non - equatorial confinements of charged particles @xcite . since both of these systems were found to be non - integrable , in which case the chaotic dynamics emerges , we subsequently focused on the dynamic properties of these orbits , trying to identify which parameters of the system trigger chaos . we found that within the off - equatorial lobes , the dynamics are mostly regular and chaos typically appears only when the energy of the particle is sufficiently increased to the level that corresponds to cross - equatorial confinements @xcite . more recently , a static model of an exact einstein maxwell spacetime was considered in this context @xcite . namely , the dynamics of charged matter in off - equatorial wells above the massive magnetic dipole described by bonnor s exact solution @xcite was investigated . in such a case , the electromagnetic field affects the geometry of the spacetime and as a particular consequence of this influence , we found that the system allows off - equatorial orbits even for neutral test particles . the interplay between the chaotic and ordered motion of test particles in the exactly - given static field of a massive source encircled by a disk or a ring was also recently investigated @xcite . nevertheless , in the following we intend to focus on a more realistic model in which the rotation of the central object is included . on the other hand , we will restrict ourselves to the test field approximation in which the electromagnetic field affects charged particles , but it does not modify the metric . in this contribution we plan to investigate the dynamical properties of charged particles in the generalized model consisting of the rotating kerr black hole in the asymptotically - uniform magnetic field that is inclined with respect to the rotation axis . a corresponding test field solution was given by @xcite . to our knowledge , however , the particle motion in this setup has not yet been inspected . the paper is organized as follows . in section [ formalism ] we present formal description of the model and explicitly specify the electromagnetic four - potential @xmath5 . then we review the hamiltonian formalism which is employed to derive equations of motion . application of the method of the maximal lyapunov exponent as an indicator of chaos is also briefly discussed . analysis of the motion of particles affected by the inclination of the magnetic field is presented in section [ dynamics ] . three distinct classes of orbits are treated separately in corresponding subsections . possible ways for observational verification are discussed in section [ discussion ] . results of the numerical analysis are summarized and discussed in section [ conclusions ] . in appendix [ appa ] we comment on the method of the effective potential and its limited applicability in given context . appendix [ appb ] provides an estimate of radiation power generated by the investigated system . the kerr metric describing the geometry of the spacetime around the rotating black hole may be expressed in boyer lindquist coordinates @xmath6 as follows @xcite : @xmath7 ^ 2+\frac{\sigma}{\delta}\,{\rm d}r^2+\sigma\ , { \rm d}\theta^2,\end{aligned}\ ] ] where @xmath8 we stress that geometrized units are used throughout the paper . values of basic constants ( gravitational constant @xmath9 , speed of light @xmath10 , boltzmann constant @xmath11 , and coulomb constant @xmath12 ) thus equal unity @xmath13 . test field solution corresponding to the aligned magnetic field ( of the asymptotic strength @xmath14 ) was derived by @xcite . this solution was later generalized by @xcite to describe the field which is arbitrarily inclined with respect to the rotation axis ( specified by two independent components @xmath14 and @xmath15 ) . here we also consider a non - zero electric charge @xmath16 of the black hole generating the electromagnetic field of the kerr newman black hole , though in the test field regime ( metric remains unaltered by @xmath16 ) . the resulting vector potential @xmath17 is given as follows : @xmath18\label{empot4}\\ & & \nonumber - b_x\sin\theta\cos\theta\big[\delta\cos\psi\\ & & \nonumber+\frac{(r^2+a^2)m}{\sigma}\left(r\cos\psi - a\sin\psi\right)\big]+\frac{qra\sin^2\theta{}}{\sigma},\end{aligned}\ ] ] where we use the azimuthal coordinate @xmath19 of the kerr ingoing coordinates , which is related to boyer lindquist coordinates as follows : @xmath20 with @xmath21 denoting the outer and the inner horizon , respectively . we notice that @xmath22 . we have previously investigated the structure of the electromagnetic field emerging in a more general case of the kerr black hole immersed in the asymptotically - uniform magnetic field ( 1 ) which is inclined with respect to the rotation axis and ( 2 ) in which the black hole is drifting with the constant velocity and in the arbitrary direction @xcite . we found complex structures of both the magnetic and the electric field lines in this model . in particular , we observed that due to combined effects of the translational motion and the frame dragging ( caused by the black hole s rotation ) , the null points of magnetic field may arise . these may be highly relevant for the acceleration processes of ionized matter in the vicinity of an accreting object . nevertheless , the motion of charged particles exposed to such a field has not been inspected in the aforementioned studies and within this paper we consider the inclination of the magnetic field but not the translational motion of the black hole . employing the hamiltonian formalism , we first construct the super - hamiltonian @xmath23 : @xmath24 where @xmath25 is the charge of the test particle ( of rest mass @xmath26 ) , @xmath27 is the generalized ( canonical ) momentum , @xmath28 is the metric tensor , and @xmath17 denotes the vector potential of the electromagnetic field . the latter is related to the electromagnetic tensor @xmath29 by @xmath30 . hamilton s equations of motion are given as @xmath31 where @xmath32 is the affine parameter ( dimensionless in geometrized units ) , @xmath33 denotes the proper time , and @xmath34 is the standard kinematical four - momentum for which the first equation reads @xmath35 , and thus the conserved value of super - hamiltonian is equal to @xmath36 . moreover , the system is stationary since the hamiltonian is independent on the coordinate time @xmath37 . its conjugate momentum @xmath38 is therefore integral of motion . namely , it expresses ( negatively taken ) the energy of the test particle @xmath39 . the mass of the black hole @xmath40 is used to scale all quantities , which is formally equivalent to setting @xmath41 in the equations . we also switch to specific quantities @xmath42 and @xmath43 when describing the test particle . from the above equations , we conclude that independent parameters of the investigated system are thus @xmath44 , @xmath45 , @xmath46 , and @xmath47 . indeed , the quantities @xmath15 , @xmath14 , and @xmath16 only appear in the product with the particle s charge in the equations of motion ( [ hameq ] ) since the electromagnetic field is treated in the test field approximation . besides that , particular trajectory is further specified by its initial position in the phase space @xmath48 , @xmath49 , @xmath50 , @xmath51 , @xmath52 , and @xmath53 whose values are , however , bound by the normalization condition @xmath54 . therefore , we do not set all the components arbitrarily and choose to compute the value @xmath52 using the normalization of the four - momentum and its remaining components instead . maximal lyapunov exponent @xmath55 is commonly used as a basic quantitative indicator of chaotic dynamics ( e.g. , * ? ? ? its value directly captures the tendency of nearby orbits to diverge as the system evolves . in other words , it allows us to express how sensitive the given orbit is on the initial condition whilst the high ( exponential ) sensitivity is a defining property of chaos . the exponent @xmath55 is defined as follows is defined as an asymptotic measure ( [ mle ] ) . in the numerical application we actually compute the quantity usually denoted as _ finite time lyapunov exponent _ , which depends on the integration variable , instead of the limit . however , in this text we disregard such distinctions and use the term lyapunov exponent for both quantities as it can not cause any confusion within the scope of the paper . ] @xmath56 where we choose to use the standard @xmath57 ( euclidean ) norm to measure the length of the deviation vector in the phase space @xmath58 . in this context , we have also experimented with norms reflecting the curvature of the spacetime , however , it appeared that for a given application the choice of the norm is not crucial @xcite . although lyapunov exponents within the relativistic framework are generally not invariant under coordinate transformations @xcite , signs of the exponents are preserved @xcite . therefore , the distinction between chaotic and regular orbits may be drawn invariantly . the usual method of determining the evolution of the deviation @xmath59 consists in solving variational equations which restricts us to the linear term in a corresponding taylor expansion @xcite . one can therefore express the variational equations in the following matrix form with column vector @xmath60 and its derivative @xmath61 with respect to the affine parameter @xmath62 : @xmath63 where @xmath64 is the hessian matrix composed of second derivatives of the super - hamiltonian @xmath23 with respect to phase space variables and @xmath65 is constant matrix with following block form : @xmath66 , in which @xmath67 and @xmath68 denote @xmath69-dimensional identity and zero matrices , respectively . in our case @xmath70 . detailed derivations of variational equations ( [ vareq ] ) and their properties are given in a recent review on lyapunov exponents by @xcite . matrix @xmath64 is evaluated along the orbit and explicitly depends on the current position in the phase space ; therefore it is necessary to integrate variational equations simultaneously with corresponding equations of motion ( [ hameq ] ) . we set the initial deviation as follows @xmath71 . the theory of lyapunov spectra guarantees that setting random initial deviation results in the computation of the maximal exponent @xmath55 with a probability of one and the set of initial deviations for which we would obtain different lyapunov exponents has zero measure @xcite . the maximal lyapunov exponent @xmath55 corresponds to the most unstable direction in the phase space , however , exponents related to the complementary phase space directions can also be determined to reveal the whole lyapunov spectrum . general properties of lyapunov exponents imply that an autonomous hamiltonian system of three degrees of freedom has four non - zero exponents out of which , however , only two are independent as they always appear in pairs of opposite signs but equal absolute values @xcite . the computation of the whole spectrum of lyapunov exponents can be performed using , e.g. , the standard method developed by @xcite . although knowing the whole spectrum provides additional information about a given orbit , for the purpose of detecting chaotic dynamics it is sufficient enough to only compute the maximal exponent . for chaotic orbits , exponent @xmath55 attains a positive value , while for regular trajectories it tends to zero as @xmath72 . therefore , in the logarithmic plot @xmath73 , the regular trajectory appears as a linearly - decreasing function while the chaotic orbit , sooner or later , leaves the trend of linear decreasing and converges to the positive value . the lyapunov time @xmath74 may thus be used to estimate the time after which the chaoticity of a given orbit manifests . as given by equation ( [ mle ] ) . in practical numerical applications we evaluate @xmath75 of a chaotic orbit for such a value of @xmath62 for which @xmath55 appears to attain its limit , i.e. it does not evolve anymore . ] in particular , our operational criterion for _ full _ or _ saturated _ chaos will be the approximate relation @xmath76 , assuring that the trajectory exhibits chaotic properties on the time scale of its azimuthal period or even faster . however , the convergence of @xmath55 may become very slow , and in some cases extremely long integration times are needed to reveal the chaotic nature of the trajectory . for example , some chaotic orbits get stuck in close vicinity of a regular orbit for a long time before they escape to the larger chaotic domain in the phase space ( so called stickiness effect see e.g. , * ? ? ? * ) . the computation of lyapunov exponents may also be obstructed by numerical errors , especially when the long integration is required , the proper choice of the integrator becomes crucial ( e.g. , * ? ? ? * ) . however , if one proceeds with caution , the method usually gives reliable results . in general , when performing the long - term integration of the hamiltonian system , the method of choice would be a symplectic integrator ( e.g. , * ? ? ? * ) . however , it is not possible to use a symplectic scheme for the simultaneous integration of the variational equations , and therefore we have to apply some non - symplectic method . namely , we employ the matlab integrator ode87 which is an explicit 7th8th order runge kutta solver based on the dorman prince formulas . the precision of this integrator with adaptive step - size is controlled by setting the relative tolerance ( @xmath77 ) , which specifies the highest allowed relative error in a single integration step . we were able to run our integrations with @xmath78 . during each integration we checked for the cumulative error ( causing artificial excitation or damping of the system ) by expressing the relative change of energy of the particle which maximally reached @xmath79 for the longest runs common characteristics of previously investigated models are stationarity and axial symmetry . such a system has two degrees of freedom in the four - dimensional spacetime . considering the oblique magnetic field , however , we break the axisymmetry and the resulting system has three degrees of freedom , which has several important implications regarding the techniques we can apply for the analysis . first of all , we can no longer use the method of effective potential to localize the regions of bound orbits and , in particular , to find the stable , circular , off - equatorial orbits corresponding to the local minima of the potential ( see appendix [ appa ] ) . besides that , the method of poincar surfaces of section , which is commonly used to visualize the trajectories and distinguish between regular and chaotic dynamics in systems of two degrees of freedom , fails to deliver unambiguous results in this case . we will explore the behavior of the maximal lyapunov characteristic exponent @xmath55 as a primary indicator of chaotic dynamics instead . inapplicability of the effective potential method leaves us without the usual tool for a systematical search for stable orbits throughout the configuration space and parameter space . in this place , we take advantage of our former study of axisymmetric versions of given systems @xcite and use the particular , previously - investigated , confinements of particles in the system with aligned magnetic fields as the starting point of our current analysis . we will gradually incline the originally - aligned field and observe its impact upon the dynamics of particles . such an approach has obvious limitations . we can not set arbitrary inclination @xmath80 as we need to maintain the confinement of stable orbits whose presence at a given location non - trivially depends on its value . actually , we shall fix all other parameters and increase the inclination of the field as long as we find the confinement of stable orbits . after reaching the critical inclination angle , the confinement disintegrates allowing the particles to escape to infinity or to fall onto the horizon . we are , however , only interested in the family of bound orbits . in the following we shall separately treat three distinct classes of orbits which we detected in the aligned field considering both charged and neutral black holes . namely , in section [ regoff ] we shall explore the impact of inclination of the field upon the regular orbits in off - equatorial confinement . in section [ chaoscross ] , we consider chaotic orbits in the cross - equatorial lobe and in section [ regeq ] , the class of regular , equatorial trajectories is investigated . the distinction between equatorial and cross - equatorial orbits is based upon the different topology of effective potential defining the confinement . in the case of equatorial orbits , the underlying potential has a local minimum in the equatorial plane , while the cross - equatorial lobe is defined by the pair of symmetric off - equatorial minima separated by the saddle point residing in the equatorial plane . according to the results of our previous analysis , the latter configuration generally appears to host mainly chaotic orbits while the dynamics in the equatorial wells remains predominantly regular as long as the energy is raised only _ slightly _ above the level of the corresponding potential minimum . as we have already mentioned , we choose examples of both charged and neutral black holes . in general , the presence or absence of test charge @xmath16 ( generating kerr newman test field which we superpose to the magnetic field see equations ( [ empot1])-([empot4 ] ) ) does not seem to act as the factor systematically shifting the dynamics to become more or less chaotic . this is not surprising if we recall that the kerr newman field alone does not represent a non - integrable perturbation for charged particles , as it admits the fourth integral of motion carter s constant @xmath81 @xcite . although it evidently affects the dynamics of ionized particles ( e.g. , changes the location and shape of the particle confinements ) it does not act as a trigger for chaos . considering the aligned magnetic field in @xcite , we classified off - equatorial lobes according to the behavior of effective potential in the vicinity of its off - equatorial minima into four distinct classes ia id ( see fig . [ wald_abcd ] ) . closed lobes define regions of possible confinement of particles , whereas the lobes extending down to the horizon allow particles to be accreted onto the black hole ( types ia c ) . on the other hand , regions extending to large distance represent the case when particles can move far out ( e.g. , along the axis in type i d ) . curves of constant energy suggest a different destiny of a particle on the off - equatorial circular orbit when its energy increases above the level of corresponding potential minimum . in type ia the symmetric off - equatorial minima first merge via equatorial saddle point , allowing stable cross - equatorial motion . increasing its energy further , the particle would reach another saddle point in the equatorial plane which allows it to fall freely onto the horizon . the class ib does not permit any cross - equatorial motion off - equatorial confinements connect to the horizon directly when the energy increases sufficiently . topology of type ic allows merging of off - equatorial lobes , however , at a merging point , these are already opened toward the horizon and the motion in the cross - equatorial confinement is thus never stable . type i d allows stable cross - equatorial motion . as the energy increases sufficiently the isopotential curve opens and particles may escape to infinity ( unlike the case of ia when they felt onto horizon instead ) . only two cases ( namely types ia and i d ) exhibit the transition from the bound off - equatorial motion to the bound cross - equatorial motion when the energy is increased . remaining types ib and ic do not allow closed cross - equatorial structures . in the following we analyze the impact of the inclination of the magnetic field on the originally regular off - equatorial orbit of type i d . in particular , to distinctly illustrate its influence we depart from the very same configuration as we used in @xcite . namely , we explored the orbit with parameters @xmath82 , @xmath83 , @xmath84 , @xmath85 and with the initial condition : @xmath86 , @xmath87 , @xmath88 and @xmath89 . for @xmath90 we observed regular off - equatorial motion while with @xmath91 the motion was cross - equatorial and chaotic . first we analyze the former case while the latter is inspected in section [ chaoscross ] . gradually increasing the inclination @xmath80 and checking if the motion remains allowed and the orbit is still bound , we conclude that up to @xmath92 the allowed region remains closed , for values up to @xmath93 , it still exists as an open region allowing the particle to escape while even higher values forbid the motion for given parameters . in the left panel of fig . [ off_reg_3d ] we illustrate two examples of analyzed trajectories in the three - dimensional view . most importantly , it shows that the inclination of the field extends the allowed region and converts the off - equatorial orbit into cross - equatorial in this case . moreover , it suggests that the originally regular trajectory turns chaotic , which we confirm by means of the maximal lyapunov exponent @xmath55 in the right panel of fig . [ off_reg_3d ] . for very small inclinations up to @xmath94 , the trajectory remains regular and @xmath55 tends to zero limit ( @xmath95 decreases linearly as a function of @xmath96 ) . in particular , for the case of @xmath97 , we have checked this trend by integrating up to @xmath98 . however , setting @xmath99 the exponent @xmath55 attains a positive value which marks the onset of chaos . the asymptotic value of @xmath55 of the given orbit further rises as the inclination increases . for @xmath100 , the trajectory becomes cross - equatorial and @xmath55 almost saturates as it does not significantly rise when the field is further inclined . the trajectory is now _ fully chaotic _ as its lyapunov time roughly equals the orbital period @xmath101 ( numerically averaged along the actual orbit see appendix [ appb ] ) . values @xmath102 correspond to the opened allowed region and the particle typically escapes before @xmath103 . the analyzed example reveals the typical behavior of the originally regular off - equatorial orbits perturbed by inclining the magnetic field . most importantly , it shows that the inclination of the field induces continuous transition to chaos as measured by means of the maximal lyapunov exponent @xmath55 . here we inspect the impact of the inclination on the cross - equatorial orbit which is already chaotic in the aligned field . we choose to analyze the example analogical to the previous case of the off - equatorial orbit . namely , we start with the trajectory which has a higher energy @xmath91 while all other parameters and initial conditions remain fixed at values specified above . in the left panel of fig . [ ekv_chaos_3d ] , we compare the three - dimensional views of two typical chaotic orbits : the original aligned trajectory ( red ) and an inclined trajectory with @xmath104 ( blue ) . higher inclinations are not permitted for given setup . inclined field lines are traced by the trajectory which clearly shows the direction of the field . however , besides the inclination of the trajectory there is no apparent qualitative difference between compared orbits in fig . [ ekv_chaos_3d ] . quantitative comparison in terms of the lyapunov exponent @xmath55 confirms that , in this case , the inclination of the magnetic field does not significantly change the dynamics ( right panel of fig . [ ekv_chaos_3d ] ) . we observe that unlike the previous case of off - equatorial regular orbits , here the asymptotic values of @xmath55 are not stratified with respect to the inclination angle . a higher inclination angle does not necessarily correspond to a higher maximal lyapunov exponent @xmath55 . changing the inclination between @xmath105 and @xmath106 results in comparable asymptotic values of @xmath55 . we conclude that once the system is fully chaotic , the additional perturbation consisting in the inclination of the field does not increase the chaoticity of the system . we demonstrate the influence of an inclined magnetic field on the originally regular orbit in the equatorial potential well . here we choose the orbit around the black hole without an additional electric charge as a representative example . parameters of the trajectory are @xmath83 , @xmath107 , @xmath108 , @xmath85 , and the initial condition is set as follows : @xmath109 , @xmath110 , @xmath88 , @xmath89 , and @xmath111 . in the given setup the trajectory remains bound for inclinations up to @xmath112 and the motion is allowed for @xmath113 . the three - dimensional view of the aligned trajectory ( red ) and two orbits in the oblique field with different inclinations ( green : @xmath114 , blue : @xmath115 ) are compared in the left panel of fig . [ ekv_reg_3d ] . for the aligned field , this orbit exhibits regular dynamics and the corresponding trajectory draws an axisymmetric pattern . the oblique field gradually _ inclines _ the trajectory as a whole and introduces apparent signs of disorder whose symptoms increase with the increasing inclination . the comparison of lyapunov exponents @xmath55 corresponding to orbits with different inclinations is given in the right panel of fig . [ ekv_reg_3d ] . in this case , only the aligned trajectory remains truly regular with @xmath116 falling linearly to zero as a function of @xmath96 . even for a very small inclination of @xmath117 , the trajectory eventually leaves the trend of linear decrease of @xmath116 and approaches a positive value in the limit , which is a hallmark of the chaotic dynamics . with the growing inclination , the limiting value of @xmath55 gradually grows . more inclined field triggers _ stronger chaos _ as measured by the exponent @xmath55 . we notice that in this case , the convergence of @xmath55 is remarkably smooth for all allowed inclinations which we observed in the case of off - equatorial orbits ( fig . [ off_reg_3d ] ) only for smaller inclinations which were inducing non - saturated chaos . the given example of the originally regular equatorial orbit affected by inclining the magnetic field , illustrated high sensitivity of regular dynamics upon the axial symmetry of the system . in this case , breaking the symmetry inevitably brings chaotic features to the system regardless the value of the inclination . in fig . [ lya_vs_uhel ] we compare asymptotic values of @xmath55 as a function of the inclination of the field for all three types of orbits analyzed above and illustrated separately in figs . [ off_reg_3d][ekv_reg_3d ] . the originally off - equatorial , regular orbit ( red curve ) shows a fast onset of chaos as the inclination increases and cross - equatorial motion becomes possible . however , the growth of @xmath55 quickly saturates as the system finishes the transition to the chaotic regime ( at @xmath118 ) and , beyond this point , the value of @xmath55 does not grow considerably with the increasing inclination of the field . departing from the cross - equatorial orbit which is already fully chaotic in the aligned field @xmath105 , we observe no systematic reaction of @xmath55 when the field gradually inclines ( blue curve ) . the system in fully chaotic mode does not respond to the additional perturbation by increasing its chaoticity . the originally regular equatorial orbit ( green plot ) undergoes a gradual growth of chaoticity as the field inclines . nevertheless , the value of @xmath55 does not saturate within the range of allowed inclinations ( @xmath113 ) . ( in the logarithmic scale ) as a function of the inclination of the field @xmath80 ( tangent of the inclination angle ) for three representative cases which were discussed in figs . [ off_reg_3d][ekv_reg_3d ] . the red curve shows the behavior of the exponent @xmath55 of the originally regular off - equatorial orbit ( parameters specified in fig . [ off_reg_3d ] ) . the response of the cross - equatorial chaotic orbit is plotted by the blue curve ( orbit s parameters in fig . [ ekv_chaos_3d ] ) , and the originally regular equatorial orbit from fig . [ ekv_reg_3d ] is represented by the green line . we note that the inclinations we considered in the analysis generally correspond to small inclination angles of the field given as @xmath119 , e.g. , the inclination of @xmath115 corresponds to the angle of @xmath120 . ] the scenario investigated within this paper is idealized and its direct applicability for interpretation of observed astrophysical phenomena is limited . however , the significance of our model lies in its ability to demonstrate the effect of combined magnetic and strong gravitational fields onto the dynamics of charged matter in a very clear form since the matter is treated as collisionless and thus its motion is governed solely by external fields . under such circumstances we raised the fundamental question whether a given combination of fields typically leads to chaotic or regular dynamics of orbiting particles . in the context of this analysis another natural question arises whether it is possible to observationally distinguish imprints of ordered or chaotic dynamics occurring in the investigated system . in general , the link between chaos on a microscopic scale and the macroscopic behavior of matter has not yet been fully understood and it represents an open problem of statistical mechanics . while , in some systems , the relation of transport coefficients ( e.g. , viscosity , thermal , or electric conductivity , etc . ) to lyapunov exponents is observed , there are also non - chaotic systems that show transport phenomena and thus suggest that chaos might not be necessary for the robust statistical behaviors ( see e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) . nevertheless , we are in a slightly different situation as we do not build a statistical ensemble of particles to derive its macroscopic properties in the framework of kinetic theory . as we rather consider a set of non - interacting particles , can we expect them to give characteristic imprints of their dynamic regime on the generated radiation ? we believe that the answer is , at least in principle , positive . frequency analysis of bound trajectories shows a clear difference between order and chaos @xcite . while regular trajectories have discrete power spectra with prominent lines at their fundamental frequencies , the chaotic ones instead show continuous spectra . in the case of charged particles which radiate when accelerated , such distinction in the spectra of dynamic frequencies inevitably translates into a frequency composition of generated electromagnetic signals . determining the actual form of resulting electromagnetic spectra would , however , require a detailed analysis , which is beyond the scope of this paper . we also note that the general problem of detecting chaos and quasiperiodicity in experimental data with noise has been studied thoroughly ( see e.g. , * ? ? ? * ; * ? ? ? * ) and a number of relevant methods have been developed over the last decades . not listing all standard techniques , we only mention a more recent method of recurrence analysis @xcite , which appears very useful for this task . the crucial question remains : whether the electromagnetic signal generated by our system is strong enough to deliver characteristic imprints of order and chaos to the observer . in appendix [ appb ] we thus calculate the power @xmath121 of synchrotron radiation generated by charged particles orbiting in the confinements discussed above . the total power of this radiation as given by equation ( [ powertot ] ) is negligible . the obvious way to substantially enhance the power output is to accelerate the particles to ultrarelativistic velocities since @xmath121 scales with lorentz factor @xmath122 as @xmath123 . in our model , however , the velocity ( [ linspeed ] ) is not a free parameter and it seems that ultrarelativistic velocities are beyond its scope . estimating the bremsstrahlung resulting from coulomb collisions ( free free radiation ) by equation ( [ brem ] ) , we find this radiative process to be substantially more efficient in given circumstances . however , unlike synchrotron radiation , we do not expect the bremsstrahlung to encode imprints of dynamic regime of colliding particles in its spectrum . we have shown that a charged matter in densities consistent with our collisionless description does not emit a significant amount of synchrotron radiation in which we could distinguish imprints of regular or chaotic dynamic regime of emitting particles . the mechanism of inverse compton scattering of thermal photons emitted from the accretion disk on electrons trapped in the confinements discussed within this paper may be more effective in delivering observational imprints in detectable signals ( ongoing work ) . the main motivation of our current analysis was to study the chaoticity in the dynamics of charged matter in accreting black hole systems as a fundamental theoretical aspect of its physics . regularity is a remarkable property of motion around black holes . we addressed the question of transition from regular to chaotic dynamics by including the effect of magnetic field , and we studied the importance of misalignment between the direction of the magnetic field and the spin of the black hole . in this paper we investigated some aspects of dynamics in the test particle model of the black hole magnetosphere . in particular , we considered a system consisting of a rotating black hole embedded in the large - scale ordered magnetic field . in our previous studies @xcite we found that , in the special case of an aligned field ( i.e. , the magnetic field was symmetric with respect to the rotation axis ) , the motion of matter in off - equatorial confinements was typically regular . the chaotic dynamics usually appeared for cross - equatorial orbits . considering the equatorial potential wells instead , we observed the dominance of regular dynamics for energies in a certain limited range , followed by a continuous transition to the mostly - chaotic regime when the energy was gradually increased above this range . here we generalized our model by considering oblique magnetic fields . in particular , we investigated the impact of the inclination upon the dynamics of charged matter orbiting around a black hole in three distinct types of confinements , namely in off - equatorial , cross - equatorial , and equatorial lobes . using the maximal lyapunov exponent @xmath55 as a basic indicator of chaotic dynamics , we found that regular orbits are , in general , highly sensitive to the perfect alignment of the field . the inclination of @xmath124 ( @xmath125 ) was typically sufficient to perturb the regular dynamics and introduce prominent chaotic features which we illustrated in several representative examples . in particular , we found that in the case of originally regular , off - equatorial , motion the system remains regular only for very small inclinations ( up to @xmath126 in analyzed example ) and with increasing inclination it quickly shifts to the chaotic regime . considering the class of regular equatorial orbits instead , we observed that these did not oppose the perturbation at all as they became chaotic for inclinations as low as @xmath127 . also in this case the system undergoes gradual transition to full chaos as the inclination increases . on the other hand , if the original trajectory is already fully chaotic in the aligned configuration , the further perturbation to the system by inclining the field does not significantly affect the dynamics . the trajectory as a whole inclines along the field ( see fig . [ ekv_chaos_3d ] ) , however , the chaoticity as measured by the lyapunov exponent @xmath55 does not increase further . we conclude that within the given model the stability of the regular motion of charged particles depends critically on the the perfect alignment of the large - scale magnetic field with the rotation axis of the black hole . once the field is slightly inclined and the symmetry of the system breaks , the chaotic regime dominates the dynamics . o.k . is grateful to obtain support from the czech science foundation ( ga r ref . 202/09/0772 ) and acknowledges the postdoctoral program of the czech academy of sciences . v.k . acknowledges the czech science foundation ( 13 - 000703 ) . in the classical mechanics , the notion of the effective potential is usually introduced when dealing with the central force problem which concerns the motion of the particle of negligible mass ( _ test particle _ ) which is attracted or repelled by a static massive center . the magnitude of this force depends solely on the distance from the center . a prominent instance of such a setup in which the central force is represented by the gravitational attraction is a kepler problem ( and a related gravitational two - body problem which can be reformulated as two one - body problems , one trivial and the other of the kepler type ) . effective potential @xmath128 is then given as a sum of the gravitational potential and the centrifugal term related to the angular momentum @xmath129 of the orbiting body . analysis of the effective potential provides valuable overall information about the dynamics without needing the actual integration of particular trajectories . its significance arises from the defining relation @xmath130 where @xmath131 is the classical total energy of the particle and @xmath132 represents its kinetic part which comprises only the radial term @xmath133 in this case since @xmath134 is expressed in terms of @xmath129 . thus , the effective potential ( as a function of @xmath135 and @xmath129 ) expresses the energy of the particle at which the radial turning points occur ( apocenter and pericenter in the case of bound orbits ) . it defines the boundary ( in the extended configuration space ) of the regions of the allowed motion . in particular , it allows us to locate stable , circular orbits which correspond to the minima of the effective potential . in general relativity , however , we have no clear distinction between the kinetic and potential energy . nevertheless , in many cases we may still derive function analogous to the classical effective potential which allows us to explore the dynamics of test particles and , in particular , to locate regions of stable orbits . in the general case of the charged test particle of the rest mass @xmath26 and charge @xmath25 in the spacetime with metric @xmath28 and electromagnetic field @xmath17 we depart from the super - hamiltonian @xmath23 whose conserved value is given by the normalization of the four - momentum : @xmath136 in the special case of the stationary and axisymmetric background of a kerr or kerr newman black hole with an additional , stationary , electromagnetic test field obeying the same symmetry in which @xmath137 and @xmath138 are constants of motion ( system therefore has two degrees of freedom ) and neither @xmath139 nor @xmath17 depend on boyer lindquist coordinates @xmath37 and @xmath140 , we obtain by straightforward manipulations @xmath141 where @xmath142\\ { \nonumber}\gamma&=&-g^{\varphi\varphi}(l - qa_{\varphi})^2-g^{tt}q^2a_{t}^2 + 2g^{t\varphi{}}qa_{t}(l - qa_{\varphi})-m^2.\end{aligned}\ ] ] since both coefficients @xmath143 and @xmath144 are positive above the outer horizon @xmath145 to which region we restrict our study , the zero point of the left hand side of ( [ eff ] ) occurs at the simultaneous turning point of motion in both the radial and latitudinal direction , and defines the boundary of allowed motion . a function which specifies the value of energy corresponding to the turning point can be regarded as a generalization of the classical effective potential @xmath128 . we can therefore express the two - dimensional effective potential @xmath146 as follows : @xmath147 where the positive square root has to be chosen to correspond with the future - pointing four - momentum @xcite . since @xmath148 above the horizon the motion is allowed just if @xmath149 . a method of effective potential has been applied to locate confinements ( both equatorial and off - equatorial ) of charged matter in several stationary and axisymmetric models in our previous works @xcite . potential @xmath128 was investigated as a function of two configuration variables @xmath135 and @xmath150 , angular momentum @xmath129 , and particular parameters of the given system . in the context of present paper , our question is whether we could also apply this method for the system of three degrees of freedom in which the axial symmetry is broken and @xmath151 but @xmath152 . in this case , the trajectory manifold spans five out of eight dimensions of the phase space due to the stationarity and autonomy of the system . effective potential reduces the number of dimensions by imposing the constraint of type @xmath153 which locates the turning point in @xmath154-direction . in this case , we seek the simultaneous turning point in all three directions @xmath135 , @xmath150 , and @xmath140 which would result in two - dimensional submanifold . for a fixed value of @xmath140 we should therefore obtain one - dimensional isopotential curves specifying the allowed region in a given meridional plane described by coordinates @xmath155 as we previously did in the case of axisymmetric systems . indeed , we can derive the expression formally analogous to ( [ eff ] ) : @xmath156 where the coefficients are now given as @xmath157\\ { \nonumber}\gamma^{\star}&=&-g^{\varphi\varphi}g^{t\varphi}g_{t\varphi}(\pi_{\varphi}-qa_{\varphi})^2-g^{tt}q^2a_{t}^2(1+g^{t\varphi}g_{t\varphi})+2(g^{t\varphi{}})^2g_{t\varphi}qa_{t}(\pi_{\varphi}-qa_{\varphi})-m^2.\end{aligned}\ ] ] the left - hand side of ( [ eff2 ] ) has the proper form necessary for expressing the effective potential ( @xmath158 is positive ) . nevertheless , the coefficients @xmath159 and @xmath160 depend on the azimuthal component of canonical momentum @xmath161 which used to be the integral of motion @xmath129 in the axisymmetric system , however , here it changes along the trajectory . the evolution of @xmath162 is not known a priori and one has to integrate the equations of motion of a given particle to reveal it . therefore , it is not possible to express the effective potential from the above equation as a function of @xmath135 , @xmath150 , and @xmath140 coordinates ( and parameters of the metric and electromagnetic field ) . we conclude that the given technique leads to the derivation of the effective potential only in the case of aligned magnetic fields . for oblique fields , the method fails to provide the potential since the simultaneous turning points and the boundaries of allowed regions are actually not captured by resulting formula ( [ eff2 ] ) . it has been proposed that a distinction between chaos and regular motion in a system of radiating particles can be revealed in the power spectra of the resulting signal . regular trajectories will contribute to localized features in the power - density spectrum ( pds ) , which disappear when chaos prevails . to assess the detectability , we first need to estimate the strength of the outgoing signal from the system . here we estimate the power of synchrotron radiation generated by charged particles orbiting in confinements discussed in this paper . to this end we make approximations whose appropriateness will be discussed . first we pick one individual trajectory ( _ typical orbit _ ) whose properties will characterize the whole ensemble of particles in a given confinement . we choose an equatorial trajectory with parameters @xmath83 , @xmath107 , @xmath108 , @xmath85 , @xmath163 , and initial condition @xmath109 , @xmath110 , @xmath88 , @xmath89 and @xmath111 ( fig . [ ekv_reg_3d ] ) . for this trajectory , we determine linear velocity @xmath164 with respect to the locally non - rotating frame @xcite with the tetrad basis @xmath165 as follows @xmath166 calculating @xmath164 over a sufficiently - long integration period ( @xmath167 revolutions around the center ) we find that the mean of the azimuthal component , which dominates the motion , reads @xmath168 . since we only consider small inclinations of the magnetic field , the azimuthal component @xmath169 approximately equals the component perpendicular to the field @xmath170 also in the case of oblique fields . the corresponding lorentz factor reads @xmath171 which justifies the cyclotron regime . as a next step , we specify the charge of the particles under consideration as well as the strength of the magnetic field , @xmath172 where the quantities without subscript @xmath173 are dimensionless ( expressed in geometrized units and scaled by mass of the black hole @xmath40 ) and the length @xmath174 reflects the solar mass in geometrized units , @xmath175 . inserting the value @xmath85 and fixing the specific charge @xmath176 to electron , i.e. @xmath177 , we find that for the black hole of @xmath178 , the corresponding magnetic field reads @xmath179 . such a value is consistent with non - thermal filaments in the galactic center @xcite . the larmor radius of the particles gyration @xmath180 is then given as follows @xmath181 which reads @xmath182 for electron with @xmath183 and @xmath178 . however , we stress that , in our case , it is not the actual radius of the orbit which is determined not only by the magnetic field , but also by the gravitational pull of the black hole . indeed , the radius of our typical orbit roughly equals @xmath109 which corresponds to @xmath184 . the azimuthal period of a typical orbit ( computed as a mean of @xmath167 revolutions ) reads @xmath185 while the latitudinal period is @xmath186 . also checking for trajectories with nonzero magnetic inclinations within the considered range of angles , we find that these values vary up to @xmath187 . we need to determine the limiting number density of particles @xmath69 , which is consistent with the collisionless description . the general formula relating density @xmath69 to the mean free path @xmath0 , the cross section of relevant interaction @xmath188 , and the corresponding impact parameter @xmath189 reads @xmath190 where all quantities are in physical units ( we omit the subscript @xmath191 for the rest of the discussion ) . as already commented on in section [ intro ] , we demand the particle mean free path to ( at least ) equal the characteristic length scale of the system . therefore , we set @xmath192 for the black hole of ten solar masses . the interaction between particles is described as a classical coulomb collision . the analysis of elastic coulomb scattering of two equal particles of mass @xmath26 and charge @xmath193 reveals ( e.g. , * ? ? ? * ) that the deflection angle @xmath55 of the incident particle with impact parameter @xmath189 and original velocity @xmath194 ( measured in the laboratory frame in which the target particle is initially at rest ) reads @xmath195 where @xmath196 denotes the permittivity of the vacuum . the relative loss of kinetic energy of the incident particle ( which is transferred to the target particle ) @xmath197 equals @xmath198 . the number density @xmath69 may thus be expressed as follows @xmath199 ^ 2.\ ] ] setting @xmath200 , electron mass @xmath201 , electron charge @xmath202 and @xmath203 , we obtain the density @xmath69 as a function of @xmath204 . if we allow the particle to lose maximally @xmath205 percent of its kinetic energy in a single collision and set @xmath206 ( maximal deflection angle @xmath207 ) , the resulting number density of electrons is @xmath208 . the total power of radiation @xmath121 emitted by non - relativistic point charge @xmath193 with acceleration @xmath44 is given by larmor formula ( e.g. , * ? ? ? * ) @xmath209 the special case of radial acceleration ( perpendicular to the velocity ) is usually denoted as cyclotron radiation . this is often produced by charges accelerated solely by the lorentz force in the magnetic field , and in such a case one sets @xmath210 in a larmor formula . in our case , however , the particles are also accelerated gravitationally , and we obtain the actual acceleration @xmath44 by approximating our typical orbit by a trajectory of uniform circular motion and setting @xmath211 . to obtain the radiation power of whole ensemble of @xmath212 electrons in the confinement we employ the dipole approximation @xcite which allows us to ignore differences in the retarded times of each particle as long as the typical size of the system @xmath213 , and the typical time scale of changes within the system , @xmath214 , fulfills @xmath215 . the overall radiation power is then given as @xmath216 . the volume of the confinement @xmath217 is estimated as the interior of the torus of inner radius @xmath218 and outer radius @xmath219 which reads @xmath220 . a counter - intuitive dependence on the specific charge of particles @xmath25 is due to the fact that we constrain the number density by only considering coulomb collisions which become less effective with decreased specific charges of colliding particles . the validity of the above formula is thus limited to the particles with @xmath222 . the case of particles with lower @xmath25 would demand further discussion . we also note that although the volume of the confinement increases as @xmath223 , the radiation power of a single particle decreases as @xmath224 and the number density as @xmath225 . thus , the total power @xmath226 is small , and it does not scale with the mass of the black hole @xmath40 . besides the radiation related to acceleration due to external fields , we may also estimate the power of bremsstrahlung ( free free radiation ) of colliding particles @xcite @xmath227 where we substituted the value of impact parameter @xmath189 which corresponds to the highest - allowed relative loss of kinetic energy @xmath206 . a substantially higher value compared to ( [ powertot ] ) looks more promising , nevertheless , in this case we do not expect the resulting radiation to encode dynamical frequencies of the orbit in pds , and the chance of distinguishing dynamic regime of colliding particles is thus minimal . 99 bardeen , j. m. , press , w. h. , & teukolsky , s. a. 1972 , , 178 , 347 - 370 benettin , g. , galgani , l. , giorgilli a. , & strelcyn , j. m. 1980 , meccanica , march , pp . 21 - 30 bik , j. , & jani , v. 1985 , , 212 , 899 - 915 blandford , r. d. , & znajek , r. l. 1977 , , 179 , 433 - 456 bonnor , w. b. 1966 , z. phys . , 190 , 444 camenzind , m. 2007 , compact objects in astrophysics : white dwarfs , neutron stars and black holes ( springer ) carter , b. 1968 , phys . , 174 , 1559 castiglione , p. , falcioni , m. , lesne , a. , & vulpiani , a. 2008 , chaos and coarse graining in statistical mechanics ( cambridge university press ) cencini , m. , cecconi , f. , & vulpiani , a. 2010 , chaos : from simple models to complex systems ( world scientific ) cencini , m. , cecconi , f. , falcioni , m. , & vulpiani , a. 2008 , lect . notes phys . , 728 , 123 - 149 contopoulos , g. , & harsoula , m. 2010 , celest . 107 , 77 contopoulos , g. 2002 , order and chaos in dynamical astronomy ( springer ) cremaschini , c. , kov , j. , slan , p. , stuchlk , z. , & karas , v. 2013 , , 209 , 15 dettmann , c. p. , frankel , n. e. , & cornish , n. j. 1995 , fractals , 3 , 161 - 181 ferrire , k. 2010 , astron . nachr . , 331 , 27 - 33 henry , r. c. 2000 , , 535 , 350 jackson , j. d. 1999 , classical electrodynamics ( john wiley & sons ) kalapotharakos , c. , kazanas , d. , harding , a. , & contopoulos , i. 2012 , , 749 , i d . 2 kaltchev , d. , & dragt , a. j. 2013 , physica d , 242 , 1 - 15 karas , v. , kopek , o. , & kunneriath , d. 2012 , classical quant . grav . , 29 , 035010 karas , v. , & kopek , o. 2009 , classical quant . grav . , 26 , 025004 karas , v. , & vokrouhlick , d. 1992 , gen . relat . gravit . , 24 , 729 kopek , o. , karas , v. , kov , j. , & stuchlk , z. 2010 , , 722 , 1240 kopek , o. , kov , j. , karas , v. , & stuchlk , z. 2010 , in proc . mathematics and astronomy : a joint long journey , eds . m. de len , d. m. de diego & r. m. ros , ( springer ) , pp . 278 - 287 kov , j. , kopek , o. , karas , v. , & kojima , y. 2013 , classical quant . 30 , 025010 kov , j. , kopek , o. , karas , v. , & stuchlk , z. 2010 , classical quant . , 27 , 135006 kov , j. , stuchlk , z. , & karas , v. 2008 , classical quant . grav . , 25 , 095011 larosa , t. n. , nord , m. e. , lazio , t. j. w. , & kassim , n. e. 2004 , , 607 , 302 - 308 li j. , spitkovsky a. , & tchekhovskoy a. 2012 , , 746 , i d . 60 lichtenberg , a. j. , & lieberman , m. a. 1992 , regular and chaotic dynamics ( new york : springer ) lukes - gerakopoulos , g. , apostolatos , t. a. , & contopoulos , g. 2010 , , 81 , 124005 marwan , n. , carmen romano , m. , thiel , m. , & kurths , j. 2007 , phys . , 438 , 237 misner , c. w. , thorne , k. s. , & wheeler , j. a. 1973 gravitation ( san francisco : freeman ) motter , e. m. , & saa , a. 2009 , phys . lett . , 102 , 184101 motter , e. m. 2003 , phys . lett . , 91 , 231101 nayfeh , a. h. , & balachandran , b. 1995 , applied nonlinear dynamics : analytical , computational , and experimental methods ( new york : wiley ) penna , r. f. , mckinney , j. c. , narayan , r. , tchekhovskoy , a. , shafee , r. , & mcclintock , j. e. 2010 , , 408 , 752 rybicki , g. b. , & lightman , a. p. 2004 , radiative processes in astrophysics ( weinheim : wiley - vch ) sadowski , a. , narayan , r. , penna , r. , & zhu , y. 2013 , mnras , 436 , 3856 semerk , o. , & sukov , p. 2010 , , 404 , 545 semerk , o. , & sukov , p. 2012 , , 425 , 2455 seyrich , j. , & lukes - gerakopoulos , g. 2012 , , 86 , 124013 sukov , p. , & semerk , o. 2013 , , 436 , 978 skokos , ch . 2010 , in dynamics of small solar system bodies and exoplanets , ed . j. souchay and r. dvorak , lect . notes phys . 790 ( springer : berlin ) , pp . 63 - 135 sprott , j. c. 2003 , chaos and time - series analysis ( new york : oxford university press ) wald , r. m. 1974 , , 10 , 1680 yoshida , h. 1993 , celest . 56 , 27 - 43

penrose @xcite coined the term `` counterfactuals '' for describing quantum interaction - free measurements ( ifm ) @xcite . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ counterfactuals are things that might have happened , although they did not in fact happen . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ he argued that in the ifm , an object is found because it might have absorbed a photon , although actually it did not . the idea of the ifm has been applied to `` counterfactual computation '' @xcite , a setup in which one particular outcome of a computation becomes known in spite of the fact that the computer did not run the algorithm . noh @xcite created counterfactual cryptography , a method for secret key distribution using events in which the particle was not present in the transmission channel . noh used a random choice of orthogonal input states ( like in @xcite ) in contrast with the non - orthogonal states of bb84 cryptographic protocol @xcite . it was argued @xcite , that a modification of the counterfactual computation proposed above , which includes quantum zeno effect , can achieve counterfactuality for _ all _ outcomes of the computation . recently , this idea has been used for performing `` counterfactual communication '' @xcite , which supposedly allowed to send information from bob to alice without transferring any particle between them . the transmission happens in a counterfactual way : the mere possibility of transmitting the particle allows transmitting the value of the bit . i find all these results very paradoxical since they contradict physical intuition of causality : information is usually transmitted continuously in space . i argued @xcite that to resolve the paradoxical feature of the ifm one has to adopt the many - worlds interpretation ( mwi ) of quantum mechanics @xcite , in which the particle touches the object in a parallel world restoring causality at least within the complete physical universe which includes all the worlds . however , i believe that a protocol which can transmit both values of a bit without any particle present in the transmission channel is impossible , irrespectively of the interpretation of quantum mechanics one adopts . i have expressed this opinion already @xcite , but more protocols were suggested @xcite and the controversy remains open @xcite . the clarification of these conceptual issues is particularly important due to the recent increasing interest in the applications of counterfactual protocols @xcite . here i discuss this issue in more detail and try to resolve the controversy . the plan of the paper is as follows . in section ii i introduce the general setup of quantum communication protocols . in sections iii and iv i describe two recent protocols which are claimed to be counterfactual . in section v i analyze various possible definitions of counterfactuality and define my preferred criterion which is based on the magnitude of the trace left in the transmission channel . in section vi i calculate the trace left by a single particle present in the channel , i.e. the trace of a non - counterfactual communication protocol . in sections vii - ix i show that the trace in the protocols claimed to be `` counterfactual '' is not less than the trace in a non - counterfactual protocol . in section x i analyze the security of `` counterfactual '' protocols against an eavesdropper . section xi is devoted to counterfactuality in the framework of the bohmian interpretation . i summarize the results in section xii . there is a surprisingly low bound on the number of bits which can be sent using 1 qubit : the holevo bound of 1 , when the qubit is not entangled @xcite , and 2 , when entanglement is allowed @xcite . this is when we transmit one particle with an internal structure of a qubit such as a polarization state of a photon . in this paper i analyze protocols in which the particle does not have an internal structure : the information is encoded in the presence or absence of the particle . let alice and bob be on the two separate sides of a region , see fig . 1 . bob has a mirror on his side which causes particles sent by alice to be bounced back to her . for a bit value of 1 , bob places a shutter which absorbs alice s particles , while for 0 , the shutter is absent . quantum mechanics , via ifm , allows , at least sometimes , to transmit the bit 1 in a counterfactual manner , i.e. , without having any particle in the transmission channel , see fig . 2 . alice arranges a mach - zehnder interferometer ( mzi ) tuned to destructive interference in one of the ports , say @xmath0 , such that one arm of the interferometer crosses the place where bob s shutter might be . detection of the particle in the dark port of the interferometer tells her with certainty that the bit is equal to 1 ( the shutter is present ) . never clicks if the paths are free . alice knows that bit 1 has been chosen ( bob blocked the path ) when she observes the click in @xmath0 . , width=272 ] the simplest argument that in this case the particle was not present in the transmission channel is : `` if the particle were in the transmission channel , it could not be detected by alice '' . in my view , this argument can not be used for claims about quantum particles @xcite . i , instead , suggest to rely on the fact that the particle does not leave any trace in the transmission channel . note that counterfactual transmission of just one bit value can be achieved using a classical particle @xcite . alice and bob agree in advance that at a particular time , for a particular value of a bit , bob sends the particle to alice , while for the other bit value , he sends nothing . note , however , that this classical protocol can not achieve the task of the quantum ifm . in the ifm , alice learns about the shutter in bob s place without prior agreement with bob and without bob knowing that she acquired this information . in the ifm shown in fig . 2 , alice does not obtain a definite information about the classical bit 0 . without the shutter , the click in the bright port happens with certainty , but it might happen ( with probability 25% ) with the shutter too . this protocol is also not the most efficient method for communication of the bit 1 . when the particle is detected by a bright port ( probability 25% ) we get no decisive information , and in half of the cases the particle is lost ( then we get the information that the bit is 1 but not in a counterfactual manner ) . the quantum method can be modified to be a reliable transmission of both values of the bit . to this end , instead of the shutter , bob inserts a half - wave plate ( hwp ) , see fig . 3 . this transforms the dark port to bright port and vice versa . however , half of the wave always passes through the communication channel , so one can not argue that this is a counterfactual communication . . bob communicates the bit to alice by changing the destructive interference to detector @xmath1 by inserting the hwp in the right path of the particle . , width=309 ] . b ) . bob blocks the interferometers and then , due to zeno effect , the particle reaches detector @xmath0 with probability close to 1 . , width=272 ] consider next what happens when we combine the quantum zeno effect with the ifm @xcite . it allows to perform a counterfactual transmission of bit 1 with probability arbitrary close to 1 . the device consists of a chain of @xmath2 interferometers with identical beam splitters with small transmittance @xmath3 , see fig . each one of the beam splitters in the chain performs the following unitary evolution of the state of the particle : @xmath4 a straightforward calculation shows that @xmath5 beam splitters perform the following evolution of the wave packets of the particle entering the chain : @xmath6 for the particular choice of the transmittance parameter , @xmath7 , after passing all @xmath2 beam splitters , the wave packet of the particle moves from one side to the other : @xmath8 , @xmath9 . the zeno effect takes place when the right arms of the interferometers are blocked , fig . the state remains @xmath10 with probability close to 1 when @xmath2 is large , @xmath11 . in summary , for bit 0 bob does nothing and alice gets the click with certainty at the right port in the detector @xmath1 . for bit 1 bob blocks the interferometers and alice gets the click with a very high probability in the detector @xmath0 . when the right path of the inner interferometer is blocked . , title="fig:",width=234 ] when the right path of the inner interferometer is blocked . , title="fig:",width=234 ] in this section i describe the recent protocol by salih et al . @xcite which followed the idea of counterfactual computation @xcite . let us first consider a mzi nested in another mzi , see fig . the inner interferometer is tuned for destructive interference toward the second beam splitter of the external interferometer , see fig . the external interferometer is tuned for destructive interference towards @xmath1 when the lower path of the inner interferometer is blocked , see fig . this configuration provides ( sometimes ) definite information about value 0 of the bit , namely the absence of the shutter . indeed , click in @xmath1 is possible only if the shutter is not present . one can naively argue that alice obtains this information in a counterfactual way , since the particle can not pass through bob s site and reach @xmath1 . however , as detailed in section [ criteria ] , the presence of a weak trace inside the inner interferometer contradicts it . salih et al . @xcite further argued that a scheme with numerous nested interferometers leads to a protocol for transmitting both values of the bit in a counterfactual way with an efficiency which is arbitrarily close to 100% . in the protocol , @xmath12 chains of the @xmath13 interferometers described in fig . 4 are parts of another chain of interferometers with @xmath14 beam splitters having transmittance , @xmath15 . to simplify the analysis , i modify salih et al . protocol making it slightly less efficient , but still good according to their line of argument . i replace the mirrors of the external chain of interferometers by highly reflective beam splitters . transmitted waves are lost , but the modification balances the losses in the inner chains when shutters are introduced , such that the states of particles which are not lost are still described by the same equation ( [ n - steps ] ) . the setup is described in fig . . the external chain of interferometers has @xmath14 beam splitters with transmittance @xmath16 and the transmittance of the side beam splitters serving as mirrors is @xmath17 . all right mirrors of the internal chains are in bob s territory . he knows that alice sends a particle at a particular time at the top of the external chain in the state @xmath18 . for communicating bit 1 , bob blocks all inner interferometers , see fig . then , after @xmath19 beam splitters of the external interferometer , the _ normalized _ quantum state is @xmath20 and after all @xmath21 beam splitters the state is @xmath22 in both equations `` ... '' signify states orthogonal to the term which is shown . if @xmath23 , then the norm of the leading term in ( [ m - step ] ) is close to 1 : @xmath24 thus , in the limit of large @xmath2 , bob s choice of bit 1 leads to a click of alice s detector @xmath1 . since the state @xmath25 is orthogonal to the state @xmath10 at the output port of the interferometer , there is zero probability for a click of @xmath0 . the particle can be lost in alice s or bob s territories , and then none of the detectors click , but the probability of such a case vanishes for large @xmath2 . if bob wants to communicate the bit 0 , instead , he does nothing , fig . then , every wave packet entering any of the inner chains of the interferometers follows evolution ( [ n - steps ] ) inside this chain and does not come back to the external interferometer . at the output of the interferometer , the normalized quantum state is @xmath26 under the condition @xmath27 , the norm of the leading term in ( [ m0-step ] ) is also close to 1 : @xmath28 therefore , the detection of the particle in the left port by @xmath0 tells alice that bob sent bit 1 . note , that the probability for a failure might become large if the condition @xmath27 is not fulfilled . the particle might be lost or detected by @xmath1 . however , if the condition holds , the probability for a failure is vanishingly small . the probability for loosing the particle and getting no result is of order @xmath29 for bit 1 and @xmath30 for bit 0 . the click of @xmath1 tells alice with certainty that the bit is 1 , and the click of @xmath0 tells her that the bit is 0 with only a very small probability for an error : about @xmath31 . this is a good direct communication protocol . as in the simple example in sec . [ comqp ] ( fig . 3 ) , using hwps instead of absorbing shutters leads to a communication protocol which is theoretically free of errors . @xcite suggest such a protocol and argue that it has `` arbitrarily small probability of the existence of the particle in the transmission channel '' . the configuration is similar to the experiment of salih et al . @xcite : a chain of @xmath12 interferometers with inner chains of @xmath13 interferometers ( but now @xmath32 have to be even numbers ) . without absorbers , the evolution is unitary and the zeno effect is not used in this protocol . there is no need to modify the protocol by replacing mirrors with beam splitters , because there are no losses to compensate . the new protocol is different also in the transmittance of the beam splitters in the inner chains . the parameter @xmath33 is bigger by a factor of 2 : @xmath34 . as a result , the chain ( without the hwps ) works as two consecutive inner chains of the protocol discussed in the previous section . the first half of the chain moves the wave packet to the right side and the second brings it back to the left . from ( [ n - steps ] ) we obtain the transformation of the wave packet in the inner chain of the interferometers @xmath35 , see fig . b ) . hwps `` undo '' the transformation on every second interferometer and the wave packet ends up in the original state on the left without additional phase . , width=234 ] when the hwps are inserted in every interferometer of the inner chain , see fig . 7b , they cause a @xmath36 phase change of every state @xmath37 and , consequently , every second beam splitter reverses the operation of the previous one : @xmath38 since every chain has an even number of beam splitters , the transformation of the wave packet in the inner chain is @xmath39 . the setup for sending bit 0 is described on fig . bob leaves the inner interferometers untouched . then , each inner chain of the interferometers changes the phase of the quantum state of the particle : @xmath40 . a state @xmath10 of the inner interferometer is a state @xmath37 of the external interferometer . thus , the operation of the first external interferometer is @xmath41 since the number of beam splitters in the external chain is even , at the end of the process the particle is on the left side and it is detected by detector @xmath0 with certainty . if bob wants to communicate the bit 1 , instead , he inserts hwps in all the interferometers of the inner chains fig . now , after every two beam splitters of the inner chain , the wave packet comes back to the left side without acquiring additional phase . thus , every inner chain works as a mirror and the external chain of the interferometers moves the particle from left to right , to be detected with certainty by detector @xmath1 . alice knows with certainty the choice of bob by observing which detector clicks . this is an ideal direct communication protocol : theoretically , when there are no losses , there is zero probability for an error . the question i want to answer in this paper is : can the protocols of sections [ directcf ] and [ directnonzeno ] be considered as _ counterfactual _ communication protocols ? let us consider the following three statements which try to capture the meaning of counterfactuality . \1 ) the probability of finding the particle in the transmission channel is zero or can be made arbitrarily small . \2 ) the particle did not pass through the transmission channel . \3 ) the particle was not present in the transmission channel ( or the probability of its existence in the transmission channel can be made arbitrarily small ) . in the papers on counterfactual communication @xcite all these statements were considered to be interchangeable , i.e. all are true and each one of them represents counterfactuality . i argue that the situation is more subtle and clarification is needed . statement ( 1 ) . a non - demolition measurement of the presence of the particle in the transmission channel disturbs completely the operation of the communication protocols which are considered . when such a measurement is added to the protocol , bob does not transmit information to alice by his actions . so , the truth or falsehood of this statement is not a decisive indication of the counterfactuality of the protocols . however , since there is a separate controversy about the validity of this statement for the two protocols under discussion , it should be clarified too . the papers on counterfactual communication claim that this is a correct statement while i @xcite claim that in these protocols the probability of finding the particle in the transmission channel is 1 . the source of this controversy is our different assumptions . the communication protocols involve preparation and detection of the particle . i consider the probability of finding the particle in the transmission channel under the condition of the same final detection as in the protocol without intermediate measurement . in this case , the probability of finding the particle is exactly 1 , since had it it not been found , the result of the final detection could not be that of the undisturbed protocol . on the other hand , without the condition on the result of the final detection , the probability to detect the particle in the transmission channel is vanishingly small . these are two correct statements about the probability of finding the particle in the intermediate measurement : the probability is 1 when both the proper preparation and the proper final detection are done , and it is vanishingly small when only the preparation of the particle is assumed . none of these statements shed much light on the issue of counterfactuality of the protocols _ without _ intermediate non - demolition measurements . statement ( 2 ) . in contrast to such a claim for a classical particle , the meaning of this statement for a quantum particle is not well defined . for a classical particle , the operational meaning of ( 2 ) is ( 1 ) , but as discussed above , statement ( 1 ) is not helpful in the quantum case . in a double slit experiment with a screen , there is no good answer through which slit the particle passed and through which it did not pass . however , if the detector which finds the particle is placed after one of the slits , and the wave packet passing through the other slit does not reach the detector , then it is frequently declared , following wheeler @xcite , that the particle did not pass through the second slit . ( note that this contradicts the textbook picture , attributed to von neumann , according to which the wave passes through both slits and then collapses to the location of the detector . ) if we adopt wheeler s definition , then statement ( 2 ) is correct for the protocol of section [ directcf ] . the wave packets of the particle passing through the transmission channel do not reach the detector which detects the particle in this protocol . i , however , argued that we should not adopt wheeler s definition for discussing the past of a quantum particle @xcite . the concept of a quantum particle passing through a channel has no clear meaning in the standard quantum mechanics in which particles do not have trajectories . it is rigorously defined in the framework of bohmian mechanics @xcite . however , since the authors of papers on counterfactual communication never mentioned bohmian mechanics , it is not particularly relevant to the current controversy . still , due to to its conceptual importance i , following the advice of a referee , will provide the analysis in the framework of the bohmian interpretation , but only at the end of the paper , in section [ bohm ] . statement ( 3 ) . apart from bohmian mechanics , quantum mechanics does not provide a rigorous meaning also for statement ( 3 ) . without a clear ontological definition i suggest to introduce an operational meaning . we can not rely on an operational definition based on statement ( 1 ) , since strong , even nondemolition , measurements change the setup we want to analyze . so , my proposal is to look at the _ weak _ trace the particle leaves . all particles interact locally with the environment . if the particle is present in a particular place , it leaves some trace there , and it does not leave any trace where it was not present . therefore , we can run the protocol we want to analyze , and then look at the trace left in the environment . if in the transmission channel there is a non - zero trace , we will say that the protocol is not counterfactual . the counterpart of ( 3 ) , a definition of counterfactual as a process without a local trace , is less robust , because , although very unlikely , it is possible that the particle changed the local environment via some local interaction , but then changed it back to its original state . since we are all along analyzing interference experiments , the trace left by the particles has to be small , as otherwise the interference is destroyed . when the trace is small , one may argue that it can be neglected . i , however , claim that it can be neglected only if it is small compare to the trace which a single particle with the same coupling being at the same place would leave . hence , the remaining task is the comparison between the trace left in the transmission channel in the `` counterfactual '' communication protocols @xcite and the trace left by a single particle passing through the channel . in the next section i will analyze the minimal trace left by a single particle being in a transmission channel and the comparison will be made in the following sections . in the framework of standard quantum mechanics there is no rigorous way to decide if statement ( 3 ) holds , that is : whether or not the particle was present in the transmission channel . in a two - slit experiment it is not clear whether the particle was in a particular slit . if a particle starts on one side of a plate with two slits and it is found later on the other side , we do not know if the particle was in the two slits together , or in one of them . however , we firmly believe that it can not be that the particle was not present in both . ) . , width=272 ] consider first a single - path transmission channel with a single particle in the form of a single localized wave packet . the wave packet passes from alice to bob , see fig . let us model the interaction of the particle with the transmission channel as von neumann measurement with a gaussian probe . the initial wave function of the pointer is @xmath42 when a particle is present in the transmission channel , the interaction shifts @xmath43 by @xmath44 . thus , the state of the measuring device after the interaction is @xmath45 where @xmath46 is orthogonal to the initial pointer state and @xmath47 . how to quantify the strength of the trace ? one option is to consider the probability of detecting the change in the state of the channel in an idealized experiment . the probability equals @xmath48 . another option is just to use the shift of the pointer via the parameter @xmath49 . for a strong trace , the probability criterion does not represent the trace well : it remains almost 1 for @xmath50 and also for @xmath51 . in practice , however , it is plausible that in a realistic experiment , when in addition to the quantum uncertainty of the pointer there is an uncertainty of the grid on which we read the pointer , only very large @xmath49 can be observed . for a small value of @xmath49 , the probability of detection is proportional to the square of this parameter . the transmission of two particles doubles the shift , but increases the probability of detection by a factor of 4 . the linear response seems to be a better representation of the trace . consider now a single particle passing through the transmission channel which consists of @xmath2 identical paths as above . the quantum state of the particle is an equal weight superposition of wave packets in all paths , @xmath52 , where @xmath53 signifies the wave packet of the particle inside path @xmath54 in the transmission channel , see fig . after the particle passes the transmission channel , the state of the particle and the pointers representing the transmission channel becomes @xmath55 the probability to detect the particle in the transmission channel is the probability to find one of the states @xmath56 . it is @xmath48 as in the case of the single - path channel . the sum of the expectation values of the shifts of the @xmath57s is also the same , @xmath58 . it is important to consider the post - selection measurement of the state of the particle made by bob . let bob select the undisturbed state @xmath59 . this corresponds to detection of the particle by bob s detector in fig . 10 . for a good , low noise channel , the probability to find this state is very close to 1 . the state of the transmission channel then becomes @xmath60 at this stage , the probability to find one of the states @xmath56 is reduced dramatically . this is because the failure of post - selection by bob implies that the probability to find one of the states @xmath56 is 1 . for small @xmath61 , the probability to detect the particle in the transmission channel after the successful post - selection is approximately @xmath62 . it is interesting that the post - selection of the particle state does not change the expectation value of the sum of the pointer variables @xmath63 . one way to see this is to note that @xmath63 is proportional to the weak value @xcite of the sum of the projections on all parts of the channel which equals 1 because the initial state is the eigenstate of the sum of the projections with the eigenvalue 1 @xcite . for describing the magnitude of the trace , the directions of the shifts are not important . so the relevant parameter is @xmath64 . we have found a lower bound , @xmath65 . i analyse next the trace left in the transmission channel in communication protocols discussed above . all protocols are based on interference , therefore , when they work properly , only a small trace can be left . i use the same model : in every path of the transmission channel , the presence of the particle shifts the gaussian pointer , see ( [ a ] ) . i assume that the coupling is weak : @xmath66 , and consequently , @xmath67 . for simplicity , i consider the trace created by particles moving from alice to bob and disregard the trace created on the way from bob to alice . then , the example described in fig . 1 , is identical to that of fig . and the trace in the communication channel is the shift @xmath44 of the pointer and the probability to discover the presence of the particle by observing the trace , is @xmath48 . in the communication of the bit 0 using ifm , fig . 2 , and in the protocol with hwps , fig . 3 the trace is of the same order of magnitude . in the ifm communication of the bit 1 , see fig . 2 , after the interaction with the transmission channel , the state of the particle and the pointer is @xmath68.\ ] ] if the particle is absorbed by bob , the trace in the channel is exactly as in the single - path channel of section [ presence ] : shift by @xmath44 and the probability to find the trace is @xmath48 . if the particle is detected by alice , then there is no trace in the transmission channel . the wave packet `` tagged '' by an orthogonal state of the channel , @xmath46 can not reach alice . now i consider the ifm experiment with the chain of the interferometers , fig . 4 , starting with the communication of bit 0 , the case in which bob leaves the interferometers undisturbed . the exact expressions are complicated , but for weak coupling , only the first order contibution in @xmath61 is significant . neglecting the coupling to the channel , we obtain from ( [ n - steps ] ) the state of the particle in the @xmath5th interferometer @xmath69 the wave packet @xmath37 , `` tagged '' at the @xmath5th interferometer by the state @xmath70 in the transmission channel , interferes only with itself and reaches detectors in the state @xmath71 in this experiment , the particle ends up in detector @xmath1 ( state @xmath37 ) with the probability close to 1 . the state of the transmission channel then is @xmath72 with normalization @xmath73 close to 1 . therefore , the probability to detect the particle in the transmission channel , i.e. , to find at least one of the states @xmath70 , @xmath74 is much larger than the minimal probability to find a single particle present in this multiple - path channel , which can be as low as @xmath62 . thus , the case of the bit 0 is definitely not a `` counterfactual '' communication . this can also be seen by calculating the pointers shifts . these shifts are proportional to the expectation value of the projection on the paths of the transmission channel . the shift in path @xmath5 is @xmath75 and the sum of all shifts , @xmath76 , is much larger than @xmath44 , the standard for the presence of a single particle in the channel . the situation is different for communication of bit 1 , when bob blocks the paths of the interferometers , fig . 4b . due to zeno effect , the probability of absorbtion by bob is negligible . detector @xmath0 clicks with probability close to 1 telling alice that the bit value is 1 . in this case , there is no trace in the communication channel . it is , therefore , a counterfactual communication for bit value 1 . let us turn now to the case of nested interferometers , fig . the case which is particularly interesting is described in fig . bob does not put the shutter in , and detector @xmath0 clicks . alice knows that the bit is 0 , and naively , it seems to be a counterfactual communication since the particle `` could not pass through the transmission channel '' . indeed , the wave packet entering the nested interferometer does not reach detector @xmath0 . the interferometer was defined only by demanding destructive interferences in particular situations . to make quantitative predictions , we have to specify the beam splitters of the interferometer . in a possible implementation of the interferometer @xcite , the first two beam splitters transform the localized wave packet entering the interferometer into a superposition : @xmath77 where states @xmath53 signifies a wave packet entering detector @xmath78 , ( @xmath54=1,2,3 ) . it is is easy to see that these rules ensure the required destructive interferences . after the interaction , the joint state of the particle and the channel is @xmath79.\ ] ] from ( [ abc2 ] ) we see that the detection of the particle in detector @xmath0 post - selects the state @xmath80 therefore , the state of the channel after the post - selection is @xmath81 this is exactly the same state of the channel as in the case that a single particle passed through it , see ( [ a ] ) . thus , contrary to the naive expectation , the scheme with nested interferometers does not provide counterfactual communication @xcite . now we are ready to analyze the trace left in the direct counterfactual quantum communication " . the case of bit 1 , fig . 6a , is simple . the trace in the communication channel is correlated with the final location of the particle . if it is absorbed by alice , which happens with a probability close to 1 and corresponds to the proper operation of the protocol , the trace is zero . the wave packets `` tagged '' by orthogonal states of the channel , @xmath82 , can not reach alice . the trace is present only if the particle is absorbed by one of the bob s shutters which happens with vanishing probability . this is a counterfactual communication protocol for bit 1 . the more interesting case is that of bit 0 , when bob leaves the interferometer undisturbed , fig . we assume that the interaction with the channel is small , @xmath83 , @xmath84 , so only the first order in @xmath61 should be considered . the amplitude of the wave packet of the particle in the @xmath5th path of @xmath19th chain of the inner interferometers @xmath85 , is @xmath86 a particle present in the path @xmath85 changes the state of the corresponding pointer according to ( [ a ] ) : @xmath87 the wave packet @xmath88 `` tagged '' by the orthogonal state @xmath82 interferes only with itself and leaves the inner chain in the state , see ( [ n - n steps ] ) : @xmath89 state @xmath37 is lost and the state @xmath10 of the last inner interferometer of the chain @xmath19 enters from the right the remaining @xmath90 large interferometers . the transformation of this state ( which is named @xmath37 in the following equation ) in the remaining interferometers is : @xmath91 where `` ... '' signifies the wave packets which do not reach the detectors . in the protocol , the particle is found with probability close to 1 by detector @xmath0 . after the detection of the particle , the amplitude of the term @xmath82 corresponding to detection of the particle in the path @xmath85 can be found by collecting the factors in ( 23 - 26 ) . it is @xmath92 as explained in section [ directcf ] , the protocol works properly if @xmath93 , so that @xmath94 and the probability that one of the orthogonal states @xmath82 will be found in the transmission channel is , approximately , @xmath95 the number of paths in the channel is @xmath96 . we have seen that a single particle present in such a channel can be found with probability as low as @xmath97 which is smaller than the probability of detection of the particle in the protocol by a factor of approximately @xmath98 . since the protocol works well only when @xmath99 , the trace in the protocol is larger than the trace of a single particle passing through the channel . another criterion of counterfactuality is the sum of displacements of the pointers in all paths of the channel , the standard for which is @xmath44 . it can be found by calculating the absolute values of weak values of all projections : @xmath100 with pre- and post - selection specified by the protocol . the scalar product in the denominator of the weak value is close to 1 since the probability of the post - selection is close to 1 . the amplitude of the forward evolving state at path @xmath85 is given by ( [ ampnm ] ) and , similarly , the amplitude of the backward evolving state is @xmath101 thus , the weak value of the projection on the path @xmath85 can be approximated as @xmath102 and the sum of all shifts of pointers is @xmath103 since @xmath99 , the trace in the protocol is much larger than the standard of the trace of a single particle present in such multiple - path channel . let us turn to the `` direct quantum communication with almost invisible photons '' protocol @xcite . when bob transmits 0 , i.e. does nothing , fig . 8a , the amplitude in the path @xmath85 is @xmath104 the wave packet @xmath88 , `` tagged '' by the orthogonal state @xmath82 , interferes only with itself and it leaves the inner chain in the state @xmath105 the state @xmath37 is lost and the state @xmath10 of the last inner interferometer of the chain @xmath19 enters from the right the remaining @xmath90 large interferometers . using the fact that `` tagging '' takes place only for odd @xmath19 , and that the number of the large interferometers is odd ( @xmath14 , the number of beam splitters is even ) , and that after every second beam splitter the wave function repeats itself , we can conclude that the wave packet leaves the last beam splitter of the last inner chain with the same amplitude . the wave packet entering the last beam splitter of the large interferometers transforms into @xmath106 in the protocol , the particle is detected with probability close to 1 by detector @xmath0 which detects the state @xmath10 . thus , after detection of the particle , the amplitude of the term @xmath82 corresponding to the detection of the particle in the path @xmath85 can be found by collecting factors from ( [ inv1]-[inv3 ] ) and a factor @xmath61 due to the interaction : @xmath107 this expression holds only for odd @xmath19 , the amplitude in the paths with even @xmath19 vanishes . summing the probabilities of finding the record the particle leaves in all the paths , i.e. summing on odd @xmath19s up to @xmath12 and on integers @xmath5 up to @xmath13 , we obtain the probability of detection of the particle : @xmath108 a single particle passing through this transmission channel can be found with probability of order @xmath109 . depending on the ratio @xmath110 , it can be smaller or larger than ( [ ampnmapprox ] ) , so we can not yet decide on the counterfactuality of the protocol . compare now the sum of the pointer shifts in the channel . it can be found by calculating the absolute values of weak values of all projections : @xmath100 with the pre- and post - selection specified by the protocol . in this case , the pre- and post - selected states are identical , so the weak values are expectation values : @xmath111 and @xmath112 the sum of the pointer shifts when there is one particle in the transmission channel equals @xmath44 , so the ratio @xmath110 tells us when the sum of displacements in the protocol ( [ wvproj0 ] ) is smaller or larger than that of a single particle present in the channel . we can see that the criterion of the pointer shifts is in agreement with the criterion of the probability of detection . let us now repeat the analysis for the case of bit 1 , when bob puts hwps in every inner interferometer . now the amplitude in the path @xmath85 is @xmath113 the wave packet @xmath88 , can be `` tagged '' by the orthogonal state @xmath82 only for odd @xmath5 . thus , due to the presence of the hwps , the wave packet comes back unchanged every second beam splitter . it leaves the chain of the inner interferometers in the state @xmath114 state @xmath37 is lost , and the state @xmath10 of the last inner interferometer of the chain @xmath19 enters from the right the remaining @xmath90 large interferometers . in the chain of the large interferometers it performs usual evolution ( [ n - steps ] ) and ends up in the state @xmath115 in the protocol , the particle is detected with probability close to 1 by detector @xmath1 which detects the state @xmath37 . thus , after the detection of the particle , the amplitude of the term @xmath82 corresponding to the detection of the particle in the path @xmath85 can be found by collecting factors from ( [ inv4]-[inv6 ] ) and the factor @xmath61 due to the interaction : @xmath116 this expression holds only for odd @xmath5 , the amplitude in the paths with even @xmath5 , vanishes . summing the probabilities of finding the record the particle leaves in all the paths , we obtain the probability of detection of the particle : @xmath117 again , as in the case of bit 0 , the ratio @xmath110 tells us if it is larger or smaller than the minimal probability of detection in case of a single particle present in the transmission channel . however , the dependence is opposite : if for bit 0 the probability of detection in the protocol was smaller than single - particle standard , for bit 1 it will be larger , and vice versa . the pointer shifts criterion of counterfactuality is in agreement with these results . the sum of pointers shifts in all paths of the channel is proportional to the sum of the weak values of all projections : @xmath100 . also in this case , the pre- and post - selected states are identical and the weak values are expectation values : @xmath118 and @xmath119 the ratio @xmath110 tells us when the sum of displacements in the protocol ( [ wvproj1 ] ) is smaller or larger than that of a single particle . if it is smaller for bit 1 , it is larger for bit 0 , and vice versa . one of the motivations for `` counterfactual '' protocols , in which no particles are present in the transmission channel , is that it is secure against eve who is trying to eavesdrop the communication : she has no particles to look at @xcite . the obvious cryptographic weakness of the protocols with shutters which are 100% counterfactual is that eve can use an active attack detecting the presence of the shutters . moreover , the shutter can be detected by eve in a counterfactual way @xcite , and recently there has been a claim of a very efficient attack of this kind @xcite . if eve uses an active attack , probing bob s bit by sending her particles , the counterfactuality property does not help . so , for the analysis of counterfactuality we should only consider passive attacks in which eve `` eavesdrops '' , i.e. measures the presence of the particle in the transmission channel . under this condition , the counterfactual protocols with shutters are secure . consider the following counterfactual key distribution protocol . there are two identical chains of interferometers as in fig . one of the chains is defined as bit 0 , and the other as bit 1 . on each run of the protocol , alice randomly sends a single particle through one of the interferometers and bob randomly chooses one of the interferometers and places the shutters in all of its paths . every time alice detects the particle in detector @xmath0 of one of the interferometers , she makes a public announcement . detector @xmath0 can click only if alice and bob chose the same interferometer , i.e. they chose the same bit . this creates a common key . the multiple shutters bob placed represent the weak point of the protocol due to the reason above , although we can improve it using detectors instead of shutters and telling alice to send particles not every time , but only at some random times ( using ideas of @xcite ) . anyway , we made a postulate here that eve only performs some ( weak ) nondemolition measurements of the presence of the particle running in the interferometer . let us see if eve can get some information about the key in this way . if eve detects the particle in one of the interferometers , it can not be the one which generates a correct bit in the key . for a correct bit generation event , alice and bob have to choose the same interferometer . eve can detect the particle only if it is present , i.e. , it has to be chosen by alice . if bob also chooses this interferometer , then after detection by eve , the particle has to be absorbed by bob , so it will not reach alice . only if the interferometer is not chosen by bob , the particle seen by eve in the nondemolition measurement can reach alice , and there is a nonzero probability that the detector @xmath0 will click and thus alice will declare generation of a bit in the key . but this will be an error bit , since bob has chosen the other interferometer . eve introduces errors , and the only information she gets is about these error bits . let us turn to the `` direct counterfactual quantum communication '' protocol . when bob places the shutters , fig . 6a , eve can not get information about the correct bit because eve s detection causes the loss of the particle . ( it is not surprising , since in this case the protocol is counterfactual . ) if the bit is 0 , and the interferometer is free , fig . 6b , eve s detection will not necessarily lead to the loss of the particle . detection of the particle by eve in the last chain of inner interferometers will lead to part of its wave packet to enter the last beam splitter from the right , so most probably it will create an error : @xmath1 clicks , but bob s bit is 0 . however , eve s detection of the particle in any of the first @xmath120 chains of inner interferometers will lead to part of the particle wave packet to enter the last beam splitter from the left side , so most probably @xmath0 will click , corresponding to a successful transmission of the correct bit . thus , sometimes , eve gets a reliable information about the transmitted bit . eve , who eavesdrops only by observing the original particle of the communication protocol , can not learn any correct bit of real counterfactual protocols with shutters which transmit bit 1 , but she does learn some correct bits in `` counterfactual '' communication protocols of bit 0 . this provides another argument why such protocols should not be named counterfactual . it seems that bohmian mechanics @xcite , which postulates that every particle has a trajectory , should provide the ultimate answer regarding the counterfactuality of a protocol . there are unambiguous answers to criteria ( 2 ) and ( 3 ) of sec . [ criteria ] : either the bohmian trajectory passes through the transmission channel , or it does not . however , the fact that the bohmian trajectory does not pass through the transmission channel does not tell us that eve , who has an access to this channel , can not get some information about this communication . for completeness of the counterfactuality analysis , i will consider the following technical question . is the bohmian trajectory present in the transmission channel of the protocols discussed in this paper ? bohmian position of a particle can not be in a place where the ( forward evolving ) wave function vanishes . therefore , the successful ifms of the presence of an opaque object , described in fig . 2 and fig . 4b , are counterfactual according to the bohmian trajectory criterion . moreover , the ifm of the absence of an opaque object , fig . 5a , is counterfactual too . direct counterfactual communication @xcite , as its predecessor @xcite and their variation presented in fig . 6 , are counterfactual . in all these protocols there is no continuous path with non - vanishing wave function between the source and the detector the particle reached . when the shutter is present . thus , if detector @xmath1 clicks , we know that path @xmath121 is free . although naively , the particle reaching @xmath1 can not pass through the inner interfereometer , the bohmian trajectory of the particle ( solid line ) does pass through @xmath121 . , width=272 ] in the `` direct quantum communication with almost invisible photons '' @xcite , see fig . 8 , the probability that the bohmian particle will pass through the transmission channel can be found from the maximal amplitude in the paths passing through the channel . for bit 0 , the amplitude is given by ( [ inv1 ] ) , and therefore , the maximum probability is approximately @xmath31 . for bit 1 , the amplitude is given by ( [ inv4 ] ) , and the maximum probability is approximately @xmath122 . thus , for large @xmath14 and @xmath2 , the probability that the communication is counterfactual is close to 1 for both bit values . however , it should be mentioned , that in most cases when the communication is not counterfactual , the bohmian particle crosses the transmission channel , not once , but many times . given equal probability for bit values , the expectation value of the number of crosses of the transmission channel by the bohmian particle can be obtained from the sum of the probabilities on all paths . the amplitudes are given by ( [ inv1 ] ) and ( [ inv4 ] ) . consequently , the expectation value of the number of crosses is , approximately , @xmath123 , and it can not be much less than 1 for any choice of @xmath14 and @xmath2 . note that in the `` direct counterfactual communication protocol '' _ experiment _ @xcite , the expectation value of the number of crosses of the transmission channel is also of order 1 , but all these events happen when the particle is lost , and these cases are discarded according to the protocol . or @xmath1 possible . only empty wave passes through the transmission channel when alice s detector clicks . , width=272 ] i mentioned above that the protocol presented in fig . 5a , which demonstrates the absence of an object in the apparently interaction - free manner , is counterfactual according to bohm . it is interesting that a slight modification of this setup , similar to the experiment actually performed @xcite , is _ not _ counterfactual . consider a setup presented in fig . 11 . a straightforward calculations of bohmian trajectories ( similar to that of bell @xcite , or in a simplified way @xcite ) show that _ all _ particles detected by @xmath1 passed through arm @xmath121 . in fact , the experiment @xcite indicated the presence of the weak trace in @xmath121 , but it had nothing to do with the presence of the bohmian trajectory there . the weak trace appeared also in @xmath124 and @xmath125 where the bohmian particle was not present . empty waves cause observable difference when they `` catch '' bohmian particles . this allows a very simple communication protocol which is counterfactual according to bohmian criterion , see fig . particles are sent through a mzi without a second beam splitter at particular times . one of the mirrors is in bob s place and he has two options for placing it , such that the wave bouncing off the mirror will end up in detector @xmath0 or @xmath1 on alice s side . the wave packet going through another arm ends at detector @xmath126 on bob s side . if the particle does not reach @xmath126 , bob knows that it reaches alice and that it he has chosen the detector which will detect the particle . every time @xmath0 or @xmath1 clicks the bohmian particle does not pass through bob s place . only an empty wave was directed by bob s mirror . indeed , the wave packets overlap at point @xmath127 or @xmath128 and the bohmian particle must `` change hands '' in the overlap , so bohmian particles reaching alice s detectors never cross communication channel . probably the majority would not consider the communication protocol described in fig . 12 as counterfactual . according to wheeler s common sense argument @xcite , the particle reaching alice could come only through bob s site . this example , and englert et al . setup @xcite , in which a strong trace ( observed at a later time ) is left in a place where the bohmian particle was not present , explain why the bohmian criterion of counterfactuality does not agree with the intuition of most physicists . the standard quantum formalism , in contrast to bohmian mechanics , does not specify the position of a quantum particle . thus , it does not provide an unambiguous answer to the question : is a particular communication protocol counterfactual ? i.e. : was the particle present in the transmission channel ? in this paper i analyzed an approach to answering this question based on the weak trace the particle leaves in the channel . i compared the trace left in the channel in recently proposed protocols claimed to be counterfactual with the trace in the protocol constructed to transmit a single particle in the same channel . in the analysis , i considered two criteria for comparing the traces . first , the probability of finding a conclusive evidence for the presence of the particle , and second , the expectation value of the sum of total shifts of some variables of the channel . the question of counterfactuality of the protocols is considered in cases the protocols work properly , i.e. when the particle is detected by the right detector . it means that the particle is pre- and post - selected . the protocols were compared with the transmission of a single pre- and post - selected particle . in all these cases the probability of post - selection was closed to 1 . the protocol direct counterfactual quantum communication " @xcite is fully counterfactual for bit value 1 . the trace is identically 0 . nothing changes in the transmission channel and therefore there is zero probability to detect the particle in the transmission channel . passive eavesdropping provides no information about the transmitted bit . however , the protocol is not counterfactual for the bit value 0 . it is true that by increasing the number of paths in the channel , the probability of finding a conclusive evidence of the presence of the particle reduces , but increasing the number of paths also reduces the probability to find a single quantum particle when it passes the channel . the probability to find the presence of the particle in the transmission channel in the event of successful operation of the protocol is larger than the probability to detect a particle successfully passing through this channel . eve , using passive attack , obtains some information about the transmitted bit . the criterion of the shifts of variables of the channel tells us the same . the expectation value of the sum of the shifts is zero for bit 1 , but for bit value 0 it is larger than the sum of the shifts when a single particle passes the channel . the protocol `` direct quantum communication with almost invisible photons '' is not fully counterfactual for any bit value . some trace is always left in the transmission channel . however , if we are ready to consider a protocol as counterfactual when it leaves a trace which is much smaller than the trace of a single particle passing through this channel , then we can arrange that it will be counterfactual for one of the bit values . by playing with the numbers @xmath2 and @xmath14 of the inner and the external interferometers respectively , the protocol can be made counterfactual for value 0 or for value 1 of the bit . it _ can not _ be made counterfactual for both . the analysis of the trace left by a particle passing through a @xmath2-path channel of section vi showed a surprising result : the probability of detection in the channel of the successfully transmitted particle is reduced by the factor of @xmath129 . it helped me to analyse the counterfactuality of the protocols , but it might also open new avenues for useful quantum communication applications . i thank yakir aharonov , eliahu cohen and shmuel nussinov for helpful discussions . this work has been supported in part by the israel science foundation grant no . 1311/14 and the german - israeli foundation grant no . i-1275 - 303.14 .

the relational data base model , introduced by codd in @xcite , has been particularly successful since it is a mathematically elegant model well suited to describe almost all `` real world '' situations . since the query languages associated to such model ( the _ relational algebra _ and the _ relational calculus _ ) have a formal and simple definition , an interesting field of research is to study the expressive power of such language . codd @xcite has proved that the relational algebra is equivalent to the _ relational calculus _ , in the sense that both query languages can compute the same set of relations . a breakthrough in this field @xcite has been a syntactic characterization of the set of relations that can be computed in a give data base . these results , also known as _ bp - completeness _ , are based on the principle of data independency from the physical representation : the information that can be extracted from the data base is completely determined at the logical level of such data base . this fact can be stated in a simple way : a relation @xmath0 can be computed from a data base @xmath1 if and only if all permutations over the elements of @xmath1 which preserve @xmath1 ( that is , all permutations that produce a data base isomorphic to @xmath1 ) , also preserve @xmath0 . an interesting interpretation of this property is that only the information given by the structure of the data can be used to differentiate data values ; consequently , a query is expressible if and only if it does not add any additional differentiation to the one initially available @xcite . this idea can be rephrased by stating that the result of a query is invariant w.r.t . permutations of indistinguishable values ; such a permutation was captured with the notion of automorphism in @xcite . while the @xmath2-criterion is a natural requirement , it refers to properties of relations in a given data base instead of queries as a whole . we recall that a query is an expression of the query language that can be applied to different data bases leading to possibly different results . thus it has been extended to a property of queries as partial functions from data bases to data bases , which is known nowadays as _ @xcite : it has been recognized as the capability of the calculus to preserve isomorphisms between data bases , rather than automorphisms . genericity is a common requirement for query languages and it is traditionally related to the _ data independence _ principle that assumes that the data base is constructed over an abstract domain which is independent from the internal representation of data . subsequent research has shown that this approach to the analysis of the expressiveness of a query language has certain shortcomings @xcite , mainly when new data models , such as the object - based model , are introduced . other notions have been proposed to analyze properties of queries in some new models @xcite pointing out the importance of extending genericity to be used in more complex models . in @xcite languages are classified w.r.t . the degree of the use of the equality predicate , by analyzing the invariance property of queries under different mappings ( not necessarily isomorphisms ) over the data domain , which are compatible with the relational structure of the data base . subsequent advances in data base theory have led to different models that take into account the limitations of the relational model when it comes to describe complex situations . most of such models have been introduced in the graph - based or object - oriented frameworks , but usually their mathematical foundations do not allow a complete study of the expressive power of the query languages introduced . in fact , to our knowledge , the only exception is the graph - based model good @xcite . in this paper we introduce a different syntactic characterization of queries computable in a data base . our characterization relies upon the notion of partitions of the domain , where each partition represents a level of _ undifferentiation _ among objects , values or vertices . notice that an automorphism also can represent a certain level of undifferentiation . initially we will exploit such notion to give two new characterizations of relations expressible in a relational data base . subsequently , we will show how to apply the new framework to analyze a simple graph - based model , hence proving that our characterization can be useful in comparing the expressive power of different data languages . following the approach of @xcite , the data models studied in this paper are domain - preserving , that is , it is not possible to create new vertices or values , but only to query an existing data base . in our framework , a binary relation over sets of data values is defined , denoted by @xmath3 , which relates those sets of values that can not be differentiated . from the relation @xmath3 we build some sets of partitions that _ respect _ @xmath3 , that is , all classes in a partition are preserved by @xmath3 . we prove that expressiveness of a query language can be stated as the conservation of some of those partitions , where the exact set of partitions that must be preserved depends on the data model . the expressibility results we obtain have the following form : _ given a data base @xmath1 , let @xmath4 be a relation or a graph over the domain set of @xmath1 . then @xmath4 can be expressed in @xmath1 if and only if @xmath5 _ , where @xmath6 and @xmath7 are two sets of partitions which depend on the model under consideration . all sets considered in this paper are assumed to be finite and nonempty . given a set @xmath8 , a _ relation _ @xmath0 over @xmath8 is a subset of the cartesian product @xmath9 ( @xmath10 times ) for some fixed integer @xmath11 , that is a set of tuples of length @xmath10 , where all components of a tuple are elements of @xmath8 . the number @xmath12 is called the _ order _ or _ arity _ of the relation . given a set @xmath13 of relations over @xmath8 , the pair @xmath14 is called a _ relational database _ ; in this setting , @xmath8 is the _ domain _ of the database , and @xmath15 is the set of relations of the database . given a relation @xmath16 of a database @xmath14 , we denote with @xmath17 the _ data domain _ of @xmath0 , that is the subset of the elements of the database domain @xmath8 that are in at least one tuple of @xmath0 . the notion of data domain is easily extended to the set @xmath15 of relations as the set union of relations data domains : @xmath18 . without loss of generality , we can assume that @xmath19 for every considered database @xmath14 . this seemingly trivial requirement is indeed very important , as it will become evident after theorem [ teo : paredaens2 ] , therefore we will omit the universe set unless it is necessary to avoid any ambiguities . just as in @xcite , when referring to a relational database , we use the _ relational algebra _ as a query language . in relational algebra two binary operators ( union and product ) and three unary operators ( projection , equality restriction and inequality restriction ) are given . in the following definition all relations are defined over the same database domain @xmath8 . [ def : relational - algebra ] let @xmath0 and @xmath4 be two relations with the same arity ; the _ union _ of @xmath0 and @xmath4 , denoted by @xmath20 , is simply the set theoretical union of the two sets of tuples . given two relations @xmath0 and @xmath4 ( not necessarily with the same arity ) , the ( cartesian ) _ product _ of @xmath0 and @xmath4 , denoted by @xmath21 , is the set of all possible concatenations of a tuple of @xmath0 with a tuple of @xmath4 : @xmath22 . the abbreviation @xmath23 is used to express the relation @xmath24 ( @xmath25 times ) . let @xmath26 be the arity of a relation @xmath0 , @xmath27 a positive integer and @xmath28 @xmath29 a function . the _ projection _ of @xmath0 over @xmath30 , denoted by @xmath31 , is the relation : @xmath32 . now , let @xmath33 and @xmath34 be two integers such that @xmath35 , where @xmath26 is the arity of a relation @xmath0 . the _ equality restriction _ of @xmath0 on @xmath33 and @xmath34 is the relation , denoted by @xmath36 , that is obtained by taking from @xmath0 all the tuples for which the @xmath33-th and the @xmath34-th components are equal : @xmath37 . analogously , the _ inequality restriction _ of @xmath0 on @xmath33 and @xmath34 , denoted by @xmath38 , is the relation obtained by taking from @xmath0 all the tuples for which the @xmath33-th and the @xmath34-th components are different : @xmath39 . the five operations just described are sufficient to generate the operations of intersection , difference , join and division , usually assumed as primitives in codd s relational algebra ; a proof of this fact can be found , for example , in @xcite . given a relational data base @xmath40 , we will denote by @xmath41 the relation which is the result of applying the expression ( of the relational algebra ) e to the data base @xmath1 . moreover a relation @xmath4 over @xmath8 is told to be _ expressible _ from @xmath15 if there exists an expression @xmath42 whose operands are all relations in @xmath15 , and such that @xmath41 is equal to @xmath4 . following @xcite , we denote with @xmath43 ( basic information contained in the set of relations @xmath15 ) the set of relations that can be expressed from @xmath15 . as observed in @xcite , @xmath43 is the set of the answers to all possible queries that can be asked to a relational datmabase that contains the relations @xmath15 . in @xcite , paredaens gives a characterization of the class @xmath43 based upon appropriate automorphisms , that is permutations of the elements of the database domain . let @xmath0 be a relation of order @xmath26 over a set @xmath8 . as in @xcite , an _ automorphism _ is a bijective function ( that is , a permutation ) on @xmath8 . we say that the automorphism @xmath44 _ respects _ the relation @xmath0 or , equivalently , that @xmath45 is @xmath0-_compatible _ if , for each tuple @xmath46 , @xmath47 . the compatibility of an automorphism @xmath44 with respect to a relation @xmath0 can be naturally extended to a set @xmath15 of relations in the following way : @xmath45 _ respects _ the relations in @xmath15 or , equivalently , @xmath45 is @xmath15-_compatible _ if @xmath45 is @xmath0-compatible for each relation @xmath0 in @xmath15 . notice that the set of automorphisms @xmath15-_compatible _ , is a group of elements , a binary associative operation on @xmath48 , and an identity element @xmath49 , such that the operation is closed and invertible in @xmath48 ] where the operation is the composition of functions and the identity is the identity function ( i.e. the function defined as @xmath51 ) . as in @xcite , we denote with @xmath52 the set of all the automorphisms @xmath44 which are @xmath15-compatible ; with a small abuse of notation , if @xmath53 , we will usually write @xmath54 instead of @xmath55 . it will be very useful to consider the following representation of @xmath52 . let @xmath14 be a relational database , with @xmath56 , @xmath57 , and let @xmath58 be the set of @xmath15-compatible automorphisms . the following relation of arity @xmath59 : @xmath60 is called the _ cogroup relation _ of @xmath14 . as we can see , each row ( tuple ) of the relation @xmath61 represents one of the @xmath15-compatible automorphisms . since we do not associate any particular meaning to the elements of the domain @xmath8 , if @xmath62 we can assume , without loss of generality , @xmath63 . we can also assume that the first tuple of @xmath61 represents the identity function on @xmath8 ( which is always present in @xmath52 , since it is compatible with every nonempty set of relations ) ; as a consequence , it can always be assumed that the first row of @xmath61 is the tuple @xmath64 . let @xmath14 be a relational data base , with : * @xmath65 * @xmath66 , with : @xmath67 it is easily verified that : @xmath68 @xmath69 if we look at the @xmath15-compatible automorphisms as permutations over @xmath8 , we can express @xmath52 as follows : @xmath70 [ ex : klein - group ] it is not difficult to see that , for a given database @xmath14 , the set @xmath52 of @xmath15-compatible automorphisms is indeed a group with respect to function composition , with the identity function over @xmath8 as unitary element . in fact , the identity over @xmath8 is always in @xmath52 , the inverse of an @xmath15-compatible automorphism is still an @xmath15-compatible automorphism , and the composition between two @xmath15-compatible automorphisms is again an @xmath15-compatible automorphism . since we can always assume @xmath71 , we can think of @xmath52 as a finite permutation group over the set @xmath72 , that is a subgroup of the symmetric group @xmath73 . in this paper we investigate the relation between expressive power and partitions of the database domain . more precisely , we investigate the possibility to characterize the expressive power of relational and graph - based databases via one or more theorems abiding to the following meta theorem . [ thm : scheme1 ] let @xmath14 be a relational database , and let @xmath4 be a relation over @xmath8 . then @xmath74 , where @xmath75 and @xmath76 are sets of partitions over @xmath8 , built from the sets @xmath15 and @xmath77 of relations respectively . the relevance of the main result in @xcite is that it is the first syntactic characterization of the relations that can be obtained from a given database @xmath14 when the relational algebra is used as a query language . more precisely , in @xcite the following theorem is proved . let @xmath14 be a relational database , and let @xmath4 be a relation over @xmath8 . then @xmath78 and @xmath79 . [ teo : paredaens ] basically , paredaens has been able to point out the fundamental relation between expressiveness in a database and the set of automorphisms in the relational model . such result has been successively extended in @xcite to define in a formal way the notion of _ genericity _ , that is computable queries @xcite have to be invariant with respect to the isomorphisms between databases . we can restate theorem [ teo : paredaens ] in a form that will be more convenient for our purposes . let @xmath14 be a relational database , and let @xmath4 be a relation over @xmath8 . then @xmath80 . [ teo : paredaens2 ] first of all , we show that @xmath81 . proving that @xmath82 is trivial as @xmath83 . the latter stems from the fact that the relations which are expressible from @xmath15 are those obtained from @xmath84 simply ignoring the relation @xmath4 . let now be @xmath85 and @xmath86 . if the expression that gives @xmath87 from @xmath84 does contain some occurrence of the relation @xmath4 , it is sufficient to replace such occurrence with the expression that gives @xmath4 from @xmath15 to conclude that @xmath88 , and thus @xmath89 . it is immediate to notice that @xmath90 implies @xmath91 . since we have established that @xmath81 , the two databases @xmath14 and @xmath92 are _ basic information equivalent _ that is , every relation of the first database can be obtained from the relations of the second database and vice versa if and only if @xmath4 is expressible from @xmath14 . a direct consequence of theorem [ teo : paredaens ] is that two databases @xmath93 and @xmath94 are basic information equivalent if and only if @xmath95 ( which are assumed to be both equal to @xmath8 ) and @xmath96 ; thus , we can conclude that @xmath80 as stated . we observe that , given our assumption that @xmath97 , in theorem [ teo : paredaens2 ] we can get rid of the inclusion between the domains , since it is implicit from the fact that @xmath4 is a relation over @xmath8 . on the other hand , we _ can not _ ignore the inclusion condition if we suppose that @xmath98 , since in such a situation it is not difficult to show two relations @xmath0 and @xmath4 such that @xmath99 but @xmath100 . a notion that seems tightly related to the expressiveness of relations in a database is that of _ indistinguishability _ between elements of the domain . intuitively , the idea is that the elements of a subset of the domain of a given database are indistinguishable if and only if no query to the database is able to divide the set in two parts , one made of the elements that occur in the relation resulting from the query and the other made of the elements that do not occur in the relation . in such a situation , we say that the set of indistinguishable elements can not be _ separated _ by any of the queries that can be presented to the database . thus , a relation resulting from a query to the database can only contain all or none of the elements of a non - separable set . theorem [ thm : scheme1 ] defines the general framework we propose to investigate the expressive power of query languages . in this framework different notions of expressible queries can be studied by considering different sets of partitions . for a given database @xmath14 , we say that a set @xmath75 of partitions of @xmath8 is a set of _ valid partitions _ if and only if it satisfies theorem [ thm : scheme1 ] . by the results in @xcite , it seems to us quite natural to define the following sets of valid partitions , namely the _ orbit partitions _ and the _ cycle partitions _ ; indeed later we will be able to prove that , in the context of theorem [ thm : scheme1 ] , they are equivalent to the characterization of relations obtainable in a relational data base of @xcite . let @xmath14 be a relational database , and let @xmath101 , @xmath102 be a partition of @xmath8 . @xmath103 is an _ orbit partition _ of @xmath8 with respect to @xmath15 if both the following conditions hold : 1 . for each relation @xmath16 and for each class @xmath104 , @xmath105 or @xmath106 ; 2 . for each class @xmath107 and for each pair @xmath108 of elements of @xmath109 there exists an automorphism @xmath110 such that @xmath111 , and @xmath112 for every class @xmath113 . we denote with @xmath114 the set of all orbit partitions of the given database @xmath14 . [ def : orbit - partition ] let @xmath14 be a relational database , and let @xmath101 , @xmath102 be a partition of @xmath8 . @xmath103 is a _ cycle partition _ of @xmath8 with respect to @xmath15 if both the following conditions hold : 1 . for each relation @xmath16 and for each class @xmath104 , @xmath105 or @xmath106 ; 2 . there exists an automorphism @xmath110 such that for each class @xmath107 and for each pair @xmath108 of elements of @xmath109 there exists an integer @xmath59 such that @xmath115 and @xmath112 for every class @xmath113 . we denote with @xmath116 the set of all cycle partitions of the given database @xmath14 . [ def : cycle - partition ] as already stated for @xmath52 , if @xmath0 is a relation we will write @xmath117 and @xmath118 instead of @xmath119 and @xmath120 respectively . the following theorem is an alternative formulation of the main result of @xcite ( the equivalence of the two formulation follows from theorem [ teo : paredaens2 ] ) which is more useful for our purposes . let @xmath14 be a relational database , and let @xmath4 be a relation over @xmath8 . then @xmath121 . let @xmath14 be a relational database , and let @xmath52 and @xmath61 be respectively the group of @xmath15-compatible automorphisms and the cogroup - relation of @xmath15 . a useful fact proved in @xcite is that the cogroup - relation is expressible from @xmath15 , that is @xmath122 . using this fact , we are able to prove the following theorem . @xmath123 . [ teo : autr = cgr ] since @xmath122 , by theorem [ teo : paredaens ] we can conclude that @xmath124 . now , let @xmath125 ; as we have already observed , @xmath126 is a permutation of the set @xmath97 , as well as of the tuples that compose the relation @xmath61 . thus , for each tuple @xmath127 , we have that @xmath128 . in particular , by letting @xmath59 be the cardinality of @xmath8 , we have : @xmath129 thus , the elements of @xmath8 are mapped by @xmath126 in such a way that the result is a row of the cogroup - relation ; so we can conclude that @xmath110 . a direct consequence of theorem [ teo : autr = cgr ] is that not only @xmath122 , as established by paredaens , but also @xmath130 for every relation @xmath16 , since @xmath131 and @xmath132 . as a corollary of theorem [ teo : autr = cgr ] , if we are interested to study the expressive power of a given relational database @xmath14 then we can work as well on the database @xmath133 , which has only one relation and , moreover , such relation is an explicit representation of the finite permutation group @xmath52 . we now turn our attention to the structure of @xmath114 and @xmath116 . first of all we observe that , thanks to theorem [ teo : autr = cgr ] , we can get rid of item 1 in definitions [ def : orbit - partition ] and [ def : cycle - partition ] since , by considering the database @xmath134 , there is only one relation and , for such relation , it holds @xmath135 for each @xmath107 . to characterize the sets of cycle and orbit partitions we need to recall some notions from basic abstract algebra . let @xmath136 be a set and @xmath137 a group . action of @xmath48 on @xmath136 _ is a map @xmath138 such that 1 . @xmath139 ; 2 . @xmath140 in group theory it is customary to omit the operators symbols from expressions when confusion does not arise ; so , the expression in item 2 above is usually written as : @xmath141 . let @xmath48 be a group acting on a set @xmath136 . for @xmath142 , let @xmath143 if and only if there exists @xmath144 such that @xmath145 . it is not difficult to see that @xmath146 is an equivalence relation on @xmath136 , and thus it induces a partition @xmath103 on @xmath136 . the classes of @xmath103 are called the _ orbits _ in @xmath136 under @xmath48 . if @xmath147 , the class containing @xmath148 denoted by @xmath149 is called the _ orbit of _ @xmath148 under @xmath48 . in other words , @xmath150 . it is not difficult to see that the partition induced by the orbits of @xmath52 on @xmath8 satisfies definition [ def : orbit - partition ] . in fact , every automorphism @xmath110 maps each orbit @xmath151 into itself and , given a pair @xmath108 of elements of @xmath8 , there exists an automorphism that maps @xmath152 to @xmath153 if and only if @xmath152 and @xmath153 are in the same orbit . moreover , if @xmath154 is a subgroup of a group @xmath48 acting on the set @xmath136 , then every orbit @xmath155 is a subset of the orbit @xmath149 ; more precisely , it is not difficult to prove that the orbits induced by @xmath154 are a _ refinement _ of the orbits induced by @xmath48 . since each partition induced by the orbits of every subgroup of @xmath52 satisfies definition [ def : orbit - partition ] , we have that @xmath114 contains the set of those partitions . vice versa , let @xmath156 . it is not difficult to see that the set of automorphisms @xmath110 that map each class of @xmath103 into itself and that map each element of a class to an element of the same class forms a subgroup of @xmath52 ; moreover , the orbit partition induced by such a subgroup is just @xmath103 . as a consequence , @xmath114 is a subset of the set of partitions induced by all the subgroups of @xmath52 ; since also the converse inclusion holds , the two sets indeed coincide . let @xmath48 be a group acting on the set @xmath136 , and let @xmath157 . for @xmath142 , let @xmath143 if and only if there exists an integer @xmath59 such that @xmath158 , where @xmath159 is the application of @xmath160 for @xmath59 times . it is not difficult to see that @xmath146 is an equivalence relation on @xmath136 , and thus it induces a partition @xmath103 on @xmath136 . the classes of @xmath103 are called the _ cycles _ of @xmath160 on @xmath136 . analogously to what said about orbits , it is not difficult to see that the partitions induced by the cycles of the automorphisms of @xmath52 satisfy definition [ def : cycle - partition ] . we observe that , while an orbit partition is induced by a subgroup of @xmath52 , a cycle partition is induced by an automorphism , that is by an element of @xmath52 . the class @xmath116 is thus the set of cycle partitions obtained by considering every element of @xmath52 . let @xmath48 be a group acting on the set @xmath136 and let @xmath160 be a permutation in @xmath48 . then the _ orbits of the _ ( cyclic ) _ group _ @xmath161 generated by @xmath160 are the cycles of @xmath160 . since @xmath161 is a subgroup of @xmath48 , we have immediately that every cycle partition of @xmath48 is also an orbit partition of @xmath48 , that is , @xmath162 . example [ ex : klein - group ] can be used to show that the converse does not generally hold : not every orbit partition is also a cycle partition . in fact we have : @xmath163 as noted above , theorem [ teo : autr = cgr ] allows us to deal only with cogroup - relations instead of sets of arbitrary relations . the same can be done when working with cycle and orbit partitions : since cycles and orbits that form the partitions in @xmath116 and @xmath114 are completely determined from the elements and the subgroups of @xmath52 respectively , by theorem [ teo : autr = cgr ] we can conclude that @xmath164 and @xmath165 . it is possible to show that both the set @xmath116 of cycle partitions and the set @xmath114 of orbit partitions of a given database @xmath14 constitute a partially ordered set ( poset ) with respect to the binary relation @xmath166 , where @xmath167 iff each class of @xmath168 is contained in some class of @xmath169 , where @xmath168 and @xmath169 are two partitions in @xmath75 , @xmath75 is equal to @xmath116 or @xmath114 . in fact , it is not difficult to see that @xmath166 is reflexive , antisymmetric and transitive : that is , @xmath166 is an order relation over both @xmath116 and @xmath114 . one notably difference between the posets @xmath170 and @xmath171 is that the first has always a maximum element , corresponding to the orbits of the entire @xmath52 , while the second may not have a maximum element , as shown above referring to example [ ex : klein - group ] , where @xmath52 is the so called klein group . instead , both the posets have a minimum element , corresponding to the cycles ( equal to the orbits ) induced by the identity element of @xmath52 : that is , the trivial partition , where each class is a singleton . in order to prove our main results we need some definitions and some well known properties of finite groups . here we just recall the notion of _ stabilizer _ ; we address the reader to an introductory book on abstract algebra , such as @xcite , for the notion of coset and its properties . let @xmath48 be a group acting on a set @xmath136 , and let @xmath147 . the subgroup @xmath172 of @xmath48 defined as @xmath173 is called the _ stabilizer _ of @xmath148 in @xmath48 . it is not difficult to see that if @xmath48 is a group which acts on the set @xmath136 , and @xmath147 , then the stabilizer @xmath172 of @xmath148 can be considered as a group which acts on the set @xmath174 . the following are two well known results in group theory : lagrange s theorem which correlates the cardinality of a given group @xmath48 and the cardinality of a given subgroup @xmath154 of @xmath48 with the number of left cosets of @xmath48 with respect to @xmath154 and a theorem which expresses the cardinality of the orbit of @xmath48 containing @xmath148 as the number of left cosets of @xmath48 with respect to the stabilizer @xmath172 . let @xmath48 be a finite group , and let @xmath154 be a subgroup of @xmath48 . then @xmath175 , where @xmath176 is the number of left cosets of @xmath48 with respect to @xmath154 , and is usually called the _ index of @xmath154 in @xmath48_. let @xmath48 be a finite group acting on a set @xmath136 , and let @xmath147 . then @xmath177 , that is there exists a one - to - one correspondence between the elements of the orbit @xmath149 of @xmath148 under @xmath48 and the left cosets of the stabilizer @xmath172 in @xmath48 . [ teo : card - orbit ] we are now able to prove the following theorem . let @xmath48 be a subgroup of the symmetric group @xmath73 , and let @xmath154 be a subgroup of @xmath48 . if the orbit partitions of @xmath48 and @xmath154 are the same , then @xmath178 . [ teo : orbit - partitions ] we prove the assertion by induction on @xmath59 . for @xmath179 the theorem can be proved by direct inspection of the subgroups of @xmath73 . now , let us suppose that the theorem is true for @xmath180 , and let us show that it holds also for @xmath59 . we first observe that since the orbit partitions of @xmath48 and @xmath154 are the same , then also the orbits @xmath181 and @xmath182 of the element @xmath59 with respect to @xmath48 and @xmath154 are the same . now , if we take all the partitions having @xmath183 as a class , we get the orbit partitions induced by the stabilizers @xmath184 and @xmath185 of the element @xmath59 with respect to @xmath48 and @xmath154 . these orbit partitions are equal and thus , by induction hypothesis , @xmath186 . by lagrange s theorem , we can express the cardinalities of @xmath48 and @xmath154 with respect to the cardinalities of their stabilizers as @xmath187 and @xmath188 . where @xmath189 and @xmath190 are the indices , respectively , of the stabilizer @xmath184 in @xmath48 and of the stabilizer @xmath185 in @xmath154 . by theorem [ teo : card - orbit ] , we can infer that @xmath191 and @xmath192 . since @xmath193 and @xmath194 , we can conclude that @xmath48 and @xmath154 have the same order , and thus @xmath195 . theorem [ teo : orbit - partitions ] allows us to show that the orbit partitions of a given database satisfy theorem scheme ii ; in fact , the following theorem provides a first characterization of expressible queries in relational databases alternative to the one originally given by paredaens . let @xmath14 be a relational database , and let @xmath4 be a relation over @xmath8 . then @xmath196 [ teo : scheme - with - orbits ] if @xmath197 , since the orbit partitions are completely determined from the subgroups of @xmath52 , we obtain that @xmath198 . for the converse , we observe that @xmath52 is a subgroup of the symmetric group @xmath73 , and @xmath199 is a subgroup of @xmath52 . by hypothesis , the orbit partitions of @xmath52 and @xmath199 are equal and thus , by theorem [ teo : orbit - partitions ] , @xmath197 . a second characterization of expressible queries in relational databases can be obtained by considering cycle partitions instead of orbit partitions . we need the following lemma . let @xmath48 be a subgroup of the symmetric group @xmath73 , and let @xmath154 be a subgroup of @xmath48 . if the cycle partitions of @xmath48 and @xmath154 are the same , then also the orbits of @xmath48 and @xmath154 are the same , that is @xmath200 for every @xmath201 . [ lem : orbits - from - cycles ] since the orbit @xmath149 is the set of elements of @xmath72 which are reachable from @xmath148 through some element @xmath160 of @xmath48 , while a cycle containing @xmath148 is the set of elements which are reachable from @xmath148 through one element @xmath160 of @xmath48 , one method to build @xmath149 from the cycle partitions of @xmath48 is given by algorithm [ alg:1 ] . result @xmath202 @xmath203 algorithm [ alg:1 ] computes the least subset @xmath204 of @xmath72 which contains @xmath148 and such that , for every cycle partition @xmath103 of @xmath48 , @xmath204 is the union of some cycles in @xmath103 ; it is not difficult to see that @xmath204 is , indeed , the orbit @xmath149 . since the cycle partitions of @xmath48 and @xmath154 are the same by hypothesis , the orbits computed by the algorithm above will be the same for @xmath48 and @xmath154 , for every choice of @xmath205 . we are now ready to prove the following theorem . let @xmath48 be a subgroup of the symmetric group @xmath73 , and let @xmath154 be a subgroup of @xmath48 . if the cycle partitions of @xmath48 and @xmath154 are the same , then @xmath178 . [ teo : cycle - partitions ] by lemma [ lem : orbits - from - cycles ] , the orbits of @xmath48 and @xmath154 are the same . thus we can prove the theorem by the same argument used for theorem [ teo : orbit - partitions ] . a direct consequence of theorem [ teo : cycle - partitions ] is that the cycle partitions of a given database satisfy theorem schema ii ; thus , the following theorem provides a second characterization of expressible queries in relational databases alternative to the one originally given by paredaens . the proof is analogous to the one given for theorem [ teo : scheme - with - orbits ] . let @xmath14 be a relational database , and let @xmath4 be a relation over @xmath8 . then : @xmath206 [ teo : scheme - with - cycles ] a final observation is due about theorems [ teo : scheme - with - orbits ] and [ teo : scheme - with - cycles ] . even though there is a strong resemblance between our meta theorem [ thm : scheme1 ] and theorem [ teo : paredaens2 ] , our results _ can not _ be expressed neither in the form @xmath207 and @xmath208 nor in the form @xmath209 and @xmath79 , as shown in the next example . let @xmath210 and @xmath211 be two relational databases , with : * @xmath212 * r = + [ cols="^,^,^,^,^ " , ] notice that @xmath54 is the cyclic group generated by the permutation @xmath213 , while @xmath214 is the cyclic group generated by the permutation @xmath215 : @xmath216 from @xmath54 and @xmath214 we can easily obtain @xmath217 . clearly , @xmath4 is not expressible from @xmath0 , since we have @xmath218 but @xmath219 ; on the other hand , @xmath220 and @xmath221 , and @xmath222 and @xmath221 . the fact that @xmath4 is not expressible from @xmath0 can be correctly determined through orbit partitions or through cycle partitions by observing that : @xmath223 or @xmath224 . in this section we study a simple graph - based model where two labeled graphs are used to model data bases . a data base consists of two distinct layers : a _ schema _ layer and a _ structure _ layer ; the objects can be found in the latter , while the former describe the data organization . each layer is a labeled weakly - connected directed graph , moreover there exists a function that maps a schema into a structure : such function will be called an _ extension_. both vertices and edges of the graphs are labeled , and we can assume that the sets of edge labels and vertex labels , as well as schema labels and structure labels , are disjoint . an example of data base is represented in figures [ fig : schema ] , [ fig : structure ] , from which it is easy to note how the schema and the structure are closely related , the following definitions only formalize the intuitive idea . a _ schema graph _ , in short _ schema _ , is a triple @xmath225 , where @xmath226 is an oriented , weakly - connected graph , and @xmath227 , @xmath228 are respectively the injective functions that maps each node ( resp . edge ) to its label . [ def_struttura ] a _ structure _ is a triple @xmath229 , with @xmath4 a colored oriented graph @xmath230 , where @xmath231 is the set of _ nodes _ of the structure , @xmath232 is the set of _ edges _ , @xmath233 , @xmath234 are respectively the injective functions that maps each node ( resp . edge ) to its label , and @xmath235 , is a labeling of the edges over the finite alphabet @xmath236 , called _ coloring _ of the structure . in the following , we will use the set @xmath237 of colors that allows to specify that a link between object instances in @xmath238 is actual or not . in the example of fig . [ fig : structure ] , only the links labeled true are represented , and the presence ( or the abscence ) of links labeled false does not change the data stored in the data base . in fig . [ fig : false ] is represented a part of the structure , where false links are represented with dotted arrows . the schema and the structure must be strongly correlated ; in fact there must exist a function , called _ extension _ ( denoted by @xmath239 ) , mapping the schema into the structure . in order to have a sound definition of extension some restrictions must be enforced , as pointed out in the following definition , where @xmath240 stands for the family of all nonempty subsets of @xmath241 . informally @xmath239 maps each vertex of the schema into some vertices of the structure and each edge of the schema into some edges of the structure . [ def_f1 ] let @xmath242 be a schema and @xmath243 a structure , where @xmath244 . then @xmath238 is an _ extensional structure _ of @xmath245 if there is a function ( the _ extension _ ) from @xmath245 to @xmath238 , @xmath246 , such that : [ condizioni_rel ] 1 . @xmath247 is a partition @xmath248 of the set @xmath249 , 2 . for every @xmath250 , the pair @xmath251 iff @xmath252 ; notice that the first point of the definition of extension implies that the function @xmath253 is well defined . in the following , if @xmath238 is the extensional structure of @xmath245 , then we write @xmath254 and we will simply say that @xmath238 is a structure of @xmath245 . given two vertices @xmath255 and @xmath256 of the schema , connected with a link @xmath257 then in the structure there must exist all links @xmath258 for @xmath259 , @xmath260 . such requirement justifies the introduction of a labeling ( and especially of a true - false labeling ) in order to have a reasonable graph - based model . [ basedati ] a _ data base _ @xmath261 is a pair @xmath262 , where @xmath245 is a schema and @xmath238 is an extensional structure of @xmath245 . the schema describes the conceptual organization of the data , while the data content or instantiation of the data base is given by the extensional structure associated to the schema . it is not hard to notice that , given a schema , there is a one - to - one correspondence between structures and extension functions , therefore we will sometimes use the pair @xmath263 as a data base . some preliminary definitions are required for introducing our query language . given a partial function @xmath264 ( i.e. a function where each element of @xmath241 can be associated to one or none of the elements of @xmath261 ) , by @xmath265 we denote the domain of @xmath266 , that is the set of elements @xmath267 such that @xmath268 is defined . let @xmath269 be two partial functions from the set @xmath241 to the set @xmath270 . then _ @xmath266 is a restriction of @xmath160 _ , denoted by @xmath271 , if @xmath272 and for every @xmath273 , @xmath274 . moreover by @xmath275 we denote the set obtained as union of all images of elements in @xmath276 : formally @xmath277 . [ morfi ] let @xmath278 be a data base . an _ instance _ of @xmath261 is a restriction @xmath266 of @xmath239 such that @xmath265 induces a weakly - connected subgraph of @xmath245 . the following notations will be used in the rest of the paper . the set @xmath279 is the set of all instances of @xmath261 . let @xmath280 be a subset of @xmath279 , then by @xmath281 , we mean the set of nodes of the structure of @xmath261 that is the union of all images of instances in @xmath282 , while @xmath283 is the union of all domains of instances in @xmath282 . an element in @xmath284 is called a _ value _ , while an element in @xmath285 is called a _ name_. then the image of a name @xmath286 is the subset @xmath241 of @xmath284 such that @xmath287 . for a value @xmath288 , the _ inverse image of @xmath289 _ , denoted by @xmath290 , is the name of @xmath285 that is mapped by @xmath239 , to a set containing the element @xmath289 . similarly , given a set @xmath241 of values , the _ inverse image of @xmath241 _ is the set @xmath291 which is union of all inverse images of the values in @xmath241 . our graph data model is proposed as a domain - preserving data base , along the same lines as other papers where the expressiveness of the relational algebra is studied @xcite , and it gives a formal embedding for languages used for the retrieval of graph - structured information @xcite . the requirement that we are dealing with domain - preserving data bases reflects in the query language : in fact we have no operation for creating new elements or modifying the schema graph , and all operations must preserve the schema and the original structure . the main consequence of the assumption that our model is domain preserving consists in the fact that we will deal with a schema which is mapped to an instance through an extensional mapping . therefore there is a complete equivalence between subgraphs of the structure and restrictions of the extensional mapping . we are now able to introduce the operations of our graph algebra : according to our reasoning above we can describe the operation as over partial functions whenever it allows a simpler formulation . [ sovrapposizione ] let @xmath292 be a data base and let @xmath293 . the _ addition _ of @xmath294 and @xmath295 , denoted as @xmath296 , is the following function over domain @xmath297 . the operation is defined only if @xmath298 : @xmath299 [ prodotto ] let @xmath292 be a data base , and let @xmath300 be two functions in @xmath279 . the _ product _ of @xmath294 and @xmath295 , denoted as @xmath301 is the instance in @xmath279 defined as follows : @xmath302 the product is defined only if @xmath303 induces a weakly - connected subgraph of @xmath245 . [ proiezione ] let @xmath292 be a data base . let @xmath266 be a function in @xmath279 and let @xmath241 be a subset of the domain of @xmath266 , such that @xmath241 induces in @xmath245 a weakly - connected subgraph . the _ projection of @xmath266 on @xmath241 _ , denoted as @xmath304 , is the instance defined as follows : @xmath305 [ differenza ] let @xmath292 be a data base . let @xmath300 be two functions in @xmath279 over the same domain @xmath241 . the _ difference of @xmath294 by @xmath295 _ , denoted as @xmath306 is the following instance : @xmath307 the difference is defined only if @xmath308 induces a weakly - connected subgraph of @xmath245 . since the coloring of the edges encodes the fact that a relation between two objects is actual or not , it is natural that the query language has some tools for exploiting such coloring . in our model we will need to extract instances where `` similar '' edges are the same color . the definition of _ selector _ is the first step in such direction . let @xmath245 be the schema of a data base @xmath261 . then a _ selector of @xmath245 _ is a pair @xmath309 consisting of a weakly - connected subgraph @xmath48 of @xmath245 and a coloring @xmath310 of the edges of @xmath48 . querying for a selector in a data base returns all subgraphs of the structure that are isomorphic to the selector : each such subgraph is indeed called a _ simple instance_. moreover , it is natural to define an operation of selection that allows to obtain instances which are compatible with a coloring of the schema over the alphabet @xmath236 . this is the last operation of our algebra . [ simple ] let @xmath263 be a data base , where @xmath311 . let @xmath312 be a selector of @xmath245 , where @xmath313 . then a _ simple instance induced by the selector @xmath314 _ is a restriction @xmath266 of @xmath239 such that @xmath315 , @xmath316 for each @xmath317 and @xmath318 for each @xmath319 . notice that all simple instaces have the same domain . [ selezione ] let @xmath292 be a data base . let @xmath266 be a function in @xmath279 and @xmath314 a selector . let @xmath320 be the set of all simple instances induced by @xmath321 that are also subinstances of @xmath266 . the _ selection of @xmath322 by @xmath314 _ , denoted as @xmath323 , is @xmath324 . given a set @xmath280 of instances , our first aim is to give a characterization of all instances that can be obtained with a query that uses only the information contained in the instances in @xmath280 , or equivalently by an expression of the algebra that has only instances in @xmath280 as operands . in such direction the main result of this section is that expressiveness in our graph algebra is equivalent to the conservation of a certain partition . it is natural to associate a notion of _ undistinguishability _ to a partition , where all elements in a set of the partition are deemed undistinguishable . we share the goals of @xcite , but we have introduced in this paper a new framework , that is we are looking for a notion of expressiveness that is coherent with our meta theorem . just as the notion of _ automorphisms _ , introduced in @xcite for relations , gives a global description of the logical dependencies among data that must be preserved when querying the data base , in our model a partition ( or an equivalence relation ) will represent all such logical dependencies . the equivalence partition over elements of the structure that we will study is called _ stability _ and is denoted by @xmath325 ( where @xmath326 is an instance ) . the simplest possible form of undifferentiation ( called _ 0-stability _ ) is based on the idea that we are able to distinguish images of different vertices of the instance and vertices of the extensional structure belonging to different functions of @xmath326 . such notion basically consists of using expressions in our algebra that do not contain any selection . [ def : split ] let @xmath241 be a subset of the image of a set @xmath282 of instances . then @xmath241 is _ split _ by @xmath282 iff there is a function @xmath327 in @xmath282 such that @xmath328 and @xmath329 . we are now able to introduce formally the definition of 0-stability , as follows : [ 0-stable ] let @xmath282 be a set of instances over a data base @xmath262 , and let @xmath241 be a subset of @xmath281 . then @xmath241 is _ 0-stable _ w.r.t . @xmath280 , if the two following conditions hold : 1 . @xmath330 , where @xmath331 , and @xmath332 is a name of the schema . 2 . for each function @xmath333 , then @xmath241 and @xmath334 are disjoint or one is contained into the other one . a more refined notion of undifferentiation is called _ @xmath335-stability _ ; informally a set @xmath241 is 1-stable w.r.t . @xmath261 if @xmath241 is 0-stable and @xmath261 is not able to distinguish two vertices of @xmath241 with edges outgoing from @xmath261 and ingoing in @xmath241 ( or outgoing from @xmath241 and ingoing in @xmath261 ) . notice that 1-stability is a binary relation over subsets of @xmath284 , while 0-stability is a unary relation . the formal definition is : [ un_passo ] let @xmath282 be a set of instances over a data base @xmath262 and let @xmath241 and @xmath261 be two disjoint subsets @xmath281 . then @xmath241 is _ 1-stable _ w.r.t . @xmath261 and @xmath280 , denoted as @xmath336 if the following conditions are verified : 1 . @xmath241 is 0-stable w.r.t . @xmath282 ; 2 . for each edge @xmath337 of @xmath338 , with @xmath339 and for each @xmath340 there exists @xmath341 such that @xmath342 ; 3 . for each edge @xmath343 of @xmath338 , with @xmath339 and for each @xmath340 there exists @xmath341 such that @xmath344 . informally 1-stability means that whenever there is an colored edge ( say a red edge ) from a vertex of @xmath241 to a vertex of @xmath261 , then all vertices of @xmath241 have a red edge ingoing in @xmath261 . in other words if we assume that @xmath261 is undistinguishable , then also @xmath241 is undistinguishable , by any single - edge path . the notion of 1-stability can be further generalized , but first we need a new definition . [ cammino ] let @xmath345 be a labeled graph . then a _ colored path _ in @xmath48 is a pair @xmath346 where @xmath347 , and for every @xmath348 , @xmath349 belongs to @xmath231 and @xmath350 is an edge of @xmath48 such that @xmath351 or @xmath352 . moreover @xmath353 is the sequence @xmath354 where @xmath355 is the color of the edge @xmath350 in @xmath48 . notice that the definition of path used in the paper is different from the one that can be usually found in a graph theory textbook , as arcs can also be in the reverse direction . the _ length _ of a path is the number of edges it contains . let @xmath356 be a data base and let @xmath346 be a colored path of @xmath338 , with @xmath357 . then the _ path schema _ of @xmath346 is the pair @xmath358 , with @xmath359 , where for every @xmath360 , @xmath361 and @xmath362 . [ path - dependencies ] let @xmath363 be two nodes of the @xmath280-structure , and let @xmath364 be a subset of @xmath281 , then the _ path dependencies _ from @xmath148 to @xmath289 in @xmath364 , denoted as @xmath365 is the set of path schemata of all paths of the @xmath280-structure that are starting in @xmath148 and ending in @xmath289 and entirely contained in @xmath364 . informally given @xmath366 , their path dependencies is obtained by removing all vertices not in @xmath364 , then computing all possible paths from @xmath148 to @xmath289 , and finally computing the respective path schemata . the next step is to generalize 1-stability to @xmath25-stability , that is taking into account length-@xmath25 paths instead of simple edges ( that is length-1 paths ) . [ k_passi ] let @xmath282 be a set of instances over a data base @xmath262 , let @xmath25 be an integer larger than one , and let @xmath241 and @xmath261 be two disjoint subsets of @xmath281 . then @xmath241 is _ @xmath25-stable _ w.r.t . @xmath261 and @xmath280 , denoted as @xmath367 if the following conditions are verified : 1 . @xmath241 is 0-stable w.r.t . @xmath282 ; 2 . @xmath241 is @xmath368-stable w.r.t . @xmath261 and @xmath280 ; 3 . for each @xmath369 there exists @xmath341 such that @xmath370 . the main idea is that when @xmath367 then if @xmath261 is undistinguishable also @xmath241 is undistinguishable when only paths no longer than @xmath25 are taken into account . our main definition follows : [ rel_collettiva ] let @xmath280 be a set of instances , and let @xmath371 be two disjoint subsets of @xmath281 . then @xmath241 is _ stable w.r.t . @xmath261 _ in @xmath280 , denoted as @xmath372 , if @xmath373 for all @xmath374 . by def . [ rel_collettiva ] , it is immediate to verify the following properties of stability : [ prop1_rel ] let @xmath280 be a set of instances , and let @xmath375 , then : 1 . if @xmath376 and @xmath377 , then @xmath372 and @xmath378 , 2 . if @xmath379 , @xmath372 and @xmath378 , then @xmath380 , 3 . if @xmath379 , @xmath381 , @xmath382 and @xmath383 , then @xmath384 . stability is a relation between disjoint subsets of the domain . the definition of expressiveness in the query language that we want to obtain is based on partitions , and now we are able to introduce the class of partitions we are interested into . a partition @xmath385 of nodes of the structure is called _ valid _ if and only if for each set @xmath241 of the partition and every set @xmath261 that is a union of sets of @xmath385 , then @xmath261 can not differentiate @xmath241 . [ partizione_accettabile ] let @xmath280 be a set of instances . a partition @xmath386 of @xmath387 is _ valid _ if for every @xmath388 , then @xmath389 . given a set @xmath282 of instances , then there may be various valid partitions of @xmath281 , and at least one valid partition always exists ( the partition where each vertex of the extentional structure is a set ) . some valid partitions are more representative of the actual undifferentiation , in fact we will assume as a measure of the undifferentiation induced by @xmath280 the coarsest valid partition , which we will call _ canonical partition _ and denote as @xmath390 . we can show that the definition of canonical partition is well - formed . every set @xmath280 of instances has a unique canonical partition @xmath390 . clearly the partition of @xmath281 into singletons is a valid partition , so there exists at least one canonical partition . now assume to the contrary that there exist two coarsest valid partitions @xmath391 and @xmath392 . let @xmath393 and @xmath394 be the equivalence relations induced by the partitions @xmath391 and @xmath392 , respectively . let @xmath395 be the transitive closure of the relation @xmath0 defined as follows : @xmath396 if and only if @xmath148 and @xmath289 are in the same set of @xmath391 or @xmath392 . we can prove that the partition @xmath385 induced by @xmath395 is a valid one of index strictly less than @xmath25 . by construction of @xmath395 each set of @xmath385 is a union of sets in @xmath391 and also a union of sets in @xmath392 , moreover each set in @xmath385 is contained in the image of a single name ( since each set must be 0-stable ) . notice that @xmath397 iff @xmath398 . let @xmath399 be a set of @xmath385 , and let @xmath400 be a class of @xmath391 contained in @xmath399 . since @xmath391 is a valid partition , and @xmath399 is a union of disjoint sets of @xmath401 , by lemmata [ rel_collettiva ] , [ prop1_rel ] , we have that @xmath402 and @xmath403 . let @xmath399 , @xmath404 be two sets of @xmath385 , with @xmath405 and @xmath406 . we have already proved that @xmath407 and @xmath408 , applying again lemmata [ rel_collettiva ] , [ prop1_rel ] and noting that @xmath406 we obtain @xmath409 , by the generality of @xmath399 and @xmath404 the partition @xmath385 is valid . moreover @xmath385 is coarser than @xmath401 , which is a contradiction . in this section we will prove our main result regarding the expressiveness of the graph - based query language by showing that a function can be computed if and only if adding such function does not change the canonical partition . in the following we will assume that the union of the images of all functions in @xmath326 is exactly the universe set ; such assumption does not violate the generality since otherwise we would simply have some sets of the canonical partition whose union consists of exactly those elements of the universe set that are not in any function in @xmath326 . [ graal - expressivess ] let @xmath410 be the set of functions that are a result of an expression of the graph algebra where only functions of @xmath280 are operand . then @xmath411 if and only if the canonical partition induced by @xmath280 is equal to the canonical partition induced by @xmath412 , that is @xmath413 . the following two properties , that are consequences of def . [ partizione_accettabile ] , will be useful to prove the main result of the paper . [ f_1 ] let @xmath280 be a set of instances and let @xmath385 be a valid partition . then the image of every instance @xmath414 is the union of sets of @xmath385 . we prove the lemma by induction on the number @xmath59 of operations of the expression for @xmath415 . if @xmath416 , then @xmath415 is a function in @xmath417 . since all sets of a valid partition are @xmath418-stable , no set of a valid partition has both an element in @xmath419 and an element not in @xmath419 , therefore the union of all sets of @xmath385 that are contained in @xmath419 is contained in @xmath419 . to prove that such containment is not strict ( i.e. such union is equal to @xmath419 ) it is sufficient to note that all elements of @xmath419 belong to some set of @xmath385 . assume now that @xmath420 and @xmath415 is obtained by the application of an operation to two expressions @xmath421 and @xmath422 in @xmath410 , or one expression for a selection . clearly , by inductive hypothesis the images of @xmath421 and @xmath422 are obtained as the union of some sets of @xmath385 . it is immediate to verify the lemma for the case that @xmath423 , @xmath424 , @xmath425 and @xmath426 . finally , assume that @xmath427 , where @xmath321 is a selector . by definition of the selection , then @xmath419 is the union of the images of all simple instances induced by @xmath321 . assume to the contrary that there exists a set @xmath241 of the valid partition such that @xmath428 and @xmath429 . now , let @xmath430 and @xmath431 . hence , by construction of selection , @xmath289 can not be in the image of any simple instance induced by @xmath321 , while @xmath148 is contained in the image of a simple instance induced by @xmath321 . by inductive hypothesis @xmath432 is union of sets of @xmath385 , moreover since @xmath241 is a set of @xmath385 , also @xmath433 is union of sets of the valid partition , implying that @xmath434 , it follows that , for each @xmath435 , @xmath436 for some @xmath437 . this implies that there is a simple instance induced by @xmath321 that has in its image @xmath289 , which is a contradiction with the above assumption . consequently , the image of @xmath415 must be union of sets of @xmath385 . we will prove that an alternative characterization of canonical partition is as the partition induced by the equivalence relation @xmath438 between elements of @xmath439 , where @xmath440 if and only if for every instance @xmath441 , @xmath442 . in the following of the paper let @xmath443 denote the partition induced by the equivalence relation @xmath438 . successively we will prove that a function @xmath322 belongs to @xmath410 if and only if @xmath275 can be obtained as union of sets of @xmath443 , completing the proof of our main result , in two steps . first we will prove that a function @xmath266 belongs to @xmath444 if and only if @xmath445 is union of sets in @xmath443 , then we will prove that @xmath446 . the following proposition is an immediate consequence of the definitions of @xmath440 and of projection . [ unique1 ] let @xmath447 such that @xmath440 . then both @xmath148 and @xmath289 belong to the set @xmath448 for some name @xmath449 . [ unique ] let @xmath450 be the partition induced by a set @xmath280 of instances . then , the inverse image of every set of the partition consists of a single vertex . [ minimo_ottenibile ] let @xmath280 be a set of instances over @xmath261 , let @xmath443 be the partition induced by @xmath282 , and let @xmath451 . then there exists an instance @xmath441 such that @xmath452 . let @xmath453 be the set of functions @xmath454 such that @xmath455 . by cor . [ unique ] all functions in @xmath453 have the same domain , therefore the expression @xmath456 is well - formed ; by construction @xmath457 . by definition of @xmath443 all functions @xmath458 are such that @xmath459 or @xmath460 , therefore all functions @xmath458 whose image intersect @xmath241 are such that @xmath455 , which in turn implies that are also in @xmath453 . hence @xmath461 , for otherwise there would be an element of @xmath241 not belonging to the image of any function in @xmath326 . [ uni2 ] let @xmath326 be a set of instances over @xmath261 , let @xmath450 be the partition induced by @xmath282 and let @xmath241 be a union of sets in @xmath443 , such that the inverse image of @xmath241 induces a weakly - connected subgraph of the schema , then @xmath241 is the image of an instance @xmath441 . let @xmath462 be the sets of @xmath443 whose union is @xmath241 , and notice that , by lemma [ minimo_ottenibile ] , it is possible to associate to each set @xmath463 the instance @xmath464 whose image is @xmath463 , moreover for each such @xmath465 , @xmath466 . for each vertex @xmath148 in the inverse image of @xmath241 we can construct the function @xmath467 as @xmath468 . then let @xmath469 ; it is immediate to note that @xmath461 . an immediate consequence of lemmata [ f_1 ] and [ uni2 ] is the following : [ canonical = obtainable ] let @xmath326 be a set of instances over @xmath261 and let @xmath333 . then @xmath470 if and only if @xmath445 is union of sets in @xmath443 and the inverse image of @xmath445 induces a weakly - connected subgraph of the schema . with lemma [ uni2 ] we have proved that all interesting unions of sets of the partition @xmath450 can be obtained with an expression of the graph algebra where all operands are taken from @xmath280 , therefore @xmath450 conveys all expressibility information . but @xmath450 is defined on the set @xmath410 , we still need to correlate the definition of canonical partition with that of @xmath450 . [ pbi_accettabile ] let @xmath280 be a set of instances . then @xmath450 is a valid partition of @xmath280 . let @xmath371 be two disjoint sets where @xmath471 and @xmath261 is union of sets in @xmath443 , we will prove that @xmath372 . first of all we will show that @xmath241 is 0-stable . remember that , by definition of @xmath443 , for each @xmath470 either @xmath472 or @xmath473 . since @xmath474 contains @xmath326 , it is immediate to not that @xmath241 is 0-stable . in the following let @xmath10 be the single - vertex inverse image of @xmath241 . if the inverse image of @xmath475 does not induce a weakly - connected subgraph of the instance , then 0-stability of @xmath241 suffices to prove that @xmath372 , therefore assume that the inverse image of @xmath475 induces a weakly - connected subgraph of the instance . let us assume that @xmath372 does not hold , then we will get a contradiction . without loss of generality we can assume that @xmath261 is a minimum set for which @xmath372 does not hold . the new assumption implies that @xmath25-stability does not hold for some @xmath25 . it follows that there are two elements @xmath476 and an element @xmath477 such that @xmath478 is not contained in @xmath479 , for every @xmath480 . now , by lemma [ uni2 ] , there is an instance @xmath414 such that @xmath481 . let @xmath482 and @xmath483 and let @xmath480 such that @xmath484 . let us consider the function @xmath485 . by construction @xmath148 belongs to @xmath486 , but @xmath289 does not ; since @xmath487 we have found a function in @xmath474 containing @xmath148 but not @xmath289 , contradicting the assumption that @xmath488 . [ pd = pbi ] let @xmath280 be a set of instances . then @xmath489 . by lemma [ pbi_accettabile ] , @xmath490 must be a valid partition of @xmath280 . by lemma [ minimo_ottenibile ] , every set @xmath491 is obtained as the image of some instance @xmath441 . clearly , since @xmath441 , by lemma [ f_1 ] the image @xmath241 of @xmath322 is the union of sets of the canonical partition of @xmath280 . hence @xmath492 . [ uguale1 ] let @xmath280 be a set of instances . then an instance @xmath415 belongs to @xmath474 if and only if @xmath493 . clearly by lemma [ pd = pbi ] it suffices to show that @xmath414 if and only if @xmath494 , moreover it is immediate to note that , by construction of @xmath443 , if @xmath414 then @xmath494 . assume now that @xmath494 . by lemma [ pbi_accettabile ] , @xmath495 must be a valid partition . by lemma [ f_1 ] , and exploiting the assumption that @xmath496 , for each @xmath497 , @xmath498 must be the union of some sets in @xmath499 , that is @xmath500 , for some sets @xmath501 . but by lemma [ minimo_ottenibile ] , for each set @xmath502 there is an instance @xmath503 such that @xmath502 is the image of @xmath504 . consequently , @xmath505 , and hence @xmath506 , which proves that @xmath414 as required . theorem [ uguale1 ] and lemma [ pd = pbi ] lead to our main result . [ cor : uguale1 ] let @xmath280 be a set of instances . then an instance @xmath415 belongs to @xmath474 if and only if @xmath507 . we have introduced the idea that partitions of the domain set can be used for characterizing the set of relations or graphs that can be extracted in a data base in the relational or in a graph - based model . by formally proving those expressiveness results we have effectively given a new framework for the analysis of data base query languages . s. abiteboul , c. beeri , m. gyssens , and d. van gucht . an introduction to the completeness of languages for complex objects and nested relations . in _ nested relations and complex objects _ , volume 361 of _ lncs _ , pages 117138 , 1989 . m. andries and j. paredaens . a language for generic graph transformations . in _ graph - theoretic conecpts in computer science , 17th international workshop , wg91 _ , volume 570 of _ lncs _ , pages 170183 , berlin , 1991 . springer - verlag . f. bancilhon . on the completeness of query languages for relational data bases . in _ proceedings , 7th symp . on mathematical foundations of computer science _ , volume 64 of _ lncs _ , pages 112123 , berlin , 1978 . springer - verlag . j. van den bussche , d. van gucht , m. andries , and m. gyssens . on the completeness of object - creating query languages . in _ 33rd annual symposium on foundations of computer science . proceedings _ , pages 372379 , 1992 .

since its prediction in 1954 @xcite , the effect of superradiance has attracted a lot of theoretical and experimental attention @xcite . its basic principle the fact that a coherent interaction of an unexcited gas with the vacuum of a common field can create a spontaneous macroscopic excitation of both matter and field subsystems appears in various incarnations in diverse branches of physics @xcite . the dicke model @xcite of superradiance resorts to a maximal simplification of the problem to capture the main features of the superradiant transition in the most transparent way . the model shows a thermal phase transition , analyzed and discussed in refs.@xcite , as well as a zero - temperature ( ground - state ) quantum phase transition ( qpt ) , which was realized experimentally with the aid of a superfluid gas in a cavity @xcite . recent theoretical analyses showed that the model exhibits also a novel type of criticality the one observed in the spectrum of excited states @xcite . these so - called excited - state quantum phase transitions ( esqpts ) affect both the density of quantum levels as a function of energy and their flow with varying control parameters , and are present in a wide variety of quantum models with low numbers of degrees of freedom @xcite . in this work , we analyze properties of a simple superradiance model interpolating between the familiar dicke @xcite and tavis - cummings @xcite hamiltonians . a smooth crossover between both limiting cases is achieved by a continuous variation of a model parameter , which allows one to observe the system s metamorphosis on the way from the fully integrable ( hence at least partly understandable ) to a prevailingly chaotic ( so entirely numerical ) regime . one of the aims of our work is to survey the phase - transitional properties of the extended model and to investigate the nature of its _ quantum phases_the domains of the excited spectrum in between the esqpt critical borderlines . a usual approach to characterize different phases related to the ground state of a quantum system makes use of suitable order parameters , i.e. , expectation values of some observables . we show that an unmistakable characterization of phases involving excited states is not achieved through the expectation values alone but rather through their different smoothed energy dependences ( trends ) . the main emphasis of our analysis is nevertheless put on the entanglement properties of excited states across the whole spectrum and their potential links to esqpts and quantum phases of the model . it is known that a continuous ground - state qpt in many models ( including the present one ) is characterized by a singular growth of entanglement within the system , which can be seen as a quantum counterpart of the diverging correlation length in continuous thermal phase transitions @xcite . a question therefore appears whether there exist any entanglement - related signatures of esqpts . the extended dicke model is rather suitable for a case study of this type since it allows one to analyze at once various types of entanglement that between the field and all atoms , and that between individual atoms . it should be stressed that robust , systematic studies of entanglement in excited spectra of any systems are rather scarce . we hope that results of our analysis , which is unavoidably model - specific , will initiate similar studies in other relevant systems . the plan of the paper is as follows : basic quantum and classical features of the model are described in sec.[edm ] . thermal and quantum critical properties and a classification of thermodynamic and quantum phases are presented in sec.[pt ] . the atom field and atom atom entanglement properties are investigated in sec.[ent ] . conclusions come in sec.[sum ] . consider single - mode electromagnetic field with photon energy @xmath0 ( polarization neglected ) interacting with an ensemble of @xmath1 two - level atoms , all with the same level energies @xmath2 . the free hamiltonian reads as @xmath3 where the first term describes the field of photons , created and annihilated by bosonic operators @xmath4 and @xmath5 , respectively , and the second term represents the ensemble of atoms , with @xmath6 standing for the third pauli matrix acting in the space of the @xmath7th atom . in the following we use all pauli matrices @xmath8 with the subscripts @xmath9 or @xmath10 . the interaction term is taken in the dipole approximation , with the dipole operator for the @xmath7th atom proportional to @xmath11 . the size of the atomic ensemble is assumed to be much smaller than the wavelength of photons ( cavity size ) . then @xmath12 where the coupling parameter ( in units of energy ) can be written as @xmath13 , with @xmath14 standing for the electric dipole moment matrix element , @xmath15 for the vacuum permitivity , and @xmath16 for the cavity volume . the hamiltonian @xmath17 represents the well known dicke model @xcite . it is commonly invoked to explain basic features of the transition of matter to superradiant phase @xcite . indeed , as shown below , if the coupling strength @xmath18 is sufficiently increased , the superradiant phase with @xmath19 and @xmath20 is created at low temperatures . note that the diamagnetic term @xmath21 is neglected here though it turns out to be very important in the discussion of superradiance in realistic situations connected to cavity electrodynamics @xcite . we emphasize that recent experimental realizations of the dicke hamiltonian @xcite do not implement the above electrodynamic content , nevertheless in this presentation we keep this original formulation as a widely comprehensible background . introducing collective quasi - spin operators for the whole atomic ensemble ( which is possible due to the long wavelength approximation ) , @xmath22 one immediately sees that the dicke hamiltonian conserves the squared quasi - spin @xmath23 with eigenvalues @xmath24 , where @xmath25 is integer for @xmath1 even or half - integer for @xmath1 odd . the full atomic hilbert space is the span of all @xmath26 possible configurations of atoms , thus @xmath27 , where @xmath28 is the space of the @xmath7th atom . however , due to the conservation of @xmath29 , the dynamics can be investigated separately in any of the single-@xmath25 subspaces @xmath30 with dimension @xmath31 . the decomposition reads as follows @xcite : @xmath32 where @xmath33 enumerates replicas ( their number is @xmath34 ) of the space with given @xmath25 differing by the exchange symmetry of atomic components . in the following , we will investigate thermodynamic properties of the model in the full atomic space @xmath35 , as well as quantum properties in a single-@xmath25 space @xmath36 . the most natural choice in the latter case is the unique ( @xmath34=1 ) subspace with maximal value @xmath37 , which is fully symmetric under the exchange of atoms and therefore emphasizes the collective character of the superradiance phenomenon . a general-@xmath25 subspace has a mixed exchange symmetry such that only a number of @xmath38 atoms can be excited independently , while excitations of the remaining @xmath39 atoms have to compensate each other ( in ref.@xcite the quantum number @xmath25 is called a cooperation number of the atomic gas ) @xcite . the reduced single-@xmath25 model has only two effective degrees of freedom , one associated with the bosonic field , the other with the su(2 ) algebra of collective quasi - spin operators ( [ jbul ] ) , hence @xmath40 . in contrast , in the full ( all-@xmath25 ) model each su(2)@xmath41 algebra of atom @xmath7 brings an independent degree of freedom , so the whole atom field system has @xmath42 . the dicke model is sometimes simplified by omitting the terms @xmath43 and @xmath44 , that for @xmath45 and very small @xmath18 yield negligible contributions to the transition amplitudes @xcite . the reduced model , in case of @xmath46 atoms called the tavis - cummings model @xcite , has an advantage of being completely integrable because of the existence of an additional integral of motion : @xmath47 where @xmath48 is the number of bosons and @xmath49 the number of excited atoms ( taking values @xmath50 $ ] , where @xmath51 is the @xmath52 quantum number ) . this conservation law follows from a u(1 ) symmetry under gauge transformations consisting of a phase shift @xmath53 and a simultaneous rotation @xmath54 , where @xmath55 is the rotation matrix by angle @xmath56 around axis @xmath57 . the quantity ( [ m ] ) is not conserved in the dicke model , although the full hamiltonian has a discrete z(2 ) symmetry with conserved parity @xmath58 . in this work , we analyze properties of the single - mode superradiance model across the transition between the tavis - cummings and dicke regimes . following refs.@xcite , we introduce a second interaction hamiltonian , @xmath59 where @xmath60 $ ] is an additional control parameter , and compose the total hamiltonian as @xmath61 , so : @xmath62 the tavis - cummings and dicke limits of the model are obtained for @xmath63 and 1 , respectively . we see that an increase of @xmath64 from zero breaks the u(1 ) symmetry while preserving the z(2 ) symmetry of the @xmath65 model . extended dicke model as a function of @xmath18 for @xmath66 and @xmath67 . the panels correspond to ( a ) the tavis - cummings limit @xmath68 , ( b ) intermediate regime @xmath69 and ( c ) the dicke limit @xmath70 . ] figure [ f_spe ] shows quantum spectra of hamiltonian with @xmath37 depending on the interaction parameter @xmath18 for three different values of @xmath64 . the evolution of the ground state , namely its sudden drop to lower energies above a certain value of @xmath18 , indicates the superradiant transition at zero temperature . we note that the eigensolutions of the model for a given @xmath25 are in general obtained by a numerical diagonalization of hamiltonian in the tensor product of atomic and field hilbert spaces , @xmath71 , where @xmath72 is the finite atomic space from eq . spanned by basis vectors @xmath73 with quasi - spin projections @xmath74 and @xmath75 is the infinite - dimensional space of the boson field spanned by vectors @xmath76 with boson numbers @xmath77 . since the diagonalization procedure requires a truncation of @xmath75 , the eigensolutions must be carefully checked for convergence . a general state vector of the model for any fixed @xmath25 is expressed as @xmath78 where @xmath79 are expansion coefficients satisfying the normalization condition @xmath80 . note that for even- and odd - parity states , respectively , the sums in eq . go either over even or odd values of @xmath81 . any vector can be visualized as a wave function @xmath82 , where the angle @xmath83 comes from the quasi - spin representation of the atomic subsystem and variable @xmath84 from the oscillator representation of the bosonic field . for integer @xmath25 ( even @xmath1 ) , the wave function can be obtained by substitutions ( up to normalization constants ) @xmath85 and @xmath86 , with @xmath87 standing for the hermite polynomial . for half - integer @xmath25 ( odd @xmath1 ) , however , the wave function @xmath82 acquires a spinorial character . a possible representation can be obtained by coupling an integer angular momentum @xmath88 ( the even core of @xmath89 atoms ) with @xmath90 ( the odd atom ) to the total @xmath25 as in the case of spinor spherical harmonics . this leads to a mapping of @xmath73 to a 2-valued function @xmath91 , which is @xmath92-periodic , with @xmath93 denoting the corresponding clebsch - gordan coefficients . examples of squared wave functions of the @xmath37 hamiltonian eigenstates for an even @xmath1 in the medium-@xmath64 regime are depicted in fig.[f_wav ] . they are taken at the parameter and energy values indicated by dots in fig.[f_qpd]d ( @xmath94 ) . all eigenstates have a good parity , which means that their wave functions are symmetric or antisymmetric under the transformation @xmath95 and @xmath96 . of selected eigenvectors of the hamiltonian with @xmath97 , @xmath66 for @xmath69 taken at the @xmath98 values indicated by dots in fig.[f_qpd]d . panels ( a)(f ) , respectively , correspond to the vertical row of points at the right of fig.[f_qpd]d from bottom to top [ panel ( a ) shows the ground state ] , panels ( g),(h ) to the left pair of points . note that variable @xmath99 is @xmath92 periodic . ] finding eigensolutions of the model is much simpler in the @xmath63 limit . in that case , the conserved quantity from eq . splits the hilbert space into a sum @xmath100 of dynamically invariant subspaces @xmath101 with different values of quantum number @xmath102 . each of these subspaces is spanned by states @xmath103 where @xmath51 goes from @xmath104 to @xmath105 and @xmath106 takes values from @xmath107 to @xmath108 , so its dimension is : @xmath109 note that the saturation of the @xmath101 dimension at @xmath110 for @xmath111 results from the finiteness of the atomic space @xmath36 . the full spectrum for @xmath63 is comprised of a number of mutually non - interacting spectra with different values of @xmath108 , each of them being obtained by truncation - free diagonalization in a finite dimension . as seen in panel ( a ) of fig.[f_spe ] , the @xmath63 spectrum indeed shows numerous crossings of levels with different @xmath108 . in contrast , an increase of @xmath64 in panels ( b ) and ( c ) leads to strengthening of level repulsion , which is a quantum signature of chaos . spectrum ( for @xmath97 ) in the @xmath112 subspace for the @xmath68 model with @xmath66 ( panel a ) , @xmath113 ( panel b ) and @xmath114 ( panel c ) . the wave - function entropies @xmath115 of individual eigenstates in the unperturbed basis @xmath103 are encoded in the varying shade of the corresponding lines . ] the structure of the spectrum for @xmath63 can be further elucidated by considering the hamiltonian with @xmath116 ( the tuned case ) . in a single-@xmath108 subspace it reads as @xmath117 where @xmath118 is the energy in absence of interaction and @xmath119 are the interaction matrix elements in the unperturbed basis @xmath120 enumerated by @xmath121 . a simple calculation yields : @xmath122 . the eigenvectors of this hamiltonian obviously do not depend on @xmath18 since both terms in eq . the eigenvalues have the form @xmath123 , where @xmath124 are eigenvalues of the matrix in the second term including the @xmath125 factor . since the characteristic polynomial of this matrix contains either even ( for @xmath126 even ) or odd ( for @xmath126 odd ) powers of @xmath127 , the eigenvalues come in pairs with opposite signs . for @xmath126 odd , there is an additional unpaired eigenvalue @xmath128 . therefore , the spectrum linearly expands with increasing @xmath18 around the non - interacting energy @xmath118 , while the eigenvectors remain unchanged . both these conclusions concerning the tuned system with @xmath63 are verified in panel ( a ) of fig.[f_cary ] , which besides the energy spectrum of a single-@xmath108 subspace ( the linearly diverging bundle of levels ) shows also the wave - function entropy @xmath129 of individual eigenstates ( encoded in the shade of the respective line ) . this quantity measures the localization of instantaneous eigenstates in the unperturbed basis @xmath103 , ranging from @xmath130 for any unperturbed state to @xmath131 for a uniformly distributed superposition @xcite . in sec.[ent ] we will see that for @xmath63 the wave - function entropy quantifies the atom field entanglement contained in the given state , but here we use it just as an indicator of structural evolution of eigenstates with varying @xmath18 . in fig.[f_cary ] we see in accord with the above analysis that the wave - function entropies of all states remain constant . it is clear that the full spectrum in fig.[f_spe]a results just from a pile up of the trivial spectra similar to that in fig.[f_cary]a with all values of @xmath108 . in contrast , the spectrum of a detuned hamiltonian ( with @xmath132 ) is less trivial . examples of level dynamics in a single-@xmath108 subspace and the corresponding wave - function entropies for @xmath133 and @xmath134 are shown in panels ( b ) and ( c ) of fig.[f_cary ] , where we see that the eigenvalues are no more straight lines and that the eigenvectors change with @xmath18 . the peculiar shapes of these spectra will be subject of our interest in sec.[qpt ] . here we only point out that the tuned situation is restored in an approximate sense for very large @xmath18 , when the last term in the detuned hamiltonian @xmath135 , with @xmath136 , becomes negligible . therefore , investigating properties of the model for large coupling strengths , one can make an assumption that @xmath45 . this conclusion , which holds independently of @xmath64 , will be seen in action in sec.[afe ] . a straightforward way to the classical limit of the extended dicke model leads via the following operator to c - number mappings : @xmath137 the spherical angles @xmath138 and @xmath139 $ ] determine the orientation of the classical quasi - spin vector @xmath140 . note that angle @xmath141 is measured here from the south to the north pole , thus a fully de - excited state of atoms ( @xmath142 ) corresponds to @xmath143 while a maximally excited state of atoms ( @xmath144 ) is associated with @xmath145 . it can be shown that the pair of variables @xmath99 and @xmath146 , respectively , represents canonically conjugate coordinate and momentum associated with the collective states of the atomic subsystem . similarly , quantities @xmath147 and @xmath148 in eq.([map2 ] ) are suitable coordinate and momentum of the field subsystem . let us point out that while the atomic phase space for each @xmath1 is finite ( it is a ball with radius @xmath25 ) , the field phase space covers the whole plane , expressing the fact that the number of bosons is unlimited . the classical hamiltonian resulting from substitution of eqs.([map1 ] ) and ( [ map2 ] ) to reads as ( see also ref.@xcite ) : @xmath149 \label{hcl } \,,\ ] ] since the classical limit of the model is reached in the limit @xmath150 , the above expression needs an appropriate scaling . it is achieved by introducing quantities @xmath151 , @xmath152 , @xmath153 and @xmath154 that for any value of the size parameter take values of order @xmath155 . this leads to the following size - invariant expression : @xmath156 \label{hclscal } \,.\ ] ] note that eq . contains an effective frequency @xmath157 and an effective coupling strength @xmath158 , which both increase linearly with @xmath25 and reach values of the bare parameters @xmath159 and @xmath18 for the maximal @xmath160 . stationary points of the classical hamiltonian will be analyzed below to determine critical values of control parameters and energy separating distinct quantum phases of the model . we know that for @xmath63 the hamiltonian is integrable as it preserves the quantity @xmath108 from eq .. this introduces a specific correlation between the degrees of freedom associated with both atomic and field subsystems . as a consequence , the system with @xmath40 degrees of freedom is for any fixed value of @xmath108 reduced to a system with just a single effective degree of freedom . to make this explicit , we employ a transformation : @xmath161 where @xmath162 is conserved . it can be checked that this transformation is canonical , so @xmath163 and @xmath164 are new pairs of conjugate coordinates and momenta . the @xmath163 variables for each @xmath165 satisfy a constraint @xmath166 $ ] , so they form a disc ( for @xmath167 ) or an annulus ( for @xmath168 ) . the classical tavis - cummings ( @xmath63 ) hamiltonian in terms of the new variables becomes : @xmath169 } \ x ' \label{hcl1d } \,,\ ] ] which for a constant @xmath165 depends only on @xmath163 , so has effectively @xmath170 degree of freedom . this not only guarantees the full integrability of the @xmath63 model , but also results in some emergent critical phenomena which will be discussed below . maps of the single-@xmath25 model with @xmath66 for six values of parameter @xmath171 $ ] . the tones of blue indicate the classical regular fraction . the curves correspond the ground state energy @xmath172 ( full line ) and the critical esqpt energies @xmath173 ( dotted ) , @xmath174 ( dashed ) , and @xmath175 ( dash - dotted ) from eqs.([ec1])([ec3 ] ) . acronyms d ( dicke ) , tc ( tavis - cummings ) , n ( normal ) and s ( saturated ) indicate different quantum phases of the model ( see sec.[qpt ] ) . dots in panel ( d ) mark places where wave functions in fig.[f_wav ] are taken . ] variation of parameter @xmath64 from 0 to 1 breaks the integrability of hamiltonian and leads to increasingly chaotic dynamics . the evolution of a classical measure of chaos with @xmath64 is demonstrated in fig.[f_qpd ] . it displays maps of so - called regular fraction @xmath176 in the plane @xmath98 for various @xmath64 . the regular fraction is defined as the volume of the phase space domain filled with regular trajectories relative to the total phase space volume available at given energy @xmath177 : @xmath178 here , @xmath179 stands for a characteristic function defining the regular part of the phase space , i.e. , the domain where classical dynamics is regular in the sense of vanishing lyapunov exponents ( @xmath180 in the regular part and @xmath181 in the chaotic part ) . the way how the regular fraction is obtained from numerical simulation of classical dynamics is described in ref.@xcite . the value of the regular fraction in fig.[f_qpd ] is encoded into the tones of blue : the full color indicates perfectly regular areas ( @xmath182 ) and white completely chaotic ones ( @xmath183 ) . we observe that the degree of chaos exhibits an overall increase with @xmath64 . however , even in the dicke limit the model is not entirely chaotic . the most chaotic domain in all @xmath184 panels is that with parameter @xmath18 above the superradiant transition and energy @xmath177 exceeding a certain value above the ground - state @xmath172 . besides the main trends in the dependence of @xmath176 we also observe some surprising fine structures for instance the ribs in panels ( d)(f ) . note that various aspects of chaos in the extended dicke model were recently studied in refs.@xcite . the extended dicke hamiltonian ( [ h ] ) has a complex phase structure @xcite . in this section , we analyze thermal phase transition of the full model with atomic hilbert space ( [ hil ] ) and the number of degrees of freedom @xmath42 . following the standard approach described in ref.@xcite , we use the glauber coherent states @xmath185 , where @xmath186 is an eigenvalue of the @xmath5 operator , @xmath187 is a normalization factor , and @xmath188 denotes the field vacuum . an average number of bosons @xmath189 in the state @xmath190 represents an indicator of the phase transition from normal to superradiant phase : it is zero in the normal phase , while in the superradiant phase we expect @xmath191 . on the way to the thermodynamic limit @xmath192 , it is convenient to perform the scaling transformation @xmath193 ( cf.sec.[cl ] ) , so that the scaled coherent - state variable @xmath194 taking values between 0 and @xmath195 becomes a suitable complex order parameter of the superradiant phase transition . we start by introducing two special values of the atom field coupling strength : @xmath196 while the first one will be shown to represent a critical strength for the occurrence of the superradiant phase , the second value will turn out to be a strength at which a specific form of the superradiant phase occurs for @xmath197 . to get an equilibrium field configuration at a given temperature @xmath198 ( measured in energy units ) , we minimize in variable @xmath199 a scaled free energy @xmath200 , where @xmath201 is the partition function , calculated for a given @xmath199 from a partial trace of the hamiltonian exponential over the full atomic hilbert space ( [ hil ] ) . for hamiltonian ( [ h ] ) this procedure yields : @xmath202 \\ \lambda\left[(1\+\delta){\bar\alpha'}\!\-i(1\-\delta){\bar\alpha''}\right ] & \frac{1}{2}\omega_0 \end{array}\right ) \right\ } \label{z } \,.\ ] ] the two eigenvalues of the matrix in the exponential are : @xmath203}\equiv\pm e_{{\bar\alpha } } \label{e } \,,\ ] ] which leads to a simple expression of the scaled free energy : @xmath204 stationary points of @xmath205 can be obtained from a simple analysis . the first derivative in variable @xmath206 or @xmath207 , respectively , vanishes at the points satisfying the first or second line of the following array : @xmath208 the point @xmath199=0 is always a trivial solution of both lines , but additional solutions appear for @xmath209 . this coupling strength represents a critical point where the normal phase @xmath210 becomes unstable at the lowest temperatures and a new , superradiant equilibrium is created at some non - zero values of @xmath199 . the critical temperature for the superradiant transition , i.e. , the upper temperature limit for the existence of the @xmath211 solution in the region @xmath209 , is : @xmath212 for @xmath213 , the stable equilibrium of the system is again only the normal solution @xmath210 . model ( @xmath66 , @xmath69 ) and sample landscapes of the scaled free energy at the @xmath214 cut of the phase diagram ( vertical line ) . the full curve in the phase diagram marks the critical temperature @xmath215 for the transition to the normal ( n ) phase , while the dotted curve indicates the temperature @xmath216 separating the tc and d superradiant phases ( with or without saddles of free energy ) . the free energy landscapes in the complex-@xmath217 plane are , for selected temperatures , visualized by contour - shade plots ( darker areas indicate lower values ) and by the cuts along real and imaginary axes . ] for @xmath63 ( the tavis - cummings limit ) , both right - side conditions in eqs.([con1 ] ) and ( [ con2 ] ) become identical and yield a solution @xmath218 that grows from zero with increasing difference @xmath219 . therefore , the superradiant minimum of the free energy forms a circle around @xmath210 . for @xmath184 , however , a simultaneous solution of both right - side conditions in ( [ con1 ] ) and ( [ con2 ] ) is no more possible . the superradiant equilibrium is then represented by a pair of points on the real axis , @xmath220 , where @xmath221 solves the right - side equation in ( [ con1 ] ) . if @xmath222 and @xmath223 , a pair of saddle points appears for low temperatures on the imaginary axis , @xmath224 , where @xmath225 solves the right - side equation in ( [ con2 ] ) . these unstable solutions exist for temperatures below @xmath226 a possibility of thermodynamic quasi - equilibrium states associated with the saddle points of free energy was recently discussed in ref.@xcite . a thermal phase diagram in the @xmath227 plane for @xmath94 is shown in fig.[f_tpd ] along with samples of the free energy landscapes in various phases and at @xmath228 . the critical temperature @xmath215 from eq.([tc ] ) determines the phase transition between the superradiant and normal ( acronym n ) phases of the model . the superradiant phase exists in two forms : the tavis - cummings phase ( acronym tc ) with the saddles in the free energy landscape and the dicke phase ( acronym d ) without the saddles . while the tc phase is the only type of superradiant phase in the tavis - cummings limit ( where @xmath229 and @xmath230 ) , the d phase is exclusive in the dicke limit ( in which @xmath231 ) . for intermediate @xmath64 both phases coexist , being separated by temperature @xmath216 from eq.([t0 ] ) . let us stress that the tc@xmath232d transition ( in contrast to d@xmath232n ) is not a phase transition in the standard sense since it does not affect the global minimum of the free energy . we return now to the analysis of the extended dicke model with the restriction to a single-@xmath25 collective subspace of atomic states . we assume in particular @xmath37 , but in the following we keep @xmath25 as a changeable parameter . the number of relevant degrees of freedom of the single-@xmath25 model is @xmath40 , independently of the size parameter @xmath1 , which implies that the infinite - size limit , @xmath192 , coincides with the classical limit @xcite . both ground - state and excited - state quantum phase transitions can be predicted from the classical version of the model , namely from the behavior of stationary points of the classical hamiltonian . in particular , the esqpts result from singularities ( non - analyticities ) in the semiclassical density of states @xmath233 where @xmath234 is the step function ( @xmath235 for @xmath236 and @xmath237 for @xmath238 ) , that appear at the points satisfying @xmath239 ( with @xmath240 standing for the gradient in the phase space ) . it has been shown @xcite that the esqpts caused by non - degenerate stationary points those which are locally quadratic as their hessian matrix of second derivatives has only non - zero eigenvalues can be classified by a pair of numbers @xmath241 , where a so - called index of the stationary point @xmath242 is a number of negative eigenvalues of the hessian matrix . in particular , for @xmath243 , the first derivative of the level density in a vicinity of a stationary - point energy @xmath244 behaves as : @xmath245 hence exhibits either a jump ( for @xmath242 even ) or a logarithmic divergence ( for @xmath242 odd ) at the critical energy @xmath244 . from eq.([trans ] ) in the phase space of the atomic subsystem defined by spherical angles @xmath141 and @xmath99 for @xmath66 and @xmath69 . various panels correspond to the indicated values of the coupling strength @xmath18 . each contour - shade polar plot ( with @xmath246angle and @xmath247radius ) is accompanied by the corresponding horizontal and vertical cuts . ] in the present case , the evaluation of the level density according to eq . can be simplified with the aid of the volume - preserving substitution @xmath248 which transforms the classical hamiltonian to the form @xmath249 with fully separated variables of the field and atomic subsystems . although the mapping does _ not _ represent a canonical transformation ( so @xmath250 is not a new coordinate - momentum pair otherwise the model would be separable and thus fully integrable for any @xmath64 ) , it simplifies the analysis of the level density . in particular , the integration in eq . over variables @xmath250 , on which the transformed hamiltonian depends quadratically , can be performed explicitly ( cf . ref.@xcite , where an analogous calculation is performed for a hamiltonian with a quadratic kinetic term ) . the result is a simplified expression @xmath251 in which the integration , involving function @xmath252 defined in eq . , goes only over the 2-dimensional collective phase space of the atomic subsystem a ball with radius @xmath25 . eq . is proportional to an area of the ball region where @xmath253 takes values less than ( or equal to ) the chosen energy @xmath177 , hence it can be visualized as flooding of a landscape with profile @xmath253 on a globe . in analogy to eq . , the @xmath253 function can be expressed in a scaled form , @xmath254 , with parameters @xmath159 and @xmath18 replaced by their effective values @xmath255 and @xmath256 depending on @xmath25 . the @xmath253 function is for selected values of control parameters @xmath18 and @xmath64 depicted in fig.[f_impo ] . , for the single-@xmath25 model with @xmath66 and @xmath69 as a function of @xmath18 and @xmath257 . the shade diagram in the left panel was obtained from eq . ; darker areas represent larger values and vice versa . panels ( a ) and ( b ) show cuts ( full curves ) at ( a ) @xmath258 and ( b ) @xmath214 in comparison with finite - size results ( dashed curves ) based on numerical diagonalization of the hamiltonian for @xmath97 and gaussian smoothening of the spectrum ( @xmath259 and @xmath260 for @xmath258 and @xmath261 , respectively ) . scales in panels ( a),(b ) are arbitrary and not the same . ] it is clear from the expression that the stationary points of @xmath262 in all four variables correspond to @xmath263 and @xmath264 determined as stationary points of the function @xmath253 . since the quadratic minimum in variables @xmath250 has a null index , any stationary point of @xmath262 has an index @xmath242 equal to that of the corresponding stationary point of @xmath253 . for the determination of esqpts it is therefore sufficient to find and classify stationary points of the function @xmath253 on a ball . non - degenerate stationary points of @xmath253 with index @xmath265 , @xmath266 , and @xmath267 cause , respectively , an upward jump , logarithmic divergence and a downward jump of @xmath268 . inserting the effective parameters @xmath256 and @xmath255 from the scaled hamiltonian into eq . , we obtain the following @xmath25-dependent values of the critical couplings : @xmath269 these will play the roles of @xmath270 and @xmath271 in individual subspaces of states with fixed values of @xmath25 . note that in the highest subspace with @xmath160 both expressions in eq . yield the bare values of critical couplings . a straightforward analysis leads to the following conclusions : \(i ) stationary points with @xmath265 represent the global minimum of both @xmath253 and @xmath262 functions , demarcating the ground state of the @xmath192 system . the minimum appears at @xmath272 ( @xmath99 arbitrary ) for @xmath273 , and at @xmath274 , @xmath275 and @xmath276 ( a pair of degenerate minima ) for @xmath277 . the ground - state energy is given by the following formula : @xmath278 & { \rm for\ } \lambda\in[\lambda'_c,\infty)\ , , \end{array } \right . \label{e0}\ ] ] which exhibits a second - order qpt from normal to superradiant ground - state phase at the critical coupling @xmath279 , where @xmath280 has a discontinuity . \(ii ) stationary points with @xmath281 represent saddles of @xmath253 . they are located at @xmath272 ( @xmath99 arbitrary ) for @xmath282 , and at @xmath283 , @xmath284 and @xmath285 for @xmath286 . these stationary points correspond to an esqpt ( a logarithmic divergence of @xmath268 ) at energy : @xmath287 & { \rm for\ } \lambda\in[\lambda'_0,\infty)\ , . \end{array } \right . \label{ec1}\ ] ] for @xmath288 we have @xmath289 and eq.([ec1 ] ) is reduced to its first line . \(iii ) stationary points with @xmath290 are maxima of @xmath253 at @xmath272 ( @xmath99 arbitrary ) for @xmath286 and at @xmath291 ( @xmath99 arbitrary ) for @xmath292 . related esqpts ( downward jumps of @xmath268 ) appear at energies @xmath293 the second maximum of @xmath253 at energy is the global one , so for @xmath294 the formula yields a constant ( saturated ) value of the level density equal to @xmath295 . subspace of the @xmath296 tavis - cummings model ( @xmath68 ) with @xmath113 as a function of @xmath18 and @xmath297 ( shade plot on the left , with dark areas indicating larger values and vice versa ) and three dependences of the corresponding hamiltonian function on @xmath298 for @xmath299 ( on the right ) . an esqpt due to an inflection point of the classical hamiltonian above the qpt critical point results in a logarithmic divergence of the semiclassical level density ( the dark band in the shade plot ) , cf.fig.[f_cary]b . ] all esqpt critical borderlines @xmath173 , @xmath174 and @xmath175 from eqs. for various values of @xmath64 are demarcated in fig.[f_qpd ] above . their existence is numerically verified in fig.[f_dender ] , which depicts the @xmath98 dependence of @xmath300 for @xmath37 somewhere in between the dicke and tavis - cummings limits . the shade plot was obtained through the phase - space integration in eq . and a comparison with results obtained by a numerical diagonalization is shown in the two panels on the right . we stress that due to the @xmath40 character of our model , the esqpt singularities occur in the _ first derivative _ of the level density with respect to energy . however , the conservation of quantity in the @xmath63 limit and the corresponding reduction of the number of effective degrees of freedom to @xmath170 ( see sec.[cl ] ) leads to a possibility to generate an esqpt singularity in the level density itself . such an effect can really be identified within a particular _ single @xmath108-subspace _ of the full model . the lowest - energy state of any of such subspaces for @xmath192 coincides with the global minimum of classical hamiltonian within the available ( for a given @xmath108 ) domain of the phase space . it turns out that the most interesting @xmath108-subspace is the one with @xmath301 . this particular subset of states exhibits for @xmath133 a second - order ground - state qpt at the coupling strength @xmath18 equal to a critical value : @xmath302 at this coupling , the main minimum of function moves from @xmath303 away to @xmath304 while @xmath305 becomes an inflection point . the inflection point is present for @xmath306 and generates an esqpt ( a logarithmic divergence of the level density in the @xmath301 subspace ) at energy @xmath175 from eq .. note that for @xmath134 the esqpt appears for @xmath307 at the upper energy of the unperturbed spectrum . this effect , described already in ref.@xcite , is demonstrated in fig.[f_tcdens ] . it displays a shade plot of the semiclassical density of levels with @xmath301 ( obtained by the @xmath170 phase space integration ) and the underlying forms of the hamiltonian . a finite - size sample of the predicted singularity in the level density was seen in panel ( b ) of fig.[f_cary ] above . a more detailed analysis of eq . shows that the present type of criticality is absent in the @xmath308 subspaces . this can be intuitively understood from the evaluation of matrix elements @xmath309 of the interaction hamiltonian @xmath310 with @xmath63 in the unperturbed basis ; see eq . and the text below it . the interaction matrix elements quantify the mixing induced by @xmath310 in the unperturbed eigenbasis of @xmath311 . neglecting the trivial state with @xmath312 which does not mix at all , the matrix elements @xmath309 are particularly small in the @xmath313 subspace ( dimension 2 ) and for the @xmath314 state of the @xmath301 subspace ( dimension @xmath31 ) . only the latter state can develop a singularity in the @xmath150 limit . it is the state @xmath315 , which for @xmath133 represents the lowest state of the @xmath301 subspace , while for @xmath134 it is the highest state ; cf.panels ( b ) and ( c ) of fig.[f_cary ] . although the @xmath308 subspaces show no quantum critical effects , it is interesting to realize that a pile up of all @xmath108-subspaces produces the downward jump of @xmath300 ( with @xmath316 standing for the total semiclassical level density ) , as observed in the @xmath317 esqpt of the @xmath63 system . the critical borderlines @xmath173 , @xmath174 and @xmath175 in the @xmath98 plane separate spectral domains that we consider to constitute different quantum phases of the model ( in analogy to thermodynamic phases ) . they are denoted by acronyms d ( dicke ) , tc ( tavis - cummings ) , n ( normal ) and s ( saturated ) , see fig.[f_qpd ] . both phases d and tc , which coexist for @xmath64 in between 0 and 1 , contain quantum states of a superradiant nature because their energy is lowered with respect to the minimal non - radiant value @xmath318 . in the limits of @xmath64 , the model shows only one type of the superradiant phase : tc for the tavis - cummings limit @xmath63 and d for the dicke limit @xmath319 . in contrast , both phases n and s contain states that resemble excitations in the fully non - radiant regime at @xmath320 . the s phase above @xmath321 yields a constant , saturated value of level density @xmath316 . ( panel a ) and @xmath322 ( panel b ) in individual eigenstates of the @xmath296 hamiltonian ( [ h ] ) for @xmath97 , @xmath66 , @xmath214 . the upper panels depict lattices for the limiting values of @xmath68 and 1 , whereas the lower panels correspond to @xmath69 . the insets of the @xmath69 panels show smoothed dependences of both observables on energy ( averages over 20 neighboring eigenstates and their piecewise fits in various quantum phases ) . ] while the definition of esqpts is obvious from the behavior of quantities like level density at the critical energies , the meaning of quantum phases in between the critical energies is not a priori clear . it should be looked for in the structure of the energy eigenstates in the corresponding energy domains . however , sample wave functions in fig.[f_wav ] [ where panels ( a ) , ( b ) , ( d ) , ( f ) and ( h ) correspond to phases d , d , tc , n and n , respectively ] indicate that individual eigenstates do not show sufficient clues for the identification of phases . we have to resort to a more efficient visualization tool allowing us a bulk inspection of the eigenstate properties . the approach employed in this work is based on the method of so - called _ peres lattices _ @xcite . it was proven useful in various models ( see ref.@xcite and references therein ) including the dicke model @xcite . a general peres lattice shows expectation values @xmath323 of a selected observable @xmath324 in individual energy eigenstates @xmath325 ( enumerated by integer @xmath326 ) arranged into lattices with energy @xmath327 on one of the axes . examples of peres lattices for observables ( a ) @xmath328 and ( b ) @xmath329 are shown in the respective panels of fig.[f_pere ] . a comparison of both @xmath94 lattices ( in the main panels ) with the map of chaos in fig.[f_qpd]d ( for @xmath18 demarcated by the second vertical line ) yields an overall agreement with the original conjecture concerning the correlation of orderly arranged domains of a generic lattice with more regular regions of classical dynamics and vice versa @xcite . also the limiting @xmath63 lattices in the upper panels of fig.[f_pere ] are perfectly regular , while those for @xmath319 above a certain energy are almost fully chaotic . the @xmath94 peres lattices in fig.[f_pere ] cross all four quantum phases of the model . there exist apparent similarities between parts of both @xmath94 lattices located in the tc and d phases , respectively , and the corresponding tavis - cummings and dicke lattices displayed in the @xmath63 and 1 upper panels . however , a more specific distinction of quantum phases results from an averaging of both lattices over the neighboring eigenstates , as shown in the insets of both lower panels of fig.[f_pere ] . points in these _ smoothed lattices _ represent averages @xmath330 and @xmath331 of the respective expectation values @xmath332 and @xmath333 over 20 neighboring eigenstates . smoothed dependences of the averaged quantities on energy are given by lines , resulting from piecewise fits within the four quantum phases . we observe that various quantum phases are recognized by different characters of the energy dependences , namely : ( i ) the phase d by slowly descending dependences of both @xmath330 and @xmath331 averages , ( ii ) the phase tc by roughly constant dependences , ( iii ) the phase n by linearly increasing dependences , and ( iv ) the phase s by saturated @xmath334 and linearly increasing @xmath331 . hamiltonian ( [ h ] ) with @xmath97 , @xmath66 and @xmath69 at @xmath214 . quantum phases d , tc , n and s are distinguished by different smoothed energy dependences of the average slopes see the piecewise fits indicated by full black lines . ] peres lattices for both quantities @xmath52 and @xmath322 are connected with the lattice of the interaction hamiltonian @xmath335 via the energy conservation relation . using the hellmann - feynman formula , @xmath336 , one derives the following expression for the slope of individual energy levels : @xmath337 the slopes averaged over bunches of neighboring 20 levels , as in the insets of fig.[f_pere ] , are presented in fig.[f_slo ] for the same values of control parameters . piecewise fits of the energy dependences in individual quantum phases are again indicated by lines . we see that any of the d , tc , n and s quantum phases is characterized by a specific , roughly invariant energy dependence of the averaged level slope @xmath338 within the corresponding energy domain . abrupt changes of the character of these dependences coincide with critical esqpt energies . this is in accord with a general relation between the density and flow signatures of esqpts , namely with the fact that a typical esqpt generates the same type of non - analyticity in the energy dependences of both quantities @xmath316 and @xmath338 @xcite . indeed , the discontinuities of @xmath339 at the critical energies @xmath174 and @xmath175 , as observed in fig.[f_slo ] , are consistent with the analogous behavior of the @xmath300 ( fig.[f_dender ] ) . on the other hand , the anticipated point of a singular growth of the level slope ( logarithmic divergence of its energy derivative ) at @xmath173 is smoothed out in the finite spectrum for a moderate system s size . we turn to the study of quantum entanglement properties of individual eigenstate of the model hamiltonian and their links to the esqpts . our analysis includes two types of bipartite entanglement : ( a ) that between the bosonic field and the set of all atoms ( atom field entanglement ) , and ( b ) that between any pair of individual atoms ( atom atom entanglement ) . the entanglement of type ( a ) is an important ingredient of superradiance since the interaction term of the hamiltonian carries a direct coupling between the atomic and field subsystems . in contrast , the entanglement of type ( b ) appears only due to an indirect coupling of individual atoms via the bosonic field , so it may be expected to be just a higher - order effect . we start with a brief description of the measures used to quantify both types of entanglement . in the following , the atom field and atom atom entanglement will be evaluated in individual eigenstates @xmath340 of hamiltonian . some examples of the eigenstate wave functions were shown in fig.[f_wav ] . since the wave - function arguments @xmath99 and @xmath341 , respectively , correspond directly to the atomic and field coordinates , a compound state @xmath342 is factorized with respect to the atom field partitioning of the system if it has a product wave function @xmath343 , where @xmath344 and @xmath345 are arbitrary atomic and field wave functions . the method to quantify a departure of a given pure state @xmath342 from exact factorization , i.e. , an amount of atom field entanglement involved in @xmath342 , makes use of the von neumann entropy corresponding to the reduced density operators @xmath346 and @xmath347 of the atomic and field subsystem , respectively . these operators are obtained by partial tracing of the total density operator @xmath348 over the irrelevant part of the total hilbert space , that is @xmath349 ( trace goes over @xmath350 ) for the atomic subsystem and @xmath351 ( trace goes over @xmath352 ) for the field subsystem . von neumann entropy of the pure compound state @xmath353 is by definition zero , but entropies of both reduced density operators @xmath346 and @xmath347 satisfy : @xmath354 . the case @xmath355 implies a separable compound state , while the case @xmath356 indicates that the reduction of @xmath353 to a single subsystem leads to a loss of information on mutual entanglement between subsystems . a maximally entangled state yields @xmath357 , where @xmath31 is a dimension of @xmath358 , the smaller of both subspaces . we therefore define a normalized atom field entanglement entropy in a compound state @xmath342 as : @xmath359}{\ln(2j+1 ) } = -\frac{\textnormal{tr}\left[\rho_{\textnormal{f}}\,\textnormal{ln}\rho_{\textnormal{f}}\right]}{\ln(2j+1 ) } \label{s } \,.\ ] ] it changes between @xmath360 for separable states and @xmath361 for maximally entangled states . quantifying the atom atom entanglement , i.e. , quantum correlations between a randomly chosen pair @xmath362 of atoms , is a more complex problem . the use of the above entropic approach is disabled by the fact that for any state @xmath342 of the whole atom field system , an arbitrary pair of atoms generically occurs in a _ mixed _ quantum state . it is known that mutual entanglement of a pair of objects in a mixed compound state can not be recognized by a non - zero entropy of the reduced density operators @xcite . indeed , a mixed compound state of the pair generates mixed reduced states of the objects even in absence of entanglement . in such cases , the evaluation of an entropy - based measure of entanglement ( so - called entanglement of formation ) has to be performed with respect to all possible decompositions of the compound density operator into statistical mixtures of pure states @xcite , which is from the computational viewpoint a difficult task . a way to bypass this obstacle for two - qubit systems was proposed in refs.@xcite in terms of a quantity called concurrence . the idea was applied and further elaborated @xcite for multi - qubit systems in fully symmetric states , like our ensemble of @xmath1 two - level atoms with @xmath160 . the atom atom entanglement in this case is characterized by a scaled concurrence : @xmath363 where @xmath364 are eigenvalues ( real and non - negative ) of a non - hermitian matrix @xmath365 , with @xmath366 denoting a reduced density matrix of the pair of atoms @xmath362 and @xmath367 the corresponding spin - flipped ( conjugate under the time reversal ) state ( with @xmath368 denoting the tensor product of operators for the selected atoms and @xmath369 a complex conjugated density matrix in the @xmath370 eigenbasis ) . the argument @xmath371 in the definition again reminds the compound state @xmath342 in which the concurrence is calculated . it has to be stressed that the full symmetry of @xmath342 under the exchange of atoms , which is guaranteed in the @xmath160 subspace , ensures that the reduced density matrix @xmath366 , as well as @xmath372 and @xmath373 , are the same for _ any _ pair of atoms , hence do not in fact depend on @xmath374 . model in the plane of control parameters @xmath18 and @xmath64 ( for @xmath66 , @xmath375 ) . the red curve indicates the qpt critical coupling @xmath279 . left : atom field entanglement measured by the entropy . right : atom atom entanglement measured by the concurrence . the darker areas indicated larger entanglement . ] matrix elements of the reduced two - atom density matrix @xmath366 for any symmetric state of an @xmath1-qubit system can be expressed through expectation values @xmath376 , @xmath377 and @xmath378 of collective quasi - spin operators in a given state see formulas ( 4 ) and ( 11 ) in ref.@xcite . this makes it possible to perform a straightforward numerical computation of @xmath373 in the @xmath160 subspace of our model . moreover , as shown in ref.@xcite , the scaled concurrence in this subspace is related to the so - called kitagawa ueda spin - squeezing parameter @xmath379 though the relation @xmath380 , so it varies between @xmath381 for separable states and @xmath382 for the states that exhibit the maximal entanglement allowed for a given @xmath1 . an entropy corresponding to the entanglement of formation of a single atomic pair reads as @xcite : @xmath383 \label{saa } \,.\ ] ] for @xmath381 we have @xmath384 , while for @xmath382 we obtain @xmath385 , which decreases from @xmath386 for @xmath387 to zero in the classical limit @xmath388 . with @xmath389 , the maximal entropy behaves as @xmath390 . note that the @xmath389 maximal atom field entanglement entropy _ per atom _ decreases as @xmath391 , see eq . , so it exceeds the atom atom entanglement entropy about @xmath1 times . these scaling properties are verified theoretically and by large-@xmath1 numerical calculations @xcite . the atom field and atom atom entanglement properties have been studied for the ground state of the dicke model @xcite . as in other models of similar nature , see e.g. refs.@xcite , the second - order qpt was shown to induce a singularity in both entanglement measures and . figure [ f_eng ] verifies this conclusion in our extended model with variable parameter @xmath64 . we observe that both @xmath392 and @xmath373 measures exhibit an increase at about the critical coupling @xmath279 from eq .. for @xmath184 , the atom atom entanglement drops back to nearly zero values with @xmath393 . the atom field entanglement saturates at a value @xmath394 for @xmath395 , which is due to an irreducible atom field coupling in the lowest positive parity state in the strong coupling limit @xcite . in the @xmath396 region , close to the integrable tavis - cummings limit , both entanglement measures show roughly a step - like increase at @xmath279 . this is due to a specific mechanism , in which the ground state at each @xmath18 is formed via unavoided crossings of levels with different values of the conserved quantum number @xmath108 , see fig.[f_spe]a . we will analyze the integrable case in more detail below . in the following , we present results of a numerical study of the atom field entanglement in individual eigenstates of hamiltonian with @xmath37 . we start by analyzing the integrable tavis - cummings limit @xmath63 . this simple setting will allow us to obtain some insight into the entanglement properties from quasi - analytic solutions , which will serve as a useful reference for the less trivial @xmath184 case . we know that for @xmath63 the hilbert space splits into the subspaces @xmath101 with fixed values of quantum number @xmath108 , see eq . , with the corresponding dimensions @xmath126 given by . the reduced density matrix of the atomic subsystem within each @xmath101 reads as @xmath397 where @xmath398 is a number of bosons associated with quasi - spin projection @xmath51 for a given value of @xmath108 . this expression implies that for @xmath63 the entanglement entropy of any eigenstate is equal to the wave - function entropy corresponding to its expansion in the non - interacting basis . model with @xmath66 for @xmath296 and @xmath97 ( left panel ) , and the atom field entanglement entropies @xmath392 in individual eigenstates corresponding to @xmath399 cut of the spectrum ( right panel ) . the lowest v - shaped chain of points ( connected by the green curve ) in the entropic spectrum corresponds to the lowest states from various @xmath108-subspaces , the second chain to the second states etc . the points corresponding to two selected @xmath108-subspaces ( @xmath400 and 8) are highlighted . ] let us focus first on the tuned case @xmath116 . the wave - function entropies for the @xmath301 spectrum of a tuned @xmath63 hamiltonian were shown in panel ( a ) of fig.[f_cary ] . we notice that the entropy of individual eigenstates does not change with @xmath18 and that it has an apparent symmetry with respect to the vertical reflection of the spectrum around the centroid energy @xmath118 . the former feature was explained by constancy of eigenstates of the simple hamiltonian , the latter follows from a recursive relation for the eigenstate components in eq . which yields the same distributions @xmath401 , hence the same entropies , for the pair of levels with opposite slopes . the full spectrum of the tavis - cummings model is obtained by combining the spectra for all values of @xmath108 . this is for @xmath116 seen in fig.[f_pileup ] . it shows the energy spectrum as a function of @xmath18 ( in the left panel ) and the atom field entanglement entropy of individual eigenstates for one particular value of the coupling strength ( the right panel ) . as follows from the above theoretical analysis , the energy spectrum in the left panel of fig.[f_pileup ] is formed by straight lines . the qpt critical coupling @xmath402 coincides with the point where the energy of the @xmath312 ground state is crossed by the lower state from the @xmath313 space . a further increase of @xmath18 above @xmath402 leads to a sequence of consecutive level crossings in which the lowest states from subspaces with increasing @xmath108 become instantaneous ground states of the system . a similar mechanism applies also in the spectrum of excited states , where we observe a sequence of separated caustic structures formed by states with increasing ordinal numbers within each @xmath108-subspace ( the @xmath322th caustic structure in the vertical direction represents an envelope of lines corresponding to the @xmath322th states from individual @xmath108-subspaces ) . the entropic spectrum on the right of fig.[f_pileup ] arises in a similar way . it is composed of several mutually shifted v - shaped chains of points , each of them containing a collection of states from different @xmath108-subspaces . for example , the lowest v - chain of points is formed by the lowest - energy states from all @xmath108-subspaces , the second lowest chain by second lowest - energy states and so on . to indicate a contribution of a single @xmath108-subspace to the entropic spectrum , we highlighted the points corresponding to two values of @xmath108 . for @xmath403 , the contribution has a seagull - like form ( see the @xmath404 example in the figure ) , whose reflection symmetry around @xmath118 results from the above - discussed properties of eq .. the instantaneous ground state at the given @xmath405 , which is well above the critical point @xmath402 , has a high value of @xmath108 and therefore carries a considerable amount atom field entanglement ( as in general expected in superradiance ) . on the other hand , the trivial one - dimensional subspace with @xmath312 formed by the state @xmath406 , which was the ground state of the @xmath63 hamiltonian at @xmath407 and becomes an excited state with energy @xmath408 for @xmath409 , has apparently zero atom field entanglement . we will see below that some remnants of this state remain present in entropic spectra also in the non - integrable @xmath184 regime . model with @xmath296 and @xmath97 in the detuned case ( @xmath113 ) for several values of the coupling strength @xmath18 . ] let us consider now the detuned case @xmath132 . field entropic spectra of a hamiltonian with @xmath63 and @xmath410 are displayed in fig.[f_detun ] for increasing values of the coupling strength @xmath18 ( the critical coupling is @xmath411 ) . we know ( see sec.[ha ] ) that in the strong coupling limit , @xmath412 , the detuned situation becomes very similar to the tuned one . this is obvious from a comparison of the @xmath413 entropic spectrum in fig.[f_detun ] with that in fig.[f_pileup ] . on the other hand , for small coupling strengths , @xmath414 , the detuning @xmath415 plays an important role . a detuned system shows generally lower values of the atom field entanglement than the tuned one , which can be attributed to a smaller perturbation efficiency of the interaction hamiltonian @xmath310 due to larger energy gaps between the unperturbed ( factorized ) states ( the energy spread of a single-@xmath108 spectrum at @xmath320 increases with @xmath415 ) . this overall lowering of entanglement entropies at small @xmath18 ( more apparent for @xmath133 than @xmath134 ) can be verified by simple numerical calculations , which we do not present here . it is clear that the factorized state @xmath406 with @xmath312 at critical energy @xmath416 is an invariant eigenstates of any @xmath63 hamiltonian , irrespective of @xmath18 and @xmath0 s . so the @xmath360 point at the place of the unperturbed ground state is present in all entropic spectra of fig.[f_detun ] . however , for an @xmath132 system , a local decrease ( converging to @xmath360 with @xmath150 ) of the atom field entanglement is present also at the highest critical energy @xmath417 . this is due to quantum critical properties of the @xmath301 subspace discussed in sec.[qpt ] . we consider here the @xmath133 case ( see fig.[f_cary]b ) , but the same phenomenon takes place also in the reversed detuning hierarchy @xmath134 ( fig.[f_cary]c ) . for @xmath418 , where @xmath419 is the critical coupling , the lowest state of the @xmath301 subspace at energy @xmath175 keeps the factorized form @xmath315 ( mind the sign difference in @xmath51 from the other factorized state , which means that now a maximal possible , for a given @xmath25 , number of atoms is in the _ upper _ level ) . distinct lowering of the atom field entanglement entropy at @xmath317 is observed in the @xmath420 and 0.4 panels of fig.[f_detun ] ( in which @xmath421 ) . this qpt - related effect becomes sharper with increasing @xmath1 and @xmath25 , and in the @xmath422 limit it results in a similar @xmath360 cusp as that at energy @xmath174 . for @xmath134 , the above factorized state , causing the same effect , is the highest state of the @xmath301 subspace at @xmath320 . subspace of the @xmath68 detuned ( @xmath113 ) model with @xmath296 : the ground state ( panel a ) and excited states which at @xmath423 have 15% ( panel b ) and 24% ( panel c ) of the maximal excitation energy . the dashed curves are for @xmath97 , the full ones for @xmath424 . the energy of the corresponding level ( for @xmath97 ) is shown at the bottom of each panel . the log - log plot in panel ( b ) inset shows the minimum value of @xmath392 in the dip as a function of @xmath25 for the level at 15% of the spectrum ( for this level , the dependence is roughly @xmath425 ) . ] an analogous lowering of the atom field entanglement entropy at the upper critical energy @xmath175 appears also for the coupling strengths _ above _ the qpt critical value @xmath419 . the effect in this domain is caused by the @xmath170 type of esqpt in the @xmath301 subspace . indeed , when excited states within this subspace cross the critical energy , they get temporarily localized in the coordinate region around the point @xmath426 , i.e. , at the position of the factorized state . this is due to dwelling of the @xmath317 classical trajectory and a semiclassical wave function at an inflection point of the classical hamiltonian . an analogous phenomenon was observed also in other quasi @xmath170 models , see e.g. ref.@xcite . the lowering of the eigenstate entropy in the @xmath427 domain was visible already in panels ( b ) and ( c ) of fig.[f_cary ] , but it is not apparent in fig.[f_detun ] . to see the effect in the pile up of all-@xmath108 spectra , one would have to go to higher values of @xmath1 and @xmath25 , when the localization becomes stronger and the entropy decreases to its lower limit @xmath360 . entropic spectra of the atom field entanglement with increasing parameter @xmath60 $ ] at @xmath214 , @xmath428 , @xmath97 . the esqpt energies are indicated by vertical lines . the spectrum contains @xmath429 well converged eigenstates . ] the esqpt - based localization effect is nevertheless well documented in fig.[f_entrotc ] , which depicts the evolution of the atom field entanglement for three selected levels ( ground state and two excited states ) from the @xmath430 subspace for @xmath431 and 1000 . we see that the passage of the selected excited states through the esqpt energy @xmath175 is correlated with a local decrease of the entanglement entropy [ sharp dips in the dependences in panels ( b ) and ( c ) ] . as demonstrated by the curves for two values of @xmath1 , the drop of entropy becomes sharper and deeper with increasing size of the system . the decrease of @xmath392 at the lowest point of the dip with @xmath37 for one of the excited states is shown in the inset of fig.[f_entrotc]b . the depicted log - log plot indicates a very slow power - low decrease according to @xmath432 . the exponent @xmath433 is rather small and depends on the selected level ( e.g. , for the displayed level at 15% of the spectrum we obtain @xmath434 , while for the level at 20% we would have @xmath435 ) note that the entanglement entropy for excited states for large @xmath18 saturates at a value close to @xmath436 , where @xmath437 is the dimension for @xmath301 . this expression follows from the use of the random matrix theory for estimating an average wave - function entropy for @xmath126 orthonormal states in the limit of strong mixing @xcite . note also that the critical coupling @xmath419 of the ground - state qpt coincides with the sharpest increase of the ground - state entanglement entropy ( fig.[f_entrotc]a ) . now we are ready to turn our attention to the non - integrable @xmath184 model . figure [ f_sdelta ] shows the evolution of the entropic spectrum with parameter @xmath171 $ ] for @xmath116 and @xmath438 , well above the critical coupling @xmath402 for any @xmath64 . the vertical lines demarcate the esqpt energies , the zones between them corresponding to the d , tc , n and s quantum phases . one can see that the departure from integrability ( an increase of @xmath64 ) gradually destroys regular patterns in the entropic spectra , in analogy with peres lattices . orderly patterns in the entropic spectrum occur in the domains of regular dynamics , while disorderly scattered points indicate chaotic domains . note that the accumulation of points in a narrow band close to ( slightly below ) the maximal entropy @xmath439 for high energies is a consequence of ergodicity of chaotic dynamics @xcite : a common eigenstate in the chaotic part of the spectrum contains virtually all atom field configurations , whose dominant part is strongly entangled . on the other hand , the regular low - energy part of the entropic spectrum tends to a minimal value @xmath440 , corresponding to states with a good parity quantum number @xcite . , @xmath69 ( cf.fig.[f_wav]e ) and ( b ) the factorized @xmath441 eigenstate for @xmath68 . the other parameters are the same as in fig.[f_sdelta ] . ] an interesting aspect of the entropic spectra in fig.[f_sdelta ] is the evolution of the atom field entanglement in transition between the superradiant and normal phases at the critical energy @xmath318 . for @xmath184 , this energy corresponds to @xmath173 for @xmath442 $ ] and to @xmath174 for @xmath443 . we know that in the @xmath63 limit , there always exists the fully factorized trivial eigenstate with @xmath444 located right at this energy and a group of weakly entangled eigenstates with low values of @xmath108 located nearby ( see figs.[f_pileup ] and [ f_detun ] ) . to what extent do these structures survive an increase of parameter @xmath64 ? the answer can be read out from fig.[f_sdelta ] . we see that the eigenstates with lowered entropy and even a single state with @xmath445 are preserved in the entropic spectra as far as the tc phase is present for the chosen value of the coupling strength @xmath18 . this is so if @xmath446 , hence @xmath447 , see eq .. if @xmath448 $ ] , that is for @xmath64 large enough to avoid the existence of the tc phase for a given @xmath18 , the decrease of entropy no longer takes place . these observations allow us to say that the occurrence of states with lowered atom field entanglement entropy is a signature of the esqpt between the n and tc phases , but not of that between the n and d phases . the wave function corresponding to the @xmath449 eigenstate with the lowest entanglement entropy for @xmath94 can be seen in panel ( e ) of fig.[f_wav ] . its detail is depicted in panel ( a ) of fig.[f_wavdet ] , in comparison with the fully factorized state with @xmath312 shown in panel ( b ) . although the fine structures of the wave function for @xmath184 ( panel a ) apparently disable its full factorization ( in contrast to the @xmath63 case ) , a distant view in fig.[f_wav]e makes clear that this state is indeed very similar to the @xmath312 state . following our findings for @xmath63 , one might expect an analogous decrease of the atom field entanglement entropy also at the critical energy @xmath175 ( in transition between the s and n phases ) in a detuned ( @xmath132 ) system with @xmath184 . a numerical evidence of this phenomenon is however hindered by the above - presented ( cf . the inset in fig.[f_entrotc]b ) slowness of the decrease of the entanglement entropy dip to zero . to see an effect in the @xmath184 system , one would have to perform a truncated - space diagonalization of the hamiltonian for a very large value of @xmath1 , which is computationally demanding . model with @xmath66 , @xmath214 , @xmath296 , @xmath97 . the contribution of selected @xmath108-subspaces ( the u - shaped chains of points ) is highlighted . ] for @xmath214 , @xmath66 , @xmath296 , @xmath97 . the concurrence spectra contain 5000 well converged states . ] we turn now to a brief numerical analysis of the atom atom entanglement , which is limited to the symmetric @xmath37 subspace of the atomic hilbert space . as in the previous case , we focus at first on the integrable tavis - cummings model . the scaled concurrence for individual eigenstates ( the concurrence spectrum ) of the tuned hamiltonian with @xmath63 and @xmath409 is plotted in fig.[f_concint ] . the highlighted points denote states belonging to selected @xmath108-subspaces . like in the entropic spectrum ( fig.[f_pileup ] ) , the states corresponding the same @xmath108-subspace form a pattern which is always symmetric with respect to the reflection around the central energy @xmath118 . trivially , the @xmath312 state @xmath406 has zero concurrence as the atomic state with minimal quasi - spin projection is fully factorized ( all atoms in the lower level ) . so this state shows no entanglement in either atom field or atom atom sense . in contrast , some states with low values of @xmath108 , which all have only weak ( increasing with @xmath108 ) atom field entanglement , show relatively large ( decreasing with @xmath108 ) atom atom entanglement . consider as an example a doublet of states with @xmath313 . the scaled concurrence for these states is @xmath450 , which changes between @xmath451 for @xmath387 and the maximal value @xmath382 for @xmath192 ( see the upper pair of points in fig.[f_concint ] ) . as follows from eq . , the @xmath313 eigenstates are expressed by the superpositions @xmath452 , whose first term is fully factorized in the atomic subspace ( as in the @xmath312 case ) while the second term is maximally entangled ( a symmetric state with one atom in the upper level and the rest of atoms in the lower level ) . it turns out that for increasing @xmath453 , only the states with the largest positive and negative slopes within the given @xmath108-subspace yield a non - vanishing atom atom entanglement . these states form the feelers of the u - shaped dependences of @xmath373 in each @xmath108-subspace . the scaled concurrence of these states decreases with @xmath108 , forming together the left and right chains of @xmath454 points in fig.[f_concint ] . the rest of the @xmath63 eigenstates has @xmath381 . figure [ f_cdelta ] shows an evolution of the concurrence spectrum with increasing @xmath64 for a fixed @xmath455 . apparently , an overall trend of the atom atom entanglement in individual eigenstates is a decrease with increasing @xmath64 . we stress that the concurrence of an absolute majority of states for any @xmath64 is zero ( the spectrum contains 5000 states ) and both the number of states with @xmath454 and the values of @xmath373 in these states further decrease with @xmath64 . the only states in which the atom atom entanglement remains significant even for a relatively large values of @xmath64 appear in a narrow energy interval around the esqpt critical energy @xmath174 demarcating the transition between the tc and n phases . these states partly coincide with those in which we previously observed a reduced atom field entanglement , see fig.[f_sdelta ] . for @xmath64 large enough to avoid the existence of the tc phase , the atom atom entanglement of all states vanishes . model with @xmath456 for two values of @xmath64 . the number of principal components of an actual eigenstate in the unperturbed basis is encoded in the shades of gray of the respective line ( black means perfect localization , @xmath457 , white complete delocalization , @xmath458 ) . revival of some @xmath459 eigenstates at @xmath460 is seen in both spectra . ] we analyzed properties of a generalized dicke model of single - mode superradiance allowing for a continuous ( governed by parameter @xmath64 ) crossover between the integrable tavis - cummings limit and the prevailingly chaotic dicke limit . we considered three constraints on the model hilbert space @xmath461 , namely : ( i ) the all-@xmath25 model with the full @xmath462 containing the complete @xmath26-dimensional atomic hilbert space @xmath35 , ( ii ) a single-@xmath25 model whose atomic hilbert space is restricted to a subspace @xmath463 with fixed quasi - spin quantum number @xmath25 ( usually @xmath37 ) and ( iii ) a single-@xmath464 model with only a single subspace @xmath465 of quantity @xmath108 , which is conserved in the integrable limit @xmath63 . we determined the classical limit and critical properties of the model . the all-@xmath25 model has @xmath42 degrees of freedom , the single-@xmath25 model @xmath40 and the single-@xmath464 model @xmath170 . the thermal phase diagram of the all-@xmath25 model for @xmath197 shows a coexistence of the tc and d types of superradiant phase ( with and without saddles in the free energy landscape ) , which contract to a single type tc or d in the limits @xmath63 or 1 , respectively . the quantum phase diagram of a single-@xmath25 model contains the ground - state superradiant quantum phase transition and three types of excited - state quantum phase transitions , characterized by singularities ( jumps and a logarithmic divergence ) in the first derivative of the semiclassical level density as a function of energy . the quantum phase diagram of a critical single-@xmath464 model with @xmath301 shows in the detuned case ( @xmath132 ) an excited - state quantum phase transition characterized by a singularity ( logarithmic divergence ) directly in the level density . we defined the spectral domains in between the esqpt critical borderlines of the single-@xmath25 model as the tc , d , n and s quantum phases . these phases were characterized by different shapes of a smoothed energy dependence of some expectation values , which show abrupt changes at the esqpt energies . a natural choice of the phase - defining quantity is the interaction hamiltonian , whose expectation value @xmath466 determines the slope @xmath467 of the given level @xmath326 . according to the density flow relation of the esqpt signatures @xcite , a smoothed level slope should exhibit the same type of non - analyticity as the semiclassical level density . therefore , quantum phases of systems with @xmath40 ( like our single-@xmath25 model ) are distinguished by @xmath468 , while phases in other systems are expected to differ in the @xmath469th energy derivative of @xmath470 . we analyzed entanglement properties of the single-@xmath25 and single-@xmath464 models . in general , an absolute majority of eigenstates in the spectrum for any choice of model parameters has large atom field entanglement but vanishing atom atom entanglement . exceptional in this sense for @xmath63 are the eigenstates with low values of @xmath108 , which yield a weak atom field entanglement and simultaneously an increased atom atom entanglement . we showed that remnants of these states exist relatively far in the non - integrable regime with @xmath184 . this concerns also the @xmath63 state with @xmath312 located at the esqpt energy @xmath174 , which has zero atom field entanglement . for @xmath132 , a state with reduced atom field entanglement entropy , originating in the critical @xmath301 subspace , exists also at the esqpt energy @xmath175 . a numerical evidence of its partial survival for @xmath184 is however difficult because of the observed slow decrease of the entropy dip with the size parameter @xmath1 . it needs to be stressed that the above states with exceptional entanglement properties form only a small subset of all states ( its relative size goes to zero in the @xmath388 limit ) . even so is their occurrence a significant phenomenon . the remnants of the @xmath63 states with low @xmath108 s appear in the range of parameter @xmath64 where the tc quantum phase still exists in the spectrum . so if @xmath64 is fixed at a non - zero value and the coupling strength @xmath18 increases from 0 to values above @xmath402 and @xmath471 , these states exhibit a surprising evolution . they first ( not later than at @xmath402 ) disappear from the spectrum , but later above the critical coupling ( somewhere in the @xmath472 region ) some of them reappear again . indeed , a systematic analysis of the spectrum discloses approximate revivals of a number of weak - coupling ( @xmath414 ) eigenstates in the strong - coupling ( @xmath473 ) regime . this is illustrated in fig.[f_fog ] where we again show spectra of the tuned model with @xmath37 for two values of @xmath64 . the fullness ( shade of gray ) of each level encodes the number of principal components @xmath474 of the corresponding hamiltonian eigenstate @xmath475 in the eigenbasis @xmath476 taken at @xmath477 infinitesimally displaced from the degeneracy point @xmath320 . the number of principal components varies from @xmath478 for a perfectly localized state ( identical with one of the @xmath479 eigenstates ) to @xmath480 for totally delocalized states . the revival of several @xmath479 eigenstates is clearly observed in fig.[f_fog ] as reappearance of certain dark lines in the spectrum for large @xmath18 in the tc and n phases . one can only speculate that the mechanism responsible for this surprising phenomenon can be perhaps related to a specific partial dynamical symmetry @xcite . the authors thank to t.brandes , n.lambert , j.hirsch and m.a.bastarrachea-magnani for illuminating discussions . this work was supported by the czech science foundation under the project no.p203 - 13 - 07117s . dicke , phys . rev . * 93 * , 99 ( 1954 ) . m. gross and s. haroche , phys . rep . * 93 * , 301 ( 1982 ) . benedict ( ed . ) , _ super - radiance : multiatomic coherent emission _ , taylor & francis , new york , 1996 . t. brandes , phys . rep . * 408 * , 315 ( 2005 ) . bekenstein and m. schiffer , phys . d * 58 * , 064014 ( 1998 ) . n. auerbach and v. zelevinsky , _ rev . phys . _ * 74 * 106301 ( 2011 ) . jaynes and f.w . cummings , proc . ieee * 51 * , 89 ( 1963 ) . m. tavis and f.w . cummings , phys . rev . * 170 * , 379 ( 1968 ) . wang and f.t . hioe , phys . rev . a * 7 * , 831 ( 1973 ) . k. hepp and e.h . lieb , phys . rev . a * 8 * , 2517 ( 1973 ) . k. hepp and e.h . lieb , ann . * 76 * , 360 ( 1973 ) . k. rzazewski , k. wodkiewicz and w. zakowicz , phys . 35 * , 432 ( 1975 ) . k. baumann , c. guerlin , f. brennecke and t. esslinger , nature * 464 * , 1301 ( 2010 ) k. baumann , r. mottl , f. brennecke and t. esslinger , phys . lett . * 107 * , 140402 ( 2011 ) p. prez - fernndez , p. cejnar , j.m . arias , j. dukelsky , j. e. garca - ramos and a. relao , phys . rev . a * 83 * , 033802 ( 2011 ) . p. prez - fernndez , a. relao , j.m . arias , p. cejnar , j. dukelsky and j. e. garca - ramos , phys . e * 83 * , 046208 ( 2011 ) . t. brandes , phys . e * 88 * , 032133 ( 2013 ) . bastarrachea - magnani , s. lerma - hernndez and j.g . hirsch , phys . rev . a * 89 * , 032101 , 032102 ( 2014 ) . p. cejnar , m. macek , s. heinze , j. jolie and j. dobe , j. phys . a * 39 * , l515 ( 2006 ) . caprio , p. cejnar and f. iachello , ann . ( n.y . ) * 323*,1106 ( 2008 ) . p. cejnar and p. strnsk , phys . e * 78 * , 031130 ( 2008 ) . p. strnsk , m. macek and p. cejnar , ann . phys . * 345 * , 73 ( 2014 ) . p. strnsk , m. macek , a. leviatan and p. cejnar , ann . phys . * 356 * , 57 ( 2015 ) . p. strnsk and p. cejnar , phys . a * 380 * , 2637 ( 2016 ) . p. cejnar and p. strnsk , phys . * 91 * ( 2016 ) 083006 . a. osterloh , l. amico , g. falci and r. fazio , nature * 416 * , 608 ( 2002 ) . j. vidal , g. palacois and r. mosseri , phys . a * 69 * , 022107 ( 2004 ) . j. vidal , r. mosseri and j. dukelsky , phys . a * 69 * , 054101 ( 2004 ) . n. lambert , c. emary and t. brandes , phys . lett . * 92 * , 073602 ( 2004 ) . n. lambert , c. emary , and t. brandes , phys . a * 71 * , 053804 ( 2005 ) . p. cejnar and j. jolie , phys . e * 58 * , 387 ( 1998 ) . p. cejnar and p. strnsk , aip conf . proc . * 1575 * , 23 ( 2014 ) . m.a . bastarrachea - magnani , b. lpez - del - carpio , s. lerma - hernndez and j.g . hirsch , phys . a * 90 * , 068015 ( 2015 ) . bastarrachea - magnani , b. lpez - del - carpio , j. chvez - carlos , s. lerma - hernndez and j.g . hirsch , phys . e * 93 * , 022215 ( 2016 ) . j. chvez - carlos , m.a . bastarrachea - magnani , s. lerma - hernndez and j.g . hirsch ( to be published ) ; see arxiv:1604.00725 [ nlin.cd ] ( 2016 ) . bastarrachea - magnani , s. lerma - hernndez and j.g . hirsch ( to be published ) ; see arxiv:1605.05357 [ quant - ph ] ( 2016 ) . a. peres , phys . rev . lett . * 53 * , 1711 ( 1984 ) . p. strnsk , p. hruka and p. cejnar , phys . e * 79 * , 066201 ( 2009 ) . a. peres , phys . * 77 * , 1413 ( 1996 ) . bennett , d. divincenzo , j. smolin and w.k . wooters , phys . a * 53 * , 3824 ( 1996 ) . s. hill and w.k . wootters , phys . * 78 * ( 1997 ) 5022 . wootters , phys . * 80 * ( 1998 ) 2245 . x. wang and k. mlmer , eur . j. d * 18 * 385 ( 2002 ) , x. wang and b.c . sanders , phys . a * 68 * , 012101 ( 2003 ) . s. heinze , p. cejnar , j. jolie and m. macek , phys . c * 73 * 014306 ( 2006 ) . y. alhassid and a. leviatan , j. phys . a : math . gen . * 25 * , l1265 ( 1992 ) . a. leviatan , prog . . phys . * 66 * , 93 ( 2011 ) .

the discovery of a higgs boson @xcite and the lack of evidence for beyond the standard model ( sm ) physics has placed natural models of electro - weak symmetry breaking under stress . most prominently , the framework of supersymmetry ( susy ) for addressing the hierarchy problem is looking more and more constrained @xcite . as this situation continues , the possibility that the hierarchy problem , which after all is a theoretical one , is not addressed in the traditional `` natural '' sense should be considered . we are led to consider other problems facing the sm within frameworks in which the electro - weak scale is tuned . in this work we will consider a class of models that attempt to explain the flavor structure of fermion masses and mixings . there are many possible frameworks to get exponential fermion hierarchy from input parameters that are of order one . these include froggatt - nielsen models @xcite , fermion profiles in flat @xcite or warped @xcite extra dimensions , and models with large anomalous dimensions @xcite . in fact , within these frameworks it is quite generic to have hierarchies that are too large . here we will focus on models of radiative fermion masses @xcite . within this framework the heavy fermions receive their mass at the tree level and light fermions receive a mass through loops of new heavy particles . there are many realizations of this idea both in the context of supersymmetric models @xcite and non - supersymmetric models @xcite . the common ingredient of all these models is the inclusion of new states that mediate flavor violation . the variety of possibilities for quantum numbers and interactions leads to a plethora of models . an attractive feature of the framework of radiative fermion masses is that the typical mass hierarchy between fermions of adjacent generations is roughly a loop factor ( with variations among models ) , in qualitative agreement with the observed spectra . a larger hierarchy is , of course , possible , but this requires additional small numbers or approximate symmetries . in our study we will use susy as a guiding principle for the particles and interactions that mediate flavor violation . because our starting point is a tuned electro - weak scale , one could have expected that we can forget about susy altogether . however , as a unique extension of the space - time symmetries of the s - matrix in four dimensions , susy is well motivated at high scales , irrespective of its relevance for setting the electro - weak scale . indeed , we will find that introducing susy at a high scale ( a pev or above ) automatically includes all of the necessary ingredients to mediate flavor violation from the third generation fermions to those of the first and second families . a lower susy scale would also yield a similar fermion mass spectrum but at the cost of introducing unacceptably high rates of flavor changing neutral current processes . in this sense , when enumerating the benefits of supersymmetry , we are trading a few orders of magnitude in its solution to the hierarchy problem for an explanation of flavor . the minimal model we will consider is remarkably simple - it is a scaled up version of the minimal supersymmetric standard model ( mssm ) , with supersymmetry broken at or above a pev and with tree level yukawas just for the third generation . supersymmetry breaking is assumed to be flavor anarchical and flavor violation is mediated from the third to the lighter generations at 1-loop . this simple setup will yield a spectrum that is qualitatively similar to observation , with a fully hierarchical spectrum and hierarchical mixings . the minimal setup will have a few quantitative problems which will be addressed subsequently . our framework differs from many other models of radiative fermion masses in that both the second and first generations receive a mass at the 1-loop level . one may worry that this generically does not give a hierarchical spectrum , but one in which the first and second generations are nearly degenerate . however , this is not the case . as we shall see , thanks to a novel mechanism , the typical spectrum in our framework is fully hierarchical , typically giving a ratio of @xmath0 for first to second generation masses . ratios of order one are not achievable . it should be noted that this small ratio is not the result of a small parameter or of a symmetry at the lagrangian level . instead it arises due to symmetry properties of the loop _ integrand _ which mediates flavor violation , which are approximately preserved at the integral level . we will discuss the main mechanism qualitatively in section [ sec : mechanism ] . in section [ sec : spectrum ] we will discuss the fermion spectrum in the minimal model quantitatively , both in the mass insertion approximation and by performing a numerical scan . we also discuss the constraints that vacuum stability place on the model . an examination of quark mixing will be presented in section [ sec : mixing ] , where we will consider the cases of flavor conserving and flavor violating @xmath1-terms separately . in section [ sec : mod ] we will consider modifications of the minimal model that allow for milder hierarchies in the fermion spectrum , particularly raising the muon mass . in section [ sec : unification ] we comment on the compatibility of our frameworks with gauge and yukawa coupling unification . in section [ sec : conclusion ] we conclude . experimental values for the quark and lepton masses as well as ckm parameters are collected in appendix [ app : masses ] . a brief discussion of vacuum stability constraints in the presence of generic sfermion flavor mixing can be found in appendix [ app : vac ] . finally , in appendices [ app : flavor ] and [ app : higgs ] we discuss constraints from low energy flavor observables and the higgs mass . we will begin by considering the most minimal setup to demonstrate the mechanism for radiative generation of flavor by sfermions . this simple setup will yield a fermion spectrum that is qualitatively similar to that observed in the sm , that of a hierarchical spectrum of fermion masses . in this section we will focus on the origin of the hierarchy . we will asses its success qualitatively in the next two section . this minimal setup has been considered briefly in the context of tev scale susy @xcite , only to be dismissed because of excessive flavor changing effects and/or the lack of naturalness . giving - up on naturalness frees us to consider this model and some of its virtues in greater detail . the particle content of the minimal setup includes just the particles of the mssm with a high susy breaking scale and a tuned electro - weak scale ( the spectrum is _ not _ split as in split - susy @xcite , distinguishing our model from @xcite ) . in the susy limit we assume only a single generation of fermions participates in a yukawa interaction with the higgs . we can thus rotate to a basis in which the yukawa matrices take the form @xmath2 this form can easily be arranged by assuming that matter interacts with the higgs only through mixing with a single generation of heavy vector - like chiral superfields which ensures the yukawa matrix is of rank 1 ( see for example @xcite ) . this yukawa interaction respects a global @xmath3 flavor symmetry , with one @xmath4 acting on the first two generations for each of the the @xmath5 and @xmath6 representations . in addition to generating a mass for the third generation , the couplings of equation ( [ eq : yukawa - tree ] ) induce non - holomorphic scalar trilinear terms of the form @xmath7 which obviously respect the same symmetry . the mass scale of all superpartners , including scalars , gauginos and higgsinos , will be set by a common susy breaking scale , @xmath8 which we will take to be at or above 1000 tev . for now , we will assume that susy breaking trilinear terms such as @xmath9 are aligned in flavor space with the tree level yukawa matrix of equation ( [ eq : yukawa - tree ] ) , respecting the @xmath3 flavor symmetry @xmath10 with @xmath11 , @xmath12 , and @xmath13 also of order @xmath8 . this assumption will be relaxed in section [ sec : aterms ] . in contrast , we assume that the scalar masses are _ completely anarchical _ , with a texture @xmath14 as a result , after susy breaking the flavor symmetry is completely broken . flavor violation will be mediated to the first and second generation fermions by loops of squarks and sleptons and the full yukawa matrix is generated at 1-loop . the dominant diagrams are shown in figure [ fig : diagrams1 ] . we note that in the case mixing in the up - top and charm - top sector , quark field renormalization diagrams contribute at the same order as the vertex correction shown in figure [ fig : diagrams1](a ) . however , the affect of these 1 - 3 and 2 - 3 entries on the fermion mass spectrum will be suppressed by the large top mass and will not affect the discussion of section [ sec : mechanism2 ] . in the down and lepton sectors , the field renormalization diagrams are suppressed by @xmath15 compared to the vertex corrections and can be neglected . we will now examine the fermion mass matrices that result from this minimal setup in greater detail and show that the mass spectrum is fully hierarchical . run over the mass eigenstates of left- and right - handed sfermions . contributions to second generation masses as well as flavor off - diagonal masses are generated by analogous diagrams.,title="fig:",scaledwidth=30.0% ] run over the mass eigenstates of left- and right - handed sfermions . contributions to second generation masses as well as flavor off - diagonal masses are generated by analogous diagrams.,title="fig:",scaledwidth=30.0% ] run over the mass eigenstates of left- and right - handed sfermions . contributions to second generation masses as well as flavor off - diagonal masses are generated by analogous diagrams.,title="fig:",scaledwidth=30.0% ] given our setup above it is clear that there will be a hierarchy of at least a loop factor between the third generation and the lighter two . one may worry , however , that because both the first and second generation masses are generated at 1-loop , they will be roughly degenerate and that additional flavor structure in equation ( [ eq : scalar - mass ] ) will be needed to explain the first - second generation mass splittings . we show here that this is not the case . in fact , even with fully anarchical scalar masses , _ our model is incapable of producing a degenerate electron and muon , and a mass splitting of order 100 is typical_. the same holds for quarks . for concreteness , let us consider the mass matrix of up - type quarks . the dominant contribution to the quark mass matrix , @xmath16 , has the form @xmath17 by construction , @xmath18 is a matrix of rank 1 which can be written in a basis in which only its bottom - right entry is non - zero as in equation ( [ eq : yukawa - tree ] ) . we will now show that @xmath19 is approximately a rank 1 matrix that is not aligned with @xmath18 and thus gives a mass dominantly to the second generation . the first generation mass arises only to the degree that @xmath19 deviates from rank 1 . evaluating the loop integral from diagram ( a ) in figure [ fig : diagrams1 ] , we can write the loop induced mass matrix @xmath19 as @xmath20 where the @xmath21 s are the mixing matrices from the left - handed and right - handed squarks , and we have already performed a wick rotation . it is useful to re - write the loop induced mass as follows @xmath22 where @xmath23 and @xmath24 . in equation ( [ eq : int ] ) we also converted the @xmath25 integral to an integral over @xmath26 ( recall that we already wick rotated ) . the object @xmath27 is a vector in flavor space which depends on @xmath28 and decides which fermion receives a mass from any given loop scale . it is interesting to note that for large loop momenta , @xmath29 , @xmath30 points purely in the third generation direction due to the unitarity of the mixing matrix @xmath31 . to further investigate the properties of @xmath32 , we normalize the @xmath30 s to unit vectors and rewrite @xmath33 where the unit vectors are @xmath34 in the last step we have used the unitarity of the squark mixing matrix . the magnitude of the mass contributed by the momentum shell @xmath28 is @xmath35 ) and ( [ eq : deltamk ] ) : * the loop _ integrand _ has the form @xmath36 , which is a rank 1 matrix , and thus respects a @xmath37 symmetry . because the vectors @xmath38 change with @xmath39 , the symmetry is broken `` collectively '' by the integration over all momentum scales . in other words , every fixed slice of loop momentum contributes mass to just one linear combination of fermions , and the integral is required to give mass to other fermions . * the magnitude of the loop integrand , @xmath40 is a peaked function , dominated at the susy breaking scale @xmath8 . it rises like @xmath41 for low @xmath39 and falls like @xmath42 at high @xmath39 s . this means that the radiatively generated fermion mass will be approximately aligned with the integrand evaluated at the scale @xmath43 . combining these observations gives us the following narrative for the hierarchy among the second and first generations : the loop momenta that dominate the loop integral , which are of order @xmath43 , give a mass to just one linear combination of fermions . we can define this combination as the second generation . as we vary @xmath39 away from @xmath8 , by the time the @xmath30 s are sufficiently mis - aligned to contribute mass to the first generation the loop integrand is already small and the first generation mass is suppressed . note that this mechanism is independent of the flavor structure in the scalar sector and is present even when the scalar masses are fully anarchic . it is worth noting that the generated hierarchy does not arise from a small parameter in our theory , but from the properties of the loop integral and collective symmetry breaking in loop momentum space . only one linear combination of fermions receives mass . the flavor vector @xmath44 determines the combination of fermions while the function @xmath45 sets the magnitude of the mass . due to the strong localization of @xmath45 the resulting mass matrix is approximately rank 1.,title="fig:",scaledwidth=54.0% ] + only one linear combination of fermions receives mass . the flavor vector @xmath44 determines the combination of fermions while the function @xmath45 sets the magnitude of the mass . due to the strong localization of @xmath45 the resulting mass matrix is approximately rank 1.,title="fig:",scaledwidth=52.0% ] + to demonstrate this mechanism more visually , we plot @xmath44 and @xmath40 for a generic choice , @xmath46 , @xmath47 , @xmath48 , @xmath49 , @xmath50 , in units of pev@xmath51 . the gluino mass is chosen to be 1 pev . the qualitative features discussed here do not depend on this choice . ] of scalar masses in figure [ fig : fi ] ( where we assumed @xmath52 for simplicity ) . we see that the contribution to the fermion mass is indeed `` localized '' to momenta of order @xmath53 . the unit vectors @xmath44 do not change drastically in this region of loop momentum and define the second generation . the first generation receives a mass only to the degree that the @xmath54 s manage to change in the 1 - 2 plane in this region . having a qualitative understanding of our mechanism for fermion hierarchy , we continue to investigate the spectrum more quantitatively in the next section . we would like to get a clearer understanding of the fermion spectrum in our model and to further quantify the suppression effect of first generation masses . in subsection [ sec : minimal_analytic ] we will do this first analytically . to keep the formulae manageable we will work in the mass insertion approximation and take into account only the dominant 1-loop contributions shown in figure [ fig : diagrams1 ] . in subsection [ sec : stability ] we will consider the constraints that vacuum stability places on the parameter space . in subsection [ sec : minimal_numeric ] we will perform a numerical scan in the mass basis to see that the mass insertion results are robust . we parameterize the soft masses of squarks and sleptons as @xmath55 @xmath56 where @xmath57 are dimensionless matrices that encode the flavor breaking and mass splittings , and whose elements are all allowed to be @xmath58 . following @xcite , we use @xmath3 flavor rotations that leave the yukawa couplings invariant , and rotate to a basis where the @xmath59 and @xmath60 entries of all @xmath61 s vanish @xmath62 @xmath63 assuming for simplicity a susy spectrum with degenerate sfermions and gauginos , @xmath64 , we find the following fermion mass spectrum at leading order in the mass insertion approximation @xmath65 the @xmath66 and @xmath67 factors in the down - type quark and lepton mass ratios arise from the loop corrections to the tree - level bottom and tau masses . as they are @xmath68 enhanced , they can have a non - negligible effect these loop corrections can dominate the bottom and tau masses @xcite . in this case the top - bottom and top - tau hierarchies are also explained by loops . this is indeed one of the interesting regions of parameter space of our model . ] . using the same approximations as above , we have @xmath69 pev , the stop trilinear coupling is @xmath70 and @xmath68 is varied between @xmath71 . the mass insertions are varied in the range @xmath72 . shown are also the observed @xmath73 fermion masses at the electro - weak scale , with their @xmath74 uncertainties.,scaledwidth=60.0% ] the generic predictions of the loop induced spectrum ( [ eq : radmass1 ] ) - ( [ eq : radmass6 ] ) are summarized in figure [ fig : chart_mssm ] and compared to the experimental values of the quark and lepton masses . the gray regions are obtained by setting all susy masses to a common value @xmath75 pev and the stop trilinear coupling to @xmath70 . all mass insertions are varied between @xmath72 and @xmath71 . we now discuss some of the features of this spectrum in more detail . [ [ section ] ] the top yukawa coupling is of order one by construction . the masses of the bottom and the tau are @xmath15 suppressed . it is thus attractive to take a large value for @xmath68 , allowing for bottom and tau yukawa couplings of order one . [ [ section-1 ] ] the second generation mass , as expected , is a 1-loop factor below the third generation . it is , however , worth taking note of the @xmath68 scaling . the loop induced second generation mass is proportional to the trilinear connecting the up - type higgs to the corresponding scalars . for the up - type quarks this is the supersymmetry breaking @xmath11 , while for the down - type quarks and leptons it is dominated by the non - holomorphic trilinear , scaling like @xmath76 . we thus find that the @xmath68 scaling which was suppressing the bottom and @xmath77 masses is un - done for the second generation . the strange quark and muon masses are thus parametrically similar to the charm quark mass ( at least as far as the @xmath68 scaling is concerned ) . we view this as a desirable feature . the trilinear couplings discussed here are constrained by vacuum stability , and the naive hierarchy between third and second generation fermions are a bit too large . we will address this in detail later . [ [ section-2 ] ] as anticipated , a hierarchy between the first and second generation masses is generated automatically even for a completely anarchic flavor structure of the sfermions @xmath78 comparing to the experimentally determined ratios collected in the appendix ( [ eq : masses1])([eq : masses3 ] ) , we see that adjusting the involved mass insertions by suitable @xmath58 factors allows to accommodate the observed ratios of first and second generation masses easily . before plugging in numbers and getting detailed spectra , we pause to consider the constraint that vacuum stability places on our parameter space . a supersymmetric theory invariably introduces many colored and charged scalars and care must be taken to ensure our electro - weak breaking vacuum is stable , or at least meta - stable . the first consideration is to ensure none of the charged and colored scalars are tachionic , namely , the eigenvalues of the anarchic scalar mass - squared matrices must all be positive . as off - diagonal entries in a hermitian matrix always cause a lower determinant , it is clear that requiring positive eigenvalues bounds them from above . the analytic form of the upper bound on any particular flavor violating mass insertion depends on the other entries of the mass matrix and is not enlightening for our discussion . in practice we find that the flavor violation mass insertions tend to be less than 1/2 . in the next subsection we will take a pragmatic approach , and scan over all anarchic scalar mass matrices . matrices that produce one or more tachyons in their spectrum are simply discarded . we note that the 2 - 3 and 1 - 2 fermion mass hierarchies are set by the off - diagonal mass insertions @xmath79 and @xmath80 respectively . the vacuum stability requirement thus pushes our fermion mass spectrum to be somewhat more hierarchical than it would have been without this constraint . even if all scalars have a positive mass - squared , we also need to consider more distant color and charge breaking vacua to which our vacuum can decay . the radiatively induced fermion masses depend sensitively on the stop trilinear coupling @xmath11 ( in the case of the up - type quarks ) and on the non - holomorphic trilinear which is set by @xmath81 ( in the case of down - type quarks and the leptons ) . like any trilinear term , these interactions may induce new minima , or runaway directions . in particular , it is well known @xcite , that a large stop trilinear can induce a deep charge and color breaking minimum in the mssm scalar potential along the d - flat direction @xmath82 . requiring that the electro - weak vacuum is the deepest minimum in the mssm scalar potential leads to a bound on the stop trilinear coupling @xmath83 this bound is derived under the assumption of flavor conserving squark soft masses . in the general case with @xmath58 squark mixing , the bound can be stronger as discussed in the appendix [ app : vac ] . if one allows the electro - weak vacuum to decay , and only demands that its life - time be longer than the age of the universe , the bound gets slightly relaxed @xcite . the upper bound on the stop trilinear coupling constrains the size of the radiatively induced charm and up masses . note that the constraint ( [ eq : atbound ] ) on the relative size of the trilinear coupling with respect to the soft masses is independent on the overall mass scale . this statement remains approximately true also if one allows for a meta - stable vacuum . for large values of @xmath81 , vacua with sbottom or stau vevs can appear in the mssm scalar potential @xcite . even if there are no exact d - flat directions involving only the up - type higgs and sbottoms or staus , the trilinear couplings of the up - type higgs with the sbottoms and staus can lead to charge and color breaking minima , if the product of higgsino mass and bottom / tau yukawa is of the order of the squark / slepton soft masses . simple analytical bounds for the higgsino mass equivalent to ( [ eq : atbound ] ) can only be obtained for a fixed direction in field space . for any given field direction , the bounds are only necessary but usually not sufficient to guarantee the absence of charge and color breaking minima . in the large @xmath68 regime , we find that in the optimized field direction @xmath84 is usually a bit larger than the sfermion fields . for definiteness we chose here exemplarily @xmath85 and @xmath86 . in this case we find the necessary conditions @xmath87~ , \\ \label{eq : mubound2 } |\mu|^2 & \lesssim & ( m_{\tilde b_l}^2 + m_{\tilde b_r}^2 ) \left [ \frac{1}{2 } + \frac{9}{32 } \frac{v^2}{m_b^2 } \left|\frac{1}{t_\beta } + \epsilon_b \right|^2(g_1 ^ 2 + g_2 ^ 2 ) \right]~,\end{aligned}\ ] ] where we expressed the yukawa couplings in terms of fermion masses , and included the @xmath68 enhanced threshold corrections @xmath67 and @xmath66 . depending on the phase of the @xmath88 , the threshold corrections can significantly strengthen or loosen the constraints . the fermion masses as well as the gauge couplings in ( [ eq : mubound ] ) and ( [ eq : mubound2 ] ) should be evaluated at the pev scale . in the presence of generic flavor mixing of the squarks and sleptons , the bounds can become stronger as discussed in the appendix [ app : vac ] . analogously to the case of the stop trilinear coupling , requiring only meta stability will relax the bounds slightly . nonetheless , it is clear from ( [ eq : mubound ] ) and ( [ eq : mubound2 ] ) that for large @xmath89 the higgsino mass , @xmath81 , can not exceed the sbottom and stau masses by more than a factor of few . how do these constraints affect the fermion mass spectrum ? to obtain a charm mass ( [ eq : radmass1 ] ) in agreement with the data ( see appendix [ app : masses ] ) , maximal mass insertions and a large @xmath11 term close to the vacuum stability bound ( [ eq : atbound ] ) are required . a large enough strange quark mass ( [ eq : radmass3 ] ) is only possible in the large @xmath89 regime and for a higgsino mass at least of the order of the squark masses . even for large @xmath68 and maximal mass insertions in the slepton sector , the muon mass is typically one order of magnitude below the measured value . in order to obtain a muon mass from ( [ eq : radmass5 ] ) in agreement with observation , the higgsino mass needs to be @xmath90 where in the last step we chose @xmath91 , neglected the @xmath67 term and set all mass insertions generously to 1 . a higgsino mass as large as in ( [ eq : mu ] ) is strongly excluded by the vacuum stability requirements discussed above . one may hope for an accidental cancellation between the @xmath92 and the @xmath67 terms , which can bring equations ( [ eq : mubound ] ) and ( [ eq : mu ] ) into apparent agreement . we note first that such an accidental cancellation would be rare indeed , particularly in light of the fact that the phase of @xmath67 is arbitrary . in addition , we note that equation ( [ eq : mubound ] ) considers a particular direction in field space . this is not the direction of greatest instability in the case of extreme cancellations in the potential . in section [ sec : mod ] we will return to the problem of the muon mass in our model and propose solutions . the conclusions of subsections [ sec : minimal_analytic ] and [ sec : stability ] are based on the mass insertion approximation and include only the leading diagrams which contribute to fermion masses . one would like to explore the parameter space beyond these approximations and to investigate for example the possibility of cancellations among various diagrams . to investigate our parameter space fully , in this subsection we will analyze the radiatively induced fermion masses numerically . we will do this without making use of the mass insertion approximation and taking into account the complete set of 1-loop contributions that consist of the gluino and bino loops shown in figure [ fig : diagrams1 ] , as well as wino loops , higgsino loops , wino - higgsino loops , and bino - higgsino loops . in order to find the ranges of fermion masses that can be accommodated in the discussed setup , we perform a numerical scan of the model parameters , the results of which are shown in figures [ fig : scan_masses_mssm ] and [ fig:1st2nd_mssm ] . apart from small renormalization group effects to be discussed below , the loop induced fermion mass matrices are independent of the overall scale of susy particles @xmath8 that therefore can be set to an arbitrary value @xmath75 pev . we scan all mass parameters of the model ( gaugino masses , higgsino mass , trilinear couplings , diagonal and off - diagonal sfermion masses ) within one order of magnitude around the common scale @xmath93 $ ] using flat distributions . we also assign random phases between @xmath94 and @xmath95 to all complex parameters . the third generation yukawa couplings are fixed such that the third generation masses agree with the observed values . we impose the absence of tachyons as well as the approximate vacuum stability bounds in ( [ eq : atbound ] ) , ( [ eq : mubound ] ) , and ( [ eq : mubound2 ] ) . masses at the electro - weak scale . shown are also the observed masses with @xmath74 uncertainties . ] in comparing our results to observation we also take into account renormalization group effects . in particular , we evaluate the gauge couplings at the scale @xmath75 pev @xmath96 these values are obtained using sm 2-loop renormalization group equations ( rges ) @xcite . integrating out the susy particles , we obtain ratios of quark and lepton masses at the susy scale @xmath97 . in order to compare the ratios to the experimentally measured mass ratios given in the appendix [ app : masses ] , we run them down to the electro - weak scale using again sm 2-loop rges . at 1-loop , the effect of the gauge interactions cancel in the ratios . only the top yukawa leads to a correction of the ratios @xmath98 and @xmath99 . in a 1-loop leading log approximation we find @xmath100 ~.\ ] ] for @xmath75 pev , the rge correction in ( [ eq : running ] ) is approximately @xmath101 and therefore hardly relevant . all other mass ratios are to an excellent approximation rge invariant . the value of @xmath68 is fixed to @xmath102 in the scan . smaller values of @xmath68 lead to smaller masses of the down - type quarks and leptons . larger values of @xmath68 lead to stronger constraints from stau and sbottom vevs . we find that @xmath103 does not lead to appreciably larger down - type quark and lepton masses . the distributions for the various masses are shown in figure [ fig : scan_masses_mssm ] compared to their observed values . as anticipated , the model , qualitatively , is remarkably successful . a fully hierarchical spectrum , with significant 1 - 2 and 2 - 3 splittings , is a typical outcome of this framework . as expected , the muon mass prediction can be satisfied only at the extreme tail of our distribution , a problem we will address later . masses at the electro - weak scale . the black stars indicate the observed values . the diagonal bands show the observed ratios of masses with @xmath74 uncertainties.,title="fig:",scaledwidth=45.0% ] masses at the electro - weak scale . the black stars indicate the observed values . the diagonal bands show the observed ratios of masses with @xmath74 uncertainties.,title="fig:",scaledwidth=45.0% ] + masses at the electro - weak scale . the black stars indicate the observed values . the diagonal bands show the observed ratios of masses with @xmath74 uncertainties.,title="fig:",scaledwidth=45.0% ] it is also enlightening to inspect the two - dimensional distribution of first and second generation masses . this unveils any non - trivial correlations among these parameters . a scatter of scan points is shown in figure [ fig:1st2nd_mssm ] for the up - type and down - type quarks as well as for leptons . we see that the up - type sector is in good agreement with observation . for the down - type sector , observation is at the edge of the model prediction and we would benefit from increasing the down type mass . in the lepton sector , the situation can improve by increasing both muon and electron masses , while keeping their ratio fixed . we now turn to examining quark mixing , i.e. the misalignment of the left - handed up and down quarks in flavor space . we will find that in the minimal model quark mixing is generated radiatively , but that the cabibbo angle is typically too small . we will then show that this can be addressed simply by allowing for flavor violating trilinear terms . just like fermion masses , the ckm matrix is generated radiatively in our framework . we will first analyze quark mixing in the minimal model as it has been described so far . in line with the previous subsections , we will first get a qualitative understanding of the result , and then proceed to consider the results analytically ( in the mass insertion approximation ) and numerically ( without approximation ) . as was the case for the fermion masses , it is enlightening to inspect the loop induced contribution to the ckm matrix at the _ integrand _ level , because symmetries that are exact for the integrand are manifested approximately at the integral level due to the dominance of a single scale . purposely dropping the loop integral , the up - type and down - type mass matrices have the following structure @xmath104 to see the parametric structure of the ckm matrix we can pick a convenient basis . obviously , we will want to work in a basis where only the 33 entry of @xmath18 is non - zero . we can further do @xmath4 rotations between the lighter generations such that @xmath105 . in this basis both the up and the down matrices in equation ( [ eq : up - and - down ] ) are non - vanishing only in the lower two - by - two diagonal block . these matrices can thus be diagonalized with a rotation between the second and third generation . the relative left handed rotation between them , which is the ckm matrix , will also be purely a 2 - 3 rotation . recalling that the structure of equation ( [ eq : up - and - down ] ) is merely approximate , we expect a suppression of the 1 - 2 and 1 - 3 ckm entries . the ckm matrix is fully determined by four parameters . we chose the cabibbo angle @xmath106 and the absolute values of the ckm elements @xmath107 and @xmath108 as well as the ckm phase . our expectation from above is that @xmath108 and @xmath109 will be suppressed . using the same approximations as in section [ sec : minimal_analytic ] , we find the following analytic expressions @xmath110 these expressions have to be compared to the measured values that are collected in the appendix [ app : masses ] . the physical phase of the ckm matrix is generically of @xmath58 in our model , in agreement with observation . the smallness of the ckm element @xmath107 is explained by a loop factor . its measured value can be easily accommodated in the large @xmath68 regime . as explained above , the loop induced quark mass matrices are approximately rank 1 , and mixing between the first and third generation is therefore suppressed by a small numerical factor . the resulting expression for the ckm element @xmath108 is smaller than @xmath107 by one order of magnitude , exactly as observed . finally , the cabibbo angle @xmath106 is not loop suppressed because it parametrizes the mixing between the first and second generation of quarks , which both acquire mass at the same loop level . however , again due to the fact that the loop induced quark mass matrices are approximately rank 1 , the cabibbo angle is suppressed by a small numerical factor . moreover , the mixing between the first and second generation of left handed quarks is proportional to @xmath111 and aligned in the up and down sectors at leading order in mass insertions . the cabibbo angle is therefore only induced at higher order in the mass insertion approximation , involving also the mixing of right handed quarks . this double suppression leads to a very small value of @xmath106 . even setting the mass insertions in ( [ eq : lambda ] ) to 1 , @xmath106 is almost one order of magnitude below the observed value . . ( right ) : distributions of the ckm parameters as obtained by a parameter scan of the minimal model . shown are also the observed values of the ckm parameters with @xmath74 uncertainties.,title="fig:",scaledwidth=43.0% ] . ( right ) : distributions of the ckm parameters as obtained by a parameter scan of the minimal model . shown are also the observed values of the ckm parameters with @xmath74 uncertainties.,title="fig:",scaledwidth=43.0% ] the generic predictions for the ckm parameters ( [ eq : lambda ] ) , ( [ eq : vcb ] ) , and ( [ eq : vub ] ) are also summarized in the left plot of figure [ fig : scan_ckm_mssm ] where we set all mass scales equal and vary @xmath72 and @xmath71 . in the right plot of figure [ fig : scan_ckm_mssm ] , we show the distributions of the ckm parameters that result from the parameter scan discussed in section [ sec : minimal_numeric ] . the observed values for @xmath112 and @xmath113 can be easily reproduced by the model . however , as expected from the analytical results , the cabibbo angle is generically one order of magnitude below the measured value and tends to be even smaller than @xmath112 . the next section will describe ways to increase the cabibbo angle . we now consider the possibility that all soft terms , both masses and trilinears , violate the flavor symmetry . traditionally , in supersymmetric theories the @xmath1-terms are taken to be proportional to the yukawa matrices for several reasons - constraints on flavor changing neutral currents , vacuum stability , as well as model building considerations . flavor alignment is certainly motivated from a model building point of view but is not necessary - one could imagine writing an operator of the form @xmath114 where @xmath115 is a hidden sector field that gets an @xmath116-term vev and @xmath117 is a messenger scale . the matrix @xmath118 need not be proportional to the yukawa matrix . indeed , in gravity mediation , where @xmath117 is the planck scale , the lore is that the theory of quantum gravity does not respect any flavor symmetries and one could naively expect @xmath119 in equation ( [ eq : fvaterm ] ) to be not aligned with the yukawa matrix . the constraints on flavor violating @xmath1-terms from flavor changing neutral currents are avoided in our framework due to the large supersymmetry breaking scale , but constraints from vacuum stability certainly do need to be considered . trilinear scalar terms can always drive an instability at large field values unless they are countered by a sufficiently large quadratic term . in the mssm the supersymmetric yukawa terms serve as stabilizing quartic couplings . if the @xmath1-terms are significantly not aligned with these yukawas , a runaway direction may arise . completely anarchic trilinear couplings are therefore subject to strong constraints from vacuum stability @xcite . in particular , soft trilinear couplings of the first and second generation as well as trilinears that mix the first with the second generation are strongly constrained , because f - term quartics for the first and second generation sfermions are absent at tree level in the considered framework . flavor changing trilinear couplings involving the third generation are less constrained because they are at least partially aligned with the yukawa couplings . focusing on flavor changing trilinears in the up sector , we write @xmath120 necessary bounds on the flavor changing trilinears in the up sector read @xcite @xmath121 in general it is possible to derive stronger bounds by optimizing the direction in field space . limiting ourselves to the case discussed below , namely the presence of @xmath11 , @xmath122 and @xmath123 trilinear couplings , and assuming for simplicity flavor diagonal soft masses , we find the following bound @xmath124 with @xmath125 , @xmath126 , and @xmath127 . the inequality in ( [ eq : atubound ] ) has to hold for all angles @xmath128 . in the limit @xmath129 we recover the bound on @xmath11 given in ( [ eq : atbound ] ) ; for @xmath130 and @xmath131 we obtain the first bound in equation ( [ eq : fcabound ] ) , etc . similar to the case of the flavor conserving trilinears discussed in section [ sec : stability ] , the vacuum stability bounds can become tighter , if one takes into account also flavor violation in the soft masses . the bounds will be slightly relaxed if one demands only meta - stability @xcite . our philosophy for flavor violating @xmath1-terms will be similar to flavor violating soft masses : we will allow anarchic flavor violation so long as it does not cause a vacuum instability . for the soft masses this amounted to checking that there are no tachyons in the spectrum , but for trilinears the requirement is that no runaway or instability is caused at large field values . in practice this implies that only trilinears that involve the third family can be turned on . in the following , we consider the effect of switching on a small flavor changing @xmath122 and @xmath123 trilinear coupling . it is well known that flavor changing trilinear couplings can lead to large corrections to the ckm matrix elements @xcite . in general we expect this new source of flavor violation to mildly spoil the approximate rank 1 structure described in section [ sec : mechanism ] . we thus expect the up quark mass , as well as @xmath106 and @xmath108 to be particularly sensitive to the flavor changing trilinears . to understand the dependence of these parameters on the flavor violating @xmath1-terms we employ the mass insertion approximation defined in section [ sec : minimal_analytic ] , working to leading order in @xmath132 and @xmath133 . we further assume that @xmath132 and @xmath133 are of order @xmath134 ( smaller than @xmath61 ) . we find the following ckm matrix elements @xmath135 at the considered order , only the flavor changing trilinear @xmath122 affects the ckm elements . values for @xmath122 approximately one order of magnitude smaller than @xmath11 , @xmath136 , are sufficient to generate the observed cabibbo angle . an @xmath122 of that size also leads to an @xmath58 shift in the ckm element @xmath108 . note that the size of @xmath122 , required to generate the observed cabibbo angle is well below the vacuum stability bound in equation ( [ eq : fcabound ] ) . both flavor changing trilinears , @xmath122 and @xmath123 affect the up quark mass @xmath137 for @xmath138 , the shift in the up quark mass is of @xmath58 and therefore does not change the predicted hierarchy between up and charm mass significantly . . shown are also the observed values with @xmath74 uncertainties . , scaledwidth=43.0% ] results for the ckm parameters coming from a numerical scan that includes a non - zero @xmath122 are shown in figure [ fig : scan_fvaterms ] . we randomly scan @xmath139 ( with a flat distribution ) and allow for an arbitrary complex phase of @xmath122 . the agreement between the predicted values for the ckm elements and the observed values is greatly improved compared to the scan without flavor changing trilinears . flavor violating trilinear couplings in the down sector can lead to additional contributions to the down mass matrix and in turn also affect the ckm matrix . however , these contributions are suppressed by @xmath15 with respect to the leading contributions that are induced by the flavor conserving supersymmetric trilinear couplings proportional to the @xmath81 term ( [ eq : mutrilinear ] ) . as a consequence , effects from flavor violating trilinears in the down sector are insignificant . another possibility would be to introduce flavor violating trilinear couplings in the down sector that are non - holomorphic and therefore not suppressed by @xmath15 . a non - holomorphic @xmath140 coupling with absolute values approximately 10% of the @xmath81 term would have similar effects on the ckm matrix as @xmath122 discussed above . a non - holomorphic @xmath141 coupling of the same order as the @xmath81 term could be used to enhance the down quark mass by a factor of few without affecting the ckm matrix significantly . however , flavor violating non - holomorphic a - terms in the down sector are strongly constrained by vacuum stability considerations . we find that with trilinear couplings of the type @xmath142 and @xmath143 , the mssm tree level potential is unbounded from below . the electro - weak minimum can still be metastable in such a situation @xcite , but a dedicated analysis to establish the maximal values for @xmath141 and @xmath140 compatible with meta - stability bounds is beyond the scope of this work . in section [ sec : spectrum ] we saw that our minimal setup , the mssm with third generation yukawas and anarchic soft masses , successfully generates a hierarchical fermion spectrum . when compared to the observed hierarchy we find that , if anything , the hierarchy is too large . in particular , the ratio of the muon to tau mass is predicted to be too small . in this section we will consider variations to the minimal framework which yield a milder hierarchy . the first possibility we consider , still within the minimal model , is to allow for hard supersymmetry breaking . the second alternative we will discuss , is to extend the gauge structure beyond that of the mssm . so far we have considered a scaled up version of the traditional softly broken supersymmetric standard model . introducing only soft susy breaking is certainly well motivated by naturalness . but given that we have given up on electro - weak naturalness in the first paragraph of this paper , it is reasonable to consider also hard breaking of supersymmetry . here we define hard breaking as any symmetry breaking by a dimensionless coupling . for practical purposes we will consider such breaking of supersymmetry as a limit of soft supersymmetry breaking models by taking the supersymmetry breaking vev @xmath116 to be of the same order as the messenger scale @xmath144 . then , dimensionless susy breaking operators can be categorized in powers of @xmath145 @xcite . for our purposes , we will be interested in the possibility that the bino - fermion - sfermion vertex is larger than the hypercharge gauge coupling @xmath146 . within our framework this can come from the operator @xmath147 we will assume that this new source of supersymmetry breaking is flavor blind , allowing the main mechanisms we have discussed to remain intact . we note that relaxing this assumption somewhat may be interesting as an alternative to the flavor violating trilinears discussed in subsection [ sec : aterms ] . the scale @xmath116 may be the only source of supersymmetry breaking , in which case _ the _ mediation mechanism may be low scale . alternatively , there can be two sources of susy breaking giving two different scales @xmath148 and @xmath149 . the scale @xmath148 can be coupled to the mssm via gravity mediation giving rise to the lions share of soft masses . the scale @xmath149 can be coupled via a low scale mediation , @xmath150 , which contributes up to an order one of the soft masses as well as some hard breaking of the form of equation ( [ eq : hard ] ) . a setup like this fits our framework well because the high - scale gravity mediation is expected to be flavor anarchic while the low scale mediation could be flavor blind . increasing the bino - fermion - sfermion coupling @xmath151 from its supersymmetric value @xmath146 , will modify the fermion spectrum in our model . in particular , both the electron and muon masses will be shifted by a factor of @xmath152 . this will improve the agreement of these masses with observation significantly as can be seen in the left panel of figure [ fig : hard_bl ] where we have taken @xmath151 to be twice its supersymmetric value . this distribution is similar to that in figure [ fig:1st2nd_mssm ] with all points shifted by a factor of four along the diagonal line . we note that introducing significant hard susy breaking generically also leads to larger radiative corrections to third generation masses ( the parameters we called @xmath67 ) . number . in both cases the agreement may be improved by increasing the relevant couplings , see text for details.,title="fig:",scaledwidth=45.0% ] number . in both cases the agreement may be improved by increasing the relevant couplings , see text for details.,title="fig:",scaledwidth=45.0% ] increasing @xmath151 does not affect the quark masses significantly as these were dominated by squark - gluino loops . increasing the quark masses with hard susy breaking is possible for example by increasing the quark - squark - gluino vertex @xmath153 in a similar way . increasing the first and second generation masses by a common factor does indeed bring the bulk of our scan models closer to observation . however we note that the observed quark masses are certainly achievable within our minimal framework . finally , we note that hard supersymmetry breaking may affect the higgs mass . as discussed in appendix [ app : higgs ] , our frameworks requires lowering the higgs mass compared to the standard supersymmetric prediction . we discuss this further in that appendix . we now return to the framework of traditional softly broken supersymmetry and consider and alternative solution to raising the muon mass . in our minimal framework the mass ratio of second to third generation fermions is a function of the gauge representations and the mssm gauge couplings . a simple possibility to generating a milder hierarchy in the lepton sector is to assume that additional gauge symmetries are present at the supersymmetry breaking scale . indeed , decades of studying tev scale supersymmetry has taught us that @xmath154 is a natural scale for spontaneously broken gauge symmetries . any additional gauge symmetry under which both left and right handed matter transforms will include new diagrams similar to those in figure [ fig : diagrams1 ] but with a new gaugino . there are several ways one could envision enlarging the gauge structure of the mssm and we will leave a more complete analysis of the most motivated possibilities for future study . here , as a simple example we consider gauging @xmath155 in addition to the sm gauge group . the @xmath155 gauge symmetry can naturally be spontaneously broken at the scale of supersymmetry breaking @xmath8 . due to their larger charge , lepton masses would be affected by the new @xmath155 gaugino diagrams more than quarks . as opposed to the case of hard susy breaking , where we needed to assume the new contribution is flavor universal , in this case flavor universality is ensured by @xmath155 gauge invariance and the hierarchical structure of section [ sec : mechanism ] is automatically maintained . in the right panel of figure [ fig : hard_bl ] we show the effects of including a @xmath155 gauge interaction whose coupling strength is @xmath156 at the pev scale . this value is chosen by requiring that the landau pole in the @xmath155 coupling is postponed until the gut scale , thus maintaining the success of gauge coupling unification and hypercharge that is induced by rge s @xcite or add additional matter that renders the two sets of charges orthogonal . ] . comparing this to figure [ fig:1st2nd_mssm ] , we conclude that gauging @xmath155 can bring the value of lepton masses within the range predicted by our framework . of course , one can bring the most `` likely '' values of the muon mass closer to the observed value by increasing the @xmath155 coupling . this may be done at the cost of introducing a landau pole below the the canonical gut scale . unification of gauge couplings would still be possible in this case if it is accelerated @xcite . alternatively , one could embed the @xmath155 into a larger non - abelian group which is asymptotically free before the landau pole . we reiterate that @xmath155 is brought here as an example of how adding gauge structure around the supersymmetry breaking scale can improve our predicted lepton spectrum . turning this statement around we can consider the degree of hierarchy in the various sm representations as an indication of their gauge charges at @xmath8 . taking this seriously may suggest a gauge symmetry under which leptons are charged fairly strongly , such as pati - salam models @xcite . we leave this line of inquiry for an upcoming publication . we now discuss the compatibility of our framework with unification of gauge and yukawa couplings . the discussion here applies to the minimal model and its extension in section [ sec : hard ] . as stated above , the the compatibility of the @xmath155 extension of section [ sec : b - l ] warrants further study . one of the attractive features of tev scale susy is that it is compatible with gauge coupling unification at a gut scale around @xmath157 gev @xcite . given the existing constraints on flavor changing neutral current processes , the anarchic scalar masses in our framework must be of order 1 pev or above . as is shown in figure [ fig : gaugeunification ] , gauge coupling unification is still compatible with a susy breaking scale of a pev . for the running of the gauge couplings we use 2-loop rges . up to the pev scale we use the sm rges @xcite and above the pev scale we use those of the mssm @xcite . we neglect threshold corrections at the susy scale . in that case , threshold corrections at the gut scale at the level of 1% are required for successful gauge coupling unification ( see also @xcite ) . besides gauge coupling unification , susy guts can also be compatible with third generation yukawa unification @xcite , e.g. bottom - tau yukawa unification in guts based on the @xmath158 gauge group , or top - bottom - tau yukawa unification in @xmath159 guts . we find that our framework can accommodate yukawa unification easily . to show this we run the third generation yukawa couplings to the gut scale using 2-loop rges . at the susy scale we incorporate threshold corrections to the bottom and tau yukawas , @xmath160 and @xmath161 , @xmath162 threshold corrections to the top yukawa coupling are at most at the percent level and therefore negligible . gev . in the left ( right ) plot we set @xmath91 ( @xmath102).,title="fig:",scaledwidth=45.0% ] gev . in the left ( right ) plot we set @xmath91 ( @xmath102).,title="fig:",scaledwidth=45.0% ] in figure [ fig : yukawaunification ] we show the size of the threshold corrections to the bottom and tau yukawa couplings required to achieve third generation yukawa unification at a scale of @xmath163 gev . these are shown for @xmath91 and @xmath102 . the shaded bands correspond to the regions that are compatible with top - bottom , top - tau , and bottom - tau unification with an accuracy of better than 5% . the threshold corrections required for yukawa unification can be easily accommodated in our framework . they are approximately given by @xmath164 with @xmath66 and @xmath67 given in ( [ eq : eps ] ) . note that requiring third generation yukawa unification does not spoil the successful generation of the first and second generation quark and lepton masses . we checked explicitly that enforcing bottom - tau unification by requiring @xmath160 and @xmath161 to lie within the corresponding band shown in figure [ fig : yukawaunification ] , does not change the distribution of the radiatively induced fermion masses and quark mixing angles appreciably . models of radiative fermion masses are an elegant and attractive possibility to explain the hierarchical flavor pattern in the mass spectrum and mixing angles of the quarks and leptons in terms of loop factors . in this work , we discussed setups that allow to generate the hierarchical flavor structure of the sm fermions from loops of superpartners with anarchical soft masses . given the anarchic squark and slepton masses , the discussed setups are subject to strong constraints from low energy flavor observables . in particular , cp violation in kaon mixing generically sets the strongest lower bound on the susy scale of the order of @xmath165 pev . a spectrum with all susy particles at the pev scale is still compatible with gauge coupling unification and also with third generation yukawa unification at a gut scale of around @xmath166 gev . we found that the minimal supersymmetric standard model with flavor anarchic soft masses for the squarks and sleptons contains all necessary ingredients to generate a fully hierarchical spectrum of quarks and leptons . at tree level , rank 1 yukawas couple the higgs only to the third generation fermions and leave a @xmath3 symmetry acting on the first and second generation unbroken . the anarchic sfermion soft masses , on the other hand , break the sm flavor symmetry completely . loops containing gauginos and sfermions communicate the flavor symmetry breaking from the sfermion sector in the fermion sector and yukawa couplings for the first and second generation fermions are generated at the 1-loop level . while both first and second generation yukawas are generated at the 1-loop level , we pointed out a novel mechanism that ensures a large splitting , of generically two orders of magnitude , without introducing small parameters or symmetries . we showed that the radiatively induced yukawa matrices are approximately rank 1 . in combination with the rank 1 tree level yukawas , they result in yukawa couplings that contain one very small eigenvalue ; the corresponding eigenstates are identified with the first generation and the obtained fermion mass spectrum is fully hierarchical . quark mixing is generically also very hierarchical . mixing between the second and third as well as between the first and third generation is induced at the one loop level . smallness of mixing between the first and second generation is guaranteed by the approximate rank 1 nature of the radiatively induced yukawa matrix . the qualitative picture agrees well with the observed sm fermion spectrum and quark mixing angles . at the quantitative level we find that our minimal framework can accommodate almost all observed flavor parameters . starting with mixing angles : if the only source of flavor violation is the scalar soft masses the cabibbo angle is found to be too small however , @xmath106 can easily be brought in agreement with experiment if trilinear couplings also violate flavor , to the degree that is allowed by vacuum stability . the additional flavor breaking adds an order one contribution to the up mass , which if dressed by a gluon can lead to a chromo - edm . the edm phenomenology of this scenario would thus be interesting to explore . as for the mass spectrum , we found that the minimal framework predicts a correct electron - to - muon mass ratio , but the muon mass is too small by about an order of magnitude . we discussed various extensions of the minimal setup that allow to accommodate also the measured muon mass while keeping the electron - to - muon mass ratio fixed . one way to obtain a larger muon mass is to allow for hard susy breaking in the form of lepton - slepton - bino couplings that are larger than the hypercharge gauge coupling by a factor of @xmath58 . we also discussed an extended model with a gauge group enlarged by @xmath167 . loops with the additional @xmath155 gauginos can give sizable contributions in particular to the lepton masses and bring the lepton spectrum in agreement with observation . the @xmath155 gauge coupling can be as large as @xmath168 at the pev scale without spoiling gauge coupling unification . larger values of it can improve the agreement with the observed values , but at the cost of having a landau pole before gut scale . this suggest as interesting alternative to gauge @xmath155 to extend the mssm with a non abelian gauge group . we defer this for a future work . frameworks that implement the idea of radiative fermion masses generically lead to spectra and mixing matrices that contain a minimum amount of hierarchy , given by a one loop factor . the observed masses and mixing angles in the neutrino sector show a pattern that is much less hierarchical with respect to the quarks and charged leptons . it would be interesting to see if simple extensions of our setups can accommodate also a realistic neutrino sector . studies in this direction are left for future work . we would like to thank nima arkani - hamed , zackaria chacko , savas dimopolous , stefania gori , alex kagan , graham kribs , steve martin , mathias neubert , takemichi okui , eduardo ponton , lisa randall and jure zupan for useful discussions and comments . wa would like to thank the theoretical physics group at slac for hospitality and support . this work was supported in part by national science foundation grant no . phys-1066293 and the hospitality of the aspen center for physics . fermilab is operated by fermi research alliance , llc , under contract de - ac02 - 07ch11359 with the united states department of energy . the research of wa was supported by the john templeton foundation . research at perimeter institute is supported by the government of canada through industry canada and by the province of ontario through the ministry of economic development & innovation . here , we report the ratios of fermion masses that we compare to our model predictions throughout the paper . for definiteness we quote values for @xmath73 masses at the electro - weak scale , that we take to be @xmath170 , the @xmath171 top mass at the scale of the top mass . in the case of the quark masses , we take into account 3-loop qcd running . in the case of the lepton masses we neglect the tiny qed corrections and use directly the pole masses . using input from @xcite we find @xmath172 \label{eq : masses2 } \frac{m_s}{m_b } = ( 1.9 \pm 0.1)\cdot 10^{-2 } & , & \frac{m_d}{m_b } = ( 9.9 \pm 0.8)\cdot 10^{-4}~,~ \frac{m_d}{m_s } = ( 5.2 \pm 0.5)\cdot 10^{-2}~,\\[4pt ] \label{eq : masses3 } \frac{m_\mu}{m_\tau } = 5.95 \cdot 10^{-2 } \qquad~~ & , & \frac{m_e}{m_\tau } = 2.88 \cdot 10^{-3}~,~ \qquad~~ \frac{m_e}{m_\mu } = 4.84 \cdot 10^{-3}~.\end{aligned}\ ] ] we do not quote the extremely small uncertainties on the lepton mass ratios , as they are of no relevance for our study . the measured values of the ckm parameters at the electro - weak scale are @xcite @xmath173 in this appendix we discuss vacuum stability constraints on the sfermion trilinear couplings and the higgsino mass in the presence of generic flavor mixing in the sfermion soft masses . we focus on the case of absolute stability . the analysis of meta - stability bounds , where one allows the electro - weak vacuum to decay at time scales larger than the age of the universe , is beyond the scope of this work . we consider the case of sufficiently large @xmath68 such that we can neglect the effect of the down - type higgs @xmath174 . for simplicity we will also consider only real mass insertions . in order to obtain a bound on the stop trilinear coupling in the presence of @xmath58 squark mixing , we consider the scalar potential of the mssm in the most general d - flat direction containing the up - type higgs @xmath84 as well as all 6 up - type quarks . we find the following bound on the stop trilinear @xmath175 where the masses @xmath176 and @xmath177 are given by @xmath178 and we introduced the shorthand notation @xmath179 the inequality in ( [ eq : flavoredatbound ] ) has to hold for all choices of the angles @xmath180 . equivalently , one can impose the bound with the right - hand side minimized with respect to the angles @xmath180 . for @xmath181 one recovers the bound in ( [ eq : atbound ] ) . in general , ( [ eq : flavoredatbound ] ) can be stronger than ( [ eq : atbound ] ) . in order to obtain bounds on the higgsino mass in the presence of @xmath58 squark and slepton mixing , we have to consider the scalar potential of the mssm in directions containing the up - type higgs @xmath84 as well as all 6 down - type quarks or leptons . as there is no exact d - flat direction in this case , for any given field direction , the derived bounds are only necessary but usually not sufficient to guarantee the absence of charge and color breaking minima . analogously to section [ sec : stability ] we chose exemplarily @xmath182 and @xmath183 . in this case we find the necessary conditions @xmath184~ , \\ \label{eq : flavoredmubound2 } |\mu|^2 & \lesssim & ( \bar m_{\tilde b_l}^2 + \bar m_{\tilde b_r}^2 ) \left [ \frac{1}{2 } + \frac{9}{32 } \frac{v^2}{m_b^2 } \left|\frac{1}{t_\beta } + \epsilon_b \right|^2(x_{b1 } g_1 ^ 2 + x_{b2}g_2 ^ 2 ) \right]~.\end{aligned}\ ] ] the masses @xmath185 , @xmath186 , @xmath187 , and @xmath188 are defined analogously to the up - type masses in ( [ eq : mt ] ) and ( [ eq : mt2 ] ) . the coefficients @xmath189 are given by @xmath190 the inequalities in ( [ eq : flavoredmubound ] ) and ( [ eq : flavoredmubound2 ] ) have to hold for all choices of the angles @xmath180 . in general , ( [ eq : flavoredmubound ] ) and ( [ eq : flavoredmubound2 ] ) can be stronger than ( [ eq : mubound ] ) and ( [ eq : mubound2 ] ) . the radiative mechanism of fermion mass generation discussed in this work requires @xmath58 flavor mixing in the sfermion soft masses . such flavor mixing is subject to strong constraints from low energy flavor observables @xcite . in particular cp violation in kaon mixing , given by the observable @xmath191 , is exceedingly sensitive to the mixing between the first two generations of down - type squarks . in order not to spoil the reasonable agreement between the experimental results and the sm prediction for @xmath191 , we find that the susy scale @xmath8 has to be of the order of few pev . we expect this to be the strongest bound on the susy breaking scale in our model . in the presence of large sfermion mixing it is known that also electric dipole moments ( edms ) are generically very sensitive low energy probes of heavy supersymmetric particles . current experimental bounds on edms of the neutron and electron can probe squarks and sleptons with masses of o(100 ) tev @xcite , and prospects for further experimental improvements are excellent . however , in the framework of radiative fermion masses , there is the potential of an approximate phase alignment between the light fermion masses and the corresponding dipole moments @xcite . this is because the leading contributions to the dipole come from dressing the leading 1-loop diagram for the mass by a gluon or a photon . however , in some of our models , the mass of the first generation fermions came from more than one diagram . in particular , once flavor violating trilinears were introduced in section [ sec : aterms ] , the up mass received comparable contributions from soft masses and @xmath1-terms , as seen in equation ( [ eq : up_mass ] ) . in this case we expect the mass and the dipole to be misaligned . a full exploration of the edm phenomenology is beyond the scope of our work . the gray bands exemplarily indicate the @xmath192 and @xmath74 uncertainties coming from the top mass , strong gauge coupling , and the higgs mass.,scaledwidth=80.0% ] as squarks are very heavy in the discussed framework ( at least few pev in order to avoid constraints from kaon mixing ) , the sm - like higgs is predicted to be somewhat heavier than the observed value of @xmath193 gev . in the high scale susy scenario that we are working in , the observed higgs mass is reproduced for a susy scale of around 10 tev if @xmath194 @xcite . in figure [ fig : mhiggs ] we show in the plane of the susy scale and @xmath68 the threshold corrections @xmath195 to the higgs quartic coupling that are required at the susy scale in order to obtain a higgs mass of 125.5 gev . at the susy scale we set the higgs quartic to @xmath196 \cos^2 2\beta + \delta \lambda ~.\ ] ] we use sm 2-loop rges @xcite for the evolution of the higgs quartic between the susy scale and the electro - weak scale , taking into account contributions from all gauge couplings and the top yukawa . we incorporate 2-loop threshold corrections at the electro - weak scale from @xcite . the gray band in the plot of figure [ fig : mhiggs ] exemplarily shows the @xmath192 and @xmath74 uncertainty on the required threshold corrections at the susy scale , taking into account the uncertainty on the top quark pole mass @xmath197 gev @xcite , the strong gauge coupling @xmath198 @xcite , and the higgs mass @xmath199 gev gev @xcite and the cms result @xmath200 gev @xcite . ] . for @xmath194 , the required threshold corrections are at the level of @xmath201 for a pev scale spectrum . note that sizable stop trilinear couplings @xmath202 typically lead to positive threshold corrections of the order of @xmath203 . a very large stop trilinear @xmath204 would lead to negative corrections . however , such a large @xmath11 is in conflict with vacuum stability constraints , see ( [ eq : atbound ] ) . various viable mechanisms exist that lead to negative threshold corrections . one example is to add a singlet @xmath205 with a susy mass @xmath206 that couples to the mssm higgs bosons @xmath207 the additional scalar modifies the higgs potential and leads to a shift in the higgs quartic . we find @xmath208 where @xmath209 is the soft mass squared for the scalar component of @xmath205 and we expanded in large @xmath210 . the first term in parenthesis proportional to the soft mass squared gives a positive definite contribution to the higgs quartic , but it is completely negligible in the large @xmath68 regime . the remaining terms are proportional to the supersymmetric masses @xmath206 and @xmath81 and can reduce the higgs quartic . as long as @xmath206 is not too heavy , moderate @xmath211 can easily lead to threshold corrections of the required size . another avenue to lower the tree level higgs mass is a hard susy breaking negative quartic coupling . as discussed in section [ sec : hard ] , hard susy breaking can help in raising the muon mass in our framework . the only hard susy breaking operator that enters at order @xmath145 in the mssm is @xcite @xmath212 this operator however leads to a @xmath68 suppressed quartic . to lower the higgs mass in our framework we need to write @xmath213 the value of @xmath214 needed to increase the muon mass ( of order @xmath146 ) is sufficient to lower the higss mass to its observed value . finally , we note that another option is to make the gauginos dirac fermions @xcite . notice that a sizeable majorana mass for the gauginos is a necessary ingredient for our mechanism to work , therefore one would need same order majorana and dirac fermion masses for the gauginos . g. aad _ et al . _ [ atlas collaboration ] , phys . b * 716 * , 1 ( 2012 ) [ arxiv:1207.7214 [ hep - ex ] ] . s. chatrchyan _ et al . _ [ cms collaboration ] , phys . b * 716 * , 30 ( 2012 ) [ arxiv:1207.7235 [ hep - ex ] ] . for a recent review see n. craig , arxiv:1309.0528 [ hep - ph ] . c. d. froggatt and h. b. nielsen , nucl . b * 147 * , 277 ( 1979 ) . m. leurer , y. nir and n. seiberg , nucl . b * 398 * , 319 ( 1993 ) [ hep - ph/9212278 ] . m. leurer , y. nir and n. seiberg , nucl . b * 420 * , 468 ( 1994 ) [ hep - ph/9310320 ] . n. arkani - hamed and m. schmaltz , phys . d * 61 * , 033005 ( 2000 ) [ hep - ph/9903417 ] . d. e. kaplan and t. m. p. tait , jhep * 0111 * , 051 ( 2001 ) [ hep - ph/0110126 ] . y. grossman and m. neubert , phys . b * 474 * , 361 ( 2000 ) [ hep - ph/9912408 ] . t. gherghetta and a. pomarol , nucl . b * 586 * , 141 ( 2000 ) [ hep - ph/0003129 ] . a. e. nelson and m. j. strassler , jhep * 0009 * , 030 ( 2000 ) [ hep - ph/0006251 ] . s. weinberg , phys . * 29 * , 388 ( 1972 ) . a. b. lahanas and d. wyler , phys . b * 122 * , 258 ( 1983 ) . d. v. nanopoulos and m. srednicki , phys . b * 124 * , 37 ( 1983 ) . a. masiero , d. v. nanopoulos and k. tamvakis , phys . b * 126 * , 337 ( 1983 ) . f. del aguila , m. dugan , b. grinstein , l. j. hall , g. g. ross and p. c. west , nucl . b * 250 * , 225 ( 1985 ) . t. banks , nucl . b * 303 * , 172 ( 1988 ) . a. l. kagan , phys . d * 40 * , 173 ( 1989 ) . a. l. kagan and c. h. albright , phys . d * 38 * , 917 ( 1988 ) . n. arkani - hamed , h. c. cheng and l. j. hall , nucl . b * 472 * , 95 ( 1996 ) [ hep - ph/9512302 ] . n. arkani - hamed , h. c. cheng and l. j. hall , phys . d * 54 * , 2242 ( 1996 ) [ hep - ph/9601262 ] . c. hamzaoui and m. pospelov , eur . j. c * 8 * , 151 ( 1999 ) [ hep - ph/9803354 ] . k. s. babu , b. dutta and r. n. mohapatra , phys . d * 60 * , 095004 ( 1999 ) [ hep - ph/9812421 ] . f. borzumati , g. r. farrar , n. polonsky and s. d. thomas , nucl . b * 555 * , 53 ( 1999 ) [ hep - ph/9902443 ] . j. l. diaz - cruz , h. murayama and a. pierce , phys . d * 65 * , 075011 ( 2002 ) [ hep - ph/0012275 ] . j. ferrandis , phys . d * 70 * , 055002 ( 2004 ) [ hep - ph/0404068 ] . j. ferrandis and n. haba , phys . d * 70 * , 055003 ( 2004 ) [ hep - ph/0404077 ] . a. crivellin and u. nierste , phys . d * 79 * , 035018 ( 2009 ) [ arxiv:0810.1613 [ hep - ph ] ] . a. crivellin , j. girrbach and u. nierste , phys . d * 83 * , 055009 ( 2011 ) [ arxiv:1010.4485 [ hep - ph ] ] . a. crivellin , l. hofer , u. nierste and d. scherer , phys . d * 84 * , 035030 ( 2011 ) [ arxiv:1105.2818 [ hep - ph ] ] . n. arkani - hamed , a. gupta , d. e. kaplan , n. weiner and t. zorawski , arxiv:1212.6971 [ hep - ph ] . m. baumgart , d. stolarski and t. zorawski , arxiv:1403.6118 [ hep - ph ] . s. m. barr , phys . d * 21 * , 1424 ( 1980 ) . r. barbieri and d. v. nanopoulos , phys . b * 95 * , 43 ( 1980 ) . l. e. ibanez , nucl . b * 193 * , 317 ( 1981 ) . r. barbieri , d. v. nanopoulos and a. masiero , phys . b * 104 * , 194 ( 1981 ) . b. s. balakrishna , phys . * 60 * , 1602 ( 1988 ) . b. s. balakrishna , a. l. kagan and r. n. mohapatra , phys . b * 205 * , 345 ( 1988 ) . b. s. balakrishna and r. n. mohapatra , phys . b * 216 * , 349 ( 1989 ) . e. ma , phys . lett . * 62 * , 1228 ( 1989 ) . e. ma , phys . * 63 * , 1042 ( 1989 ) . x. -g . he , r. r. volkas and d. -d . wu , phys . d * 41 * , 1630 ( 1990 ) . z. g. berezhiani and r. rattazzi , phys . b * 279 * , 124 ( 1992 ) . b. a. dobrescu and p. j. fox , jhep * 0808 * , 100 ( 2008 ) [ arxiv:0805.0822 [ hep - ph ] ] . m. hashimoto and v. a. miransky , phys . d * 80 * , 013004 ( 2009 ) [ arxiv:0901.4354 [ hep - ph ] ] . p. w. graham and s. rajendran , phys . d * 81 * , 033002 ( 2010 ) [ arxiv:0906.4657 [ hep - ph ] ] . a. ibarra and a. solaguren - beascoa , arxiv:1403.2382 [ hep - ph ] . f. r. joaquim and j. t. penedo , arxiv:1403.4925 [ hep - ph ] . h. georgi and s. l. glashow , phys . d * 6 * , 2977 ( 1972 ) . h. georgi and s. l. glashow , phys . d * 7 * , 2457 ( 1973 ) . r. n. mohapatra , phys . d * 9 * , 3461 ( 1974 ) . s. m. barr and a. zee , phys . d * 15 * , 2652 ( 1977 ) . s. m. barr and a. zee , phys . d * 17 * , 1854 ( 1978 ) . n. arkani - hamed and s. dimopoulos , jhep * 0506 * , 073 ( 2005 ) [ hep - th/0405159 ] . g. f. giudice and a. romanino , nucl . b * 699 * , 65 ( 2004 ) [ erratum - ibid . b * 706 * , 65 ( 2005 ) ] [ hep - ph/0406088 ] . j. d. wells , phys . d * 71 * , 015013 ( 2005 ) [ hep - ph/0411041 ] . y. grossman , r. harnik , g. perez , m. d. schwartz and z. e . surujon , phys . d * 71 * , 056007 ( 2005 ) [ hep - ph/0407260 ] . b. a. dobrescu and p. j. fox , eur . j. c * 70 * , 263 ( 2010 ) [ arxiv:1001.3147 [ hep - ph ] ] . w. altmannshofer and d. m. straub , jhep * 1009 * , 078 ( 2010 ) [ arxiv:1004.1993 [ hep - ph ] ] . m. claudson , l. j. hall and i. hinchliffe , nucl . b * 228 * , 501 ( 1983 ) . j. a. casas , a. lleyda and c. munoz , nucl . b * 471 * , 3 ( 1996 ) [ hep - ph/9507294 ] . a. kusenko , p. langacker and g. segre , phys . d * 54 * , 5824 ( 1996 ) [ hep - ph/9602414 ] . n. blinov and d. e. morrissey , jhep * 1403 * , 106 ( 2014 ) [ arxiv:1310.4174 [ hep - ph ] ] . r. rattazzi and u. sarid , nucl . b * 501 * , 297 ( 1997 ) [ hep - ph/9612464 ] . j. hisano and s. sugiyama , phys . b * 696 * , 92 ( 2011 ) [ erratum - ibid . b * 719 * , 472 ( 2013 ) ] [ arxiv:1011.0260 [ hep - ph ] ] . w. altmannshofer , m. carena , n. r. shah and f. yu , jhep * 1301 * , 160 ( 2013 ) [ arxiv:1211.1976 [ hep - ph ] ] . m. carena , s. gori , i. low , n. r. shah and c. e. m. wagner , jhep * 1302 * , 114 ( 2013 ) [ arxiv:1211.6136 [ 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. e. machacek and m. t. vaughn , nucl . b * 249 * , 70 ( 1985 ) . m. x. luo and y. xiao , phys . lett . * 90 * , 011601 ( 2003 ) [ hep - ph/0207271 ] . j. a. casas and s. dimopoulos , phys . b * 387 * , 107 ( 1996 ) [ hep - ph/9606237 ] . j. h. park , phys . d * 83 * , 055015 ( 2011 ) [ arxiv:1011.4939 [ hep - ph ] ] . s. p. martin , phys . d * 61 * , 035004 ( 2000 ) [ hep - ph/9907550 ] . k. s. babu , c. f. kolda and j. march - russell , phys . d * 54 * , 4635 ( 1996 ) [ hep - ph/9603212 ] . n. arkani - hamed , a. g. cohen and h. georgi , hep - th/0108089 . j. c. pati and a. salam , phys . d * 10 * , 275 ( 1974 ) [ erratum - ibid . d * 11 * , 703 ( 1975 ) ] . s. dimopoulos and h. georgi , nucl . b * 193 * , 150 ( 1981 ) . s. dimopoulos , s. raby and f. wilczek , phys . d * 24 * , 1681 ( 1981 ) . u. amaldi , w. de boer and h. furstenau , phys . b * 260 * , 447 ( 1991 ) . p. langacker and m. -x . luo , phys . d * 44 * , 817 ( 1991 ) . s. p. martin and m. t. vaughn , phys . d * 50 * , 2282 ( 1994 ) [ erratum - ibid . d * 78 * , 039903 ( 2008 ) ] [ hep - ph/9311340 ] . e. bagnaschi , g. f. giudice , p. slavich and a. strumia , arxiv:1407.4081 [ hep - ph ] . l. j. hall , r. rattazzi and u. sarid , phys . rev . d * 50 * , 7048 ( 1994 ) [ hep - ph/9306309 ] . m. s. carena , m. olechowski , s. pokorski and c. e. m. wagner , nucl . b * 426 * , 269 ( 1994 ) [ hep - ph/9402253 ] . t. blazek , r. dermisek and s. raby , phys . d * 65 * , 115004 ( 2002 ) [ hep - ph/0201081 ] . w. altmannshofer , d. guadagnoli , s. raby and d. m. straub , phys . b * 668 * , 385 ( 2008 ) [ arxiv:0801.4363 [ hep - ph ] ] . j. beringer _ [ particle data group collaboration ] , phys . d * 86 * , 010001 ( 2012 ) and 2013 partial update for the 2014 edition . f. gabbiani , e. gabrielli , a. masiero and l. silvestrini , nucl . b * 477 * , 321 ( 1996 ) [ hep - ph/9604387 ] . w. altmannshofer , a. j. buras , s. gori , p. paradisi and d. m. straub , nucl . b * 830 * , 17 ( 2010 ) [ arxiv:0909.1333 [ hep - ph ] ] . w. altmannshofer , r. harnik and j. zupan , jhep * 1311 * , 202 ( 2013 ) [ arxiv:1308.3653 [ hep - ph ] ] . d. mckeen , m. pospelov and a. ritz , phys . d * 87 * , 113002 ( 2013 ) [ arxiv:1303.1172 [ hep - ph ] ] . g. f. giudice and a. strumia , nucl . b * 858 * , 63 ( 2012 ) [ arxiv:1108.6077 [ hep - ph ] ] . p. draper , g. lee and c. e. m. wagner , phys . d * 89 * , 055023 ( 2014 ) [ arxiv:1312.5743 [ hep - ph ] ] . d. buttazzo , g. degrassi , p. p. giardino , g. f. giudice , f. sala , a. salvio and a. strumia , jhep * 1312 * , 089 ( 2013 ) [ arxiv:1307.3536 ] . atlas and cdf and cms and d0 collaborations , arxiv:1403.4427 [ hep - ex ] . g. aad _ et al . _ [ atlas collaboration ] , phys . b * 726 * , 88 ( 2013 ) [ arxiv:1307.1427 [ hep - ex ] ] . cms collaboration , cms - pas - hig-13 - 005 .

[ sec : intro ] small volumes of simple and complex fluids that occur naturally in biological contexts or that are employed in modern technology , such as e.g. , in microfluidics , are often ( partly ) confined by a free surface that may be covered by surface active agents . these so - called surfactants may be tensids , lipids , certain nano - particles , or particular polymeric compounds . because they decrease the surface tension of the free surface , gradients in their concentration correspond to gradients in the surface tension . these gradients result in tangential forces at the free surface that drive flows in the bulk liquid . this corresponds to the so - called solutal marangoni effect , that is e.g. , responsible for the tears of wine @xcite . all surface active agents are to some extent soluble in the bulk liquid , implying that a complete dynamical model needs to describe the motion of the bulk liquid , bulk concentration of surfactant , the surface concentration of surfactant and the adsorption / desorption processes that exchange surfactant molecules between the bulk liquid and the free surface . however , for many practically important surfactants , the bulk solubility is actually very small . then one speaks of `` insoluble surfactants '' and only considers the dynamics of the surfactant that is adsorbed at the free surface . here , we restrict our attention to such insoluble surfactants at concentrations at which no micelles are formed in the bulk liquid @xcite . the governing transport equations that relate the material properties of the insoluble surfactant and the resulting hydrodynamic flow are well established for low values of the surfactant surface coverage @xmath0 @xcite . in this case , the linear equation of state @xmath1 describes how the surface tension deviates from its reference value @xmath2 , for a bare free surface . the coefficient @xmath3 is a material constant that is negative for most combinations of liquid and surfactant . the resulting tangential marangoni force at the free surface is @xmath4 where @xmath5 is the derivative along the free surface and @xmath6 is the unit normal vector . for any ( linear or nonlinear ) equation of state , the surface tension gradient @xmath7 enters the tangential stress boundary condition of the momentum transport equation . the latter is accompanied by a transport equation for @xmath0 that accounts for advective and diffusive transport of the surfactant @xcite . the resulting system of equations may be simplified in order to apply them to various physical situations such as , for example , the dynamics of surfactant - laden drops or bubbles immersed in ( another ) liquid @xcite , free - standing soap films @xcite , liquid bridges covered by a surfactant monolayer @xcite , surfactant - covered vertical falling liquid films @xcite , films on horizontal solid substrates @xcite , and drawn meniscii @xcite . in particular , the latter geometry allows for an asymptotic treatment which results in a long - wave or lubrication description of the dynamics , via two coupled evolution equations for the film height and the surfactant surface coverage @xcite . in the following , we focus on this geometry , but we should emphasise that our main arguments also apply to the general case . many works only treat the case of low surfactant surface coverages and employ the linear equation of state in eq . . the surface tension driven flow is then said to result from a linear solutal marangoni effect . however , there is a growing literature where a similar approach is used to treat the dynamics of free surfaces covered by large concentrations of insoluble surfactants . it is common practice to replace the linear equation of state by a nonlinear one and leave all other terms in the dynamical equations unchanged . we argue below that this may result in governing equations that are thermodynamically inconsistent , since one must also amend the surfactant surface diffusion term . there should also be amendments to the basic equations which describe the influence of the surfactant coverage on wettability close to three phase contact lines of very thin films , the effects of phase transitions in the surfactant layer at high concentrations , and also the influence of a nearby solid substrate on such phase transitions that may result in substrate mediated condensation ( surfactant aggregation ) . the approach that we propose in this paper allows one to deal with all these cases in a thermodynamically consistent manner . the structure of the paper is as follows : first , we review in section [ sec : filmsurf - class ] the ` classical ' thin - film hydrodynamic coupled equations of motion for a thin liquid film covered by a low concentration surfactant . then , in a preparatory step we decouple and re - formulate the two individual equations in a `` thermodynamic form '' . in particular , section [ sec : pure ] gives the gradient formulation of the evolution equation for a thin - film of pure liquid on a solid substrate , for the case where capillarity and wettability are the dominant influences , while section [ sec : diff ] briefly reviews the classical diffusion equation and places it in the thermodynamic context that we employ . in section [ sec : filmsurf - graddyn ] the full coupled system is re - formulated as a gradient dynamics based on an underlying free energy functional . this thermodynamic form is used in section [ sec : filmsurf - ext ] to extend the thin - film model to consistently account for ( i ) nonlinear equations of state , ( ii ) surfactant - dependent wettability , ( iii ) surfactant phase transitions , and ( iv ) substrate mediated condensation . we also note some differences to the models in the literature . finally , section [ sec : conc ] concludes and discusses the limitations of our approach . [ sec : filmsurf - class ] [ fig : sketch ] if a hydrodynamic system involves a free surface that is covered by an insoluble surfactant , the boundary conditions for the momentum equation have to be supplemented by an evolution equation for the surfactant concentration on the free surface that accounts for transport of the surfactant by advection and diffusion and also for shape changes of the surface that act as effective source / sink terms @xcite . this equation must be solved in conjunction with the hydrodynamic equations and boundary conditions for the liquid film . these equations can be greatly simplified for the case of a thin film of liquid on a solid substrate . if all quantities in the film vary over distances with a length scale parallel to the substrate that is large as compared to all length scales perpendicular to it , one may make a long - wave approximation @xcite to obtain coupled evolution equations for the film thickness profile @xmath8 and the surfactant surface coverage profile @xmath9 , which is a dimensionless surface packing fraction ( or concentration ) , where @xmath10 is a cartesian coordinate over the surface . the surface concentration is defined as @xmath11 , where @xmath12 is the surface number density ( number per area ) and @xmath13 is the surface area per surfactant molecule when the surfactant molecules are at maximum packing on the surface ( i.e. @xmath14 is a molecular length scale ) , so that close packing corresponds to @xmath15 . for the three - dimensional physical situation illustrated in fig . [ fig : sketch ] , the equation for the film height is @xmath16 \,-\ , \nabla\cdot \left(\frac{\gamma_\gamma h^2}{2\eta}\nabla\gamma\right ) . { \label{eq : govh}}\ ] ] note that this equation is equally valid if one uses @xmath17 instead of the dimensionless @xmath0 or indeed any other measure of the surfactant surface density . only the quantity @xmath3 needs to be redefined so that the product of @xmath3 with the surface density yields a quantity with the dimensions of an energy per area [ c.f . eq . ] . as a result , many papers in the literature do not mention what units they choose for @xmath0 . the time evolution equation for @xmath0 is @xmath18 \,-\ , \nabla\cdot \left(\frac{\gamma_\gamma h\gamma}{\eta}\nabla\gamma\right ) + \nabla\cdot ( d \nabla \gamma ) , { \label{eq : govgam}}\ ] ] where @xmath19 is the liquid - gas surface tension and @xmath20 is the dynamic viscosity of the pure liquid . partial derivatives with respect to time and space are denoted @xmath21 and @xmath22 , respectively , @xmath23 is the planar gradient operator and @xmath24 is the laplace operator . the mobility @xmath25 results from poiseuille flow in the film without slip at the substrate . the pressure @xmath26 contains the laplace surface curvature contribution to the pressure and additional contributions such as a hydrostatic or a disjoining pressure @xcite . note that the latter is normally assumed to be independent of @xmath0 . exceptions are discussed below . the diffusive transport of the surfactant in eq . ( [ eq : govgam ] ) follows from fick s law for the flux @xmath27 . in most papers , it is assumed that the diffusion constant @xmath28 does not depend on the surfactant concentration , i.e. , the term @xmath29 in eq . becomes @xmath30 . note that eq . ( [ eq : govgam ] ) is an equation obtained in the long - wave approximation and therefore does not include the source - like surface dilatation term . different forms for such a term are discussed in refs . @xcite . for the same reason , the @xmath31 operator in the diffusion term is the planar operator and not the operator @xmath32 that acts tangentially to the free surface . to extend the present ideas to more general geometries , these contributions must be taken into account . to obtain eqs . ( [ eq : govh ] ) and ( [ eq : govgam ] ) , we have related the surfactant surface coverage @xmath0 to the surface tension @xmath33 by the linear equation of state in eq . ( [ eq : state - lin ] ) , i.e. , a linear solutal marangoni effect is assumed . in deriving eq . ( [ eq : govh ] ) , one also assumes that @xmath34 and that therefore the laplace pressure term ( @xmath35 ) only depends on @xmath19 . to incorporate effects of high surfactant concentration , the equations are often extended by translating @xmath36 back into @xmath37 , and then replacing the linear equation of state in eq . ( [ eq : state - lin ] ) by some non - linear equation of state . there are problems with doing this , as we show below . another extension is to incorporate a surfactant - dependent wettability into the evolution equations ( [ eq : govh ] ) and ( [ eq : govgam ] ) . this is sometimes done in an ad - hoc manner by simply replacing @xmath38 by some @xmath39 . however , it turns out that this leaves the equation incomplete and may even result in qualitatively incorrect predictions . after re - formulating the evolution equations as a gradient dynamics in section [ sec : filmsurf - graddyn ] , we discuss such extensions in section [ sec : filmsurf - ext ] . first , however , we introduce the gradient formulation for the decoupled thin - film equation ( section [ sec : pure ] ) and surfactant surface diffusion equation ( section [ sec : diff ] ) . [ sec : pure ] it was noted some time ago that the time evolution equation for the film thickness in the case without surfactant [ eq . ( [ eq : govh ] ) with @xmath40 can be written in a variational form @xcite . this allows one to appreciate that eq . ( [ eq : govh ] ) corresponds to a time evolution equation for a conserved order parameter field @xmath8 ( cf . @xcite ) that follows a dissipative gradient dynamics governed by the following equation @xmath41 . { \label{eq : varh}}\ ] ] this equation describes how the field @xmath42 evolves towards a minimum of the free energy functional @xmath43\,=\ , \int\left[\gamma_0 + \frac{\gamma_0}{2}(\nabla h)^2 + f(h ) \right]\,da { \label{eq : en1}}\ ] ] where @xmath44 , the mobility function @xmath45 ( c.f . also ref . @xcite ) and @xmath46 is a cartesian area element along the substrate . note that the free energy in eq . ( [ eq : en1 ] ) corresponds to the one that is obtained making a small slope approximation in the free energy @xmath47 , where the surface element @xmath48 is approximated using @xmath49 . the constant part @xmath50 of the free energy in eq . ( [ eq : en1 ] ) is normally omitted since it does not contribute to the dynamics , as one can see from eq . ( [ eq : varh ] ) . a similar formulation is given in the following section for the surfactant surface diffusion equation . [ sec : diff ] the diffusive transport of a species with small surface coverage @xmath0 is described by the diffusion equation @xmath51 where the diffusive flux is given by fick s law @xmath52 . the time evolution equation for the surfactant density in eq . , reduces to the diffusion equation in eq . in the limit when the liquid film thickness @xmath8 is a constant and the coefficient @xmath53 ; i.e. when there is no marangoni effect . the form in eq ( [ eq : diff ] ) can easily obscure the underlying thermodynamics , which can be seen when this equation is formulated as a gradient dynamics based on the helmholtz free energy for an ideal gas ( i.e. a system of non - interacting particles ) : @xmath54\,=\,\frac{kt}{l^2}\int \gamma [ \log(\gamma ) - 1 ] \,da , { \label{eq : en - diff}}\ ] ] where @xmath55 is boltzmann s constant and @xmath56 is the temperature . the transport equation for @xmath0 is of the same form as eq . ( [ eq : varh ] ) and reads @xcite : @xmath57 , { \label{eq : trans - diff}}\ ] ] where the mobility @xmath58 . here , @xmath59 is the molecular mobility related to the diffusion process and may in principle depend on all the independent variables , although in the following we will assume that @xmath59 is constant . the equivalence of the formulation in eq . ( [ eq : diff ] ) and in eqs . and ( [ eq : trans - diff ] ) is easily established . the advantage of the gradient dynamics form in eq . ( [ eq : trans - diff ] ) is the `` built - in '' straightforward way to extend the description , e.g. , to incorporate attractive forces between the diffusing molecules and the effects of higher concentrations . for instance , replacing the functional in eq . ( [ eq : en - diff ] ) by the one discussed by cahn and hilliard @xcite ( their equation ( 2.4 ) with ( 3.1 ) ) results in the non - linear diffusion equation ( or cahn - hilliard equation ) , eq . ( 9 ) of ref.@xcite ( when @xmath60 is expanded about @xmath61 and only the lowest order term is kept ) . more recently , marconi and tarazona @xcite showed that one can derive eq . , starting from over - damped stochastic equations of motion for the ( surfactant ) particles . they showed that the diffusive fluid dynamics is described by eq . , taken together with a suitable approximation for the helmholtz free energy functional taken from equilibrium density functional theory @xcite . this so called dynamical density functional theory @xcite is now a growing body of work , allowing one to go beyond cahn - hilliard theory , and to develop a theory which includes a microscopic ( on the scale of the particles ) description of the dynamics of particles suspended in a fluid medium , or in the present case of surfactant particles on the surface of the liquid . to our knowledge , no gradient dynamics formulation has yet been given for the evolution of a thin - film covered by insoluble surfactant as described by eqs . ( [ eq : govh ] ) and ( [ eq : govgam ] ) . since the system is relaxational , i.e. , there is no energy influx , a variational formulation in terms of a pair of coupled evolution equations for two conserved order parameter fields must exist and in fact is presented in the following section . [ sec : filmsurf - graddyn ] to construct a gradient dynamics description for the full coupled system , eqs . ( [ eq : govh ] ) and ( [ eq : govgam ] ) , we start by considering the helmholtz free energy for the system @xmath62 $ ] , that also turns out to be the lyapunov functional for a surfactant covered thin liquid film . it contains contributions that result from wettability ( adhesion ) , expressed in terms of the height profile of the film , @xmath63 , _ and _ contributions from the surface @xmath64 , where @xmath65 is the helmholtz surface free energy density ( i.e. an energy per area ) , and @xmath66da$ ] is a surface element . the contribution of the height profile are similar to those in eq . ( [ eq : en1 ] ) . in the parameter regime where the linear equation of state ( [ eq : state - lin ] ) for the surfactant is valid , i.e. when the surfactant density is low , then the surfactant layer corresponds to a two - dimensional gas of surfactant molecules on the film surface . the corresponding contribution to the free energy density is the entropic ( ideal - gas ) term @xmath67 that on its own leads to a diffusion equation for @xmath0 , as discussed in section [ sec : diff ] . however , one may add other contributions that are relevant at higher concentrations , resulting from the interactions between surfactant molecules . such contributions are discussed below in section [ sec : filmsurf - ext ] . for the non - interacting case we obtain @xmath68\ , = \,\int\left\{f(h)+g(\gamma)\,\xi\right\}\,da { \label{eq : en2}}\end{aligned}\ ] ] where @xmath69 . { \label{eq : gg}}\end{aligned}\ ] ] it should be noted that variations in @xmath42 and in @xmath0 are not independent : if locally the slope of @xmath42 changes , the area of the liquid surface changes and thus the surface coverage @xmath0 may change without any surfactant transport . to derive evolution equations that are of a gradient dynamics form , one needs a concentration variable that is independent of the film height profile @xmath42 . we introduce a surface coverage @xmath70 , as sketched in fig . [ fig : sketch - proj ] , that corresponds to the coverage @xmath0 ` projected ' onto the flat substrate ( thus , in principle , @xmath70 can become larger than one ) . it is given by @xmath71 and the coverage @xmath0 on the modulated free surface , defined in eq . . ] [ fig : sketch - proj ] using @xmath72 $ ] from eq . ( [ eq : en2 ] ) , the long - wave hydrodynamic equations ( [ eq : govh ] ) and ( [ eq : govgam ] ) are equivalent to the following general form for the time evolution equations @xmath73 \nonumber\\ \partial_t \widetilde\gamma \,&=&\ , \nabla\cdot\left[q_{h \gamma}\nabla\frac{\delta f}{\delta h}\,+\,q_{\gamma \gamma}\nabla\frac{\delta f}{\delta \widetilde\gamma}\right ] { \label{eq : coup}}\end{aligned}\ ] ] where the symmetric positive definite mobility matrix is @xmath74 q_{h \gamma } & q_{\gamma \gamma } \end{array } \right ) \,=\,\left ( \begin{array}{cc } \frac{h^3}{3\eta } & \frac{h^2\gamma}{2\eta } \\[.3ex ] \frac{h^2\gamma}{2\eta } & \frac{h\gamma^2 } { \eta}+ \widetilde{d}\gamma \end{array } \right ) . { \label{eq : mob}}\ ] ] note that we have written @xmath75 in terms of @xmath0 and @xmath42 . the justification for using @xmath0 and not @xmath70 in the long - wave approximation mobilities will become clear below . to fully appreciate the equivalence of eq . [ with eqs . , and ] to eqs . and , we calculate the variations of the free energy in eq . : @xmath76 , \nonumber\\ \frac{\delta f}{\delta \widetilde{\gamma } } \,&=&\,g ' , { \label{eq : insolsurf - variations}}\end{aligned}\ ] ] where we have introduced the local surface grand potential density @xmath77 . inserting these results into eq . , we obtain the following time - evolution equations : @xmath78\ , + \,\frac{h^2\gamma}{2\eta}\nabla g'\right ] , \nonumber\\ \partial_t \widetilde\gamma \,\approx\,\partial_t \gamma&=&\ , \nabla\cdot\left[\frac{h^2\gamma}{2\eta}\nabla\left[\partial_h f(h ) - \nabla\cdot ( \omega \nabla h)\right]\,+\,\left(\frac{h\gamma^2 } { \eta}+ \widetilde{d}\gamma\right ) \nabla g ' \right ] , { \label{eq : insolsurf - evolequ2}}\end{aligned}\ ] ] where we have used the fact that @xmath79 to approximate @xmath80 in the left - hand side of the second equation , as is appropriate in the long - wave limit . used at this stage , this approximation leads to @xmath81 . note , however , that this approximation should be applied with caution . should only be employed in the final time - evolution equation . if one were to make this approximation in the free energy equation , before taking the functional derivative , several physically important terms end up being omitted from the final time - evolution equation . ] the equations for @xmath82 and @xmath83 exactly correspond to the hydrodynamic model [ eqs . ( [ eq : govh ] ) and ( [ eq : govgam ] ) ] if ( i ) one identifies the local surface grand potential density @xmath84 with the surface tension @xmath33 ( see further discussion in section [ sec : eqstate ] below ) and ( ii ) employs eq . ( [ eq : gg ] ) for @xmath65 , i.e. , one only includes the ideal - gas contribution to the free energy , which is a reasonable approximation to make for low concentrations of surfactant . as a result @xmath85 and so one finds that the diffusion coefficient @xmath28 and the solutal marangoni coefficient @xmath3 in eqs . ( [ eq : govh ] ) and ( [ eq : govgam ] ) are given by @xmath86 and @xmath87 , respectively . note that the hydrodynamic community often assumes that the change of the surface tension with concentration is small as compared to the reference surface tension @xmath19 and therefore only uses @xmath19 in the laplace pressure term , i.e. , @xmath88 . [ sec : eqstate ] before presenting several extensions to the hydrodynamic equations ( [ eq : govh ] ) and ( [ eq : govgam ] ) , based on our reformulation in eq.([eq : insolsurf - evolequ2 ] ) , we put our gradient dynamics formulation in its proper thermodynamic context . first , we discuss the equation of state that relates surface tension @xmath33 to surfactant concentration @xmath0 , and show that the thermodynamic and hydrodynamic approaches are fully consistent for any dependence @xmath65 . to this end , we review first some elementary considerations concerning the relation between the surface tension and the surface free energy density . the latter is defined as the excess free energy per unit area that is due to the presence of a surface . in analogy to bulk thermodynamics , this excess may be defined for different thermodynamic ensembles and may therefore depend on different surface thermodynamic variables . moreover , this surface excess acts as the thermodynamic potential for surface variables ( for an enlightening general discussion of surface excesses , see ref . the surface tension is the derivative of this surface thermodynamic potential with respect to the area it is the equivalent of the pressure in bulk thermodynamics ( up to a sign convention : a positive pressure generates an outward force on the container walls , whereas a positive surface tension creates an inward force on the lines bordering a surface element ) . it is easy to show that the surface tension is always given by the surface excess grand potential density , regardless of whether the surface free energy is defined in the canonical or grand - canonical ensemble . in the canonical case , the surface free energy is the surface ( excess ) contribution to the helmholtz free energy , @xmath89 , for a surface element of area @xmath90 . the variation of this free energy is @xmath91 . the second term arises from the fact that the variation in the canonical ensemble has to be taken for a fixed number of surfactant molecules , @xmath92 ( recall that @xmath93 ) , and thus the variation of the surface area creates a variation in the local concentration equal to @xmath94 , and the surface tension becomes @xmath95 , where we have introduced the surface grand potential density @xmath96 , with @xmath97 the chemical potential . , although strictly it is defined as the derivative @xmath98 . this definition , which differs from the usual one by a factor of @xmath13 , leads to a simplification of some of our equations . ] alternatively , if the surface excess is defined in the grand - canonical ensemble , the surface free energy is directly given by @xmath99 . now , the variation has to be taken at constant chemical potential ( the surface element is connected to a reservoir of surfactant molecules ) . since @xmath0 is a function of the chemical potential in the grand - canonical ensemble , it remains fixed , and therefore the surface tension is directly given by @xmath100 . is not a function of @xmath0 , but of @xmath101 , so when we write @xmath102 , this should be read as @xmath103 . ] for flat interfaces with a small surfactant concentration , the surface - related part of the local helmholtz free energy @xmath104 is given by eq . ( [ eq : gg ] ) , and so the chemical potential @xmath105 . with this , the surface tension becomes @xmath106 i.e. , one recovers the linear dependence , eq . ( [ eq : state - lin ] ) , used in hydrodynamics with @xmath87 and @xmath107 . note that if the surface tension is defined ( incorrectly ) as the local helmholtz free energy @xmath65 , the logarithmic terms entail that one does not recover the linear dependence in eq . ( [ eq : state - lin ] ) . by identifying the surface tension with the local grand potential , the thermodynamic and hydrodynamic formulations are fully consistent for any convex local @xmath65 . in the hydrodynamic formulation ( [ eq : govh ] ) and ( [ eq : govgam ] ) ] the marangoni force contributes to the advective flux as @xmath108 , and normally the equation of state @xmath109 is given directly . with @xmath110 , the marangoni term becomes @xmath111 . this expression is identical to the term one obtains in the variational formulation ( [ eq : coup ] ) , i.e. , @xmath112 . for any @xmath113 of the form in eq . ( [ eq : en2 ] ) , this equals @xmath111 . the equivalence of the marangoni term in the hydrodynamic and thermodynamic formulation is valid for any convex local @xmath65 . in other words , the surface tension gradient @xmath114 may be expressed either as @xmath115 or as @xmath116 . this implies that eq . ( [ eq : insolsurf - evolequ2 ] ) may be written as @xmath117\ , -\,\frac{h^2}{2\eta}\nabla\gamma\right ] \nonumber\\ \partial_t \gamma&=&\ , \nabla\cdot\left[\frac{h^2\gamma}{2\eta}\nabla\left[\partial_h f(h ) - \nabla\cdot ( \gamma \nabla h)\right ] \,-\,\left(\frac{h\gamma } { \eta}+ \widetilde{d}\right ) \nabla \gamma \right ] . { \label{eq : insolsurf - evolequ3}}\end{aligned}\ ] ] thus , the diffusion term is expressed in terms of @xmath114 and the molecular mobility @xmath59 . note , however , that this argument no longer holds if the free energy functional contains non - local terms in @xmath0 , such as , e.g. , @xmath118 . then the formulation in eq . ( [ eq : insolsurf - evolequ3 ] ) can not be used and one must start directly with eqs . ( [ eq : coup ] ) . [ sec : filmsurf - ext ] up to this point , we have presented a gradient dynamics re - formulation of the hydrodynamic long - wave model for the evolution of a thin - film that is covered by a low concentration of insoluble surfactants , that has several connections to the approach taken in dynamical density functional theory , with a local approximation for the free energy . the present re - formulation really demonstrates its advantages when seeking to make ( common ) extensions of the hydrodynamic model , such as to incorporate nonlinear equations of state , surfactant - dependent wettability , phase transitions at high surfactant concentrations , or substrate mediated surfactant condensation . such effects are often included into the hydrodynamic formulation ( [ eq : govh ] ) and ( [ eq : govgam ] ) in an _ ad hoc _ manner that may result in the omission of important terms and sometimes lead to qualitatively incorrect behaviour . the thermodynamic variational framework presented here , i.e. , eqs . ( [ eq : insolsurf - evolequ2 ] ) , allows us to make extensions stemming from changes ( extra terms ) in the free energy functional ( [ eq : en2 ] ) in a systematic and thermodynamically consistent manner . in the following , we discuss several examples . the most common extension is to replace the linear equation of state in eq . ( [ eq : state - lin ] ) by various nonlinear expressions . examples include the exponential ( gaussian ) dependence @xmath119 @xcite ; a smoothed step - wise change @xmath120 $ ] @xcite ; langmuir - szyszkowski ( @xmath121 ) @xcite or frumkin ( @xmath122 ) @xcite equations applied to insoluble surfactants ; expressions related to power laws , such as , e.g. , @xmath123 @xcite ) ; and fits to experimentally obtained isotherms @xcite . in all cases the @xmath124 s represent various constants . in most works , the extension is done by solely replacing the parameter @xmath3 in eqs . ( [ eq : govh ] ) and ( [ eq : govgam ] ) by the function @xmath125 . this , however , does not take into account that in addition to the convective transport due to the marangoni force , the diffusive transport of the surfactant is also affected when the underlying free energy functional changes . thus , most works assume that the diffusion constant @xmath28 remains independent of the surfactant concentration , even when working with highly nonlinear equations of state . an exception is ref . @xcite that uses a @xmath126 . based on our thermodynamic reformulation , the proper relation between @xmath33 and @xmath28 ( discussed at the end of section [ sec : eqstate ] ) results in the amended hydrodynamic equations ( [ eq : insolsurf - evolequ3 ] ) that are expressed in terms of @xmath109 . one may define a non - constant @xmath126 by enforcing ficks s law @xmath127 to hold . from a comparison of eqs . ( [ eq : govgam ] ) and ( [ eq : insolsurf - evolequ2 ] ) one obtains @xmath128 , i.e. , @xmath129 . with this @xmath28 the hydrodynamic formulation ( [ eq : govh ] ) and ( [ eq : govgam ] ) is consistent with the gradient dynamics form ( [ eq : insolsurf - evolequ2 ] ) . to obtain @xmath28 in terms of the equation of state , one differentiates the relation @xmath130 with respect to @xmath0 . the result @xmath131 implies @xmath132 note , however , that in principle @xmath59 itself might also be a function of @xmath0 ( and other state variables ) . the relations @xmath109 and @xmath126 employed in ref . @xcite are only consistent with the general thermodynamic framework given here , if particular dependencies of @xmath59 on @xmath0 are assumed . note that fick s law in eq . is only true in the low density @xmath133 limit . more generally one should have @xmath134 where @xmath135 is a mobility coefficient @xcite . this form is universally valid , whereas eq . does not always hold ( e.g. in the ` uphill diffusion ' observed in spinodal decomposition ) . note also that many authors correctly employ the hydrodynamic form in eqs . ( [ eq : govgam ] ) and ( [ eq : insolsurf - evolequ2 ] ) with a linear equation of state , but give the marangoni flux in its general form as @xmath136 in combination with a surfactant - independent diffusion constant @xcite . whilst this approach is indeed correct for a linear equation of state , it should be stressed that it is not valid for arbitrary ( nonlinear ) equations of state . it is widely accepted that wettability depends on the surface density of the surfactants @xcite . however , the literature is less clear on how such effects may be incorporated in a hydrodynamic thin - film description by extending the model given in eqs . ( [ eq : govgam ] ) and ( [ eq : insolsurf - evolequ2 ] ) . in all the contributions we are aware of , this is done by replacing the film - thickness dependent derjaguin ( or disjoining ) pressure @xmath137 , that is contained in the pressure @xmath38 in eqs . ( [ eq : govh ] ) and ( [ eq : govgam ] ) , by a disjoining pressure @xmath138 that depends on film thickness and surfactant concentration @xcite . the influence of surfactants on the various components of derjaguin s pressure for thin - films are discussed in detail in refs . @xcite in the context of free standing ( soap ) films . for a simple model for forces between surfaces with adsorbed layers see ref . @xcite . based on our thermodynamic re - formulation , one can now see how the hydrodynamic equations ( [ eq : govh ] ) and ( [ eq : govgam ] ) must be amended to account for any dependency of the adhesion energy on film thickness and surfactant concentration . replacing @xmath139 in the free energy functional ( [ eq : en2 ] ) by @xmath140 results in the additional contributions @xmath141 to @xmath142 that affect both evolution equations . the resulting hydrodynamic form is @xmath143\ , -\,\frac{h^2}{2\eta}[\nabla \gamma -\,\gamma\nabla \partial_\gamma f(h,\gamma ) ] \right ] \nonumber\\ \partial_t \gamma&=&\ , \nabla\cdot\left[\frac{h^2\gamma}{2\eta}\nabla\left[\partial_h f(h,\gamma ) - \nabla\cdot ( \gamma \nabla h)\right ] \,-\,\left(\frac{h\gamma } { \eta}+ \widetilde{d}\right ) [ \nabla \gamma -\,\gamma\nabla \partial_\gamma f(h,\gamma ) ] \right ] . { \label{eq : insolsurf - evolequ4}}\end{aligned}\ ] ] the extra terms should be interpreted as an additional contribution to the marangoni force that must be taken into account for small film thicknesses . the effective marangoni force is @xmath144 . it also becomes the effective driving force for diffusion of the surfactant . to our knowledge these terms have not been included in any of the thin - film evolution equations that model surfactant - covered ultrathin films . however , they are necessary in any model that involves a surfactant - dependent derjaguin pressure . without them , the model may exhibit qualitatively incorrect behaviour , such as oscillatory instability modes @xcite , as discussed in more detail below . furthermore , as the system evolves in time , it does not tend to the correct equilibrium state , particularly in the contact line region . in short , to treat such effects correctly one must determine a suitable form for @xmath140 , obtained from the surfactant - dependent derjaguin pressures discussed in the literature @xcite . a third example is the description of phase separating surfactant mixtures or phase transitions at high surfactant concentrations . the simplest case of surfactant molecules that slightly attract each other is already addressed by the discussion above , as it only results in a nonlinear equation of state . for instance , for weakly attracting surfactant molecules , one replaces the purely entropic form in eq . ( [ eq : gg ] ) by @xmath145 -(a/2)\gamma^2 $ ] , where the attraction strength parameter @xmath146 . this results in @xmath147 and so the effective diffusion constant depends linearly on @xmath0 . the situation becomes more involved for surfactant layers that can undergo a phase transition when the concentration changes , e.g. , between the gaseous and the liquid - expanded or between the liquid - expanded and the liquid - condensed phases @xcite . beside a function @xmath65 that accounts for the particular surfactant isotherm , one also needs to incorporate a surface gradient term @xmath148 in the free energy functional ( [ eq : en2 ] ) to account for the finite width and line tension of the interface between the various surfactant phases . if a double - well potential is used for @xmath104 , this amounts to a description of the surfactant layer using a convective cahn - hilliard - type equation . a similar approach is employed in ref . @xcite to describe a thin liquid film covered with an insoluble surfactant in the vicinity of a first - order phase transition . however , as explained below , our formulation differs on a number of important points . as the algebra is involved , we illustrate this for the one - dimensional case , where @xmath149 is the only spatial coordinate . the free energy functional is @xmath150\ , = \,\int\left\{f(h)+g\left(\frac{\widetilde\gamma}{\xi}\right)\,\xi + \frac{\kappa}{2 } \left(\partial_x\frac{\widetilde\gamma}{\xi}\right)^2 \frac{1}{\xi } \right\}\,dx { \label{eq : en - nonlocal}}\ ] ] where we use @xmath151 , @xmath152 , @xmath153 and @xmath154 is the arc - length coordinate along the free surface . note that the final contribution to the integral in eq . is simply the term @xmath155 . we keep @xmath156 exact throughout the derivation and only use @xmath157 at the end . employing the approximation too early can lead to neglecting physically essential terms , such as the laplace pressure . details of the calculation of the functional derivatives are given in appendix [ sec : app - vari ] . the resulting expressions are @xmath158 \nonumber\\ \frac{\delta f}{\delta \widetilde{\gamma } } \,&=&\ , g ' -\kappa\partial_{xx}\gamma { \label{eq : nonlocal - variations}}\end{aligned}\ ] ] where we have used @xmath157 [ see appendix [ sec : app - vari ] in particular eqs . ( [ eq : app - vari2 ] ) and ( [ eq : app - vari3 ] ) ] and @xmath77 . the time evolution equations for @xmath42 and @xmath0 are obtained by substituting eqs . ( [ eq : nonlocal - variations ] ) into eqs . ( [ eq : coup ] ) . on inspecting the resulting equations , one notices that we have again obtained the form in eq . ( [ eq : insolsurf - evolequ3 ] ) , but now the surface tension is @xmath159 recall that above , for the case without the gradient terms in @xmath0 , we had that @xmath160 [ eq . ( [ eq : insolsurf - gamrel ] ) ] . it turns out that in the present case this result still holds . the surface grand potential density for the nonlocal case is @xmath161 . this observation implies that with the proper definition of surface tension , the evolution equations in eq . ( [ eq : insolsurf - evolequ3 ] ) are valid for both the extension to include nonlinear equations of state and the present extension that incorporates gradient terms in @xmath0 in the free energy . this issue explains the differences between our formulation and that in ref . @xcite , that starts from a hydrodynamic formulation somewhat similar to that in eq.([eq : insolsurf - evolequ3 ] ) . instead of @xmath155 for the gradient term . this leads to the factor @xmath162 in the last term of eq . ( [ eq : en - nonlocal ] ) becoming @xmath163 , resulting in the opposite sign of the @xmath118 term in the prefactor of @xmath164 in eq . ( [ eq : nonlocal - variations ] ) , when the @xmath165 approximation is used . furthermore , they use @xmath166 instead of @xmath167 . their formulation corresponds to the use of different approximations for @xmath0 in the various instances it appears in the mobility matrix in eq . ( [ eq : mob ] ) . ] we expect the formulation presented here to be useful for studying the dynamics of surfactant phase transitions on thin films , for the case of insoluble surfactants . for instance , incorporating gradient terms may enable one to explain the spatially non - monotonic distribution of a spreading surfactant drop that has been observed in recent experiments @xcite . as final example , we mention the so - called substrate - mediated phase transitions of surfactant layers that may occur when surfactant monolayers are transferred from a deep trough onto a solid substrate , i.e. , during a langmuir blodgett transfer @xcite . often , the substrate triggers a phase transition from the liquid - expanded ( le ) phase to the liquid - condensed ( lc ) phase of the surfactant monolayer . within the framework presented here , this transition may be described by replacing the surfactant contribution to the free energy ( [ eq : en2 ] ) , @xmath65 , by a term that depends on both the surfactant concentration and film height , @xmath168 . doing this , one obtains additional contributions to the free energy variations : the @xmath169 has to be replaced by @xmath170 and a term @xmath171 is added to @xmath172 in eq . ( [ eq : insolsurf - variations ] ) . for the full expressions , see appendix [ sec : app - vari ] . such a surfactant concentration and film - height dependent contribution to the free energy is employed in refs . @xcite to describe substrate - mediated condensation , but without incorporating the additional @xmath171 term . for their choice of @xmath168 , the omission is of no major consequence ; it only amounts to a redefinition of the parameters in the disjoining pressure . [ sec : filmsurf - consequ ] the advantage of the gradient dynamics formulation , besides its thermodynamic consistency , is that one may readily use general results obtained for other systems having governing equations of the form of eq . ( [ eq : coup ] ) . similar formulations exists , for instance , for two - layer thin - film systems ( where the two conserved fields are the two film thicknesses ) @xcite and for thin - films of solutions or suspensions ( where the two conserved fields are the film thickness @xmath42 and the effective solute layer thickness @xmath173 , where @xmath174 is the vertically averaged concentration @xcite ) . just as in the above - mentioned cases , one can show that the free energy functional @xmath62 $ ] in eq . is a lyapunov functional : the total time derivative of @xmath62 $ ] is @xmath175 . expressing the partial derivatives @xmath82 and @xmath83 by the expressions in eq . ( [ eq : coup ] ) and after integration by parts and assuming periodic or no - flux boundary conditions , one obtains @xmath176\,dx . { \label{lyap_dt}}\ ] ] because [ cf . @xmath177 and @xmath178 and @xmath179 , the quadratic form in eq . ( [ lyap_dt ] ) is positive definite and therefore @xmath180 , and @xmath113 is a proper lyapunov functional . furthermore , one may identify the stationary solutions of eqs.([eq : coup ] ) with the extrema of @xmath113 . next , we briefly discuss a general result for the linear stability of flat films @xmath181 , that are covered by a homogeneous layer of surfactant @xmath182 . on a system of infinite size , one may decompose any fluctuation disturbances of the film height and surfactant concentration in this homogeneous state into fourier modes and consider their time evolution . we employ the ansatz @xmath183 $ ] and @xmath184 $ ] where @xmath55 and @xmath185 are the wave number and growth rate of the harmonic mode , respectively . the overall amplitude of the disturbance is @xmath186 , while @xmath187 is the amplitude ratio of the disturbances in the surfactant concentration and film thickness profiles . in short , the amplitudes may be written in vector notation as @xmath188 . employing these ansatzes for @xmath189 and @xmath190 in eqs.([eq : coup ] ) , and then linearising in @xmath191 , as is appropriate for small amplitude disturbances , leads to the following eigenvalue problem @xmath192 where @xmath193 is the @xmath194-symmetric jacobian given by @xmath195 and where @xmath196 and @xmath197 are the matrix of the second variations of @xmath113 in fourier space and the mobility matrix , respectively , both evaluated at @xmath198 and @xmath199 . since @xmath200 for @xmath201 , eq . can be written as the generalised eigenvalue problem @xmath202 because @xmath196 and @xmath203 are both symmetric and @xmath203 is positive definite , one can deduce that all eigenvalues @xmath204 are real @xcite , as one should expect for a variational problem . inspecting eq . further indicates that the stability of the system is completely determined by the eigenvalues of @xmath196 , i.e. , by the second variations of the free energy functional . the stability threshold is given by @xmath205 . however , having @xmath196 is not sufficient to obtain the actual growth rate @xmath185 of the unstable modes ( i.e. the dispersion relation ) and the amplitude ratio @xmath187 . these are obtained by solving eq . . a remarkable effect that arises from the coupling of the two fields , i.e. , when @xmath206 , is that the system becomes unstable for a larger range of parameter values than the individual ( decoupled ) systems are on their own . this effect is discussed in many other contexts see e.g. @xcite . the fact that the present system has a non - diagonal mobility matrix @xmath207 in eq . ( as do the models in refs . @xcite ) distinguishes it from many other systems with evolution equations for two coupled order parameter fields having a gradient dynamics . this means that the equations for both fields depend on variations of the free energy with respect to each of the fields @xmath42 and @xmath0 . thus , for a non - diagonal @xmath207 and when @xmath208 , the evolution of the two fields are coupled both through the free energy functional and through the dynamical mobility coefficients in @xmath207 . in contrast , many such systems have a diagonal mobility matrix , and then the evolution of the two fields is solely coupled by the off - diagonal term in the matrix of second derivatives , @xmath209 . examples are the equations in ref . @xcite , that describe the spinodal decomposition in ternary systems , the dewetting of nanoparticle suspensions in ref . @xcite , the coupled demixing and dewetting of a binary mixture discussed in ref . @xcite , the electric field driven surface instability of two air - gap separated polymer layers in a capacitor @xcite , and the model equations employed in ref . @xcite to describe the interplay between ordering and spinodal decomposition in binary systems . [ sec : conc ] we have proposed several amendments and extensions for models describing the dynamics of liquid films that are covered by high concentrations of insoluble surfactant . after briefly reviewing the ` classical ' hydrodynamic form of the coupled evolution equations for the film height and the surfactant concentration profile , that are well established for small concentrations , we have re - formulated the model in three stages as a gradient dynamics . we refer to this as the `` thermodynamic form '' of the evolution equations . in the first stage , we have given the gradient dynamics form of the evolution equation for a thin film of a pure liquid on a flat substrate without surfactant , in the case where capillarity and wettability are the dominant influences . this formulation was discussed before , e.g. in refs . @xcite , and is of the standard form suitable for conserved dynamics , of which a classic example is the cahn - hilliard equation @xcite for the demixing dynamics of a binary mixture . in the second stage , we have briefly reviewed the classical diffusion equation and have noted the existence of a gradient dynamics formulation that puts it in the context of nonequilibrium thermodynamics and dynamical density functional theory . finally , in the third stage we have re - formulated the full coupled system of equations for the liquid film height and surfactant concentration profile in a gradient dynamics form , based on an underlying free energy functional that accounts for wettability , capillarity and entropic contributions for the surfactant . the resulting equations are equivalent to the hydrodynamic form for the case of a linear equation of state for the surfactant . based on this thermodynamic re - formulation , we have proposed amendments to the basic hydrodynamic model that account for four different physical effects that all may be included through changes to the free energy functional . in particular , we have extended the thin - film model to consistently account for ( i ) nonlinear equations of state , ( ii ) surfactant - dependent wettability , ( iii ) surfactant phase transitions , and ( iv ) substrate - mediated condensation . the ideas that we have presented can also be directly applied to films covered by monolayers of nano - sized particles that are not soluble in the liquid film @xcite , or any substance that remains on the surface of the liquid film . our results indicate that nearly all long - wave models found in the literature that extend the hydrodynamic equations for thin liquid films covered by insoluble surfactants by including non - linear equations of state are either not fully consistent or not complete . the most important differences between our model and those in the literature , as discussed above in section [ sec : filmsurf - ext ] , are : * when incorporating a nonlinear equation of state , most authors fail to note that one must also amend the surfactant diffusion term in the governing dynamical equations . * to account for a surfactant - dependent wettability , it is not sufficient to just adapt the derjaguin ( or disjoining ) pressure . the marangoni and diffusion term must also be amended . * to account for surfactant phase transitions , square gradient , or other non - local terms for the surfactant concentration must be incorporated into the free energy functional . in these , the gradient should be taken along the free surface . * when incorporating terms to describe a surfactant phase transition that depend on the distance between the film surface to the solid substrate ( e.g. , to describe substrate mediated condensation ) , the added coupling terms lead to additional terms in the equation of state _ and _ the derjaguin pressure . the corrections to models in the literature that result from following our approach will in many cases only result in quantitative ( rather than qualitative ) changes to the results , that may also be rather small . in some cases , however , the differences will be qualitative and significant . for instance , we believe that the oscillatory dewetting modes ( `` dewetting waves '' ) described in ref . @xcite for one and two - layer films with surfactant , are present in the model as a consequence of a broken variational structure of the governing equations that stems from omitting terms in the equation for the time evolution of the surfactant concentration profile . using the complete equations , all eigenvalues of the linearised problem are real , and thus all instability modes are monotonic . the various extensions that we have proposed may all be simultaneously included so as to account for more complex situations . the corresponding free energy functional is @xmath210\ , = \,\int\left\{f(h,\gamma)+g(\gamma , h)\,\xi + \frac{\kappa}{2 } \left(\nabla\gamma\right)^2 \frac{1}{\xi } \right\}\,da { \label{eq : en : full}}\ ] ] where @xmath140 is a generalised wetting interaction term and @xmath211 is a generalised local free energy of the surfactant on the free surface . note that future work should identify the connections that must exist between these more general functions @xmath212 and @xmath104 because they both arise from the same molecular interactions ; this is an issue that we have not touched upon here . our approach may also be further extended to accommodate more general terms that one should expect to be present in the free energy functional , including non - local integral ( convolution ) contributions to @xmath113 , which are commonly used in dynamical density functional theory ( ddft ) @xcite , which uses as input the free energy functionals coming from equilibrium density functional theory @xcite . we emphasise that points ( i ) and ( iii ) above are particularly important for a number of bio - physical systems such as , e.g. , the description of the surfactant layers that reside on the aqueous thin - film of the lung lining @xcite , where ( a ) the equations of state that are used are strongly nonlinear and ( b ) experiments show that phase transitions frequently occur , e.g. , in layers of porcine lung surfactant at the air - water interface at physiologically relevant concentrations and temperatures @xcite . similar results are found for calf lung surfactant , where an expanded - to - condensed phase transition is observed as the surfactant concentration is increased @xcite . the dynamics of the ongoing surfactant phase transitions and their interaction with the hydrodynamics of the thin liquid film is highly important . the formulation and extensions we present here allows one to extend the ` classical ' hydrodynamic thin - film models to include the more intricate thermodynamic effects based on equations of state ( obtained from suitable free energy functionals ) that are observed experimentally . note , however , that a real layer of lung lining is much more complicated than the idealised situations mentioned above @xcite , as it consists of mixtures of soluble surfactants . work is currently underway to extend our approach to describe soluble surfactants @xcite . the extension towards mixtures of surfactants is more straightforward , but including a third field in the system makes the algebra somewhat tedious . further extensions that should be considered in the future concern the dynamical aspects . here , we have employed the mobility matrix @xmath207 in eq . , which is derived from the well known hydrodynamic transport equations obtained for the simplest case of an insoluble surfactant with a linear equation of state . although we believe this approximation should hold over a large parameter range , at very high surfactant concentrations one must make corrections . although , slip at the solid substrate can easily be accounted for , it is not clear what changes to the mobility matrix @xmath207 should arise from incorporating surface viscosity effects and/or a no - slip condition at the surfactant - covered surface . here we have only discussed the gradient dynamics formulation in the context of surfactant - covered liquid films on solid substrates , because the mathematical formulation is most convenient . however , it is important to note that most of the effects that we mention also occur in other geometries . a prominent example where our considerations also apply are soap films based on insoluble surfactants . this is important for many systems , such as those reviewed in @xcite , that involve , for instance , surfactant - dependent derjaguin pressures and highly nonlinear equations of state . another example are surfactant - covered drops of liquid immersed in another fluid where issue ( i ) is particularly relevant when , for example , studying the shear - driven deformation and/or breakup of such droplets @xcite or liquid bridges or threads @xcite . in this case , for instance , incorporating a nonlinear equation of state should also be accompanied by the corresponding amendment of the surfactant diffusion term . [ sec : app - vari ] the free energy for the surfactant covered thin liquid film is @xmath150\,=\,\int\left\{f(h,\widetilde\gamma/\xi)+g(\widetilde\gamma/\xi , h)\,\xi + \frac{\kappa}{2 } \left(\nabla(\widetilde\gamma/\xi)\right)^2 \frac{1}{\xi } \right\}\,da . { \label{eq : en : full}}\ ] ] we define @xmath150 = f_\mathrm{wet } + f_\mathrm{surf } + f_\mathrm{grad}\ ] ] so as to be able to calculate the variations of the three terms in the free energy separately . in the following , we limit ourselves to the 1-dimensional case and we often need to use the result : @xmath213 @xmath220\nonumber\\ & & -\frac{d^2}{dx^2}\left[\frac{\kappa}{\xi^3 } \left(\partial_x\frac{\widetilde\gamma}{\xi}\right)\gamma\partial_x h \right]\nonumber\\ & = & -\frac{d}{dx}\left\ { \frac{\kappa}{\xi^3}\left [ -\frac{1}{2}\left(\partial_x \gamma\right)^2\partial_x h -\left(\partial_x\gamma\partial_x h + \gamma\partial_{xx } h -2\frac{\gamma}{\xi^2}(\partial_x h)^2\partial_{xx } h\right)\partial_x\gamma \right.\right.\nonumber\\ & & \left.\left.-\left(3\frac{\gamma}{\xi^2 } ( \partial_x h)^2\partial_{xx } h - \partial_x\gamma\partial_x h - \gamma\partial_{xx } h\right)\partial_x\gamma + \gamma\partial_x h\partial_{xx}\gamma \right]\right\}\nonumber\\ & = & \frac{d}{dx}\left\ { \frac{\kappa}{\xi^3}\left [ \frac{1}{2}\left(\partial_x \gamma\right)^2\partial_x h + \frac{\gamma}{\xi^2}(\partial_x h)^2(\partial_{xx } h)\partial_x\gamma - \gamma\partial_x h\partial_{xx}\gamma \right]\right\}\end{aligned}\ ] ] @xmath223\\ & = & -\frac{\kappa}{\xi^4}(\partial_x\gamma)(\partial_x h)(\partial_{xx } h ) + \kappa\left [ \frac{2}{\xi^4}(\partial_x h)(\partial_{xx } h)(\partial_x\gamma ) -\frac{1}{\xi^2}\partial_{xx}\gamma \right]\\ & = & \frac{\kappa}{\xi^4}(\partial_x\gamma)(\partial_x h)(\partial_{xx } h ) -\frac{\kappa}{\xi^2}\partial_{xx}\gamma\end{aligned}\ ] ] the resulting expressions are @xmath224 \nonumber\\ \frac{\delta f}{\delta \widetilde{\gamma } } \,&=&\,\frac{1}{\xi}\partial_\gamma f + \partial_\gamma g -\frac{\kappa}{\xi^2}\partial_{xx}\gamma + \frac{\kappa}{\xi^4}(\partial_x\gamma)(\partial_x h)\partial_{xx } h { \label{eq : app - vari}}\end{aligned}\ ] ] this seems the appropriate stage in the derivation to apply the long - wave approximation , i.e. , to use @xmath225 . therefore @xmath226 and one obtains @xmath227 \nonumber\\ \frac{\delta f}{\delta \widetilde{\gamma } } \,&=&\,\partial_\gamma f + \partial_\gamma g -\kappa\partial_{xx}\gamma + \kappa(\partial_x\gamma)(\partial_x h)\partial_{xx } h { \label{eq : app - vari2}}\end{aligned}\ ] ] the respective last term is @xmath228 smaller than the other terms with prefactor @xmath229 and can therefore safely be dropped , yielding @xmath230 \nonumber\\ \frac{\delta f}{\delta \widetilde{\gamma } } \,&=&\,\partial_\gamma f + \partial_\gamma g -\kappa\partial_{xx}\gamma { \label{eq : app - vari3}}\end{aligned}\ ] ] eqs . ( [ eq : nonlocal - variations ] ) in the main text are obtained by setting @xmath231 , whereas the results in eqs . ( [ eq : insolsurf - variations ] ) are obtained by setting @xmath231 together with @xmath232 . g. amarandei , p. beltrame , i. clancy , c odwyer , a. arshak , u. steiner , d. corcoran , and u. thiele . pattern formation induced by an electric field in a polymer - air - polymer thin film system . ( submitted ) . j. w. cahn . thermodynamics of solid and fluid surfaces . in w. c. carter and w. c. johnson , editors , _ the collected works of john w. cahn _ , page 379 , warrendale , 1998 . the minerals , metals and materials society . v. b. fainerman , v. i. kovalchuk , e. h. lucassen - reynders , d. o. grigoriev , j. k. ferri , m. e. leser , m. michel , r. miller , and h. mohwald . surface - pressure isotherms of monolayers formed by microsize and nanosize particles . , 22:17011705 , 2006 . m. hameed , m. siegel , y. n. young , j. li , m. r. booty , and d. t. papageorgiou . influence of insoluble surfactant on the deformation and breakup of a bubble or thread in a viscous fluid . , 594:307340 , 2008 .

the understanding of the relationship between excitation parameters and oscillation regimes is a classical topic concerning bowed string instruments : for instance shelleng @xcite , or guettler guettler , or demoucron et al @xcite proposed 2d diagrams with respect to either bow force and bow position , or bow position and bow velocity . for reed instruments , this kind of diagrams are less numerous : in ref . @xcite , dalmont et al proposed a diagram with respect to excitation pressure and reed opening , and recently almeida et al almeida proposed a diagram with respect to blowing pressure and lip force , related to the reed opening . in the minimum model of reed , clarinet - like instruments , three parameters are considered : i ) the mouth pressure , ii ) the reed opening at rest , iii ) the length of the resonator assumed to be cylindrical . in ref . dalmont:05 , a supplementary parameter was added : the loss parameter of the resonator ( using the raman model , that considers resonator losses independent of frequency ) . this allowed explaining the extinction of sound when the mouth pressure becomes very large . we notice that the agreement of the theoretical results with experimental results was qualitatively satisfactory ( see ref . dalmont:07 ) . the objective of the present paper is to revisit the paper by dalmont et al@xcite : the focus is the search for a diagram of oscillation regimes of reed instruments with respect to two parameters : the reed opening , and the loss parameter . the choice of these two parameters is justified by the fact that the third parameter , the blowing pressure , is the easiest to modify for the instrumentalist . the elements of this diagram were rather complete in ref . @xcite , but phenomena occurring for strong losses , especially at the emergence of the sound , were not investigated . the use of simplified models for the prediction of the oscillation regimes is classical for musical instruments producing self - sustained oscillations . for the calculation of the instability thresholds , linearization was used ( see wilson and beavers @xcite or silva et al @xcite ) , while for ab initio computation , the iterated map scheme was studied ( see mc intyre et al @xcite , maganza et al @xcite , taillard et al @xcite ) . the interest of the model chosen in the present paper is that analytical formulas are possible , given clear dependence of the parameters ( for other models , numerical calculations could be possible by using similar basic ideas , using for instance continuation methodskarkar or time - domain methods @xcite ) . the analytical calculations presented hereafter are slightly different than those of ref . they also are limited to the limit cycles corresponding to the two - state oscillating regime , but are based upon a generalization of the fact that for this regime , when no losses are present , the flow rate is a constant . this regime is the most similar to what musicians consider as a normal " sound . in particular the method allows studying the character of the emergence and extinction bifurcations of this regime , which are important properties related to the possibility to play pianissimo or not , and more generally to the ease of playing . in ref . @xcite it was shown that the extinction bifurcation can be direct or inverse ; here it is shown that this is true also for the emergence bifurcation . in section [ model ] , the basic model of ref . @xcite is reminded ( see also @xcite ) , and a treatment of the problem based upon a unique quantity , the pressure difference @xmath0 is presented in section [ recu ] . this leads to a simple graphical analysis of the two - state regime , explained in section [ exis ] , yielding a proof that it can not exist with reverse flow , and an easy calculation method . then some blowing pressure thresholds ( stability , existence , ... ) are calculated with respect to the parameters of interest ( loss , reed opening ) . in section [ ct ] the thresholds related to the instability of the regimes are calculated . then by making two thresholds equal , the mouth pressure can be eliminated and limits of existence and stability of the static and two - state regimes are found in section [ dia ] : this allows drawing the diagram sought . finally a discussion is proposed concerning the existence of oscillating regimes ( section [ disc ] ) , with an attempt to consider more realistic models and a discussion about musical consequences . we briefly remind the basic elements of the model , the non - linear characteristic of the exciter , and the origin of the iteration method , thanks to a simplified treatment of the resonator . in a quasi static regime , the flow @xmath1 entering the resonant cavity is modeled with the help of an approximation of the bernoulli equation , as discussed e.g. in @xcite . comparison with experiment can be found in ref . we note @xmath2 the acoustic pressure inside the mouthpiece , assumed to be equal to the one at the output of the reed channel , @xmath3 the pressure inside the mouth of the player . for small values of the difference : @xmath4the reed remains close to its equilibrium position , and the flow @xmath1 is proportional to @xmath5 ; for larger values of this difference , the reed moves and , when the difference reaches the closure pressure @xmath6 , it completely blocks the flow . these two effects are included by assuming that if @xmath7 the flow @xmath8 is proportional to @xmath9 $ ] , and if @xmath10 , the flow vanishes.introducing the dimensionless quantities : @xmath11where @xmath12 is the characteristic acoustic impedance of the cylindrical resonator , having the cross section @xmath13 ( @xmath14 is the density of air , @xmath15 the velocity of sound ) , we obtain : @xmath16with@xmath17the parameter @xmath18 characterizes the intensity of the flow and is defined as : @xmath19where @xmath20 is the opening cross section of the reed channel at rest . @xmath21 is inversely proportional to square root of the reed stiffness , contained in @xmath6 . in real clarinet - like instruments , typical values of the parameters are @xmath22 $ ] ; @xmath23 $ ] ; values @xmath24 will not be considered here , since they correspond to multi - valued functions to be solved ( see ref . @xcite ) , and this case does not seem very realistic in practice for clarinet - like instruments . the function @xmath25 is obviously non - analytic ; it is made of three separate analytic pieces , with a singular point at @xmath26 , and its derivative is discontinuous at @xmath27 . the resonator of length @xmath28 is assumed to be cylindrical , with zero terminal impedance . using the dalembert decomposition , a change in variables at the entry of the resonator can be chosen as:@xmath29with the following relationship between incoming wave @xmath30 and outcoming wave @xmath31:@xmath32where @xmath33 is the loss parameter , assumed to be independent of frequency . this is a strong assumption , necessary to obtain square signals : for certain initial conditions , all quantities remain constant in the time interval @xmath34 . the approximation is rough for certain characteristics of the signal , such as the spectrum , but it is useful for the study of the existence , stability and amplitude of the produced sound . with this assumption , the resonator is characterized by a unique ( recurrence ) relation:@xmath35where @xmath36 is the time unit . as discussed further in section [ rs ] , losses can occur either at the extremity ( radiation ) or during propagation ( e.g. in the boundary layers ) : in the latter case , if @xmath37 is the attenuation constant per unit length , @xmath38 and the dimensionless input impedance at zero frequency is @xmath39while the input impedance at the operating frequency is @xmath40 . in what follows , the losses are characterized by the parameter @xmath41 , varying between @xmath42 ( no losses ) @xmath43and @xmath44 ( very strong losses , no wave reflection ) . this parameter , the reed opening @xmath18 and the mouth pressure @xmath45 are the three parameters of the problem . several combination parameters will be useful:@xmath46@xmath47 is proportional to the input impedance at zero frequency , while @xmath48 is proportional to the input admittance at the operating frequency . two pairs of parameters can be used : either @xmath49 or @xmath50 notice that because @xmath18 and @xmath41 are smaller than unity , @xmath51 and @xmath52 , and @xmath53 . other parameters will be useful and @xmath18 are with dimension , except in the appendix , and are denoted @xmath54 and @xmath55 , respectively . @xmath47 , @xmath56 and @xmath57 are defined in the same way than in the present paper.]:@xmath58 using eqs . ( [ e1 ] ) , ( [ e3a ] ) and ( [ e49 ] ) , the recurrence relation ( [ e4 ] ) can be rewritten with respect to the quantities @xmath59 and @xmath60 . the result is:@xmath61because the flow rate @xmath62 is a function of the pressure difference @xmath0 this relation is a recurrence for the quantity @xmath63 , equivalent to the recurrence used in ref . @xcite for the quantity @xmath64:@xmath65 \text { \ \ with } \label{pa2 } \\ h(x ) & = & x+\zeta f(x)\text { , \ \ } k(x)=\gamma ( 1+\lambda ) -\lambda ( x-\zeta f(x ) ) . \notag\end{aligned}\]]the inverse of function @xmath66 can be found in ref . @xcite ( appendix a ) . ( [ pa1 ] ) is interesting in particular for the calculation of the limit cycles . for the static regime , @xmath67 is a constant , then@xmath68it is possible to calculate @xmath67 from the value of @xmath45 , or vice - versa . concerning the stability , if the iteration function is denoted @xmath69 , $ ] the classical stability condition is @xmath70 . because @xmath71 , $ ] @xmath72 , and the condition can be written as is equivalent to @xmath73:@xmath74 ^{2}<1\text { \ \ or \ } \left [ f^{\prime } ( \delta p)+\beta _ { 1}\right ] \left [ 1+\beta f^{\prime } ( \delta p)\right ] > 0\text{\ } \label{e10}\ ] ] for the two - state regime , because @xmath75 , the following expression is found by eliminating @xmath45 from the equation ( [ pa1 ] ) written for the pairs ( @xmath76 , @xmath77 ) and ( @xmath77 , @xmath78):@xmath79an important property of the two - state regime is the square shape of the signal , which can be decomposed into the sum of a mean value @xmath80 ( zero frequency component ) and an acoustic component @xmath81 ( sum of the odd harmonics of the operating frequency ) , with zero mean value:@xmath82notice that @xmath83 , and that a similar equation can be written for the flow rate . ( [ e889 ] ) generalizes the result obtained when losses are ignored ( @xmath84 , i.e. the constant flow rate , and it is nothing else than the input impedance relation for the acoustic component:@xmath85 it is possible to calculate the values of the two states without knowledge of @xmath45 , starting e.g. with the value of @xmath86 : @xmath87 and @xmath45 are successively deduced from eqs . ( [ e889 ] ) and eq . ( [ pa1 ] ) . adding the two equations ( [ pa1 ] ) for the pairs ( @xmath88 , @xmath77 ) and ( @xmath77 , @xmath78 ) , it is obtained : @xmath89 \text { , or } \label{e921a } \\ \gamma _ { ij } & = & \frac{1}{2}(\delta p_{i}+\delta p_{j})(1-\beta \beta _ { 1})+\beta h_{1}\text { , } \label{e921}\end{aligned}\]]with @xmath90 and @xmath91 , and @xmath92 . similarly for regimes with more than two states , it could be possible to start the calculation from a given state , and to deduce the other states , the prior knowledge of @xmath45 being unnecessary . ( [ e921a ] ) is the input impedance relation for the mean value component : @xmath93 for the two - state regime , the stability condition is @xmath94 , and after some algebra , the result of ref . @xcite is found:@xmath95 the previous results can be applied whatever the shape of the function @xmath96 the present section investigates the existence of the static and two - state regimes for the particular shape of the function given by eqs . ( [ e2 ] ) and ( [ e2a ] ) . for negative @xmath67 , the function @xmath25 is negative too , therefore the static regime does not exist for negative flow ( and positive excitation pressure @xmath45 ) . for @xmath97 , the static regime exists for @xmath98 . the pressure @xmath99 vanishes : this is obvious because the reed closes the input of the resonator . otherwise , for a non - beating reed , the study of the function in the right - hand side of eq . ( [ e8 ] ) shows that it increases from @xmath42 to @xmath44 when @xmath67 increases from @xmath42 to @xmath100 therefore a unique solution \right ] ^{2 } \\ \text{with } \kappa & = & \left [ -2 + 9(-1 + 3\gamma ) \beta ^{2}\right ] /\left [ 2(1 + 3\beta ^{2})^{3/2}\right ] .\end{aligned}\ ] ] ] exists for @xmath101 . for the two - state regime , two values of @xmath67 , with the same value of the function @xmath102 are sought . they do not depend on the impedance at zero frequency , i.e. on the value of @xmath103 the study of @xmath102 leads to the following results ( see fig . [ fighx ] ) : for negative @xmath104 , the derivative @xmath105 is always positive , while for positive @xmath104 , it is positive up to @xmath106 , where @xmath107 is the positive root of the following equation:@xmath108 ^{2}. \label{e912}\end{aligned}\]]for @xmath109 ( beating reed ) , the value of the derivative is @xmath56 , and is positive . two cases need to be distinguished : if @xmath110 , i.e. if @xmath111 , the derivative is always positive and it is impossible to find two values of @xmath104 with the same value @xmath112 on the contrary , for @xmath113the function decreases from @xmath114 to @xmath56 when @xmath104 increases from @xmath115 to @xmath44 , then re - increases . solutions of eq . ( [ e889 ] ) can be found in this case only , and the corresponding value of the function is necessarily larger than @xmath56 . a consequence is that eq . ( [ e889 ] ) has no solution for @xmath116 , and this is true in particular for negative @xmath102 . therefore no two - state regime can be found with negative flow . this conclusion is compatible with the general result obtained for all possible regimes in ref . @xcite . because of this result , the paper is now focused on positive pressure differences @xmath117 for @xmath118 $ ] , three values of @xmath104 , defined as @xmath119 lead to the same value of the function ( see fig . [ fighx ] ) , and three two - state regimes are possible , which can be either non - beating ( for the pair ( @xmath120 , @xmath121 ) ) or beating ( for the pairs ( @xmath120 , @xmath122 ) and ( @xmath121 , @xmath122 ) ) . the intervals of the three solutions are as follows : @xmath123 $ ] ; @xmath124 $ ] ; @xmath125 . [ cols="^,^ " , ] the radiation losses are small at low frequencies , therefore it is very simple and classical to deduce them by perturbation from the output flow rate , considering a monopole radiation . two cases have to be distinguished : losses occur at the output , or losses occur during propagation into the tube . both cases give the same input impedance , but not the same transfer functions between input and output . the latter case is more realistic , and is considered here . if the output impedance is @xmath126 and losses due to boundary layers only , the output acoustic flow rate is given by the standard transmission lines relationships : @xmath127 , where @xmath128 ( see section [ besr ] ) . therefore the amplitude relationship is the following:@xmath129the maximum output acoustic flow rate is obtained for the saturation threshold ; at the saturation threshold , because the reed is beating , @xmath130(see eq . ( [ e18h ] ) ) . for @xmath131 the saturation threshold is the beating threshold ( see section [ eb ] ) , and @xmath132 , therefore:@xmath133for a given value of @xmath18 , this is monotonously decreasing function of @xmath134 ; for a given value of @xmath41 , this is a increasing function of @xmath135 the stability conditions ( [ e10 ] ) and ( [ e301 ] ) are calculated using the expression of the derivative @xmath136:@xmath137 the condition ( [ e10 ] ) generalizes the condition @xmath138 , as discussed in ref . @xcite for the lossless case ( see page 349 ) . for the ( static ) beating reed case , which exists for @xmath139 ( see previous section ) , @xmath140 , thus the static regime is always stable . for the non - beating reed case , the first factor of inequality ( [ e10 ] ) is equal to @xmath141 , thus it vanishes when @xmath63 satisfies eq . ( [ e914 ] ) , i.e. @xmath142 . for this value , @xmath115 , the second factor of inequality ( [ e10 ] ) is @xmath143 , which is positive ) vanishes : this leads to the solution of either @xmath144 when @xmath97 , or @xmath145 , with @xmath146 . these two equations have no solution . ] . thus ( [ e912 ] ) together with eqs . ( [ e8 ] ) gives the threshold @xmath147 , given by eq . ( [ e923 ] ) . therefore the condition @xmath148 gives both the limit of existence of the two - state regime ( see eq . ( [ e923 ] ) ) , and the instability threshold of the static regime . nevertheless the nature ( direct or inverse ) of the bifurcation between the two regimes is not yet solved by this result was assumed to be very small in practice , and eq . ( [ e923 ] ) was simplified in @xmath149 , but the complete eq . ( [ e923 ] ) was given for the threshold of existence for the two - state regime . ] . the term @xmath150 is the static pressure @xmath151 in the mouthpiece ; the presence of the parameter @xmath47 indicates that the threshold depends of the impedance at zero frequency . for the beating regime @xmath152 , with @xmath153 thus , using condition ( [ e301 ] ) , @xmath154 for @xmath155 or @xmath156 . because @xmath157 , and because @xmath158 , the second inequality ( [ e301 ] ) is never valid . therefore the stability is defined by the condition @xmath159 when @xmath160 , this condition is always satisfied , and the beating two - state regime is stable ( no overcritical threshold @xmath161 exists , as noticed in section [ eb ] ) . when @xmath162 , eq . ( [ e18f ] ) is used , yielding @xmath163 . the function @xmath164 being always increasing for positive @xmath165 , the inequality @xmath166 holds if @xmath167 , or @xmath168 this distinguishes in the ( @xmath169 ) plane the two branches separated by the overcritical threshold : the upper one is stable , while the lower one is unstable : because @xmath170 , the solution @xmath171 at the overcritical threshold is larger than @xmath115 , thus it is always larger than @xmath172 instability occurs for the pair ( @xmath173 @xmath122 ) . this is confirmed by figs . [ pfigiii ] and [ pfigi ] . ] . the unstable branch is the branch joining the static regime , because the two regimes can not be stable for the same value of the parameter @xmath45 when they converge to the same point . this can be explained with mathematical arguments , based upon either a perturbation method ( see ref . @xcite ) or the topological degree ( see ref . this can be also studied by using inequalities ( [ e301 ] ) , as done in ref . @xcite . it can be noticed that because @xmath161 is larger than unity , the beating , two - state regime is always stable for @xmath174 , whatever the value of @xmath175 \i ) for the non - beating regime ( @xmath155 , @xmath176 ) , an expression of the instability threshold was given in ref . @xcite and it was explained that the threshold is given by @xmath177 . some errors were done in the derivation , and they are corrected in appendix a of the present paper . when the excitation pressure is larger than this threshold , here denoted @xmath178 , period doubling can occur , then a complex bifurcation scenario ( see ref . @xcite ) . \ii ) otherwise , it can be checked that the second condition @xmath179 leads to the subcritical emergence threshold @xmath180 : it separates two branches in the ( @xmath181 plane , the upper one being stable while the lower one is unstable . this is similar to what happens for the overcritical threshold . when it exists , the emergence bifurcation is as follows : when playing crescendo , the oscillation starts for @xmath182 , while playing decrescendo , the oscillation stops for @xmath183 when this subcritical threshold can exist ? from eq . ( [ a ] ) and ( [ cz ] ) , it is found that @xmath184 \label{cq}\]]where @xmath185 and @xmath186 , with @xmath187 two cases are possible : * @xmath188 : the emergence bifurcation is direct : stable solutions exist for @xmath189 . * @xmath190 : the emergence bifurcation is inverse , and stable solutions exist for @xmath191 when @xmath192 continues to decrease below @xmath193 , the bifurcation remains inverse , but the subcritical threshold @xmath180 becomes the beating threshold @xmath194 . this happens when @xmath195 , or @xmath196 ( notice that the inequality @xmath197 always holds ) . the discussion is extended in section [ ol ] . from the different expressions of the thresholds , it is possible to deduce the limits separating different domains in the plane @xmath199 , as shown in fig . [ pertes4 ] . above the diagonal ( region 0 ) , no two - state regime can exist . other regions of the plane are defined by the nature of the emergence and extinction bifurcations : in regions i and iii , the extinction bifurcation is inverse , while in regions ii and iv , it is direct . what is new in this paper is the separation between regions i and ii , with direct emergence bifurcation , and regions iii and iv , with inverse emergence bifurcation . finally , in region v , the two - state regime can be unstable , and can be replaced by more complicated regimes , with period doubling , chaos , intermittences , etc ... ( see ref . @xcite ) . figures [ pfigi ] , and [ pfigii ] to [ pfigiv ] show examples of bifurcation schemes for the regions i to iv . the calculation is made using the analytical formulas given hereafter , and have been verified using the iterated map algorithm ( ref . @xcite ) : obviously only stable solutions are found with the latter method . the limits corresponding to the beating threshold ( between regions ii and iv , and between i and iii ) are discussed in section [ ol ] . the four regions i to iv correspond to the different cases presented for the examples shown in fig . [ pfigi ] , and [ pfigii ] to [ pfigiv ] . the chosen values for the examples are indicated by a circle . in region v period doubling or other regimes can exist , in region 0 no sound is possible . the numbers refer to the equation number . above line ( [ e22 ] ) , no sound is possible . curve ( [ e23 ] ) distinguishes the extinction bifurcation ( direct above , inverse below ) . curve ( [ e24 ] ) distinguishes the emergence bifurcation ( direct above , inverse below ) . below curve ( [ e237 ] ) , complicated regimes can appear by destabilization of the two - state regime . the two curves close to the curve ( [ e24 ] ) correspond to equations ( [ e25 ] ) , and ( [ e236 ] ) , from to the top to the bottom ( see section [ ol]).,width=377 ] , @xmath200 ( point ii of fig . [ pertes4 ] ) . see figure caption of fig . [ pfigi ] left . both emergence @xmath201 and extinction bifurcations @xmath202 are direct . the point @xmath203 indicates the beating threshold @xmath204 of the two - state regime . because @xmath205 , saturation occurs at the beating threshold @xmath206 , @xmath207 ( point iii of fig . [ pertes4 ] . see figure caption of fig . [ pfigi ] left . both emergence and extinction bifurcations are inverse . the point @xmath208 indicates the overcritical ( extinction ) threshold @xmath209 of the two - state regime , and the point @xmath202 the closure threshold @xmath210 of the static regime . ] , @xmath211 ( point iv of fig . [ pertes4 ] ) . see figure caption of fig . [ pfigi ] left . the emergence bifurcation is inverse while the extinction one is direct . ] * when the instability threshold @xmath212(eq . ( [ e923 ] ) ) of the static regime reaches the closing threshold @xmath213 , the static regime becomes stable whatever the values of all parameters , and no sound can be expected . this happens has two solutions : @xmath214 , and @xmath215 . because @xmath47 is necessarily less than unity , the latter solution is larger than unity . ] for @xmath111 , and this confirms the result explained in section [ nsol ] that no two - state regime can exist ( this discussion is extended in the next section ) . the condition for the existence of sound can be also written as : @xmath216 * when the overcritical threshold @xmath161 ( eq . ( [ e18q ] ) ) reaches the closure threshold of the static regime @xmath213 ( eq . ( e19 ) ) , the extinction bifurcation becomes direct instead of inverse , as explained in the previous section . for @xmath217 the bifurcation is inverse : this is probably the most frequent case for real clarinets and clarinettists , and corresponds to:@xmath218between the two limits ( [ e22 ] ) and ( [ e23 ] ) , the extinction bifurcation is direct ( regions i and iv ) . * when the subcritical threshold @xmath180 ( eq . ( [ cy ] ) ) reaches the instability threshold of the static regime @xmath147 ( eq . ( e923 ) , the emergence bifurcation becomes direct . from eq . ( [ cq ] ) , this happens when @xmath219 . the bifurcation is direct in regions i and ii , with the following condition : @xmath220 finally , when the instability threshold @xmath178 of the oscillating regime given by the condition @xmath177 ( see section sb ) reaches the beating threshold @xmath194 , the limit was given in ref . @xcite ( with a small error ) . the formula can be obtained from eq . ( [ e301 ] ) , for @xmath221 , and @xmath222 , @xmath223 the limit @xmath224 is given by the following equation:@xmath225this leads to a second order equation in @xmath226:@xmath227a particular value is @xmath228 , @xmath229 , @xmath230 , @xmath231 , @xmath232 . for a given loss coefficient @xmath41 @xmath233 , the two - state regime is always stable when @xmath234 . for sake of simplicity , we remark that an excellent approximation of the limit , better than 1% , is based upon the fact that for small losses , the coefficient @xmath235 is small , therefore @xmath236:@xmath237 another limit is reached when the subcritical threshold @xmath180 becomes the beating - reed threshold @xmath194 , thus the stable two - state regime is always beating . however the unstable branch is non - beating ( see figs . [ pfigiii ] and [ pfigiv ] ) . ( [ e ] ) , it is found to be given by @xmath238 , i.e.@xmath239 a last limit is reached when the instability threshold of the static regime @xmath147 ( [ e923 ] ) reaches the beating threshold of the two - state regime @xmath194 , i.e. when @xmath240 in eq . ( [ a]):@xmath241for both limits given by eqs . ( [ e25 ] ) and ( [ e236 ] ) , the corresponding curves are therefore within regions iii and iv of figure pertes4 . they are very close to the limit given by eq . ( [ e24 ] ) , corresponding to the change in nature of the emergence bifurcation . the three limits are even equal for @xmath242 , @xmath243 , @xmath244 for a given @xmath18 , when losses are small ( @xmath41 small ) , the emergence bifurcation is direct . then , when @xmath47 reaches the limit ( e24 ) , the bifurcation becomes inverse , with a non - beating reed . then , when @xmath47 reaches the limit given by eq . ( [ e25 ] ) , the bifurcation remains inverse , but the reed becomes always beating in the two - state regime . [ figbdzoom ] shows details of the bifurcation scheme ( @xmath245 near the subcritical threshold for different cases between the non - beating case @xmath246 and the beating case @xmath247 ( for the latter case , the beating threshold is the subcritical threshold , as for the points iii and iv ) . the corresponding values of the parameter ( @xmath248 ) are extremely close together . * left up : * case @xmath249 , @xmath250 * right up : * case @xmath251 , @xmath252 ; * left down : * case @xmath253 , @xmath254 * right down : * case @xmath255 , @xmath256 , title="fig:",width=188 ] * left up : * case @xmath249 , @xmath250 * right up : * case @xmath251 , @xmath252 ; * left down : * case @xmath253 , @xmath254 * right down : * case @xmath255 , @xmath256 , title="fig:",width=188 ] * left up : * case @xmath249 , @xmath250 * right up : * case @xmath251 , @xmath252 ; * left down : * case @xmath253 , @xmath254 * right down : * case @xmath255 , @xmath256 , title="fig:",width=188 ] * left up : * case @xmath249 , @xmath250 * right up : * case @xmath251 , @xmath252 ; * left down : * case @xmath253 , @xmath254 * right down : * case @xmath255 , @xmath256 , title="fig:",width=188 ] when losses tend to infinity ( @xmath41 tends to unity ) , no sound is possible whatever the value of @xmath18 : this is in accordance with the intuition that if no reflection exists at the input of the resonator , no self - sustained oscillation can happen . nevertheless , we do not prove that other types of regimes can not exist , such as four - state , eight - state , ... in this section , this issue is discussed together with the influence of the choice of the model . moreover , some musical consequences are discussed . when the static regime is always stable ( @xmath111 ) , it has been proved that the two - state regime can not exist . it is probable that other types of regimes do not exist , but the general proof is difficult . using the calculation of the successive iterates functions for different values of the initial condition ( see ref . @xcite ) , we verified that when @xmath257 , the successive iterates converge to the unique point that is the limit cycle of the static regime , thus no other regime can exist . this can be done for every set of parameter values , @xmath45 , @xmath18 , @xmath258 : if the convergence is always to a unique state , then it is sure that no other regime than the static one can exist . obviously , this verification is not possible in practice , but the verification has been done for some set of parameter values . we conclude that no oscillating regime exists for @xmath111 , even if the rigorous sentence should be : no oscillating regime exists by destabilization of the static regime . a similar discussion could be done for the destabilization of the two - state regime into more complicated regimes ( see section [ lit ] ) . the method used in the present paper can also be used for any shape of the nonlinear characteristic , at least numerically ( the condition being that a nonlinear characteristic is static ) . all equations of sections [ model ] and [ recu ] remain valid when modifying the nonlinear function @xmath259 in particular if a smooth beating transition is chosen with no singularity , fig . [ fighx ] shows that the condition ( [ e913 ] ) , @xmath260 , can be generalized into the following condition : the function @xmath261 goes through both a maximum and a minimum . the raman model is interesting because all quantities can be determined analytically , but it is not very realistic . it is based upon two important assumptions : losses do not depend on frequency , and the reed has no dynamics . a generalization of the present study is out of the scope of the present paper , but it is interesting to note that the condition @xmath262 can be easily generalized when these assumptions are not done , as explained hereafter . when losses depend on frequency , it is possible to use the characteristic equation obtained by linearizing the nonlinear equation around the pressure of the static regime @xmath151 , and writing the approximation of the first harmonic:@xmath263where @xmath264 is the admittance of the fundamental frequency . the characteristic equation is written as:@xmath265as it is well known ( see ref . @xcite ) , this gives the condition @xmath266 , thus the playing frequency @xmath267 at the threshold can be deduced . moreover , if at this frequency , @xmath268 , the pressure threshold is given by : @xmath269because @xmath270 therefore @xmath271 satisfies eq . ( [ e914 ] ) : @xmath272as expected in section [ is ] . as a consequence , when the losses depend on frequency , the value of the threshold is the same as for the raman model , and the limit of existence of the two - state regime @xmath242 is unchanged . nevertheless the hypothesis has been done that the small oscillations are sinusoidal ( on this subject , see refs . @xcite , grand or @xcite ) , and this is not true for inverse bifurcation . when several harmonics interact , the problem becomes much more intricate , especially because of resonance inharmonicity . moreover taking into account the frequency dependence of the losses leads to a distinction between the threshold of the fundamental regime and the overblown " regimes ( see ref . @xcite ) : this distinction does not exist with the raman model , which allows a distinction based upon the initial conditions only . as a summary , it can be said that if we suppose that the impedance peak of the operating frequency is higher than the other ones , the emergence bifurcation is direct and the limit @xmath273 is valid . this is true in particular for the first register of a clarinet and a part of the second register . the large increase of radiation losses at higher frequency due to the open toneholes lattice ( see e.g. ref . @xcite ) does not affect the highest peak , therefore the main result of the present paper can be extrapolated to a large number of notes of a real clarinet . when the reed dynamics is taken into account as that of a 1 dof oscillator , the following characteristic equation has been obtained by silva et al silva:@xmath274 \label{e80}\]]where @xmath275 , @xmath276 is the reed - resonance angular frequency , and @xmath277 its quality factor . the threshold pressure and frequency can be deduced from this equation , and were studied in this paper ; here we are interested in the limit for which the static regime becomes always stable , i.e. when @xmath278 if the input impedance of the resonator is considered around a resonance frequency @xmath279 , it is possible to write:@xmath280\]]where @xmath281 is the quality factor of the resonance . for @xmath282 the real part of eq . ( [ e80 ] ) leads to the following result:@xmath283for a lossless reed , @xmath284 : the limit of the losses in the tube allowed for having a sound is increased by the reed dynamics , which favors the sound production . but the effect of the reed losses is to decrease the limit . it can be concluded that reed losses and resonator losses act in the same sense concerning the range of parameter allowing sound production ( this conclusion is valid for the direct - bifurcation case ) . the previous results can be useful in order to understand and teach important aspects of the sound control by the instrumentalist . such an objective knowledge should largely increase the pedagogical efficiency . otherwise the approach of the problems remains more subjective and the explanations can be lengthy and less clear . one of the most useful aspects is about _ pianissimo _ playing . the bifurcation diagrams show that the players have two possibilities : near the emergence and near the extinction . the first possibility is used for playing _ dolce _ , with a quasi monochromatic sound , but the sound is noisy and can not be sustained for a long time , due to high value of the airflow @xmath285 ( see fig . [ pfigi ] ) . the dynamic is not easy to control because of the steepness of the bifurcation diagram near @xmath147 . the second possibility , near @xmath161 , conducts to a clean pianissimo , with a sound richer in high harmonics . this can however only be achieved by crossing the curve ( [ e23 ] ) in fig.pertes4 , in order to reach the region ii where the extinction bifurcation is direct . this property is usually unknown by the players ( the ability of playing such a magical `` pianissimo is often attributed exclusively to the quality '' of the reed ) . the beginners reduce the reed opening ( @xmath286 by biting `` the reed and this causes unwanted effects : the pitch rises considerably and such a bending stress can cause a plastic ( irreversible ) deformation of the reed . the skilled player can reach region ii by increasing the damping due to the lip and use a high blowing pressure near the extinction threshold , playing in the reverse way ( decreasing the mouth pressure for playing louder , see fig . [ pfigii ] ) . the lip comes very near to the tip of the reed , with a moderate lip pressure . this effect is probably similar to an increase of @xmath48 , resulting in a displacement of the playing parameters on the @xmath287 plane exactly in the wanted direction , increasing the value of @xmath41 and decreasing the value of the parameter @xmath18 proportionally to @xmath288 . the decrease in @xmath18 is probably due to an increase of the reed stiffness . decreasing @xmath18 without biting '' too much could also be achieved by modifying the hydrodynamics of the airflow entering the channel , in order to increase the _ vena contracta _ , but to our knowledge no experimental evidence shows that the player can indeed modify significantly the _ vena contracta_. conversely , it seems that the parameter @xmath47 can not be significantly controlled by the player ( in another way than by modifying the length of the air column ) . in real life , this parameter of static airflow resistance may not be determined only by the length and the diameter of the bore but certainly also by hydrodynamic effects near the channel , due to the viscosity of the air . this acts in a similar but possibly stronger way than the static resistance of the bore . practical tests show that the effects predicted in the zones iii and iv are indeed observed in some pathological situations , despite the fact that our theoretical model would require many meters of a tube of small diameter to obtain such high values of @xmath47 . to include the musician mouth in the model is obviously rather complicated , even if at low frequencies , the effect of the vocal tract is not important . therefore the previous analysis requires some conjectures . besides the problems of bifurcation , the analysis of the raman model permits establishing some facts useful for the musician : * the calculation of the mean flow shows that the most economical blowing pressure is near the beating threshold in region i ( corresponding to normal playing ) this explains that skilled players can sustain the sound significantly longer than beginners . * the transients are much faster if @xmath18 is large ( see ref . kergomard:95 ) ; weak reeds help doing this , as well as using a moderate lip pressure . this simplifies the staccato learning . * the effects of leaks in the air column ( misplacement of a finger , defective pads ) increase the value of @xmath56 , so that regions 0 , and probably iii or iv can be possibly visited ( see section [ fdl ] ) . almost any control can be destroyed over the dynamics ( or at least rendering the dynamic control more difficult ) , despite of the attempts of the clarinettist to supply more energy to the instrument by opening the embouchure , increasing @xmath18 . the present paper is focused on limit cycles corresponding to two - state regimes , and is a complement to the paper @xcite , which was focused on transients . thanks to a formulation focused on the pressure difference between mouth and mouthpiece , the effect of the nonlinear function on the production of the two - state regime can be analyzed , and especially the role of the losses . the map shown in fig [ pertes4 ] can certainly be improved by using more complex models , but we think that some results are robust . when the reed opening at rest is very small or when the reed stiffness is very large ( i.e. when the dimensionless parameter @xmath18 is very small ) , losses can be too large and the sound production becomes impossible . a complement to this conclusion is the following : for @xmath18 larger than @xmath289 , when losses increase , the emergence bifurcation becomes inverse before the sound disappears , and the instrument becomes more difficult to play . for @xmath18 smaller than this value , when losses increase , there is a direct passage from the direct emergence bifurcation to the absence of sound . this work was supported by the french national agency anr within the sdns - aimv and cagima projects . we thank also the high school arc - engineering in neuchtel . finally we wish to thank philippe guillemain and christophe vergez for fruitful discussions . the instability threshold @xmath178 of the oscillating regime is given by the condition @xmath177 ( see ineqs . ( [ e301 ] ) , it is the limit of unstable solutions toward period - doubling of the two - state regime ) . this leads to the following equation , if @xmath290 and @xmath291 are defined by eq . ( [ aaa ] ) : together with eq . ( [ azr ] ) leads to a fourth - order equation in @xmath290 ; from the solution @xmath290 the value the threshold of instability @xmath293 is deduced by using eq . ( [ a ] ) . another method is to start from certain values of the parameter @xmath56 and of the solution @xmath290 , then to deduce @xmath291 using eq . ( [ azr ] ) , then @xmath47 , which is solution of a second order equation in ref . @xcite , eq . ( a23 ) was correct , but a factor 4 was missing in eq . ( a24 ) , the correct equation being the present eq . ( [ e21 ] ) . ( a25 ) needs to be corrected by introducing a factor 4 on the right - hand side , and the factor @xmath294 needs to be replaced by @xmath295 in eq . ( a28 ) and similarly @xmath296 needs to be replaced by @xmath297 in eq . ( a30 ) . concerning the limit @xmath298 , eqs . ( a32 ) to ( a34 ) of ref . @xcite are corrected in section [ lit ] of the present paper . the last equation gives the coefficient @xmath56 as a series expansion of the limit @xmath224:@xmath299 \label{216}\]]this expression corrects eq . ( a34 ) of the previous paper ( the correction is small , because the order 6 in @xmath18 only is concerned ) , but this approximation is much less accurate than the present eq . ( [ e238 ] ) . a. almeida , j. lemare , m. sheahan , j. judge , r. auvray , kim son dang , s. john , j. geoffroy , j. katupitiya , p. santus , a. skougarevsky , j. smith and j. wolfe , clarinet parameter cartography : automatic mapping of the sound produced as a function of blowing pressure and reed force proceedings of the international symposium on music acoustics , 2010 , sydney , australia . kergomard , j. , `` elementary considerations on reed - instrument oscillations '' . in _ mechanics of musical instruments _ , vol . * 335 * ( a. hirschberg/ j. kergomard/ g. weinreich , eds),of _ cism courses and lectures _ , pages 229290 . springer - verlag , wien , 1995 . hirschberg , a. , van de laar , r. w. a. , marrou - maurires , j. p. , wijnands , a. p. j. , dane , h. j. , kruijswijk , s. g. and houtsma , a. j. m. a quasi - stationary model of air flow in the reed channel of single - reed woodwind instruments . _ acustica _ , 70 , 1990 , 146154 . dalmont , j .- p . , gilbert , j. and ollivier , s. . nonlinear characteristics of single - reed instruments : quasistatic volume flow and reed opening measurements , j. acoust . _ , _ 114 , 2003 , 22532262 .

studies of cp - violation in @xmath0 decays must involve final state interaction ( fsi ) effects . unfortunately , a reliable estimate of such effects is very hard to achieve . in the analyses of @xmath9 decays ( @xmath10 - pseudoscalar meson ) only some intermediate states , believed to provide nonnegligible contributions , are usually taken into account . many authors restrict their studies to elastic rescatterring @xmath11 only . in regge language this is described in terms of a pomeron exchange . although in @xmath12 quasi - elastic rescatterring at @xmath13 contributions from other nonleading regge exchanges are much smaller , they are not completely negligible and have been included in various analyses . the main problem , however , is posed by the sequence @xmath14 involving _ inelastic _ rescatterring processes @xmath15 . arguments have been given that it is these inelastic processes that actually constitute the main source of soft fsi phases @xcite . it has been also pointed out @xcite that nonzero inelasticity strongly affects the extraction of fsi phases in models based on quasi - elastic rescatterring . thus , inelastic events affect model predictions even if rescatterring is of quasi - elastic type only . on the other hand , fsi phases are often attributed directly to short - distance ( sd ) quark - line diagrams in the hope that this will take into account all inelastic production phenomena . this belief persists despite justified scepticism about the dominance of short - distance qcd in fsi of @xmath0 decays ( see eg.@xcite ) . in fact , it is known that the resulting prescription strongly violates such tenets of strong interactions as isospin symmetry @xcite ( see also @xcite ) . the origin of the problem pointed out in ref . @xcite is the lack of any correlation between the spectator quark and the products of @xmath16 quark decay . by its very nature such correlation can not be provided by sd dynamics . what must be involved here is a long - distance ( ld ) mechanism which ensures that quarks `` know '' about each other . thus we are led to hadrons , hadron - level dynamics , and inelastic rescatterring effects . in this paper we perform a simplified analysis of corrections which should be introduced by inelastic rescatterring into the sd - based description of some nonleptonic decays of @xmath0 mesons . of course , any such analysis must be half - qualitative in nature , because - in the presence of many decay channels - it is well beyond our ability to take them accurately into account . since it is believed that @xmath0 decays into @xmath17 and @xmath3 may provide some handle on the determination of angle @xmath18 of the unitarity triangle , we shall concentrate on these decays : it is important to know the effects of fsi here . as shown in ref . @xcite , inclusion of coupled - channel quasi - elastic effects ( @xmath19 ) generates an effective long - distance penguin amplitude comparable in size to the short - distance one . one may expect that inelastic channels will also contribute to this effect . we start with an sd - based model of nonleptonic @xmath0 decays . on the basis of standard tree - dominated mechanism for these decays , enriched with the related and well - established models of semileptonic @xmath0 decays , we qualitatively estimate the types and the number of states produced in the first stage of the nonleptonic decay . as in other existing models , we take these states as composed of two ( flying apart ) resonances @xmath4 ( section 2 ) . these resonances are assumed to rescatter into @xmath1 through regge exchange . in order to provide the basis for an estimate of this rescatterring , we recall how in a regge picture the unitarity relation involving @xmath20 and other @xmath21 processes looks like . this enables us to make a rough estimate as to what part of all inelastic @xmath22 processes is due to the @xmath20 transitions and , consequently , how much the situation deviates from the case of ( quasi-)elastic rescatterring . using a rough estimate for the contribution @xmath23 from an average single inelastic intermediate channel @xmath24 , we estimate the number of inelastic channels involved ( section 3 ) . in section 4 we analyse the behaviour of the @xmath25 and @xmath26 amplitudes as a function of the number of intermediate states considered . we show that although effects due to each single intermediate state are small , the combined effect of all intermediate states is large . in particular , amplitudes of decays into @xmath3 become significantly larger than those estimated through short - distance penguin amplitudes , to the point of being comparable to the @xmath7 amplitudes . our conclusions are given in section 5 . in short - distance approaches to nonleptonic weak @xmath0 decays , the relevant amplitudes are usually expressed as sums of amplitudes corresponding to different types of quark diagrams ( @xmath27 - tree , @xmath28 - colour - suppressed , @xmath29 - @xmath30-exchange , @xmath10 - penguin , @xmath31 - annihilation , @xmath32 - penguin annihilation ) . in this paper we concentrate on @xmath33 decays of @xmath34 mesons which are initiated by a @xmath35 transition . in these decays one expects that the dominant contribution comes from the tree diagram @xmath27 , with the main corrections to it provided by the colour - suppressed and penguin diagrams @xmath28 and @xmath10 @xcite . accordingly , neglecting contributions from other diagrams , the sd amplitudes @xmath36 for @xmath37 decays into a pair of octet pseudoscalar mesons @xmath38 with total isospin @xmath39 are expressed in the su(3 ) symmetry case as @xmath40 inclusion of @xmath41 and @xmath42 into our considerations is mandatory if we want to maintain su(3 ) symmetry @xcite . from the above formulas one may find sd amplitudes @xmath43 from state @xmath44 into states @xmath45 of given isospin @xmath39 belonging to definite representations @xmath46 of su(3 ) : @xmath47 with @xmath48\nonumber \\ { \bf w}_0^t&=&[\langle ( \pi \pi ) _ 0 | w | b^0 \rangle , \langle ( k \bar{k})_0| w | b^0 \rangle , \langle ( \eta _ 8 \eta _ 8)_0 | w | b^0 \rangle]\end{aligned}\ ] ] and the matrices @xmath49 given by : @xmath50 @xmath51\ ] ] and @xmath52\ ] ] where rows correspond ( from top to bottom ) to @xmath53 for @xmath54 ; @xmath55 for @xmath56 ; and @xmath57 for @xmath58 . one expects that sd amplitudes @xmath28 and @xmath10 constitute a 10 - 20% correction @xcite . indeed , for the factor @xmath59 in the relation @xmath60 , the short - distance qcd corrections give the value of @xmath61 ( @xmath62 , @xmath63 are wilson coefficients ) , while estimates in ref . @xcite yield @xmath64 in the range of @xmath65 . dominance of @xmath27 amplitudes is expected to hold for other decays initiated by @xmath66 as well . for larger invariant masses of @xmath67 and @xmath68 systems these quark - level states should hadronize mainly as @xmath69 states @xcite . it is the rescatterring from these states to the final pp state that is of interest to us . since we want to estimate rescatterring at hadron level , we need to know what hadronic states are produced in the first stage of the decay . there are two groups of hadrons produced : one comes from the decay @xmath70 , the other one originates from the recombination of the @xmath71 quark with spectator @xmath72 ( fig . 1 ) . the values of invariant mass squares @xmath73 and @xmath74 ( see fig . 1 for definitions ) are not large . the well - known probability distribution of @xmath16-quark decay is @xmath75 which falls for increasing @xmath76 , the average @xmath77 being @xmath78 . the values of @xmath74 are smaller . estimates obtained in various models for semileptonic and nonleptonic decays yield @xmath79 around @xmath80 or so , with the distribution of @xmath81 extending to around 2.0 or 2.5 gev @xcite . when such a low - mass quark - antiquark state hadronizes , a resonance is produced . in the isgw2 model @xcite for semileptonic decays , exclusive partial widths for the production of lowest - in - mass resonances have been predicted . the resonances considered were the ground - state mesons ( pseudoscalars @xmath10 : @xmath82 and vector mesons @xmath83 : @xmath84 ) , the p - wave mesons ( tensor mesons @xmath27 : @xmath85 , axial mesons @xmath31 : @xmath86 and @xmath0 : @xmath87 , and scalar mesons @xmath88 : @xmath89 ) , and the @xmath90 and @xmath91 states . the highest ( nonstrange ) resonance mass explicitly considered in @xcite is below 1.5 gev . the total partial width for the production of all these resonances is 5 - 6 times larger than the partial width for the production of a pseudoscalar meson . thus , the average partial width into a resonance is smaller by a factor of 0.7 or so than that for the production of a pseudoscalar meson . the resonances explicitly considered in @xcite do not saturate the inclusive decay rate which is still about two times larger @xcite . thus , several other resonances of masses below 2.0 or 2.5 gev should be added to the list given above . assuming that the average contribution from each one is similar to that just estimated , one expects that the number of types of `` average '' resonances produced should be of the order of 15 . similar or even larger number of resonance types is expected from the hadronization of the @xmath67 created from the @xmath30-boson . thus , in nonleptonic b decays , apart from the @xmath1 state , many other resonance pairs must be produced : @xmath4 = @xmath92 , ... , @xmath32 , @xmath93 , @xmath94 , @xmath95 , @xmath96, ... etc . exact counting of the number of all these two - resonance states is not important for our purposes . however , it is fairly easy to give an estimate : limiting oneself to resonances of mass smaller than 2 gev , this number will definitely be greater than 10 , probably of the order of a few tens . clearly , there is no hope that one can reliably calculate the contribution from rescatterring into @xmath1 from each of these intermediate states . however , one may try to estimate their overall contribution in an average way . for the first stage of the decay process , we will assume that the only important amplitude is the tree amplitude @xmath27 and that this amplitude is approximately the same for all intermediate states considered ( for low values of @xmath77 that we shall be concerned with later , this is is indeed the case in eq . ( [ eq : bdecay ] ) ) . although this may seem a very rough assumption for @xmath74 , we shall further see that our general results should be fairly independent of it as long as there is a rather large number of two - resonance states with production amplitudes scatterred around the average . the next question is how to describe transitions @xmath97 ( or vice versa ) in an average way . this is what we shall discuss in section 3 . elastic scatterring and multiparticle production processes are related to each other through unitarity of the s - matrix . since @xmath0 has spin @xmath98 , we shall work with @xmath99 sector only . in the following we shall suppress the subscript @xmath100 and its value . in refs . @xcite it was shown that in the su(3 ) symmetry case , with the effects of coupled channels included , one should work with states @xmath101 belonging to definite representations @xmath102 of su(3 ) . the unitarity relation for @xmath1 scattering in the @xmath103 partial wave and in su(3 ) representation @xmath46 is : latexmath:[\[\label{eq : unitarity } @xmath46 , @xmath105 labels states different from @xmath101 . matrix elements occuring in eq . ( [ eq : unitarity ] ) may be expressed in terms of argand amplitudes @xmath106 as follows @xmath107 apart from the pomeron , there are other regge trajectories , whose exchange in the @xmath108-channel contributes to quasi - elastic scatterring @xmath109 . when the leading non - pomeron exchange - degenerate regge trajectories @xmath110 , @xmath111 , @xmath112 , and @xmath113 and their su(3)-symmetric partners are taken into account , one obtains the @xmath103 partial wave amplitudes @xcite : @xmath114 with @xmath115 where for the leading non - pomeron regge trajectory we use @xmath116 ie . we have put @xmath117 and , consequently , in eqs ( [ eq : as ] ) and further on both @xmath118 and @xmath108 are in @xmath119 . the amplitudes @xmath120 are independent of isospin @xmath39 as shown in @xcite , where their sizes at @xmath13 have also been estimated . the su(3)-symmetric regge residue @xmath121 is fixed from experiment as @xcite @xmath122 while for the pomeron one has @xcite @xmath123 using the above values in eqs ( [ eq : as ] ) one finds @xmath124 while the leading non - pomeron regge contributions _ alone _ are : @xmath125 contribution from elastic scatterring ( the pomeron in regge language ) is independent of @xmath46 : @xmath126 ( cf . @xmath127 in ref . @xcite ) . omitting the reggeon - pomeron interference term , the value of the contribution from the leading non - pomeron regge exchange to the unitarity relation of eq . ( [ eq : unitarity ] ) , after averaging over representations @xmath46 , is equal to @xmath128 neglecting reggeon contribution to @xmath12 one obtains from eq . ( [ eq : unitarity ] ) : @xmath129 one may conjecture that contributions to the above sum from reggeon - exchange - induced processes @xmath130 ( where @xmath131 denotes low - lying resonance ) will be of size similar or smaller than @xmath132 . thus , if all inelastic channels @xmath133 were two - resonance states @xmath4 , one would obtain from eq . ( [ eq : numericalunitarity ] ) the number of @xmath134 as the number of states contributing to the unitarity relation . this should be compared with the estimate of a few tens obtained in the previous section for the number of two - resonance states produced in weak decays of @xmath0 meson . of course , the estimate of eq . ( [ eq : totalnumber ] ) is probably too low : contributions from transitions @xmath135 for heavier resonances @xmath131 are likely to be smaller and the total number of states may be larger . for example , with average value of @xmath106 going down by a factor of 0.6 - 0.7 from 0.08 to 0.05 ( section 2 ) , the above estimate of the number of average quasi - two - body states increases from 25 to over 60 . on the other hand , two - resonance states need not saturate eq.([eq : numericalunitarity ] ) . thus , there are two important questions which should be answered : 1 . how many of states @xmath133 in the unitarity relation of eqs ( [ eq : unitarity],[eq : numericalunitarity ] ) are indeed of the form @xmath4 , so that they can contribute to rescatterring in @xmath0 decays , and 2 . can the effect of all of these states be described in some average way . at present , elastic and quasi - elastic contributions from long - distance fsi in @xmath0 decays are usually evaluated using the old language of regge theory . this language was used in the past also for the description of resonance and multiparticle production processes that are of interest to us . in particular , the question of the build - up of elastic scatterring ( pomeron exchange ) as a shadow of inelastic multiparticle production processes ( and thus the very content of the unitarity relation of eqs ( [ eq : unitarity],[eq : numericalunitarity ] ) ) was discussed extensively . therefore , we must recall the essential elements of an approach which dealt with that problem . the approach , predominantly occupied with the issue of the unitarization of dual models ( for reviews see @xcite ) , was based on general properties of dual string models and on phenomenological analysis of resonance and multiparticle production data . in this approach , multiparticle production processes occurring in hadron - hadron collisions at high energies are pictured as proceeding through the production of resonances or clusters in a multiperipheral model ( fig . 2 ) with leading non - pomeron reggeons exchanged in between the clusters . it is the _ sum _ over _ all types _ and _ numbers _ of resonances produced in this way that saturates the s - matrix unitarity relation eq . ( [ eq : unitarity ] ) . for @xmath118 above the inelastic threshold but still small enough , one may limit oneself to the production of just one pair of resonances @xmath4 . with increasing energy @xmath118 , the number of resonances produced in a single collision increases on average . although cross - section for a production of a particular number of resonances goes down at sufficiently high energy , the sum of cross sections over all possible numbers of resonances remains approximately constant . this constancy of the total inelastic cross section is ensured by a sufficiently fast increase in the number of all possible quark - line diagrams , ie . in the number of all possible states @xmath133 ( and ways in which they are produced ) @xcite . in our case , at @xmath136 the model @xcite predicts that states composed of just two resonances are produced in the fraction of @xmath137 cases approximately . further 35% comes from production of three resonances , etc . although these numbers are obtained in @xcite for resonances of any mass , the main contribution comes from production of resonances of invariant mass squared smaller than @xmath138 . ( contribution from the production of objects of mass @xmath139 is suppressed as @xmath140 for larger @xmath139 @xcite ) . the average mass of a resonance produced may be estimated in various ways to be around @xmath141 ( @xmath142 in ref . @xcite ) , in good agreement with mass @xmath143 or so @xcite , expected for average @xmath81 in the sd - based models of weak decays . contributions from rescatterring of states with larger values of @xmath77 will be suppressed because production of such states in @xmath1 collisions is not likely . thus , in a rescatterring process @xmath144 one may expect that the dominant contribution will indeed come from the rescatterring of states composed of two low - mass resonances @xmath24 . translating the above expectation of a 50% share of @xmath4 states in the unitarity relation of eqs ( [ eq : unitarity],[eq : numericalunitarity ] ) into a number @xmath145 of contributing channels which may connect to the state originally produced by the sd dynamics ( ie . @xmath146 ) , we conclude that this number should be around @xmath147 for average @xmath148 or @xmath149 for average @xmath150 . the latter estimate is probably more realistic since the average size of a contribution from a single @xmath4 channel should diminish with growing resonance masses . if one accepts that final state interactions can not modify the probability of the original sd weak decay , it follows that vector @xmath151 representing the fsi - corrected amplitudes is related to vector @xmath152 of the original sd amplitudes through @xcite @xmath153 indeed , in the basis of @xmath154-matrix eigenstates @xmath155 the above equation reduces to @xmath156 , ie . the condition of unchanged probability ( @xmath157 ) admits watson phases only . the s - matrix may be written in terms of the matrix @xmath158 of amplitudes @xmath106 : @xmath159 we assume that we may treat the fsi - induced corrections to the sd decay amplitudes in a perturbative fashion . this is in agreement with the ideas of the dominance of sd dynamics . if this assumption is incorrect , obtaining even half - quantitative predictions will be almost impossible ( cf . although this assumption may be questioned , it has an important advantage : one may study what happens when the number of contributing two - resonance intermediate states is increased to its expected share ( @xmath160 ) . in agreement with the assumption of a perturbative treatment of rescatterring ( ie . small contribution from @xmath158 ) , we expand the square root in @xmath161 and keep only the first term . this leads to @xmath162 this is in fact the k - matrix prescription for the estimate of rescatterring effects @xcite . in this prescription , final state interactions modify probabilities of the sd amplitudes as can be seen explicitly from eq . ( [ eq : kmatrix ] ) written in the basis of s - matrix eigenstates . our calculations will be based on eq.([eq : kmatrix ] ) . in previous sections we considered a simplified picture of contribution from resonance pairs in which all amplitudes @xmath163 were equal in absolute magnitude while their phases were arbitrary . indeed , when one row of the unitarity condition ( ie . ( [ eq : unitarity ] ) ) is discussed , no knowledge of phases is needed . for the purpose of studies of cp - violation the question of phases is important , however . therefore , we have to make a very rough estimate of the fsi phases appearing in the strong rescatterring amplitudes @xmath164 in @xmath165 thanks to @xmath166 invariance of strong interactions , the @xmath164 amplitudes are symmetric ( as is the @xmath167-matrix ) . in order to estimate them we have to recall what are the predictions of dual string models for the production of two resonances in high energy @xmath1 collisions . in the appendix of ref . @xcite it is shown that the dual string model predicts that the amplitude for the @xmath135 production through an uncrossed diagram of fig . 3a will pick up a rotating regge phase resulting from the expression @xmath168 similarly , for the crossed diagrams of fig . 3b one has to remove the @xmath169 sign in the above expression , ie . the amplitude is real . thus , the phase - generating factor differs from the familiar one in @xmath12 scatterring ( ie . @xmath170 ) only by a different scaling factor ( @xmath171 ) in the denominator . such a dependence of regge amplitudes on the masses of produced resonances has been confirmed in analyses of experimental data @xcite . at this point we have to take into account the fact that regge amplitudes describe scatterring of two colliding resonances @xmath172 and @xmath173 in a state of definite momenta into a similar @xmath1 state . in particular , the produced @xmath1 state is a superposition of partial waves , while we are interested in the @xmath99 sector of the @xmath88-matrix only . restriction to the @xmath98 sector is achieved by integrating the rescatterring amplitudes @xmath163 with @xmath174 ( with @xmath175 being the angular momentum of the @xmath1 pair ) over the allowed range of scatterring angle @xmath176 , or over the corresponding range of momentum transfer @xmath177 . angular momentum conservation will then admit states @xmath24 with total angular momentum @xmath98 only . this will have no effect on the assumed form of the sd decay amplitude since in the sd mechanism the decaying @xmath16 quark does not `` know '' about the spectator quark , and , consequently , about the value of @xmath100 for the whole system . one calculates that for @xmath178 the minimal value of @xmath108 is @xmath179 , while for @xmath180 one may assume @xmath181 . projecting the @xmath97 regge amplitude onto the @xmath98 sector , ie . integrating regge expressions over @xmath182 , we obtain , up to a common overall normalization factor , the following phase factors : 1 . for the uncrossed diagram ( fig . 3a ) @xmath183 2 . for the crossed diagram ( fig . 3b ) @xmath184 where @xmath185 . the overall normalization of contributions from amplitudes @xmath97 was fixed in section 3 . for the regge description to be valid , the value of @xmath186 must be large . with @xmath187 , the product @xmath171 should not be greater than @xmath188 , possibly @xmath189 , corresponding to the minimum value of @xmath190 equal to @xmath191 . for @xmath77 equal to @xmath74 , this corresponds to the maximum value of resonance mass being around 1.6 - 1.8 gev . these are still reasonable numbers when compared with our previous estimates of @xmath192 . in fig . 4 we show dependence of the phase of @xmath193 on @xmath194 for @xmath195 . if one approximates @xmath196 by their average values of around @xmath197 , one finds that the average value of the product @xmath194 is close to @xmath198 . from fig . 4 we see that for @xmath199 the phase of @xmath200 is zero . although the phase of @xmath200 at smaller and larger values of @xmath194 deviates from zero quite significantly , these deviations are on average of the order of @xmath201 and , consequently , it is still meaningful to talk about coherent superposition ( ie . with approximately similar phases ) of regge contributions from various intermediate states . because the quark - line structure of @xmath12 and @xmath202 amplitudes is the same , the eigenvalues with which the above phase factors enter into expressions for @xmath203 in definite su(3 ) representations @xmath46 are the same as those in amplitudes @xmath109 . the latter were calculated in ref . @xcite . on the basis of ref . @xcite we have therefore @xmath204 the fsi - corrected amplitudes for @xmath0-meson decays into states in definite representations @xmath46 of su(3 ) and with isospin @xmath39 are @xmath205 where @xmath206 are sd weak decay amplitudes into @xmath24 , all assumed approximately equal to @xmath43 ( eq . ( [ eq : pptor ] ) , section 2 ) . as in section 3 , we assume that _ all _ amplitudes @xmath207 are equal to some average value @xmath208 . although the phase of @xmath200 changes over the range of corresponding @xmath194 , it is very instructive first to approximate it everywhere by a constant , namely its average evaluated at , say , @xmath209 , where @xmath200 is real . the approximate reality of average @xmath200 is a consequence of the particular value of @xmath118 ( being here equal to @xmath210 ) and not an @xmath118-independent feature of the approach . at @xmath211 we get @xmath212 the sum over all @xmath145 two - resonance states will then yield the contribution from inelastic rescatterring : @xmath213 note that for fixed @xmath214 the average amplitude @xmath215 is inversely proportional to @xmath216 . thus , contribution from inelastic events in eq . ( [ eq : constphaseapprox ] ) is proportional to @xmath216 : the smaller the value of @xmath208 , the larger the summed contribution from all two - resonance states , provided their contribution in the unitarity relation ( eq . ( [ eq : unitarity ] ) ) is fixed by the same value of @xmath214 . obviously , this is a general feature of any perturbative treatment of rescatterring contribution from several channels , the reason being the _ linear _ nature of perturbatively treated fsi as compared to the _ quadratic _ nature of the unitarity relation . thus , estimates of fsi effects given here are most likely estimates from below : although amplitudes @xmath217 corresponding to rescatterring from states composed of resonances of larger masses are expected to be smaller , their combined rescatterring effect should be relatively larger . the amplitudes for @xmath0 decays into @xmath218 , @xmath219 are calculated from the inverse of eq . ( [ eq : pptor ] ) : @xmath220 in table 1 we give the values of @xmath221 calculated from the above equation with the approximation of average phase ( eq . ( [ eq : constphaseapprox ] ) ) for @xmath222 and @xmath223 ( @xmath224 respectively ) . predictions of the model without inelastic rescatterring , ie . with quasi - elastic fsi ( @xmath1 intermediate states ) only , are also given for comparison . amplitudes are given in units of input tree amplitude @xmath27 . from table 1 one can see that quasi - elastic fsi do not affect the decays @xmath225 strongly . the amplitude moduli are somewhat smaller than those without the fsi . this is due mainly to the @xmath226~$ ] factor originating from pomeron exchange . phase changes are small . for the @xmath227 decays , quasi - elastic fsi affect the amplitude significantly : the amplitude driven in sd - dynamics by the penguin diagram ( here vanishing ) receives contribution from the coupled - channel - generated long - distance penguin ( hereafter denoted @xmath228 ) @xcite . using @xmath229 to estimate the effective ld penguin , the size of @xmath228 is ( as in ref . @xcite ) of the order of 5% of the tree amplitude @xmath27 , thus permitting significant interference effects with the short - distance penguin amplitude when the latter is taken from standard sd estimates . the main results of this work are given in the last two columns of table 1 . * one can see that the @xmath225 amplitudes , when compared with their estimates taking into account quasi - elastic fsi only , increase in absolute magnitude by a factor between 1 and 2 . this is due to the additional contribution coming from rescatterring chain @xmath230 . * an important change can be seen in phase sizes : they are now one order of magnitude larger than in the quasi - elastic case . the origin of this effect is as follows . the @xmath30 amplitude is composed of two parts : one ( approximately real ) is the sd amplitude @xmath231 weakly suppressed by elastic ( quasi - elastic ) rescatterring , while the other contains the contribution from the inelastic @xmath232 rescatterring . the latter part , being proportional to @xmath233 , is mainly imaginary ( for approximately real @xmath106 ) . with a large contribution from inelastic rescatterring , the resulting phase must therefore be large . * finally , we see that the @xmath234 amplitude becomes much larger than in the case of quasi - elastic rescatterring only , and is dominantly imaginary . since the @xmath234 amplitude is fed from no - hidden - strangeness states @xmath235 , one might be tempted to compare this with ref . namely , it was shown there for the case of a simple two - channel s - matrix that the phase of amplitude in channel 1 is large if the particle originally decays to channel 2 . in our case , although a similar result holds , it is not general - it depends on the value of @xmath118 at which amplitudes @xmath163 are evaluated . note that inelastic rescatterring renders the @xmath227 amplitude larger than the sd penguin estimate : in fact , it is comparable to the @xmath236 amplitude . in our approach the full ( fsi - corrected ) amplitudes for @xmath237 decays vanish since : + 1)we have assumed that only the tree amplitude @xmath27 is nonzero , and + 2 ) the rescatterring from the isospin @xmath238 state of @xmath239 , which in principle might feed the final @xmath240 channel , is zero . state of @xmath4 is antisymmetric ( ie . of the form @xmath241 ) , while the rescatterring contribution due to the uncrossed diagram of fig . 3a is zero when evaluated in between the antisymmetric state @xmath242 and the symmetric state @xmath243 ( for the definitions of states , see eq . ( 1 ) of ref . @xcite ) . rescatterring through diagram of fig . 3b can not change the type of quarks and , consequently , can not induce a transition from the no - hidden - strangeness state @xmath244 into @xmath245 . ] + the dependence of the ld rescatterring - induced effective penguin on the isospin channel - with vanishing ( large ) effects in i=1 ( 0 ) channel - should be compared with the sd mechanism which assigns the same size and phase to penguin amplitude @xmath10 in both the @xmath246 and @xmath240 decay channels . this is a general feature of long - distance dynamics : the size and phase of a quark - level diagram depend on what isospin ( su(3 ) ) amplitude it contributes to @xcite . although the above expectation of fsi effect larger than that naively expected is fairly general , we have to analyse in some detail our assumption of replacing the @xmath171-dependent phase of @xmath200 with an average ( and vanishing ) phase . such an assumption would be well justified if a large fraction of @xmath24 states led to phases close to the average . since for a given value of the product @xmath194 the phase is fixed , we need to know the density of two - resonance states as a function of @xmath247 . regge models @xcite predict that the dependence of @xmath248 on @xmath249 is proportional to @xmath250 for larger values of @xmath251 . this fall of distribution for larger mass values should be clearly visible already at the beginning of the region where using regge description becomes sensible , ie . at @xmath252 or so . the distribution of @xmath74 in sd models vanishes even faster ( it is fairly negligible above @xmath253 ) . since we want to estimate corrections to sd models , direct use of mass distributions generated in sd - models might seem to be the simplest and most natural choice . furthermore , as rescatterring of states with larger values of @xmath77 is suppressed by the small size of the contribution from transitions @xmath254 , one might use the sd distribution of @xmath74 as that of @xmath77 as well . the problem is , however , that a significant part of the sd distribution of @xmath74 ( or @xmath77 ) corresponds to @xmath255 . on the other hand , regge amplitudes assume the form of @xmath256 only for @xmath257 . for @xmath258 , one should replace @xmath259 with @xmath260 . thus , if regge phases are to be reasonably evaluated we must use an appropriately modified distribution of @xmath74 . we model this situation in the simplest possible way : by assuming that the distribution of @xmath261 vanishes below @xmath262 and above @xmath263 , while in between these values it is given by @xmath264 this yields the average value of `` effective '' @xmath139 equal to @xmath265 , which is also the value obtained from the accmm distribution of @xmath81 ( see eg . ref . @xcite ) if contributions from @xmath266 are replaced with contributions at @xmath267 . using @xmath268 as in eq . ( [ eq : mmdistr ] ) , it is straightforward to evaluate the distribution @xmath269 of @xmath171 . this distribution ( fig . 5 ) is peaked at @xmath270 , its median being at @xmath271 , and the average @xmath272 at @xmath273 . the tail above @xmath189 contributes a few percent only . thus , the assumption of @xmath274 used in our previous discussion appears quite reasonable . with two - resonance states @xmath133 spread in @xmath171 according to distribution @xmath269 ( larger values of @xmath275 correspond to states with appropriately larger values of @xmath194 ) , it is simple to analyse the predictions of the model numerically . we are interested in the question how the fsi effects change when heavier and heavier intermediate states @xmath133 are included . for decays into states in definite representations of su(3 ) , the amplitudes of interest to us are therefore : @xmath276 where @xmath277 is the number of intermediate inelastic states considered ( @xmath278 ) and @xmath279 are given in eq . ( [ eq : deltas ] ) . in fig . 6 we present predictions of the @xmath277-dependent version of eq . ( [ eq : inverse ] ) for the absolute values and phases of the amplitudes of @xmath0 decays into @xmath280 and @xmath246 states . these predictions are given as a function of the number @xmath277 of intermediate states considered . we show the case with @xmath281 ( for @xmath282 one obtains very similar plots with features discussed below being even more pronounced ) . the point most to the right in each plot ( ie . at @xmath283 ) corresponds to a large value of @xmath171 , where regge approximation breaks down . consequently , this point should be discarded . one can see that for @xmath284 the size of the amplitude does not depend very strongly on @xmath277 and is close to the value of the input fsi - free amplitude @xmath285 , which is @xmath286 . furthermore , the fsi - induced phase is still relatively small ( of the order of @xmath287 ) . this can not be said of the @xmath227 process . here , the absolute value of the amplitude grows fast with the increasing number of intermediate states . in addition , already at @xmath288 or so , the phase becomes close to @xmath289 , not far from the previous result of @xmath290 ( table 1 ) obtained for constant phases . one can also see that when the number of intermediate states approaches its maximum allowed value , the size of the @xmath227 amplitude becomes significant when compared to the tree @xmath284 amplitude . as the number of intermediate states increases , the long - distance - induced penguin amplitude starts to dominate over the sd one ( which was estimated at 0.04 to 0.20 of the tree amplitude , @xcite ) . furthermore , assuming applicability of eq . ( [ eq : qdbdecays ] ) , one estimates from fig . 6 that at large @xmath277 the effective long - distance penguin amplitude @xmath228 is around @xmath291 . this should be compared with the input or effective @xmath27 amplitudes which are around @xmath292 . thus , the long - distance effective penguin amplitudes become really large . large rescatterring effects obtained above stem from an _ approximately coherent superposition of contributions from several intermediate channels_. one might argue that in the calculation of this paper sizes of individual rescatterring amplitudes @xmath106 ( ie . contributions from individual intermediate channels ) are overestimated . however , the trend of the results expected for the case of smaller amplitudes @xmath106 may be seen from table 1 and eqs . ( [ eq : constphaseapprox],[eq : constphaseapprox1 ] ) . the column of @xmath282 corresponds to a smaller value of average amplitude @xmath293 . the connection existing between the values of @xmath145 to @xmath294 stems from the assumption that at energy @xmath13 a two - resonance state is produced in approximately half of all inelastic @xmath1 collisions . as long as this fraction ( @xmath214 ) is kept constant and the perturbative treatment of fsi is valid , smaller values of amplitudes @xmath106 - after summing over all intermediate channels - result in rescatterring corrections to weak decays larger than naively expected . one may also try to estimate roughly the rescatterring contribution in a direct way ( ie . without using the value of @xmath214 ) by just adding rescatterring contributions with amplitudes @xmath217 assumed to be of the order of @xmath295 or so ( in agreement with experiment ) . then , for the number of average intermediate states taken as equal to the number of final states in standard models of sd decays , ie . of the order of @xmath296 or more , one is bound to obtain a large rescatterring effect . please note that our whole approach starts with the generally accepted features of the sd decay ( and resonance production ) mechanism and assumes that subsequent fsi may be treated perturbatively . we have shown that even in this case the corrections tend to become large . of course , if they are too large , the whole perturbative scheme of their estimation ( as well as the sd mechanism for the description of @xmath0 decays ) ceases to be viable . we have discussed multichannel inelastic rescatterring effects in @xmath0 decays into @xmath17 and @xmath297 . generally accepted features of short - distance decay mechanism have been assumed as part of our input . these assumptions included estimates of the types and number of resonances produced in two - resonance states initiated by the @xmath298 transition . rescatterring of these two - resonance states into the final state consisting of one pair @xmath1 of pseudoscalar mesons was evaluated under the assumption that such fsi may be treated perturbatively . the basis for this evaluation was provided by existing knowledge about how the inelastic multiparticle ( resonance ) production in @xmath1 collisions is correlated with elastic @xmath1 ( pomeron exchange ) scatterring . this knowledge permitted us to estimate that at @xmath13 @xmath1 scatter into a two - resonance @xmath4 state in approximately 50% of cases . thus , the total size of rescatterring from @xmath4 to @xmath1 was fixed . using the regge model for the description of the @xmath97 processes , we have shown that the rescatterring contributions from the individual intermediate channels add approximately coherently . as a result , the combined effect of rescatterring through many two - resonance intermediate states was shown to be quite large . this was demonstrated under the assumption that fsi may be treated perturbatively . if that assumption is overoptimistic , reliable estimate of ( presumably even larger ) fsi effects will almost certainly be much more difficult . in our calculations the amplitude for the decays @xmath234 was induced by ld rescatterring from no - hidden - strangeness isospin 0 states produced via short - distance tree diagram . this fsi - induced amplitude was shown to be larger than its short - distance penguin counterpart . the phase of the ld amplitude was estimated to be around @xmath289 . the author would like to thank prof . p. singer and the department of physics at the israel institute of technology , haifa , where some ideas of this paper have originated , for their hospitality . j. f. donoghue et al . , phys . 77 , 2178 ( 1996 ) . m. suzuki and l. wolfenstein , phys . rev . d60:074019 ( 1999 ) . m. suzuki , _ final state interactions in heavy hadron decay _ , talk given at 3rd international conference on b physics and cp violation , taipei , taiwan , 3 - 7 dec 1999 , hep - ph/0001170 . a. n. kamal and c. w. luo , phys . d57 , 4275 ( 1998 ) . a. n. kamal , phys . d60:094018 ( 1999 ) . grard , j. weyers , eur . j. c7 , 1 ( 1999 ) . l. wolfenstein , phys . d52 , 537 ( 1995 ) . p. enczykowski , phys . b460 , 390 ( 1999 ) . m. gronau , o. f. hernandez , d. london , j. l. rosner , phys . d52 , 6356 ( 1995 ) . p. enczykowski , acta phys . b28 , 1605 ( 1997 ) . m. gronau , phys . b300 , 163 ( 1993 ) . a. v. dobrovolskaya , a. b. kaidalov , k. a. ter - martirosyan , and v. r. zoller , phys . b229 , 293 ( 1989 ) . l. wolfenstein , phys . d43 , 151 ( 1991 ) . g. altarelli et al . , nucl . b208 , 365 ( 1982 ) . v. barger , c. s. kim , and r. j. n. phillips , phys . b251 , 629 ( 1990 ) . d. scora and n. isgur , phys . d52 , 2783 ( 1995 ) . i. bigi , r. d. dikeman , n. uraltsev , eur . j. c4 , 453 ( 1998 ) . m. battaglia , `` study of @xmath299 decays with an inclusive generator '' hep - ex/0002040 . hong - mo chan , sheung tsun tsou `` dual unitarization : a new approach to hadron reactions '' , rl-76 - 080 , sep.1976 , published in bielefeld sum . 1976 , 83 ; g. f. chew , c. rosenzweig , phys . rept . 41 , 263 ( 1978 ) . huan lee , phys . 30 , 719 ( 1973 ) ; g. veneziano , phys . lett . b43 , 413 ( 1973 ) . hong - mo chan , j. e. paton , sheung tsun tsou , sing wai ng , nucl . b92 , 13 ( 1975 ) . hong - mo chan , j. e. paton , sheung tsun tsou , nucl . b86 , 479 ( 1975 ) . n. sakai , nucl . b99 , 167 ( 1975 ) . p. hoyer , r. g. roberts , and d. p. roy , nucl . b56 , 173 ( 1973 ) a. c. irving , r. p. worden , phys . 34 , 117 ( 1977 ) fig.6 + dependence of fsi - induced effects on the number of intermediate channels included . + + table 1 . + effects of inelastic rescatterring on @xmath304 amplitudes in average phase approximation . amplitudes are in units of input tree amplitude @xmath27 .

the purpose of this work is to describe some aspects of the dynamics of homogeneous three - dimensional incompressible fluids . the evolution of such fluids is governed by the euler equations @xmath0 where @xmath1 and @xmath2 denote the velocity and the vorticity of the fluid . we shall focus on fluids in which remarkable structures , called vortex filaments , are present . a vortex filament is a very thin tube in which the vorticity is sharply concentrated . it can be asymptotically assimilated to a curve @xmath3 in @xmath4 , where @xmath5 denotes the time and @xmath6 the arc - length parameter . according to a suitable formal derivation , which will be described in the next section , the asymptotic motion law of the curve is the binormal flow @xmath7 equation exhibits a rich variety of motions , which will be presented in section [ sec : one - filament ] below . particular attention will be payed to motions generating a singularity in finite time . on the other hand , special solutions of the binormal flow are given by the infinite straight filaments . up to a rotation , we can assume that there are parallel to the @xmath8axis so that they can be parametrized as @xmath9 for some fixed @xmath10 , where @xmath11 can be also seen as an arc - length parameter . of interest is the evolution of perturbations of such straight filaments , namely the nearly parallel vortex filaments , which are parametrized by @xmath12 where @xmath13 is small . it turns out that linearization about leads to the linear one - dimensional schrdinger equation for @xmath14 @xmath15 next , we wish to investigate the dynamics of several vortex filaments that are all nearly parallel to the same @xmath8axis . we consider a collection of @xmath16 filaments , which are represented by the complex - valued functions @xmath17 , with @xmath5 and @xmath11 . in 1995 , klein , majda and damodaran @xcite derived , in a suitable asymptotic regime , a system of coupled equations for the filament positions @xmath18 @xmath19 here @xmath20 corresponds to the circulation of the @xmath21th filament , and the parameter @xmath22 is related to the core structure of the filament . in particular , in the case where the vortex filaments are all exactly parallel , i.e. if @xmath23 does not depend on @xmath24 , the previous system reduces to a two - dimensional system of ordinary differential equations for the positions @xmath25 @xmath26 this system is called point vortex system or kirchhoff law , and it has been intensively studied in the literature . we observe that the system combines two different aspects of the dynamics . on the one hand , the linearized self - induced motion of each filament already expressed in the equation is represented through the schrdinger operator . on the other hand , the interaction of the @xmath21th filament with the other filaments is represented by the potential field @xmath27 , which corresponds to the velocity generated by the other filaments . ] . for straight filaments there is no self - induced motion , which means that each filament moves only with the velocity induced by the other filaments according to exactly as for planar vortices , see section [ sec : point ] hereafter . the fact that both effects are taken into account in is really due to the asymptotic conditions chosen for the derivation , which relate the wavelength of perturbations , the core sizes and the separation distances between the vortex tubes in a particular way . since the system is thought to be a simplified model for the dynamics of filaments , a first natural issue is the existence and uniqueness of solutions . the potential fields are not well - defined at points when two or more filaments collide in finite time . one possibility is to consider weak distributional solutions , allowing possibly for exceptional collisions , and for which the potential is well - defined almost - everywhere ( see section [ sec : filaments ] ) . but then uniqueness is not achieved in this class . therefore , we will focus on a class of stronger solutions , for which the filaments are well - separated up to the largest time of existence of the system . hence the largest time of existence corresponds to the first collision time in this framework . it should be mentioned that , even for the simpler situation of straight filaments ( the point vortex system ) , collisions in finite time are known to occur for some initial configurations ( see section [ sec : point ] ) . therefore we can not hope for global existence in general . we shall determine sufficient conditions on the initial collection of vortex filaments that lead to large time or global existence , and we shall also study the case of finite - time collapses . the remainder of this work is organized as follows . in section [ sec : one - filament ] we present the formal derivation of the binormal flow for the motion of one single filament and we describe some of its remarkable properties . in particular we review the recent results on singularities for self - similar and almost self - similar solutions of , as well as for their perturbations . section [ sec : point ] contains a brief state - of - the - art concerning the point vortex system , as an introduction to the more complicated dynamics of almost parallel vortex filaments . then , section [ sec : filaments ] gives an overview of existence and uniqueness results concerning the system . in particular , we present some recent results from @xcite on symmetric configurations of vortex filaments . we conclude by providing several and simple examples of finite - time collapses of vortex filaments . the collapses are of different nature , and some of them are new . * notations*. in the following we will systematically identify complex numbers and real vectors . in particular , introducing the matrix @xmath28 we shall identify the @xmath29-vector @xmath30 and the complex @xmath31 . * acknowledgements*. the first author is partially supported by the anr project `` r.a.s . '' . both authors are grateful to vincent torri for his support in using scilab . in this section we briefly present classical facts about the local induction approximation and about the binormal flow , as well as recent results of singularity formation in finite time . for parts of this description the reader may also consult the book @xcite and the surveys @xcite and @xcite . we shall start with a short description of the way the binormal flow equation was find as a model for vortex filaments dynamics in a three - dimensional ideal incompressible fluid . a fluid with a vortex filament is a fluid where the vorticity @xmath32 is a singular measure supported along a curve @xmath33 , which is parametrized by arc - length parameter @xmath34 , and with density @xmath35 . here @xmath36 denotes the constant circulation along the filament . as a consequence of kelvin s circulation law vortex filaments move with the flow , so in order to get the dynamics of the vortex curve @xmath33 we have to compute the velocity of the fluid near the curve . the three - dimensional biot - savart kernel is @xmath37 so the velocity of the fluid is @xmath38 we fix @xmath39 , we suppose without loss of generality that @xmath40 and we get focused on what is happening for the velocity of the fluid near @xmath41 . for this purpose , two localizations are done , one is that @xmath3 will be approximated by a taylor development of order 2 near @xmath42 and the other is that the integral has to be considered only locally around @xmath42 , on a segment @xmath43 $ ] . we obtain that @xmath44 is approximated by @xmath45 when @xmath46 , the first term is diverging but in the same way the velocity does around one straight vortex filament . one straight vortex filament remains still in a fluid , so this first contribution should not influence the dynamics of one single vortex filament @xmath33 . finally , even after localization , the integral in the last term is still diverging as @xmath47 . a time rescaling allow to resorb this growth and then the binormal flow @xmath48 appears as an approximation for the dynamics of vortex filaments , called local induction approximation . it was discovered by da rios @xcite in 1906 and re - lightened by arms and hama @xcite in 1965 . however the rigorous derivation of from the euler equations , and in fact even the question of the stability of the vortex filaments , are open . being a crude model , the binormal flow has advantages and disadvantages . the positive points are the fact that special solutions as straight lines , translating circles , helices are on the one hand solutions of the binormal flow and on the other hand these dynamics are observed in fluid dynamics . this way other special solutions of the binormal flow may give some intuition for particular type of vortex filaments in fluids . a successful example in this direction are the traveling type solutions of the binormal flow found in @xcite and displayed in an fluid experiment @xcite . also , the binormal flow is accessible for numerics . one of the main issues of the l.i.a . is that closed curves conserve their lengths under the binormal flow which is not the case for closed vortex filaments in fluids . the binormal curvature flow also appears in the context of superfluids that are governed by the gross - pitaevskii equation @xmath49 more precisely , is conjectured to govern the asymptotic dynamics of codimension-@xmath50 submanifolds around which the jacobians of the solutions to ( a rescaled version of ) concentrate . this was proved in any dimension @xmath51 in the special case of the @xmath52-dimensional sphere @xcite . in a parallel setting , mean curvature flow governs the vortex dynamics for parabolic ginzburg - landau equation @xcite . the fact that vortex filaments exhibit a common asymptotic motion law in fluids and superfluids results from a fundamental analogy between and the euler equations in such regimes . more precisely , the madelung @xcite transform @xmath53 , as long as @xmath14 does not vanish , yields a compressible euler type system for the variables @xmath54 and @xmath55 . it is called hydrodynamical form of and it reduces formally to the incompressible euler equations for @xmath56 in the above - mentioned singular limit of . further details may be found in the recent survey @xcite and in references quoted therein . the binormal flow is a rich geometric equation . it is a completely integrable system . in the following we shall denote curvature and torsion of a solution of by @xmath57 and @xmath58 . some remarkable conserved quantities of the binormal flow are the kinetic energy @xmath59 , helicity @xmath60 , linear momentum @xmath61 , angular momentum @xmath62 . it is a reversible in time equation , invariant with respect to translations and rotations . the tangent vector @xmath63 of a solution of the binormal flow satisfies the schrdinger map equation @xmath64 which plays an important role in ferromagnetism as a simplification of the landau - lifschitz equation . let us present now the link with the schrdinger equation . as mentioned in the introduction , the linear schrdinger equation can be obtained easily as a rough approximation of the binormal flow . this can be seen by plugging in the ansatz @xmath65 to leading order when @xmath66 , we obtain that @xmath67 is a solution of the linear schrdinger equation . it was in 1972 that hasimoto @xcite has discovered the link between the binormal flow and the cubic one - dimensional nonlinear schrdinger equation . the so - called filament function constructed from curvature and torsion of a solution of , @xmath68 satisfies the 1-d cubic nls @xmath69 with @xmath70 in terms of curvature and torsion @xmath71 . more precisely , @xmath72 which might be seen as a compatibility condition . let us notice that the hasimoto transform can be seen as an inverse of madelung s transform - actually curvature and torsion satisfy an euler with quantum pressure type of system . one can remark that the non - vanishing of the curvature appears as a constraint . nevertheless , koiso gave in @xcite a way to avoid this issue by using , instead of frenet frames , parallel frames @xmath73 that satisfy @xmath74 then @xmath75 is a solution of 1-d cubic nls . for details of this kind of hasimoto transform we refer to appendix a of @xcite . as a conclusion , up to a change of phase , the filament function satisfies the 1-d cubic nls . by using the hasimoto transform the first local well - posedness results were obtained for the binormal flow - curvature and torsion were considered in high order sobolev spaces , see @xcite . a recent result of local well - posedness for the binormal flow , for less regular closed curves , was obtained by jerrard and smets @xcite by considering a weak version of the binormal flow . their method acts at the level of the tangent of the curve and it does not use the hasimoto transform . also , they have derived this way new results for the schrdinger map equation . as far as we have seen , there exists a number of intimate relations between the evolution of filaments in fluids and dispersive equations . we end this subsection by discussing other similar connections . in order to improve the validity of the binormal flow as a model for vortex filaments dynamics , an extended version of was considered by fukumoto and miyazaki @xcite , which includes an axial flow along the filament , @xmath76 here @xmath77 denotes the magnitude of the axial flow . the hasimoto transform leads then to the so - called hirota equation @xmath78 which is in link both with nls and the modified kdv equation ( see @xcite for the initial derivation of this equation ) . it actually turned out that the modified kdv equation appears as a model for the evolution of the curvature of a vortex patch in a two - dimensional fluid @xcite , see also @xcite for more details and for self - similar solutions of this model . an interesting class of solutions that is not covered by the previous well - posedness results are the self - similar solutions of the binormal flow and their small and smooth perturbations . since solutions of are considered to be parametrized by arc - length , self - similar solutions are of type @xmath79 it is then easy to compute that they form a family of solutions @xmath80 indexed by a real parameter @xmath81 and @xmath82 ( @xcite ) . they have been studied in the last decades as a frictionless model for the evolution after a line - line reconnection in the absence of counterflow for superfluid 4he and in ferromagnetism @xcite . they also appear , this time for finite length filaments with appropriate boundary conditions , in some models for the fiber architecture of aortic valve leaflets , see @xcite . their rigorous mathematical study was achieved by gutierrez , rivas and vega , who gave in @xcite a description of the profile and in particular established that @xmath83}(s),\ ] ] with @xmath84 distinct , non - opposite and @xmath85 so self - similar solutions of the binormal flow are smooth infinite curves that generate a corner in finite time . in fluid mechanics , their dynamics is close to the one of the delta - wing vortex . in @xcite numerical simulations for self - similar solutions of the binormal flow were shown to be in correlation with the physical experiment . another family of generating a singularity in finite time at the level of the frenet frame are the almost self - similar solutions @xmath86 where @xmath87 is a real @xmath88 antisymmetric matrix , see @xcite . gutirrez and vega @xcite proved that @xmath89}(s).\ ] ] the curves at times @xmath90 are spirals , some of them with a singularity point , see @xcite for some numerical simulations . both similar and almost self - similar solutions have self - similar curvature and torsion , so the filament function is also of type @xmath91 and has to solve @xmath92 therefore the data has to be homogeneous of degree @xmath93 . the case of dirac distribution leads to the self - similar solutions of , and the principal value distribution to the almost self - similar solutions respectively . the study of small and smooth perturbations of self - similar solutions was initiated by banica and vega @xcite from the point of view of the stability of the formation of the singularity . the starting point of the analysis is the hasimoto transform : the filament function of the self - similar solutions @xmath80 is @xmath94 solution of . by the pseudo - conformal transformation @xmath95 understanding the behavior at time @xmath96 for perturbations of @xmath97 , solutions of , is equivalent to understanding the large time behavior for the equation @xmath98 a scattering theory has been developed in @xcite for this problem , by strongly relying on fourier analysis . the informations on have been transported at the level of the binormal flow by treating differently the regions @xmath99 and @xmath100 . in order to discard the linear non - oscillating term , the asymptotic profile for @xmath101 has to be modified into the long - range one @xmath102 in @xcite it has been shown that wave operators exist , with a strong @xmath103 decay , for asymptotic profiles @xmath104 . by adding the extra - conditions @xmath105 and @xmath106 the existence of solutions of that develop a corner - type singularity at @xmath90 has been derived . this insures that the formation of singularity in finite time appearing in the self - similar case is not an isolated phenomenon . a similar result for the almost self - similar solutions of has been proved in @xcite . asymptotic completeness with loss of regularity was then proved in @xcite , in spaces defined in terms of the behavior of the low fourier modes . at the level of the binormal flow , this information has been translated into the fact that all small and regular perturbations at time @xmath107 of a self - similar solution @xmath80 still yield , after evolving by the binormal flow , a singularity at time @xmath90 . finally , in @xcite we give a geometrical description of the curves at the singularity time by getting a control of the evolution of weighted norms for equation . we prove the stability of the selfsimilar dynamics of small pertubations of a given selfsimilar solution , and in particular the fact that the angle of @xmath80 is recovered at time @xmath90 . [ sec : point ] this section is devoted to a short presentation of the point vortex system . the reader may find the subsequent results as well as many additional details in the surveys @xcite and in references quoted therein . the point vortex system arises as an asymptotic motion law for point singularities in two - dimensional incompressible fluids , which are governed by the incompressible euler equations , or superfluids , governed by the gross - pitaevskii equation . in contrast with the three - dimensional case , the persistence of point vortices and the validity of the point vortex model have been established rigorously for well - prepared data , at least up to the first collision time between the vortices ( see @xcite for the euler equations and , e.g. , @xcite for the gross - pitaevskii equation ) . system possesses an hamiltonian structure , since it has the form @xmath108 with the hamiltonian @xmath109 defined by @xmath110 as a consequence , we have the conservation of @xmath109 @xmath111 moreover , the point @xmath112 and the angular momentum @xmath113 are preserved in time . if @xmath114 , we define the constant center of vorticity @xmath115 it follows that the quantity @xmath116 is constant as well . a direct consequence of the previous conservation laws is that , in case of circulations @xmath117 all having the same sign , no collision between the vortices can occur in finite time and the @xmath51-vortex system admits a unique and global solution . it is well - known that the point vortex system admits particular solutions that keep the geometric shape of the polygon formed by the vortex positions . to describe these solutions we shall adopt the following definitions from , e.g. , @xcite : we say that the configuration @xmath118 is @xmath119 a fixed equilibrium if @xmath120 , @xmath121 ; @xmath119 rigidly translating if , for all @xmath39 , there exists @xmath122 such that @xmath123 , @xmath121 ; @xmath119 a relative equilibrium if there exists some @xmath124 and for all @xmath39 , there exists @xmath125 such that @xmath126 , @xmath121 ; @xmath119 a self - similar ( or homographic ) motion if there exists some @xmath124 and for all @xmath39 , there exists @xmath127 , with @xmath128 , such that @xmath129 , @xmath121 . note that in the first three cases both the shape and the size of the polygon formed by the points are preserved , since the motion is a translation in the second case and a rotation with angular velocity @xmath130 in the second case . in the case of homographic motion the configuration at time @xmath39 is obtained from the initial one by a rotation and a dilation . more precisely , it must satisfy @xmath131 , where the reals @xmath132 and @xmath133 are determined in terms of @xmath134 . when @xmath135 , which actually corresponds to the case @xmath136 , the configuration shrinks and collapses into the center of vorticity @xmath137 at time @xmath132 . otherwise we have @xmath138 , i.e. @xmath139 and the configuration expands in time . for @xmath140 the distance @xmath141 remains constant therefore the solution is global . the motion is a uniform rotation about the center of vorticity if @xmath142 , and a rigid uniform translation otherwise . for @xmath143 the motion is integrable and has been widely investigated , see e.g. @xcite . an important part of the analysis was initiated in 1877 by grbli @xcite in his phd thesis . in particular grbli noticed that the @xmath144-vortex system can be entirely formulated in terms of the three distances between the vortices . the motion can then be analyzed qualitatively through the use of trilinear coordinates and phase diagrams , see the work of aref @xcite . there exists a large variety of regimes : fixed equilibria ( if @xmath145 ) , rigidly translating equilateral triangles ( if @xmath146 and @xmath147 ) , uniformly rotating equilateral triangles ( if @xmath148 and @xmath147 ) , collinear rotating or fixed configurations . also , self - similar ( shrinking or expanding ) collapsing configurations exist @xcite . it turns out that the conditions @xmath149 and @xmath145 are both necessary and sufficient to obtain a self - similar expanding or shrinking motion , and this provides an effective and simple way to construct them explicitly . on the other hand , the linear stability of such motions has been investigated in , e.g. , @xcite . we shall come back to these collapses in section [ subsec : filaments - collapse ] , where we will use them to construct collapsing filament solutions . for @xmath150 the @xmath51-vortex problem is not integrable in general , but it still exhibits some remarkable solutions . first well - known relative equilibria are given by the uniformly rotating configuration of @xmath51 identical vortices placed on the same line in a particular way , or , when @xmath51 is odd , by @xmath151 identical vortices and a central vortex of arbitrary circulation . other relative equilibria are obtained by setting one vortex with same circulation @xmath152 at each vertex of a regular @xmath51 - polygon . the motion is then a uniform rotation with constant angular velocity @xmath153 , where @xmath154 denotes the size of the polygon . the problem of stability of these @xmath51-polygons was raised by thomson ( lord kelvin ) in the nineteenth century and has attracted much attention since then . the full answer to thomson s question that the regular @xmath51-polygon is stable if and only if @xmath155 was finally recently provided by kurakin and yudovitch @xcite . one can get other uniformly rotating relative equilibria by adding to a regular @xmath51-polygon configuration one point of arbitrary circulation @xmath156 at the center of the polygon , in which case the angular velocity is @xmath157 . in particular , choosing @xmath158 one obtains a fixed equilibrium solution . the stability of centered @xmath51-polygons has been proved to hold by cabral and schmidt @xcite for all @xmath51 provided the circulation @xmath156 lies in a suitable interval depending on @xmath51 . in particular , the range of stability includes negative values of @xmath156 for @xmath159 . finally , we mention that other relative equilibria can be constructed in the form of concentric rings of polygons with identical vortices . the polygons may be not regular , and the rings may have different circulations ( heterogeneous rings ) . relative equilibria of identical two - vortex or three - vortex rings have been determined by aref , newton , stremler , tokieda and vainchtein @xcite , and oneil @xcite constructed relative equilibria of heterogeneous three - vortex rings . the existence of collapsing solutions for @xmath150 is a difficult issue . given a set of circulations such that @xmath160 for all @xmath161 , the set of initial configurations leading to a collapse in finite time has zero lebesgue measure in @xmath162 @xcite . on the other hand , in general it is not known whether self - similar collapses exist . necessary conditions for such motions are @xmath149 and @xmath163 , but , in contrast with the case @xmath143 , there are not sufficient . nevertheless , some specific cases have been explored : four and five - vortex collapses were studied by novikov and sedov @xcite and collapses of two - vortex rings were constructed in @xcite . collapsing configurations for infinite point vortex lattices are constructed by oneil in @xcite . finally , @xcite exhibits self - similar collapses of heterogeneous three - vortex rings . the objective of this paragraph is to review known existence results concerning the system , which have been established by klein , majda and damodaran @xcite and kenig , ponce and vega @xcite . we also mention the work by zhakarov @xcite . we recall the system supplemented by initial data @xmath164 : @xmath165 the hamiltonian @xmath167 the mean angular momentum @xmath168 and the mean center of vorticity @xmath169 they are the three - dimensional analogs of the notions of hamiltonian , angular momentum and center of vorticity for point vortices , which have been defined in section [ sec : point ] . there is also conservation of the quantity @xmath170 which can be interpreted as a mean momentum . nevertheless , the previous quantities may be not well - defined , not even formally . actually , for general configurations of exactly parallel vortex filaments there are infinite . in the next section we will bypass this difficulty by the use of renormalized quantities . first rigorous results on vortex filaments were obtained by klein , majda and damodaran @xcite for pairs of filaments @xmath173 it may be convenient to introduce the new variables @xmath174 , @xmath175 and @xmath176 , @xmath177 . then @xmath178 in @xcite ( see also @xcite ) only the case of identical vortex core parameters @xmath179 is considered . then for identical circulations @xmath180 we get @xmath181 so can be decoupled into two equations for @xmath182 and @xmath14 . the solutions have the general wave - like form @xmath183 where @xmath182 is any solution to the linear schrdinger equation , and where the parameters @xmath184 , @xmath185 and @xmath134 satisfy a consistency equality . on the other hand , for opposite circulations we have @xmath186 . by section [ sec : point ] , the exactly parallel solution is given by a uniformly translating vortex pair . for nearly parallel filaments one looks for solutions with the same symmetry features , i.e. @xmath187 for all @xmath188 . the equations then yield a coupled system of equations for the real variables @xmath189 and @xmath190 . from a mathematical point of view , one can also consider the case when @xmath191 . it follows that @xmath181 so the equations are decoupled @xmath192 for identical circulations one retrieves the previous co - rotating filament pair . in the case of opposite circulations , @xmath14 satisfies the linear schrdinger equation . the linear stability of system , with @xmath179 , around the exactly parallel solution to the point vortex system is analyzed in @xcite . linear instability occurs when @xmath193 , while linear stability holds for @xmath194 . moreover , numerical computations predict global existence in the first case , and self - similar collapse in finite time in the second case . so , in contrast with the @xmath50-dimensional setting ( pairs of vortex points ) , finite time collapses of two nearly parallel filaments might happen . following zhakarov @xcite , klein , majda and damodaran suggested that such collapses should be well described by a class of self - similar solutions to the system . also the equations could maybe provide a class of self - similar collapses . to our knowledge , such constructions have not been rigorously achieved yet . the next decisive results regarding were accomplished by kenig , ponce and vega @xcite in 2003 . the article @xcite is presented in the survey @xcite . in @xcite , existence of vortex filaments in the form of @xmath195-perturbations of the exactly parallel solution @xmath25 is investigated with the ansatz @xmath196 note that , thanks to the sobolev embedding @xmath197 , the ansatz guarantees that the filaments are well - separated as long as the quantity @xmath198 is sufficiently small with respect to the distance between the point vortices @xmath25 . as long as this remains true , the @xmath199 solve a system of schrdinger equations with potentials that are lipschitz functions in the @xmath199 : @xmath200 this makes it possible in @xcite to obtain local well - posedness for for any @xmath51 , for any set of circulations @xmath117 , and for sufficiently small @xmath201 , and at least up to times of order @xmath202 , where @xmath203 is the first collision time between the point vortices . on the other hand , kenig , ponce and vega @xcite obtained global existence for sufficiently small @xmath201 in the two following special cases : @xmath140 and @xmath194 , or @xmath143 and @xmath204 is the uniformly rotating equilateral triangle solution ( see section [ sec : point ] ) . the proof of @xcite is based on the use of a suitable notion of energy . we recall that for filaments given by the ansatz the quantities defined in [ para : conserved ] do not make sense . this is the reason why @xcite introduces the renormalized quantities @xmath205 note that in the framework of we have @xmath206 when @xmath207 . moreover , in view of the properties of the point vortex system , the previous quantities are still formally preserved in time provided @xmath208 for all @xmath209 . now , in the cases under consideration in @xcite the distances @xmath213 are constant in time and all the same . therefore it turns out that @xmath214 is constant . on the other hand , for circulations having all the same sign it was noticed in @xcite that the sobolev embedding combined with the convexity inequality @xmath215 for @xmath216 $ ] imply coercivity for the energy : @xmath217 as long as the vortex filaments @xmath18 are not too far from the straight filaments @xmath218 . finally , it can be easily seen that @xmath219 therefore in the cases of small @xmath195-perturbations considered in @xcite the energy is and remains small for all time , hence the vortex filaments remain well - separated for all time and global existence follows . we stress that the vortex filaments constructed in this way remain uniformly close to the straight filaments for all time . to conclude this section , we mention that lions and majda @xcite introduced and proved the global existence of `` very weak '' solutions to for any @xmath166 , identical circulations and identical vortex core parameters . such solutions satisfy a weak formulation of with suitable test functions , there are @xmath220-periodic with respect to @xmath24 , they belong to @xmath221 and they have finite energy : @xmath222 . actually , such a definition does not exclude the possibility of collisions , since the only information is that @xmath223 moreover , uniqueness in this class is not known . we present now the results obtained recently in @xcite . in all the following we will take identical vortex core parameters @xmath224 we start with a natural remark that should in part motivate the kind of perturbations we shall consider then . as we have seen in the previous sections , the dynamics of both systems and is complicated when @xmath51 becomes large , and in fact additional symmetry conditions are needed . in this paragraph , we shall first describe a local in time result that extends the one in @xcite to energy - type spaces . we will turn then to the case of four filaments for which global results can be obtained without requiring too many symmetries on the perturbations . finally , for any @xmath16 we will present a global existence result for dilation - rotation - type perturbations of the regular n - polygon exactly parallel configuration . in this latter case the perturbations , at any height @xmath24 , have a regular n - polygon shape . moreover , traveling waves will be displayed in this setting . we shall complete this subsection with another class of simple perturbations . for a positive @xmath134 we define the energy @xmath225 the motivation for this definition will become clear in the next paragraphs . recurrent tools in @xcite , which are partially inspired by @xcite ( in particular see and ) , are the following facts ( see lemmas 2.1-.2.3 in @xcite ) . * if @xmath226 has small energy @xmath227 , then @xmath226 is close to @xmath228 , @xmath229 * if @xmath226 is small in @xmath230 norm and if @xmath39 small enough , @xmath231 * if @xmath232 then we can compare the energies : @xmath233 * if @xmath226 has small energy @xmath227 and @xmath234 is small in @xmath195 , then the energy @xmath235 is finite . more precisely we have , for absolute numerical constants @xmath236 , @xmath237and @xmath238 as recalled in [ subsec : results - filaments - before ] , local existence was proved in @xcite for for any @xmath166 and small @xmath195-perturbations @xmath239 . the relevant energy functional in that setting is @xmath240 which has been defined in . we assume next that all the circulations have the same sign , for example @xmath241 for all @xmath209 . then the energy @xmath242 satisfies the properties listed above for each function @xmath243 . it is then natural to consider the problem of perturbations with small energy as a single assumption . on the one hand for small initial energy we obtain that @xmath244 is not too far from @xmath245 and that @xmath246 . on the other hand for @xmath247 small in @xmath195 it is easy to see that @xmath246 . therefore the energy smallness assumption is weaker than the @xmath195 smallness assumption in general . for example , it allows for a larger geometric class of perturbations , namely the small rotations . indeed , take a rotation and translation type perturbation of the square configuration , @xmath248 with @xmath249 and @xmath250 . it follows that @xmath251 while @xmath252 the idea to use the energy space rather than @xmath195 is also reinforced by the fact that in the setting of , that we shall see later to be related in some cases to the gross - pitaevskii equation , the @xmath253-norm of the solutions might grow in time while the energy remains controllable . _ for an initial configuration @xmath256 with small energy , there exists @xmath257 and a unique solution @xmath258,h^1({\mathbb{r}}))^n$ ] to the system with @xmath259 and for @xmath260 sufficiently small such that @xmath261 _ the proof is based on finding a find a fixed point in the banach space @xmath262,h^1\right)^n,\quad \sup_{0\leq t\leq t } \|w(t)\|_{h^1 } \leq \frac{d}{4}\right\}\ ] ] for the operator @xmath263 defined by @xmath264 then the solution is given by @xmath265 notice here that we extend @xmath199 locally from a time @xmath132 not by a fixed point for perturbations of the initial data directly , but by a fixed point argument for small @xmath195 perturbations @xmath266 of the linear evolutions of the initial data . we must use crucially the fact that the deviation @xmath267 can be upper - bounded in @xmath268 in terms of the energy at the initial time @xmath269 . this is insured from the previous study of the gross - pitaevskii equation ( see lemma 3 in @xcite ) . we recall the short proof : the fourier transform of @xmath267 can be written as @xmath270 , so the @xmath253 norm is bounded by @xmath271 and the @xmath230 norm is bounded by @xmath272 , i.e. @xmath273 as a consequence of the local in time result we obtain that the solution @xmath274 to exists as long as the energy @xmath275 remains small . indeed note that on the one hand the norm @xmath276 can grow exponentially , but it can not blow up as long as the energy is sufficiently small . on the other hand , as long as the energy remains small , the filaments @xmath277 remain close the the straight ones @xmath25 , so no collapse occurs for . [ [ n4-local - and - global - results - around - the - square - configuration ] ] @xmath278 , local and global results around the square configuration ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ _ if @xmath284 is small enough , there exists an absolute constant @xmath285 , and there exists a time @xmath260 , with @xmath286 such that there exists a unique corresponding solution @xmath287 to on @xmath288 $ ] , satisfying @xmath289 , with @xmath290,h^1({\mathbb{r}})\right)$ ] , and such that @xmath291 , \quad \sigma\in { \mathbb{r}}.\ ] ] moreover , if the initial perturbation is parallelogram - shaped , namely @xmath292 then the solution @xmath287 is globally defined . _ the proof goes as follows . in view of the local existence argument exposed in the previous subsection , the extension of a solution is ensured by the control of the energy . in this particular setting of the square , direct computations lead to the formula @xmath293 if @xmath294 is a solution of then @xmath295 is also a solution . so if the initial perturbation is a parallelogram like the one presented in figure [ parallelogram ] , it will remain so at later times , hence @xmath296 and global existence follows . on the other hand , for a perturbation without symmetry conditions the growth of @xmath297 up to time @xmath260 is controlled by a lengthy computation based crucially on the particular properties of the underlying square configuration @xmath279 . finally , let us notice that one could try the same approach for other configurations @xmath279 . explicit computations give for instance @xmath298 for @xmath143 and @xmath279 at the ends and the middle of the segment , and @xmath299 for @xmath300 and @xmath279 a regular hexagon . nevertheless , these quantities have no reason to be conserved , unless the perturbations retain the same shape as @xmath279 , which enters precisely the framework of the next paragraph . [ [ ngeq-2-global - results - for - regular - polygon - type - configurations ] ] @xmath16 , global results for regular polygon - type configurations ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ in order to get global results for a large number @xmath51 of filaments , in @xcite symmetry conditions are imposed both on the straight filaments configuration and on the perturbations . more precisely , we let @xmath279 be the regular @xmath51-polygon configuration , with or without its center , rotating with constant velocity @xmath134 and centered at the origin . we consider filaments @xmath287 with equal circulations @xmath280 and of the type @xmath301 so in particular for any time @xmath39 and any height @xmath24 , the points @xmath302 form the same polygon as @xmath118 up to a rotation or / and dilation , see figure [ polygon ] . a straightforward computation shows that if @xmath287 is a solution of then @xmath182 is a solution of @xmath304 eq . is an hamiltonian equation , with hamiltonian given by @xmath305 in our context , as we expect that @xmath287 are small nearly parallel perturbations of @xmath279 , we have @xmath306 , so that is formally similar to the already mentioned gross - pitaevskii equation @xmath307 with corresponding hamiltonian given by @xmath308 this formal comparison turns out to be valid : as we have reminded at the beginning of this subsection , when @xmath306 the energies @xmath309 and @xmath310 are comparable . so our objective is now to solve globally the cauchy problem for the equation in the energy space , in the same spirit as previous works on , see @xcite . _ for an initial data @xmath313 with @xmath279 a regular polygon configuration , with or without its center , rotating with constant positive speed @xmath134 , and @xmath314 small , has a unique global solution of type such that @xmath315 and @xmath316 . in particular , if @xmath317 then for all time @xmath318 , so the filaments @xmath287 remain nearly parallel . _ the proof is based on the transposition of the arguments presented in [ local ] to the setting of . we notice that @xmath319 since @xmath320 is conserved it follows that the corresponding local existence result for can be iterated and global existence is achieved . in the third part of @xcite subsonic traveling waves for are constructed , still under the condition @xmath311 . more precisely , for @xmath321 sufficiently close to @xmath322 there exist solutions of of the shape @xmath323 with @xmath324 a smooth function of small energy , of even increasing modulus , exponentially increasing to @xmath228 as @xmath24 tends to @xmath325 . moreover , @xmath326 these traveling waves are the equivalents objects of the grey solitons for the 1-d gross - pitaevskii equation . the proof of the existence follows the approach in @xcite and @xcite . the equation for the profile @xmath327 @xmath328 is a little more complicated than the one for the gross - pitaevskii equation which can be solved explicitly . it is integrated in @xcite using the smallness assumption on @xmath329 . as long as the energy is small , the modulus of @xmath330 stays close to @xmath228 and the nonlinearity in enters the frameworks of @xcite ( see also @xcite ) , so these traveling waves are orbitally stable . also , the results in @xcite are valid and insure that there are no non - trivial supersonic traveling waves . finally , by reusing an argument from @xcite , we can transform these traveling waves in the following way . system is galilean invariant , which means that for any real @xmath331 and any solution @xmath287 , the family @xmath332 is also a solution . by choosing @xmath333 we get as another solution of @xmath334 which represents a stationary @xmath335-twisted @xmath51-helix filament configuration with some localized perturbation traveling in time on each of its filaments . as noticed in @xcite , if the perturbations are considered of the shifted form @xmath336 for any @xmath279 with @xmath117 all the same , it follows that @xmath337 must solve the linear schrdiger equation . for an initial data in @xmath338 , by using the dispersion inequality for the linear schrdinger equation it follows that the filaments remains separate globaly in time and the perturbations spread along the straight configuration @xmath279 . by considering less regular initial data one can get examples of perturbations @xmath337 decaying at infinity that lead to a @xmath268 dispersive blow - up for the linear schrdinger equation . for instance the homogeneous data @xmath339 with @xmath340 yields a self - similar linear schrdinger solution , smooth for positive times ( @xcite ) . so this solution leads to solutions of blowing - up in @xmath268 in finite time at height @xmath341 . recently an example of initial data in @xmath253 but not in @xmath338 that provides dispersive blow - up was given in @xcite . more precisely , the solution with initial data @xmath342 with @xmath343 blows - up in finite time at one point . this makes a second example of solutions of that blow up in finite time at a certain height but for which no collision occur . in the last part of this paper we describe configurations of nearly parallel filaments evolving towards collision in finite time . most of them are new results . they are based on perturbations of type of exactly parallel filament configurations @xmath279 . therefore the collapse for the filaments @xmath287 solutions of is linked to solutions of , with modulus initially close to one , that vanish at least at one point in finite time . so a pointwise control of solutions of is needed , which is quite unusual in the study of the schrdinger equation . let us first notice that if @xmath344 leads to a collision in finite time , the shifted perturbations introduced in [ shift ] , @xmath345 with @xmath346 in @xmath195 for instance , yield a collapse of the same kind . so collapses are not isolated phenomena . [ [ gaussian - collapses - about - a - stationary - regular - polygon - configuration ] ] gaussian collapses about a stationary regular polygon configuration ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ we start by recalling the single collapse situation described in @xcite . we let @xmath279 be the stationary regular @xmath51-polygon configuration , with circulation @xmath280 in the @xmath51 vertices of the polygon and circulation @xmath347 in its center ( see [ section : vortexpoints ] ) . note that this implies @xmath16 . since the angular velocity is @xmath348 , it follows that @xmath182 is a solution of the linear schrdinger equation . the evolution of a real gaussian @xmath349 by the linear schrdinger equation can be computed explicitly @xmath350 therefore we are insured that the non - vanishing initial condition @xmath351 yields a solution @xmath287 for the system , @xmath352 with @xmath353 , such that all filament collide together at time @xmath354 at @xmath355 . figure [ gaussian - collapse ] displays such initial perturbations for @xmath140 around the stationary configuration @xmath356 with center at @xmath357 . as a new observation , we can exhibit a similar scenario of collision for special perturbations of the previous data , namely for all initial data of type @xmath359 with @xmath346 small enough in @xmath338 and such that @xmath360 this is due to the fact that @xmath361 is then a solution of the system . in view of the dispersion inequality and the explicit formula for a linear schrdinger solution , we have @xmath362 therefore we are insured that at initial time the filaments @xmath287 are separated and they collide at time @xmath354 at @xmath341 . [ [ self - similar - collapses - about - a - stationary - regular - polygon - configuration ] ] self - similar collapses about a stationary regular polygon configuration ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ again we consider @xmath279 to be the stationary regular polygon configuration , with circulations @xmath280 in the vertices of the polygon and circulation @xmath347 in its center . this time we take advantage of the explicit self - similar solution of the linear schrdinger equation exhibited by cazenave and weissler @xcite . the initial data considered there is @xmath363 with @xmath340 . by combining ( 3.2 ) , proposition 3.3 . and corollary 3.4 . b ) of @xcite it follows that @xmath364 in particular we obtain the existence of two times @xmath365 such that @xmath366 therefore we consider as initial data the non - vanishing function @xmath367 it yields as a solution for @xmath368 which vanishes at time @xmath369 at @xmath341 . moreover , proposition 3.7 . in @xcite insures that @xmath370 belongs to some @xmath371 space , so we have indeed @xmath318 . the purpose of this paragraph is to investigate the behavior of vortex filaments around a self - similar collapsing solution to the three - vortex problem . we consider the one constructed in , e.g. , @xcite or @xcite . the initial configuration is @xmath372 the sufficient collapse conditions @xmath163 and @xmath149 are both satisfied . then if we denote by @xmath373 the corresponding center of vorticity , the configuration satisfies @xmath374 where @xmath375 the solution is explicitly given by @xmath376 as already mentioned , since the exact parallel solution @xmath204 collapses in finite time then any shifted perturbations of the form @xmath377 , for all @xmath378 , with @xmath379 , leads to collision at time @xmath380 ( see figure [ triangle - collapse ] ) . we next look for other kinds of less trivial perturbations around the triangle collapse . as we did before in section [ subsec : filaments - symmetric ] , we look for a filament solution under the form @xmath381 from now on we will assume that @xmath382 then the equation for @xmath182 is @xmath383 so finally @xmath384 in the case of an equilateral uniformly rotating triangle we would have @xmath385 real and constant and we would retrieve eq . . we rewrite as @xmath398 and observe that the discriminant in the numerator is negative for @xmath399 and positive for @xmath400 . we are led to the study of different cases which exhibit different kinds of motion . from now on @xmath401 will denote a positive constant depending only on @xmath380 and @xmath77 , which can possibly change from one line to another . we have @xmath402 . it follows that @xmath403 and @xmath404 after finding @xmath405 in , and taking into account we obtain that the filament solution constructed this way satisfies : @xmath406 if @xmath407 , this means that a blow - up of the filament solution at each level @xmath24 , without collapse , takes place at time @xmath408 ( strictly before the occurrence of the three - vortex collapse if @xmath409 and exactly at the same time if @xmath410 ) . if @xmath411 , no blow - up nor collapse occurs for the filaments at the time @xmath412 of the three - vortex collapse . actually the modulus of the filaments remain regular up to @xmath412 and only the angular velocity becomes singular at time @xmath380 . we have @xmath420 . then @xmath417 but @xmath421 ; therefore , @xmath422 is strictly contained in @xmath423 . we conclude that the filament solution evolves toward the following self - similar collision in finite time : it exists up to time @xmath424 , collapses at each level @xmath24 at time @xmath425 and satisfies @xmath426 in particular the collision between the filaments takes place before the collision between the point vortices . in the case of a uniformly rotating polygon ( see [ subsec : filaments - symmetric ] ) we have @xmath428 is constant in time , real , so that the analog of is @xmath429 thus the corresponding filament solution is given by @xmath430 for some @xmath431 and @xmath432 . it blows up at time @xmath412 but does not collide . f. bthuel , p. gravejat and j .- c . saut , _ existence and properties of traveling waves for the gross - pitaevskii equation _ , in `` stationary and time dependent gross - pitaevskii equations '' , a. farina and j .- c . saut editors , contem . math . * 473 * 55104 , american mathematical society , providence , r.i . ,

in the course of the last few years exclusive photo- and electroproduction of neutral vector mesons ( @xmath5 , @xmath6 , @xmath1 , @xmath7 ) at intermediate and high energies has attained increased theoretical interest , which has been stimulated by corresponding experimental efforts at cebaf @xcite and hera @xcite . the aim of the hera experiments is a more fundamental understanding of pomeron phenomenology in terms of qcd . vector - meson dominance in combination with pomeron exchange proves to be very successful in describing diffractive photo- and electroproduction of neutral vector mesons , provided the energy is the only large scale in the process @xcite . in their approach donnachie and landshoff @xcite assume that the pomeron couples directly to on - shell quarks in the vector meson and the proton , respectively . a qcd - inspired version of pomeron exchange , in which the pomeron is replaced by two non- perturbative ( abelian ) gluons , goes back to the same authors @xcite and is usually termed a soft pomeron . in the limit @xmath8 the soft - pomeron exchange becomes very similar to the usual pomeron exchange . this can be used to interpret the form factor at the quark - pomeron vertex @xcite . however , if in addition to the energy another scale , like the photon virtuality or the mass of the vector meson , becomes large perturbative qcd effects tend to enter the game . for large photon virtuality ( and small values of mandelstam @xmath3 ) perturbative qcd becomes applicable due to a factorization theorem which has been proved in ref . the three building blocks of the factorization formula are the hard photon - parton scattering amplitude , the distribution amplitude of the vector meson and finally a skewed gluon distribution of the proton . as a link between hard inclusive and exclusive reactions which generalizes the ordinary parton distributions , skewed parton distributions have recently attracted a great amount of interest ( see , e.g. , ref . @xcite and references therein ) . electroproduction with highly virtual photons thus offers a possibility to study such skewed parton distributions . within this perturbative formalism for vector - meson electroproduction with highly virtual photons the two - gluon aspect of the pomeron emerges automatically in a form which is related to the skewed gluon distribution in the proton . one of the aims of photoproduction experiments at cebaf is , on the other hand , to understand the transition from non - perturbative to perturbative production mechanisms . in the case of photoproduction the transverse momentum transfer @xmath9 has to be large in order that perturbative qcd becomes applicable . the pertinent factorization formula , which in general holds for arbitrary exclusive hadronic processes at large transverse momenta , is the result of a leading - order perturbative analysis . it represents a hadronic amplitude as a convolution integral of a hard - scattering amplitude with distribution amplitudes ( das ) @xcite . the process - dependent hard - scattering amplitude is perturbatively calculable and represents the scattering of hadronic constituents in collinear approximation . the process- independent das contain the non - perturbative bound - state dynamics of the hadronic constituents . there is no doubt that this hard - scattering approach ( hsa ) gives the correct description of exclusive hadronic processes in the limit of asymptotically large momentum transfer . but there is increasing evidence that the pure perturbative contribution is not the dominant one at experimentally accessible momentum transfers . a reasonable reproduction of the existing exclusive large-@xmath3 data is only achieved with very endpoint- concentrated hadron das , like those proposed by chernyak and zhitnitsky on the basis of qcd sum rules @xcite . such das , however , occur to be problematic for a perturbative calculation since they enhance contributions from the soft end - point regions in the convolution integral @xcite . some improvement with regard to the consistency of the perturbative calculation can be achieved by applying a modified factorization scheme in which transverse - momentum effects and sudakov suppressions are included . but this leads to a substantial reduction of the perturbative contribution @xcite . for the pion an end - point concentrated da meanwhile seems even to be ruled out by the recent cleo data @xcite for the @xmath10-transition form factor , since it provides a much too large perturbative result . on the other hand , it has been shown recently for electromagnetic nucleon form factors @xcite and for wide - angle compton scattering @xcite off protons that the existing data can be saturated by soft overlap contributions . furthermore , these contributions can be modeled by a nucleon light- cone wave function which gives rise to a distribution amplitude close to the asymptotic one . this , however , implies that the perturbative contribution becomes small . the approach which we are going to apply for hard exclusive photoproduction of @xmath1 mesons is still based on the hsa but includes also a certain amount of soft physics which is modeled by means of diquarks . diquarks may be considered as an effective way to account for quark - quark correlations in baryon wave functions . part of the soft overlap contributions are then absorbed into phenomenologically parameterized diquark form factors which occur at the gauge - boson diquark vertices . this hsa - based diquark model has already been applied successfully to other photon - induced hadronic reactions like magnetic and electric baryon form factors in the space- @xcite and time - like region @xcite , real and virtual compton scattering @xcite , two - photon annihilation into baryon - antibaryon @xcite , and photoproduction of @xmath11-mesons @xcite . further applications of the diquark model include the charmonium decay @xmath12 @xcite and the calculation of landshoff contributions in elastic proton - proton scattering @xcite . the purpose of the present investigation is to extend this foregoing work to @xmath13 . the reason why we concentrate on the @xmath1 channel is that it is cleaner and simpler than photoproduction of @xmath5 or @xmath6 mesons . within the hsa the valence fock state of the @xmath1 , i.e. the @xmath14-@xmath15 state , can only be produced via the exchange of two gluons between the @xmath1 and the proton . quark exchange could only happen via the knockout of an @xmath14-@xmath15 pair from the strange - quark sea in the nucleon which is , however , power suppressed within the hsa . in the following section we start with a short outline of the diquark model , show how feynman diagrams are properly grouped according to their propagator singularities ( sec . [ sec : elamp ] ) , sketch how we deal with the propagator singularities numerically ( sec . [ sec : propnum ] ) , and present our treatment of hadron - mass effects ( sec . [ sec : mass ] ) . section [ sec : helamp ] summarizes how photoproduction of vector mesons is described in terms of helicity amplitudes . furthermore , the fixed - angle scaling behavior of these helicity amplitudes is discussed and related to the treatment of mass effects . analytical expressions for the hard part of the hadron - helicity - conserving amplitudes are also given in this section . our predictions for photoproduction observables along with predictions from pomeron and soft - pomeron exchange are presented in sec . [ sec : result ] . at the end of this section we comment on the prospects of extending our calculation to photoproduction of @xmath7 mesons . finally , a summary and concluding remarks can be found in sec . [ sec : sum ] . as already mentioned before , within the hsa an exclusive scattering amplitude at large momentum transfer is expressed as a convolution integral of a hard - scattering amplitude with hadron distribution amplitudes @xcite . for the photoproduction reaction @xmath16 , which we are interested in , this integral takes on the particular form @xmath17 the distribution amplitudes @xmath18 are probability amplitudes for finding the valence fock state in the hadron @xmath19 with the constituents carrying certain fractions of the momentum of their parent hadron and being collinear up to a factorization scale @xmath20 . in our model the valence fock state of an ordinary baryon is assumed to consist of a quark ( @xmath21 ) and a diquark ( @xmath22 ) . we fix our notation in such a way that the momentum fraction appearing in the argument of @xmath23 is carried by the quark with the momentum fraction of the other constituent ( either diquark or antiquark ) it sums up to 1 ( cf . fig . [ kinem ] ) . for our actual calculations the ( logarithmic ) @xmath20 dependence of the das is neglected throughout since it is of minor importance in the restricted energy and momentum - transfer range we are interested in . the hard scattering amplitude @xmath24 is calculated perturbatively . intrinsic transverse momenta of the constituents are thereby neglected , i.e. , one assumes that the momenta of the constituents are collinear to those of their parent hadron . @xmath24 consists in our particular case of all possible tree diagrams contributing to the elementary scattering process @xmath25 . eight of the , altogether , sixteen diagrams which contribute to photoproduction of ( heavy ) quarkonia are depicted in fig . the subscript @xmath26 of @xmath27 represents a set of possible photon , proton and vector - meson helicities . for our purposes it appears to be more convenient to express the analytical results in terms of massless mandelstam variables @xmath28 , @xmath29 , and @xmath30 than in terms of the usual ( massive ) ones . a detailed account of our treatment of hadron - mass effects is given in sec . [ sec : mass ] below . the diquark model , as applied in refs . @xcite-@xcite , comprises scalar ( @xmath31 ) as well as axial - vector ( @xmath32 ) diquarks . the dynamics of @xmath31 and @xmath32 diquarks is determined by their coupling to photons and gluons . the expressions for the photon - diquark and gluon - diquark vertices correspond ( almost ) to the most general form for the ( parity and time - reversal invariant ) coupling of a spin-1 gauge boson to a spin-0 or a spin-1 particle , respectively @xcite . thereby @xmath32 diquarks are allowed to possess an anomalous ( chromo)magnetic moment @xmath33 . in applications of the model feynman graphs are calculated first with the corresponding feynman rules for point - like diquarks . as a second step , feynman graphs with @xmath34 gauge bosons coupling to the diquark are combined in a group which we call n - point contribution . finally the composite nature of diquarks is taken into account by multiplying each n - point contribution with phenomenological vertex functions , the diquark form factors . the particular choice @xmath35 for 3-point contributions and @xmath36 for n - point contributions ( @xmath37 ) ensures that in the limit @xmath38 the scaling behavior of the diquark model goes over in that of the pure quark hsp . the factor @xmath39 ( @xmath40 for @xmath41 ) provides the correct powers of @xmath42 for asymptotically large @xmath43 . for the running coupling @xmath44 the one - loop result @xmath45 is used with @xmath46 . in addition @xmath47 is restricted to be smaller than @xmath48 . @xmath49 and @xmath50 are strength parameters which allow for the possibility of diquark excitation and break - up in intermediate states where diquarks can be far off - shell . having sketched how diquarks are treated perturbatively , a few words about the choice of the das ( which incorporate the bound - state dynamics of the hadronic constituents ) and the spin - flavor wave functions of the proton and the @xmath1 are in order . if one assumes zero relative orbital angular momentum between quark and diquark and takes advantage of the collinear configuration ( @xmath51 and @xmath52 ) the valence fock - state wave function of an octet baryon b may be written as @xmath53 the two terms in eq . ( [ wfcov ] ) represent states consisting of a quark either a scalar or a vector diquark . the pleasant feature of the covariant wave - function representation , eq . ( [ wfcov ] ) , is that it contains , besides @xmath54 and @xmath55 ( the lorentz index of the vector - diquark polarization vector ) , only baryonic quantities ( momentum @xmath56 , helicity @xmath57 , baryon mass @xmath58 ) . assuming an su(6)-like spin - flavor dependence , the flavor functions @xmath59 for a proton take on the form @xmath60 } \ , , \;\;\ ; \chi_v^p = \phantom{- } [ u v_{\{u , d\ } } -\sqrt{2 } d v_{\{u , u\ } } ] / \sqrt{3 } \ , . \label{flavp}\ ] ] in eq . ( [ wfcov ] ) the color part of the quark - diquark wave function has been neglected . since the quark - diquark state has to be a color singlet the diquark has to be in a color antitriplet state ( like an antiquark ) thus the quark - diquark color wave function is simply @xmath61 . like the baryon wave function , eq . ( [ wfcov ] ) , also the @xmath62-@xmath63 wave function of a vector meson @xmath64 can be represented in a covariant way @xmath65 with @xmath66 denoting the polarization vector of the vector meson . the flavor function of the @xmath1 is @xmath67 . in refs . @xcite-@xcite a quark - diquark da of the form ( @xmath68 for @xmath31 diquarks ) @xmath69 \ , , \label{dap}\ ] ] turned out to be quite appropriate for octet baryons @xmath70 . the origin of the phenomenological baryon da , eq . ( [ dap ] ) , is a non - relativistic harmonic - oscillator wave function ( cf . therefore the masses appearing in the exponential have to be considered as constituent masses ( 330 mev for light quarks , 580 mev for light diquarks , strange quarks are 150 mev heavier than light quarks ) . the oscillator parameter @xmath71 gev@xmath72 is chosen in such a way that the full wave function gives rise to a value of @xmath73 mev for the mean intrinsic transverse momentum of quarks inside a nucleon . the exponential does not only provide a flavor dependence , but also suppresses the end - point regions @xmath74 in the convolution integral , eq . ( [ convol ] ) . one thus avoids to pick up sizable contributions from the soft end - point regions and results become less dependent on details of the baryon da . in the actual data fitting the exponential plays only a minor role . therefore , the constituent masses and the oscillator parameter have not been considered as free parameters but have been fixed in advance . we stress that the constituent masses do not appear in the hard scattering amplitudes ( cf . [ sec : mass ] below ) . a model for the @xmath1 meson da which fulfills qcd sum - rule constraints has been proposed by benayoun and chernyak @xcite . since we also want to investigate to which extent our calculations for @xmath1 production may be generalized to @xmath7 production we modify this da by attaching an exponential factor which provides a flavor dependence . the effect of this factor is negligible for @xmath1s , but it becomes crucial for @xmath7s . the form of the exponential factor is inspired by a simple relativistic treatment of heavy mesons @xcite and provides a flavor dependence in accordance with heavy - quark effective theory . for longitudinally polarized @xmath1s and @xmath7s the da reads ( @xmath75 ) @xmath76\ , . \label{dabc}\ ] ] the da for transversally polarized @xmath1s and @xmath7s contains an additional polynomial in @xmath77 which enhances the maximum at @xmath78 and makes it narrower ( cf . @xcite ) @xmath79 \nonumber \\ & & \times ( 5225.2\ , z_1 ^ 6 ( 1 - z_1)^6 + 0.39\ , z_1 ( 1 - z_1 ) ) \ , . \label{dabcl}\end{aligned}\ ] ] for the oscillator parameter @xmath80 we take a value of @xmath81 gev@xmath82 in accordance with estimates for the radius of the @xmath7 meson @xcite . the usual normalization condition , @xmath83 , fixes the constants @xmath84 and @xmath85 in eqs . ( [ dap ] ) , ( [ dabc ] ) , and ( [ dabcl ] ) . the quantity @xmath86 showing up in eq . ( [ wfcovv ] ) is related to the experimentally determinable decay constant of the meson @xmath64 . in case of the @xmath1 we get from the radiative decay width of the @xmath87 decay @xmath88 mev . the analogous constants @xmath89 and @xmath90 for the @xmath21-@xmath22 das of baryon @xmath70 are free parameters of the diquark model . they are , in principle , determined by the probability to find the @xmath21-@xmath22 state ( @xmath91 ) in the baryon @xmath70 and by the transverse - momentum dependence of the corresponding wave function . our present study is performed with the set of diquark - model parameters found in ref . @xcite by fitting elastic electron - nucleon scattering data . the numerical values of the parameters are : @xmath92 for further details of the diquark model we also refer to the paper of jakob et al . @xcite . the introduction of diquark form factors already requires a decomposition of the hard - scattering amplitude @xmath93 into 3- and 4-point contributions @xmath94 } } \widehat{t}_{\{\lambda\}}^{(s)}(x_1,y_1,z_1;\hat{s},\hat{t } ) \nonumber \\ & + & e_s \widehat{t}_{\{\lambda\}}^{(q , v)}(x_1,y_1,z_1;\hat{s},\hat{t } ) \ , . \label{tn}\end{aligned}\ ] ] @xmath95 and @xmath96}}$ ] are the electric charges of the quark @xmath21 and the s diquark @xmath97}$ ] in units of @xmath98 , respectively . the superscripts indicate whether the photon couples to the heavy quark @xmath62 in the meson , or the light quark @xmath21 in the proton with the diquark @xmath99 acting as spectator ( 3-point contributions ) , or whether the photon couples to the @xmath31 or @xmath32 diquark in the proton ( 4-point contributions ) . note that the only @xmath32- diquark contribution is @xmath100 . the reason is that the charge - flavor factor in front of @xmath101 vanishes and @xmath102 represents already a mass correction of second order , which is neglected in our calculation . for the numerical treatment of the convolution integral , eq . ( [ convol ] ) , it is advantageous to further split the @xmath103-point contributions into two parts which differ by their propagator singularities : @xmath104 f_d^{(3)}(-x_2 y_2 \hat{t } ) \ , , \nonumber \\ \widehat{t}_{\{\lambda\}}^{(q , d)}(x_1,y_1,z_1;\hat{s},\hat{t } ) & = & \left [ \frac{f_{\{\lambda\}}^{(q , d)}(x_1,y_1,z_1;\hat{s},\hat{t } ) } { ( \hat{g}_1 ^ 2 + i \epsilon ) ( \hat{q}_2 ^ 2 + i \epsilon^\prime ) } + \frac{g_{\{\lambda\}}^{(q , d)}(x_1,y_1,z_1;\hat{s},\hat{t } ) } { ( \hat{g}_1 ^ 2 + i \epsilon ) ( \hat{q}_3 ^ 2 + i \epsilon^\prime ) } \right ] f_d^{(3)}(-x_2 y_2 \hat{t } ) \ , , \nonumber \\ \widehat{t}_{\{\lambda\}}^{(d)}(x_1,y_1,z_1;\hat{s},\hat{t } ) & = & \left [ \frac{f_{\{\lambda\}}^{(d)}(x_1,y_1,z_1;\hat{s},\hat{t } ) } { ( \hat{g}_2 ^ 2 + i \epsilon ) ( \hat{q}_5 ^ 2 + i \epsilon^\prime ) } + \frac{g_{\{\lambda\}}^{(d)}(x_1,y_1,z_1;\hat{s},\hat{t } ) } { ( \hat{g}_2 ^ 2 + i \epsilon ) ( \hat{q}_4 ^ 2 + i \epsilon^\prime ) } \right ] f_d^{(4)}(-x_2 y_2 \hat{t } ) \ , . \label{tfg}\end{aligned}\ ] ] the functions @xmath105 and @xmath106 correspond to gauge - invariant subgroups of the 16 feynman diagrams which contribute to the hard - scattering amplitude @xmath93 ( cf . [ feyn ] ) . the diagrams entering the @xmath106s are related to those entering the @xmath105s by interchange of the two gluons . the @xmath107 and @xmath108 denote just those quark and gluon propagators which can go on- shell when integrating over @xmath54 , @xmath109 , and @xmath77 . explicitly , the propagator denominators read : @xmath110 as already indicated in eq . ( [ tfg ] ) propagator singularities are treated by means of the usual @xmath111 prescription . the resulting imaginary part in the amplitudes is considered as a non - trivial prediction of perturbative qcd which is unaffected by long - distance effects @xcite . as we will see below , it gives rise to non - vanishing predictions for polarization observables which require the flip of a hadronic helicity . for analytical expressions of the functions @xmath105 and @xmath106 and their symmetry properties we refer to sec . [ sec : helamp ] . the numerical difficulties in performing the convolution integral , eq . ( [ convol ] ) , are mainly caused by the occurrence of propagator singularities in the range of the @xmath112-integration . these give rise to principal value integrals @xmath113 as explained in some detail in ref . @xcite such integrals can still be treated by means of a rather fast and stable fixed - point gaussian quadrature after carefully separating the principal value parts and doing the corresponding integrations analytically . here we only want to give a short account of our integration procedure . we first perform the @xmath54-integration for which the various contributions to the hard scattering amplitude ( cf . ( [ tfg ] ) ) lead to integrals of the general form @xmath114 in order to simplify notations we have neglected helicity labels and the dependence on the mandelstam variables @xmath28 and @xmath29 . furthermore , the distribution amplitudes have been absorbed into the function @xmath115 . by applying a partial fractioning , exploiting eq . ( [ pv ] ) , and separating the principal value part eq . ( [ convol2a ] ) may be rewritten to give @xmath116 @xmath117 and @xmath118 are the zeroes of the propagator denominators @xmath119 and @xmath120 , respectively , and @xmath121 stands for either @xmath122 or @xmath123 , depending for which of the two the limit @xmath124 has been taken . as long as the two zeroes @xmath125 and @xmath126 ( which still depend on @xmath109 and @xmath127 ) do not coincide in the @xmath128-integration region this is the case for the contributions of @xmath129 , @xmath130 , @xmath131 , and @xmath132 the @xmath112-integration can be performed very efficiently with the help of eq . ( [ convol3a ] ) . after analytic integration of the two principal - value integrals in eq . ( [ convol3a ] ) ( which is a simple matter ) all the integrands of the remaining integrations are regular functions of the integration variables and can be treated by fixed - point gaussian quadrature . for @xmath133 and @xmath134 , however , the situation becomes a little bit more complicated . after having performed the @xmath54-integration in eq . ( [ convol3a ] ) one observes that the two propagator singularities @xmath125 and @xmath126 coincide along a trajectory in the @xmath128-integration region . this leads to a quadratic zero @xmath135 of @xmath136 . a closer inspection of eq . ( [ convol3a ] ) reveals that this quadratic zero in the denominator of @xmath137 does not cause any further problem when performing the @xmath128-integration for the real part of @xmath137 . the reason is that it is completely compensated by a corresponding quadratic zero in the numerator of @xmath138 . in @xmath139 , however , the zero of @xmath140 is only partly compensated by the numerator and one still encounters a single pole . at the pole @xmath139 becomes proportional to @xmath141 = \{1/(y_1- y_1^{(k_1,k_2 ) } + i \hat{\epsilon } ) + 1/(y_1-y_1^{(k_1,k_2)}- i \hat{\epsilon})\}/2 $ ] . as a consequence @xmath142 becomes a pure principal - value integral ( cf . ( [ pv ] ) ) @xmath143 which can be treated numerically analogous to the principal - value integral in @xmath54 . one only has to separate the principal - value part and do the corresponding integration analytically . for this purpose the analytical expression of the residue of @xmath144 at the position of the pole @xmath145 is needed . the remaining integrations are then again amenable to gaussian quadrature . thus it is possible , by carefully separating the singular contributions , exploiting @xmath146 functions , rewriting principal - value integrals as ordinary integrals plus analytically solvable principal - value integrals , to do all the numerical integrations by means of fixed - point gaussian quadrature . whereas the authors of ref . @xcite , who have used a different integration procedure , were only able to obtain reliable results for the imaginary part of the convolution integral , eq . ( [ convol ] ) , our numerical results for the imaginary * and * the real part occur to be very stable . going from an @xmath147 grid of @xmath148 points to a @xmath149 grid changes the results for the amplitudes ( over the whole angular range ) by less than 0.1% . our numerical calculations were performed on an alphaserver 1000 4/266 . for the larger grid size the evaluation of @xmath150 ( taking into account the 8 hadronic non- and single - flip amplitudes ) took less than 7 seconds per energy point and angle . our calculation of the hard - scattering amplitude involves an expansion in powers of @xmath151 which is performed at fixed scattering angle . we keep only the leading order and next - to - leading order terms in this expansion . hadron masses , however , are fully taken into account in flux and phase - space factors . the reasons why we want to include hadron - mass effects in the hard scattering amplitude are twofold . on the one hand , we want to apply our model already at momentum transfers of only a few gev , where hadron masses may still play a role . on the other hand , we also want to make predictions for spin observables which require the flip of hadronic helicities . within usual perturbative qcd such observables would vanish since a spin-1 gauge boson which couples to a nearly massless ( current ) quark can not flip the helicity of the quark . in the original diquark model @xcite violations of hadronic helicity conservation are generated by massive v- diquarks on the account of introducing a mass parameter for the v - diquark . in this paper we adopt another strategy which allows for a more consistent treatment of mass effects without introducing new mass parameters for the hadronic constituents . like the authors of ref . @xcite we assume for every hadronic constituent that its 4-momentum is proportional to the 4-momentum of its parent hadron . this requires to assign to every constituent of the hadron @xmath19 an effective mass @xmath152 , where @xmath153 is the fraction of the four- momentum of the hadron @xmath19 carried by the constituent . keeping in mind that the momentum fractions are weighted by the hadron das in the convolution integral , eq . ( [ convol ] ) , this means that quarks and diquarks acquire ( on the average ) a mass which is rather the mass of a constituent than a current particle . this may be interpreted in such a way that at intermediate momentum transfers a quark or diquark surrounded by a cloud of @xmath154 pairs and gluons is not resolved but rather acts like a single effective particle with a corresponding mass . this prescription for the constituent masses applies to the ( on - shell ) particles which correspond to the external legs of the feynman diagrams . it has already been tacitly exploited when casting the hadronic spin wave functions into covariant form ( cf . ( [ wfcov ] ) and ( [ wfcovv ] ) ) . taking the quark - diquark da of eq . ( [ dap ] ) the average values of the effective masses of quarks and diquarks in a proton become @xmath155 analogously the effective mass of the heavy ( anti)quark in the meson becomes on the average half of the meson mass . the choice of the masses which occur in the propagators of internal particles is not that obvious , since , e.g. , a quark in the incoming proton has mass @xmath156 whereas the same quark in the outgoing proton has mass @xmath157 . so what should we take for the mass of this quark between two vertices ? in order to explain our recipe we , e.g. , consider the momentum of the heavy quark between the two gluon vertices in the third diagram of fig . it is @xmath158 with @xmath159 , @xmath160 and @xmath161 representing the 4- momenta of the hadrons as indicated in fig . [ kinem ] . the quark mass in the propagator related to @xmath162 is now obtained by replacing the 4-momenta in eq . ( [ propmom ] ) by the masses of the corresponding particles , i.e. @xmath163 taking again into account that the momentum fractions are weighted by the hadron das , eq . ( [ dap ] ) and eq . ( [ dabc ] ) , the average heavy quark mass in the propagator becomes @xmath164 . extending this recipe to all the quark and diquark propagators and using the covariant wave - function representations , eqs . ( [ wfcov ] ) and ( [ wfcovv ] ) , for the external particles , leads to analytical expressions for the feynman diagrams which contain ( apart of the momentum fractions @xmath165 , @xmath166 , and @xmath167 ) only hadronic momenta , masses , spinors , polarization vectors , and helicities . the occurring traces and spinor products can be evaluated in the usual way . the final step is then to reexpress the ( massive ) mandelstam variables @xmath14 , @xmath3 , and @xmath168 in terms of massless ones @xmath169 and to make an expansion in terms of the ( small ) parameters @xmath151 , keeping the scattering angle @xmath170 fixed . in order to facilitate the treatment of propagator singularities we retain only the leading and next - to - leading order term in this expansion . taking again as an example the propagator related to the momentum @xmath162 ( cf . ( [ propmom ] ) ) the propagator denominator becomes @xmath171 hence the quark propagator is given by @xmath172 i.e. , by neglecting mass corrections of @xmath173 we end up with propagator denominators which still appear like those of massless particles ( cf . ( [ prop ] ) ) . apart from the simplifications in the treatment of propagator singularities and in the analytical expressions for the hard part of the amplitudes our approximate treatment of mass effects has a few other nice features . by including mass corrections of @xmath174 we obtain non - vanishing results for hadron - helicity - flip amplitudes and thus for polarization observables which require the flip of a hadronic helicity ( see the discussion in the next section ) . like the leading - order hadron - helicity - conserving amplitudes also the hadron - helicity - flip amplitudes are @xmath175 gauge invariant with respect to the photon and @xmath176 gauge invariant with respect to the gluon . finally , crossing relations for the hadronic helicity amplitudes are fulfilled up to @xmath174 , as has been shown for compton scattering off baryons and its crossed process @xmath177 @xcite . for exclusive photoproduction of vector mesons @xmath178 one finds , altogether , 24 complex helicity amplitudes . by virtue of parity invariance only 12 of these helicity amplitudes are independent . following the notation of ref . @xcite we denote them by @xmath179 with @xmath180 . for our normalization of the helicity amplitudes the unpolarized differential cross section takes on the form @xmath181 as examples for single spin observables , for which our model provides nontrivial predictions , we will consider the beam and target asymmetry . the corresponding expressions are @xcite @xmath182 for the photon and @xmath183 for the proton , respectively . within the hard scattering approach the energy dependence of the helicity amplitudes at fixed cm angle and large @xmath14 is roughly @xmath184 depending on whether the helicity of the hadronic constituents is conserved or flipped by one or two units . this scaling behavior is modified by logarithms due to the running coupling @xmath44 and , eventually , the evolution of the hadron distribution amplitudes . further deviations are due to the diquark form factors . for small momentum transfers diquarks appear nearly point- like and the decay behavior of the helicity amplitudes is weakened by one power of @xmath14 . the energy dependence in eqs . ( [ scaling ] ) is only approached at large enough momentum transfer , where the diquark form factors become fully operational . for the helicity amplitudes our treatment of mass effects ( cf . [ sec : mass ] ) leads to an expansion in powers of @xmath185 . as compared to the hadron - helicity - conserving amplitudes @xmath186 and @xmath187 , the leading - order term of the single - flip amplitudes @xmath188 , @xmath189 , @xmath190 , and @xmath191 is suppressed by @xmath174 . the leading - order term of the double - flip amplitudes @xmath192 and @xmath193 is even suppressed by @xmath185 and thus has the same order as the first ( non - vanishing ) mass - correction term of the hadron - helicity - conserving amplitudes . to simplify matters we only consider non- and single - flip amplitudes to leading - order and neglect double - flip amplitudes at all . in the following we quote the analytical results for the functions @xmath105 and @xmath106 which have been obtained with the help of the symbolic computer program mathematica , in particular with the two program packages ` feynarts`@xcite and ` feyncalc`@xcite . in case of the hadron - helicity conserving amplitude @xmath187 , i.e. for the helicity combination @xmath194 , @xmath195 , @xmath196 , @xmath197 , the functions @xmath105 read : [ fs1 ] @xmath198 \nonumber \\ & & \left . + x_1 y_2 \hat{t}^2 \left [ 4 z_1 ^ 2 ( y_1 - z_1 ) + y_1 ( x_2 + z_2 ) + z_1 ( 2 x_1 z_1 - 3 y_1 ) \right ] \right\}\ , , \\ % f_{\left\ { 0 , - \frac{1 } { 2 } ; + 1 , - \frac{1 } { 2 } \right\}}^{(q , s ) } & = & - \frac{6 m_p^2 } { \kappa \hat{t } } f_{\left\ { 0 , - \frac{1 } { 2 } ; + 1 , - \frac{1 } { 2 } \right\}}^{(q , v ) } = - 4 v \frac{1 } { x_2 y_1 \hat{t } } \left\ { x_2 z_1 \hat{u } + y_2 z_2 \hat{s } \right\}\ , , \\ % f_{\left\ { 0 , - \frac{1 } { 2 } ; + 1 , - \frac{1 } { 2 } \right\}}^{(s ) } & = & - 4 v \frac{1 } { x_1 y_1 \hat{t } } \left\ { x_1 z_1 \hat{u } + y_1 z_2 \hat{s } \right\}\ , , \end{aligned}\ ] ] with @xmath199 where @xmath200 denotes the color factor and @xmath55 the fine - structure constant , respectively . the functions @xmath106 are related to the @xmath105s in a simple way : [ gs1 ] @xmath201 vector 4-point contributions @xmath202 and @xmath203 are , in general ( for all helicity combinations ) , suppressed by @xmath173 or even stronger . relations analogous to also valid between the @xmath105s and @xmath106s entering the other @xmath204 amplitudes @xmath188 , @xmath186 , and @xmath189 . the functions @xmath105 contributing to @xmath186 the other hadron - helicity conserving amplitude ( helicity combination @xmath205 , @xmath206 , @xmath196 , @xmath207 ) are obtained from those entering @xmath187 by interchange of the mandelstam variables @xmath208 and the momentum fractions @xmath209 : [ fs2 ] @xmath210 the analytical expressions for the hadron - helicity - flip amplitudes @xmath188 , @xmath189 , @xmath190 , and @xmath191 are of similar length as the expressions quoted above and can be obtained from the authors on request . like in eqs . ( [ fs1 ] ) the vector 3-point contributions to @xmath190 and @xmath211 are proportional to the corresponding scalar ones . it is also interesting to observe that due to the ( anti)symmetry properties of the feynman graphs under interchange of the momentum fractions @xmath77 and @xmath212 ( cf . ( [ gs1 ] ) ) , only diagrams in which the photon couples to the @xmath14 quark contribute to the @xmath204 amplitudes . actually we have computed all the functions @xmath105 and @xmath106 , and relations like ( [ gs1 ] ) and ( [ fs2 ] ) served as a check of our analytical expressions . furthermore , we have verified @xmath175 gauge invariance with respect to the photon and @xmath176 gauge invariance with respect to the gluon . the proof of gauge invariance is facilitated by the observation that not only the sum of all 16 tree diagrams provides a gauge invariant expression , but rather each of the functions @xmath105 and @xmath106 is gauge invariant by itself . finally , we have recalculated a few of the diagrams by hand to confirm the outcome of mathematica . fig . [ result ] shows the diquark model predictions ( solid line ) for @xmath213 $ ] at @xmath214 gev together with the outcome of the pomeron- exchange mechanism ( dotted line ) @xcite and results of a two - gluon - exchange model ( dash - dotted and dashed line ) @xcite . a direct comparison of our predictions with experiment is not possible yet since data are only available at low momentum transfers ( @xmath215 gev@xmath216 ) where the perturbative photoproduction mechanism is certainly not the dominant one . the experimental situation , however , will hopefully improve as soon as the data analysis of the cebaf-93 - 031 experiment will be completed . nevertheless , fig . [ result ] shows that the diquark model provides a reasonable extrapolation of the low-@xmath3 data . thereby one should keep in mind that the cross section which decays exponentially in forward direction is expected to flatten around @xmath217 ( @xmath3 and @xmath168 large ) . the few available data for photoproduction of pseudoscalar mesons and also photoproduction of @xmath5 and @xmath6 mesons which extend to larger values of @xmath3 exhibit just such a behavior @xcite . as can be seen from fig . [ result ] the forward cross section for @xmath218 production is reasonably well reproduced by simple pomeron phenomenology ( dotted line ) . as a mechanism for photoproduction of vector mesons donnachie and landshoff @xcite proposed that the photon fluctuates into a quark - antiquark pair which subsequently recombines to a vector meson . the resulting quark loop is connected to a quark in the proton via the exchange of a single pomeron which couples to quarks like an isoscalar photon . the qcd - inspired version of the pomeron exchange of donnachie and landshoff @xcite , in which the pomeron is replaced by two non - perturbative ( abelian ) gluons , has been applied by laget and mendez - galain @xcite to photoproduction of @xmath1 mesons ( dash - dotted line ) . the latter authors consider the two - gluon - exchange model as a link which connects the diffractive with the hard - scattering domain . at @xmath219 gev@xmath216 the two - gluon - exchange cross section exhibits a characteristic node which can be understood as an interference effect between the two feynman diagrams which enter the photoproduction amplitude . however , this picture changes drastically if the two gluons are allowed to couple to different quarks in the proton . recently laget @xcite tried to estimate such contributions by modeling quark - quark correlations inside a proton via a simple wave function . he found that the node in the cross section is completely washed out and that the contributions in which the two gluons couple to different quarks inside the proton start to dominate the photoproduction of @xmath1 with increasing @xmath3 ( dashed line ) . actually , for @xmath220 gev@xmath216 the result of laget becomes comparable with the diquark - model result . this is not surprising since we know from the perturbative analysis that the hard - scattering mechanism , i.e. contributions from diagrams without loops in which all the hadronic constituents are connected via gluon exchange , should become dominant for large values of @xmath3 and @xmath168 . correlations between hadronic constituents are automatically accounted for within such an approach and hence also within our diquark model . to check whether our model provides a reasonable energy dependence we have also calculated the differential cross section at a much larger photon energy ( @xmath221 gev ) . the result is depicted in fig [ result2 ] ( solid line ) along with a two - parameter fit of @xmath222 gev@xmath216 data from zeus @xcite . the hatched area indicates the uncertainties in the forward cross section and the slope parameter . keeping again in mind that the cross section is expected to flatten with increasing @xmath3 the magnitude of our prediction occurs still to be within the range of an extrapolation of the low-@xmath3 data . the dashed line in fig . [ result2 ] represents a leading - order calculation in which mass effects are neglected . this means that only the two hadron - helicity conserving amplitudes @xmath186 and @xmath187 are taken into account in the leading - order cross section . a comparison between the full calculation ( solid line ) and the leading - order calculation ( dashed line ) reveals that mass effects are still sizable and amount to @xmath223-@xmath224 of the full differential cross section . they increase in forward and backward direction and are smallest around @xmath217 . at this point we want to comment on another attempt to calculate photoproduction of @xmath1 mesons within the diquark model . the authors of ref . @xcite use a different parameterization of the model . only the hadron - helicity - conserving amplitudes @xmath186 and @xmath187 are taken into account in their calculation . with their numerical method they encounter difficulties in calculating the real parts of these amplitudes . but they assert that the real parts are small for their parameterization and therefore neglect them . their results for differential cross sections are about one order of magnitude smaller than ours . this is plausible if one looks at their values of @xmath225 and @xmath226 , i.e. the normalization of the quark - diquark das , and the cutoff masses in the diquark form factors . both are considerably smaller than ours . in contrast to the authors of ref . @xcite we think that the present experimental situation in @xmath1 photoproduction does not allow to discriminate a particular parameterization of the diquark model . however , we want to point out that our parameterization has already been applied successfully to a variety of other exclusive processes @xcite . on the other hand , it is still unclear whether these processes can also be described satisfactorily with the parameterization of carimalo et al . their parameterization has been determined by means of the proton magnetic form factor . apart from @xmath1 photoproduction it has only been applied to charmonium decays into the @xmath227 final state @xcite with the result that the data could not be reproduced . in the kinematic region , where @xmath3 and @xmath168 are large , the differential cross section at fixed scattering angle is expected to exhibit a scaling behavior of approximately @xmath228 , which is supposed to indicate that the hard photoproduction mechanism becomes relevant . [ scalingfig ] shows @xmath229 for photon energies of @xmath230 , @xmath231 , and @xmath232 gev . 5 gev is presently the upper limit of cebaf , 10 gev will be reached by the planned upgrade of cebaf and 20 gev is a value which could be reached at future facilities like elfe or a further extension of cebaf which are presently still under discussion . comparing the diquark - model results for these energies one observes that one is still far away from the @xmath228 scaling behavior . the reasons have already been mentioned in sec . [ sec : helamp ] . a closer inspection shows that the cross section at @xmath233 scales rather like @xmath234 than @xmath228 for @xmath235 gev . starting with point - like diquarks and a constant @xmath236 one would expect that the leading - order cross section ( without mass effects ) should scale like @xmath237 . by including the diquark form factors and cross - section contributions from hadron - helicity - flip amplitudes , which are already suppressed by one power of @xmath14 as compared to the leading - order contributions , the scaling behavior changes already to nearly @xmath228 . finally , the approximate @xmath234 decay is caused by the fact that @xmath236 is taken as a running coupling constant ( with argument @xmath238 ) . [ effects ] shows how the differential cross section ( at @xmath214 gev ) is influenced by different contributions within our model . a comparison of the solid ( full calculation ) and the short - dashed ( mass effects neglected ) line reveals , e.g. , the influence of mass effects . for @xmath214 gev and @xmath239 they amount to @xmath240 of the full cross section . the dash - dotted line represents the pure s - diquark part of the cross section . it varies between @xmath241 and @xmath242 if one goes from backward to forward angles . since s- and v - diquark contributions add coherently ( and constructively , cf . ( [ fs1 ] ) ) this means that the corresponding amplitudes are of the same order of magnitude . the differential cross section for the photoproduction of longitudinally polarized @xmath1s is given by the long - dashed curve . it amounts to @xmath243 of the full cross section at @xmath239 . this fraction increases in backward direction and decreases in forward direction . finally , we have also examined the influence of the exponential in the @xmath1 da ( cf . ( [ dabc ] ) ) . neglecting this exponential one ends up with the dotted curve which is nearly indistinguishable from the full calculation . this also shows that we are not very sensitive to the @xmath244 endpoint regions when performing the convolution integral eq . ( [ convol ] ) . as examples for single - polarization observables the beam asymmetry @xmath245 and the target asymmetry @xmath246 are depicted in fig . [ asymmetry ] . the beam asymmetry would be already non - zero , if only the leading - order helicity amplitudes @xmath186 and @xmath187 were taken into account ( cf . ( [ gasy ] ) ) . it would not even be necessary that the helicity amplitudes acquire a non - trivial phase . on the other hand , a non - trivial target asymmetry demands for both , non - vanishing hadron - helicity - flip amplitudes and non - trivial phases of the amplitudes ( cf . ( [ pasy ] ) ) . as already mentioned at the end of sec . [ sec : elamp ] phases are generated through propagator singularities , hadron - helicity - flip amplitudes occur in our model as mass corrections . as it turns out within our model , also the beam asymmetry is considerably affected by the inclusion of the hadron - helicity - flip amplitudes . in a recent paper @xcite the influence of nucleonic resonance effects on polarization observables in @xmath1-meson photoproduction has been investigated within a constituent - quark model in which the diffractive contribution is produced by t - channel pomeron exchange . it is interesting to observe that the result of these authors for the target asymmetry at a rather small photon energy of @xmath247 gev is comparable in size to our result at @xmath214 gev . this is remarkable since two completely different concepts are at work : on one hand nucleonic resonance contributions , on the other hand a purely perturbative mechanism . also for the beam asymmetry some similarities can be found . the corresponding predictions of both approaches are again comparable around @xmath233 . a ( somewhat smaller ) negative beam asymmetry has also been found in another model which is based on pomeron exchange , pion exchange , and a direct knockout production from the strange - quark sea in the proton @xcite . it would be interesting to see experimentally , whether polarization observables like the target asymmetry @xmath248 , the recoil polarization @xmath249 , or the vector - meson polarization @xmath250 , which vanish in leading - order perturbative qcd , are still sizable at momentum transfers of a few gev . such observables would be a more sensitive tool than the unpolarized differential cross section to figure out whether the kind of nonperturbative ingredients ( diquarks , effective constituent masses ) which are taken into account in our perturbative approach , are already sufficient , or whether much more non - perturbative physics is still at work at a few gev of momentum transfer . at the end of this section we want to comment on the perspective to extend this calculation to photoproduction of @xmath7 . it is a simple matter , since we have already assumed a flavor dependent da for the vector meson ( cf . ( [ dabc ] ) ) . the only thing to do is to replace the @xmath1 mass by the @xmath7 mass and the decay constant @xmath251 by @xmath252 . one , however , has to keep in mind that we have applied an expansion in the parameter @xmath151 . in order that this parameter is approximately of the same size in @xmath1 and @xmath7 production the photon lab energy has to be increased approximately by a factor @xmath253 when going from @xmath1 to @xmath7 . a second criterion for the reliability of our approximate treatment of mass effects is that the coefficient of the first mass correction term should become small compared to the leading order . this coefficient is angular dependent and becomes large at small scattering angle . that means , if the size of the mass corrections relative to the leading order should be about the same in @xmath1 and @xmath7 production ( @xmath151 comparable ) , one has to consider much larger momentum transfers for @xmath7 than for @xmath1 production . a more detailed discussion of this point along with some numerical results can be found in ref . a possible way to apply the perturbative approach to @xmath7 production at a few gev of momentum transfer would be to retain our prescription for the choice of the effective masses , but to refrain from the final expansion in @xmath151 . this has not been attempted yet since the analytical expressions for the amplitudes become very lengthy and the treatment of propagator singularities in the convolution integral , eq . ( [ convol ] ) , becomes much more involved . we have investigated the reaction @xmath254 in the few - gev momentum transfer region where unpolarized differential cross - section data from the cebaf-93 - 031 experiment at jlab are expected to become available very soon . our theoretical approach is based on perturbative qcd supplemented by the assumption that baryons can be treated as quark - diquark systems . the same approach has already been successfully applied to other photon - induced hadronic reactions @xcite . the motivation for introducing diquarks is , above all , to extend the perturbative approach from large down to moderately large momentum transfers . there are mainly two reasons why this can be achieved with diquarks . first , the overall momentum transfer has to be shared between less hadronic constituents than in the pure quark hsa . hence the gluons which keep the hadronic constituents collinear are on the average harder and thus the running strong coupling constant becomes smaller . the second reason is that a certain class of non - perturbative effects , namely quark - quark correlations inside a baryon , are effectively accounted for by means of diquarks . the fact that a heavy quarkonium is produced in the final state entails a considerable reduction in computational effort . only a small fraction of the feynman diagrams which contribute in general to photoproduction of arbitrary mesons occurs in @xmath1 production . to be more specific , these are just those diagrams in which the photon fluctuates into the heavy quark - antiquark pair which is then connected to the quark and the diquark in the proton via the exchange of two gluons , respectively . both , the scalar and the vector - diquark component of the proton have been taken into account . we have also improved the diquark model in the sense that hadron - mass effects in the hard - scattering amplitude have been included in a systematic way . this has been accomplished by assigning to each hadronic constituent an effective mass @xmath255 ( @xmath256 longitudinal momentum fraction of the constituent ) and expanding the scattering amplitude in the small parameter @xmath151 . only the leading and next - to - leading order terms in this expansion have been kept . the leading - order term provides for the hadron - helicity - conserving amplitudes and agrees , of course , with the massless result . the next - to - leading order term represents a power- correction and gives rise to non - vanishing hadron - helicity - flip amplitudes . it enables us to make predictions for polarization observables which require the flip of a hadronic helicity like , e.g. , the target asymmetry or the recoil polarization . we want to point out that our treatment of mass effects still preserves gauge invariance with respect to the photon and the gluon . our numerical studies have been performed with the diquark - model parameters and the quark- diquark das proposed in ref . the da of the @xmath1 meson has been taken from the literature @xcite . it fulfills qcd sum - rule constraints . hence we have been able to make predictions without introducing new parameters . we have paid special attention to the correct and numerically robust treatment of propagator singularities . in the absence of large-@xmath3 data ( @xmath257 gev@xmath216 ) our results can only be compared with the available data which go up to @xmath258 gev@xmath216 . our predictions for the differential cross section at two rather different photon energies ( @xmath214 gev and @xmath259 gev ) which are typical for cebaf and hera , respectively , occur to be still within the range of an extrapolation of the existing low-@xmath3 data . it is worth mentioning that the predictions of a two - gluon - exchange model , in which the gluons are considered as a remnant of the pomeron @xcite , become comparable to our result for @xmath257 gev@xmath216 if the gluons are allowed to couple to different quarks in the proton . this is not surprising because contributions in which all the hadronic constituents are ( approximately ) kept collinear by means of gluon exchange to prevent the hadrons from breaking up become increasingly important as soon as one enters the hard - scattering domain . just those contributions are assumed to be the dominant ones within our hsa - based diquark model . we have also investigated within our model whether the ( asymptotic ) @xmath228 scaling behavior of the differential cross section at fixed angle is already realized for @xmath14 values between @xmath231 and @xmath260 gev . after dividing out the strong running coupling ( which occurs to the sixth power in the cross section ) an approximate @xmath228 scaling has been found . our investigation of hadron - mass effects has revealed that they are still non - negligible in the few- gev momentum transfer range and photon energies of the order of 10 gev . correspondingly , also polarization observables which require the flip of a hadronic helicity , like the target polarization , occur to be sizable . finally , we have considered the possibility to extend the present calculation to photoproduction of @xmath7 mesons . in order that our treatment of mass effects remains reliable for @xmath7s one has to go to much larger energies and momentum transfers where experimental data can not be expected . the situation with respect to the momentum transfer could perhaps be improved by refraining from the final expansion in @xmath151 . this , however , will considerably complicate the treatment of propagator singularities and make the analytical expressions for the amplitudes much more complicated . therefore it has not been attempted as yet . our diquark - model calculations for @xmath261 will meet the first severe test in the near future when the data analysis of the cebaf 93 - 031 experiment will be completed . with these data , which go up to @xmath3 values of about 5 gev@xmath216 , one will get a first glimpse how different production mechanisms ( pomeron exchange @xmath262 gluon exchange @xmath262 hard - scattering mechanism ) in high - energy photoproduction evolve with increasing momentum transfer . however , in order to really touch the hard - scattering domain much higher photon energies than those which can be reached at the present stage of cebaf would be desirable . in addition to unpolarized differential cross sections , polarization observables which require the flip of hadronic helicities , could serve as an indicator for the relative importance of hard and soft photoproduction mechanisms , since hard scattering is closely connected with hadronic helicity conservation . c. f. berger would like to thank the paul - urban - stipendienstiftung for supporting a visit at the institute of theoretical physics at the university of graz during which part of this paper has been completed . email address : carola.berger@sunysb.edu email address : wolfgang.schweiger@kfunigraz.ac.at g. audit _ et al . _ , cebaf proposal no . 93 - 031 ( 1993 ) . j. a. crittenden , _ exclusive production of neutral vector mesons at the electron - proton collider hera _ , springer tracts in modern physics * 140 * ( springer , berlin heidelberg , 1997 ) . a. donnachie and p. v. landshoff , phys . b * 348 * , 213 ( 1995 ) . a. donnachie and p. v. landshoff , phys . b * 185 * , 403 ( 1987 ) . a. donnachie and p. v. landshoff , nucl . b * 311 * , 509 ( 1989 ) . j. c. collins , l. frankfurt , and m. strikman , phys . d * 56 * , 2982 ( 1997 ) . m. diehl , t. feldmann , r. jakob , and p. kroll , eur . j. c * 8 * , 409 ( 1999 ) . for a comprehensive review of hard exclusive processes and further references see , e.g. , s. j. brodsky and g. p. lepage , in _ perturbative quantum chromodynamics _ , edited by a. h. mueller ( world scientific , singapore , 1989 ) . v. l. chernyak and a. r. zhitnitsky , phys 112 * , 173 ( 1984 ) . n. isgur and c. h. llewellyn smith , nucl . b * 137 * , 526 ( 1989 ) . et al . _ , c * 66 * , 267 ( 1995 ) . j. gronberg _ et al . _ , d * 57 * , 33 ( 1998 ) . j. bolz and p. kroll , z. phys . a * 356 * , 327 ( 1996 ) . a. v. radyushkin , phys . d * 58 * , 114008 ( 1998 ) . r. jakob , p. kroll , m. schrmann , and w. schweiger , z. phys . * a347 * , 109 ( 1993 ) . p. kroll , th . pilsner , m. schrmann , and w. schweiger , phys . b316 * , 546 ( 1993 ) . p. kroll , m. schrmann , and p. a. m. guichon , nucl . a * 598 * , 435 ( 1996 ) . c. f. berger , diploma thesis , technological university graz ( 1997 ) ; p. kroll , m. schrmann , k. passek , and w. schweiger , phys . d * 55 * , 4315 ( 1997 ) . r. jakob , phys . d * 50 * , 5647 ( 1994 ) . t. huang , nucl . [ proc . suppl . ] b * 7 * , 320 ( 1989 ) . m. benayoun and v. l. chernyak , nucl . b * 329 * , 285 ( 1990 ) . m. wirbel , b. stech , and m. bauer , z. phys . c * 29 * , 637 ( 1985 ) . t. feldmann and p. kroll , phys . b * 413 * , 410 ( 1997 ) . g. r. farrar , g. sterman , and h. zhang , phys . * 62 * , 2229 ( 1989 ) . c. carimalo , n. arteaga - romero , and s. ong , preprint hep - ph/9712485 ( 1997 ) . m. anselmino , f. caruso , and f. murgia , phys . d * 42 * , 3218 ( 1990 ) . m. pichowsky , . avkli , f. tabakin , phys . c * 53 * , 593 ( 1996 ) . j. kblbeck , m. bhm , and a. denner , comp . . commun . * 60 * , 165 ( 1990 ) . r. mertig , m. bhm , and a. denner , comp . * 64 * , 345 ( 1991 ) . laget and r. mendez - galain , nucl . a * 581 * , 397 ( 1995 ) . laget , preprint dapnia / sphn-98 - 54 ( 1998 ) . j. ballam et al . , phys . d * 5 * , 3150 ( 1973 ) . barber et al . , z. phys . c * 12 * , 1 ( 1983 ) . r. l. anderson et al . , phys . d * 14 * , 679 ( 1976 ) . m. derrik et al . , phys . b * 377 * 259 ( 1996 ) . c. carimalo and s. ong , z. phys . c * 52 * , 487 ( 1991 ) . q. zhao , j .- didelez , m. guidal , and b. saghai , preprint nucl - th/9906013 ( 1999 ) . a. i. titov , y. oh , s. n. yang , and t. morii , phys . c * 58 * , 2429 ( 1998 ) . c. f. berger and w. schweiger , preprint hep - ph/9909270 , to appear in proceedings of the int / jlab workshop on _ exclusive and semi - exclusive processes at high momentum transfer _ , newport news , usa , may 1999 .

the general goal of quantum information processing is to manipulate the information coded in a particular quantum system , while simultaneously trying to isolate it from undesired perturbations due to the external environment . for instance , in the fields of quantum computation @xcite and quantum simulation @xcite , the processing stage usually relies on a designed unitary evolution which , preserving the coherence of the system , provides a quantum state that contains the outcome of the computation , or the properties of the quantum phase of matter under study . therefore , the experimental progress in these fields has been typically associated to technological developments that minimize experimental imperfections and maximize environmental isolation @xcite . however , the coupling of a system to its surrounding environment is not necessarily a disadvantage . in fact , the dissipation of a quantum system , if judiciously exploited , can act as a resource for quantum information processing . this interesting change of paradigm started with the recognition that dissipation can assist the generation of entanglement between distant atoms , either in free space @xcite , or trapped inside cavities @xcite . by measuring the presence of a photon spontaneously emitted by the atoms @xcite , or its absence @xcite , it is possible to project an initially uncorrelated atomic state into a maximally entangled one , as has been recently experimentally demonstrated @xcite . these schemes can be improved further by tailoring the atom - cavity interaction @xcite or by modifying the measurement techniques @xcite , and can also be generalized to provide a route towards universal quantum computation @xcite . since the dissipation does not always render the desired result in these approaches , the measurement is required to select only the successful outcomes . therefore , the combination of dissipation and measurement can be considered as a probabilistic resource for quantum information . another promising approach is to assess whether , by designing the system - environment coupling in a particular way , the dissipation could provide the desired quantum state with certainty . this _ quantum reservoir engineering _ finds its roots in the theory of laser cooling of atoms @xcite and , more related to this work , of trapped ions @xcite . laser - cooling schemes try to design the system - environment coupling such that the dissipation yields a stationary state with reduced kinetic energy . these ideas can be taken a step further in order to non - probabilistically produce stationary states that display non - classical aspects , or a certain amount of entanglement @xcite . for instance , by mimicking superradiance phenomena @xcite , certain engineered decay channels yield partial entanglement in the stationary mixed state @xcite . however , it would be highly desirable to devise dissipative protocols that provide _ maximally - entangled pure states _ asymptotically , the so - called dark states , showing fidelities comparable to those obtained by more standard unitary schemes @xcite . this approach has been recently pursued by different groups which have shown that , provided that one can engineer a dissipation that acts quasi - locally on different parts of the system , it is possible to design dissipative protocols to produce a number of paradigmatic dark states @xcite , or to perform universal dissipative quantum computation @xcite . for instance , by engineering a particular dissipation that acts on pairs of adjacent atoms in an optical lattice , the system can be driven dissipatively into a superfluid phase @xcite . in addition , as initially proposed for rydberg atoms @xcite and first realized in experiments of trapped ions @xcite , by concatenating multi - qubit gates with a controlled dissipation of ancillary qubits , a variety of multipartite entangled states ( i.e. bell , greenberger - horne - zeilinger , and dicke states ) have been dissipatively generated with reasonably high fidelities . let us note that this stroboscopic time - evolution corresponds to a markovian master equation in the limit of many gates and dissipation steps , where the dissipative jump operators correspond to multi - qubit jump operators . from a fundamental point of view , it is of interest to address whether similar dissipative protocols can also work in an _ ( i ) analog _ and _ ( ii ) global _ fashion ( i.e. using always - on couplings without individual addressing ) . additionally , from a more pragmatic point of view , such global analog schemes would not be limited by the accumulation of the errors in each step of the stroboscopic protocols . however , finding particular schemes to provide multi - qubit jump operators in an analog manner seems to be a tremendous task from both a theoretical and experimental point of view . therefore , we impose a further constraint , the jump operators should be composed of _ ( iii ) single - qubit operators_. in this article , we develop an instance of a global analog dissipative protocol that generates multipartite entangled states corresponding to ground states of a quantum spin chain . the underlying idea is to engineer jump operators that are a particular sum of single - spin operators , in analogy to the models of collective spontaneous emission @xcite . we show that , if the dissipation of the spins is mediated by a common harmonic mode , the spin chain sees a structured markovian environment . by structured environment we refer to a spectral density exhibiting a sharp peak at a certain frequency ( so called breit - wigner resonance ) , corresponding to a weakly damped harmonic oscillator . this structure will allow us to design jump operators in such a way that the stationary state of the system corresponds to the ground state of the quantum spin chain . in particular , we consider an isotropic xy spin chain , which describes a critical phase of matter . we explain in detail how to implement this protocol with trapped ions , relying on tools that have already been achieved experimentally . these tools are the so - called state - dependent forces @xcite , and sympathetic resolved - sideband cooling developed for quantum computation @xcite . we test the scheme for realistic parameters , and show that the fidelities that can be achieved are comparable to the unitary protocols that produce multipartite entangled states @xcite . our procedure performs well for small chains of trapped ions , which is actually the regime in most of the current experiments . this article is organized as follows . in sec . [ dissipative_protocol ] , we describe the dissipative protocol . we start by introducing the model under consideration in sec . [ model ] , and then move onto an analytic discussion , supported by numerical results , of how to engineer a structured environment that allows us to prepare multipartite entangled states dissipatively ( sec . [ cooling_spin_wave ] ) . in sec . [ spin_gs_cooling ] we explain how the method can be used to cool the system to the many - body ground state of the quantum spin chain . in sec . [ sec : ions ] we show how crystals of trapped atomic ions are ideally suited for an experimental implementation of our ideas . section [ ion_crystal ] describes the trapped - ion setup , a two - species coulomb crystal . in secs . [ ion_xy_model ] , [ ion_spin_boson ] and [ ion_damping ] we introduce the trapped - ion toolbox required to implement the dissipative protocol . section [ numerical ] contains a numerical analysis of the trapped - ion dissipative protocol . finally , we present our conclusions and an outlook in sec . [ conclusions ] . and @xmath0 going respectively up and down the spectrum of spin - chain excitations with quasi - momentum @xmath1 and energy @xmath2 ( the model parameters are defined in the text ) . ] let us start by introducing the model under consideration : an interacting spin-@xmath3 chain coupled to a damped bosonic mode ( see fig . [ fig_scheme]*(a ) * ) . this model , which shall be referred to as the _ damped spin - boson chain _ ( dsbc ) in the rest of the manuscript , is a many - body generalization of a system considered recently for the dissipative generation of two - qubit entanglement @xcite . the dsbc is described by the following master equation ( @xmath4 ) @xmath5+\mathcal{d}_{\rm b}(\rho),\ ] ] where @xmath6 is the total density matrix , and @xmath7 is the dissipative liouvillian . we consider a finite chain of @xmath8 spins evolving under the isotropic xy hamiltonian @xcite @xmath9 where @xmath10 is an antiferromagnetic interaction strength , and we introduced the raising and lowering operators @xmath11 . the bosonic hamiltonian consists of a single mode @xmath12 where @xmath13 and @xmath14 are respectively the creation and annihilation operators ( the oscillator frequency @xmath15 can actually correspond to a detuning @xmath16 , as will become clear from the experimental implementation ) . the spins and the boson are coupled through @xmath17 in the above expression , we have introduced the site - dependent coupling strength @xmath18 , and the pauli matrices @xmath19 . finally , we also consider weak damping of the bosonic mode via a lindblad - type dissipator @xmath20 where @xmath21 is the damping rate . we note that all these individual ingredients can be realized with the state - of - the - art technology of trapped - ion crystals , as described in sec . [ sec : ions ] . the objective of this work is to understand the interplay of these dynamics , such that the liouvillian @xmath7 generates stationary entanglement between the initially uncorrelated spins . so far , we can already appreciate two of the properties of the protocol outlined in the introduction : _ ( i ) _ it is _ analog _ , since the couplings - are considered to be switched on during the whole protocol . _ ( ii ) _ it is _ global _ , since the couplings - address all the spins of the chain . let us note that the spin - chain hamiltonian alone could already generate entangled states unitarily at given instants of time . nonetheless , these correlations have a transient nature , whereas the interest of this work is the onset of stationary correlations . therefore , we explore the interplay of this hamiltonian part with the irreversible dynamics introduced by the damped boson through the spin - boson coupling . in fig . [ fig_scheme]*(a ) * , we represent schematically the dsbc . the spin chain is considered to be our system @xmath22 , whereas the bosonic mode , together with the markovian reservoir where it dissipates , provide an effective structured environment @xmath23 for the spin chain . at first sight , the dephasing - like spin - boson coupling seems to introduce a source of decoherence in the spin chain , thus hindering rather than assisting the generation of stationary entanglement . in the following section , we will show that this nave intuition is not always valid , and explain the subtle mechanism that allows for generation of stationary entanglement in a particular regime of the system parameters . we consider an initial state of the system @xmath24 , where @xmath25 is an arbitrary state of the harmonic mode , and the uncorrelated spin state @xmath26 has a single spin excitation at the left edge of the chain . in this section , we discuss how the system liouvillian @xmath7 acts on this state , allowing for the onset of stationary entanglement . as a consequence of the xy hamiltonian , the initially localized spin excitation is exchanged between different neighbors . in fact , the spin chain corresponds exactly to a tight - binding model where the excitation hops along the sites of the underlying chain according to @xmath27 in this expression , @xmath28 results from the projection of the xy hamiltonian onto the single - excitation subspace @xmath29 , where @xmath30 is the projector onto an @xmath8-dimensional subspace spanned by the states @xmath31 , such that @xmath32 labels the position of the spin excitation in the chain . this tight - binding model gives rise to the following band structure ( see fig . [ fig_scheme]*(b ) * ) @xmath33 where each energy is associated to a `` spin - wave excitation '' @xmath34 the goal of this section is to find the conditions to generate dissipatively one of these multipartite entangled states , @xmath35 . we consider a spatial modulation of the spin - boson coupling strength in the form @xmath36 , where @xmath37 . in the single - excitation subspace , the spin - boson coupling becomes @xmath38 where the ladder operators @xmath39 are responsible for ascending or descending along the ladder of spin - wave excitations ( see fig . [ fig_scheme]*(b ) * ) . accordingly , the spin - boson coupling connects different spin - wave excitations , while simultaneously creating or annihilating bosonic quanta . moreover , since the harmonic mode dissipates irreversibly into a markovian environment , the ladder operators @xmath40 introduce irreversible dynamics in the spin chain . in terms of these ladder operators , the single - excitation spin - chain hamiltonian reads @xmath41 where for simplicity we reset the zero of energy at the bottom of the band . we show below how to generate mesoscopic entangled states by controlling the relative strengths of the dissipative processes that take the system up and down the _ spin - wave ladder_. since we are interested in the stationary properties of the dsbc , we consider long times @xmath42 , such that the bosonic mode has enough time to relax under the action of the damping . besides , for @xmath43 , this relaxation is much faster than the process of energy exchange between the boson and the spins . in this limit , the boson thermalizes individually , and the spin chain evolves on a much slower timescale under such a bosonic background . to obtain the effect of the boson background on the slower spin dynamics @xcite , we must `` integrate out '' the bosonic degrees of freedom from the spin - boson coupling . we start with a state of the form @xmath44 , where @xmath45 is the vacuum of the harmonic mode , and fulfills @xmath46 . we then expand the state at a later time @xmath47 , with @xmath48 , in powers of the coupling constant @xmath49 between the spin system and the oscillator , keeping up to second order in @xmath49 . tracing over the mode , we obtain : @xmath50 where the `` hat '' indicates that we work in interaction picture with respect to @xmath51 , and where we have introduced @xmath52 $ ] . using the explicit form of @xmath53 and @xmath54 , after some algebra and one integral in @xmath55 we find : @xmath56.\end{gathered}\ ] ] here , we introduced the _ collective jump operators _ @xmath57 with @xmath58 $ ] , being @xmath59 the energy difference between two neighboring spin - wave excitations in the spin - wave ladder ( see fig . [ fig_scheme]*(b ) * ) . the integrand in eq . ( [ first_integral ] ) contains some terms that oscillate rapidly in time . we now perform the remaining integral , assuming that the time interval @xmath60 is short compared to the time scale given by @xmath61 , but long enough such that @xmath62 . this condition restricts the values of @xmath49 for which this treatment is valid . to perform the integral , we group the frequencies @xmath63 in such a way that within each group the frequencies are equal , and the difference between the frequencies in different groups is finite ( we note that this is not possible in the limit of infinite sizes , where the energies become a continuum ) . it is worth noticing that the presence of degenerate frequencies is typical in this model and makes the grouping necessary . if one keeps only the dominant terms , which are the terms in the integrand that are constant in time , back in schrdinger picture one finds the following master equation governing the coarse - grained evolution over the time scales of interest : @xmath64+\mathcal{d}_{\rm s } ( \rho_{\rm s}),\ ] ] with @xmath65\,,\end{gathered}\ ] ] and where @xmath66\,.\end{gathered}\ ] ] in the expressions above , the sum over @xmath67 runs over the different transition frequencies in the system , @xmath68 denotes an anticommutator , and the lindblad operators are defined as : [ jump_operator ] j_^+ = _ n/_n= l_n^+ = j_^-^where the sum is over all the values of @xmath69 such that @xmath70 . the action of the different lindblad operators is weighted by the spectral density of the effective environment , @xmath71}.\ ] ] according to these expressions , the spins are subjected to a lorentzian reservoir centered at the boson detuning @xmath15 with a width given by the boson damping rate @xmath72 . therefore , the dissipation on the spins is not equal at all frequencies , but stronger at frequencies matching that of the bosonic mode ( i.e. they do not see a totally flat environment , but a structured one @xcite ) . as announced previously , the ladder operators @xmath40 are responsible for introducing the irreversible dynamics in the spin chain . in particular , they determine the collective jump operators , where the adjective collective emphasizes that they act over all the spins in the chain . however , we remark that these jump operators are a sum of single - spin operators , as opposed to the multi - spin nature of some other engineered dissipation protocols considered recently @xcite . with this discussion , we show the third property of the protocol outlined in the introduction : jump operators are composed of sums of _ ( iii ) single - qubit operators_. this draws an analogy to the models of collective spontaneous emission @xcite , but we will show that the special dissipation mediated by the boson mode allows us to use them to prepare dissipatively the ground state of the spin chain . , the jump operators that descend the ladder govern the dissipative dynamics ( i.e. @xmath73 ) . this can be understood in terms of the effective lorentzian spectral density @xmath74 of the environment seen by the spin chain , which overlaps minimally with the processes that climb up the ladder . * ( b , d ) * in the regime of negative oscillator detuning @xmath75 , the jump operators that climb up the ladder dominate the dissipative dynamics ( i.e. @xmath76 ) , since in this case the spectral density of the environment seen by the spin chain overlaps minimally with the processes that descend the ladder . ] we are searching for the steady state @xmath77 of eq . , @xmath78 , with three properties : being pure , displaying multipartite entanglement , and being unique . a pure steady state @xmath79 necessarily belongs to a hamiltonian eigenspace @xcite , which in our case is given by the spin - wave excitations @xmath80 . additionally , the steady state should also belong to the kernel of the dissipator , @xmath81 . by inspection of the jump operators , we identify two regimes where this can happen , which depend on the boson detuning with respect to the spin - wave energy difference : @xmath82 @xmath83 for the regime of positive detunings , we can then approximate the evolution using that @xmath84 is negligibly small for all @xmath69 , while all the @xmath85 are non - negligible . in that case the non - unitary part of the evolution is _ s _ s 2_j_a ( ) ( j_^- _ s j_^+ - \ { j_^+ j_^- , _ s } ) , taking the system to the lowest state in the manifold of spin waves ( the hamiltonian part does not induce transitions between spin waves ) . thus , the unique pure steady state corresponds to the lowest spin wave @xmath86 . conversely , for negative detunings , the steady state corresponds to the highest spin - wave @xmath87 . following @xcite , it can be shown that in these limits the particular form of the lindblad operators does not allow for mixed steady states . we thus get the desired result : a unique pure state displaying stationary multipartite entanglement . in fig . [ fig_2 ] , we represent schematically the two regimes of interest . when conditions are fulfilled , for @xmath88 , the dissipative processes that go down the spin - wave ladder dominate since @xmath89 . effectively , the dissipation induced by the damped bosonic mode `` cools '' the spin state to the lowest - energy spin wave , @xmath90 ( see fig . [ fig_2]*(a ) * ) . the bath spectral density peaks at the frequencies corresponding to the dissipative processes descending the spin - wave ladder ( see fig . [ fig_2]*(c ) * ) and therefore , the bath absorbs very efficiently the energy dissipated by the spin chain during its relaxation to the spin wave @xmath91 . conversely , when the system parameters fulfill conditions , for negative detunings @xmath92 , it is the dissipative processes that climb up the ladder which dominate , @xmath93 . in this limit , the bosonic mode drives the spin state to the highest - energy spin wave , @xmath94 ( see fig . [ fig_2]*(b ) * ) . the bath spectral density is maximal for the ascending dissipative processes ( see fig . [ fig_2]*(d ) * ) , and the bath provides the required energy to climb up the spin - wave ladder . we may thus conclude that in both regimes , the structured reservoir singles out a unique spin wave as the steady state of the chain , assisting the generation of stationary multipartite entanglement . in the light of these results , we can also understand the regime @xmath95 . in this limit , the reservoir spectral density becomes essentially flat , and both up / down processes contribute equally @xmath96 . it is straightforward to see that if @xmath97 is constant , the dissipator in eq . ( [ complete_dissipator ] ) satisfies @xmath98 , and the totally mixed state @xmath99 becomes a steady state . this regime corresponds to the nave argument of section [ model ] , following which one expects that the dephasing - like term can only decohere the spin chain . it is now clear that there are other regimes , - , where we can profit from a structured environment to assist the generation of stationary entanglement . we now comment on the time required to reach the aforementioned steady states . in the regime where the effective liouvillian was derived , it is given by @xmath100 , where @xmath101 . for a particular experimental setup , where the spin couplings can not reach arbitrarily large values , this preparation time may turn out to be too long for practical purposes . we emphasize , however , that the same dissipative preparation of entangled states can be obtained in a non - markovian regime where @xmath102 , or @xmath103 . since simple master equations can not be derived analytically for this regime , we shall explore it numerically . let us note that although the spin - wave excitations are genuinely multipartite entangled , the weights for the excitation of each of the spins in the system are different . a small modification of the scheme also allows for the dissipative generation of @xmath8-partite w - like states , where the w - state is defined as |w=(|+| + | ) . let us leave for a moment the constraint to consider purely @xmath104 couplings , and assume that it is possible to add external transverse fields that act locally at the two edges of the chain @xmath105 in the subspace with one spin excitation , the ground state of this hamiltonian is of the form |_n = _ i=1^n ( -1)^i |i , which is locally equivalent to the above @xmath8-partite w - state ( we note that the actual w - state can be obtained if the spin - spin coupling is ferromagnetic instead of antiferromagnetic ) . as opposed to the ground state of the spin system considered so far , the w - state has the property that it is fully symmetric under particle interchanges . in the state @xmath106 , the excitation is equally distributed over all the sites , and each of the particles is equally entangled with the rest this state can be prepared with the same method described before , at the price of introducing individual addressability in the trapped - ion proposal of sec . [ sec : ions ] . * approach to the asymptotic state : * time evolution of the error @xmath107 , where @xmath108 is the fidelity with the inhomogeneous spin - wave excitation @xmath109 for a chain of three spins . the curves correspond to the resonant case @xmath110 , while the other parameters are @xmath111 , @xmath112 ( red ) , @xmath113 , @xmath114 ( blue ) , and @xmath115 ( green ) . for these numerical calculations , the oscillator was truncated to three levels , with trace preserved to order @xmath116 . ] + in this section , we analyze numerically the validity of our previous analytical derivations by integrating directly the master equation of the damped spin - boson chain . we obtain the fidelity with the desired stationary entangled state @xmath117 where @xmath26 , we consider that the boson mode is initially in the vacuum state @xmath118 , and @xmath119 is the particular entangled state that the protocol targets . we test the robustness of this fidelity against a non - optimal choice of the system parameters - , a limited protocol time @xmath120 , and an increasing particle number @xmath8 . in figure [ fig : fidelity_dynamics ] , we consider the time evolution of the fidelity for the preparation of the lowest - energy spin - wave excitation of an @xmath121 spin chain @xmath122 for these calculations , we set the oscillator detuning @xmath15 to the value @xmath123 which is resonant with the spin - wave transitions , and vary the ratio between the spin - boson coupling strength @xmath49 and the damping rate @xmath72 . the plot indicates that , for similar values of the final fidelity , the convergence is much faster when @xmath49 and @xmath72 are of the same order . this implies that for practical purposes it is most convenient to work in the deeply non - markovian regime . the optimal ratio @xmath124 , however , depends in general on the number of spins and the value of @xmath15 . it is important to note that the resonance condition can be relaxed to some extent . in figure [ fig : fidelity_n3 ] we analyze the fidelity as a function of the detuning @xmath15 and of the coupling @xmath49 , setting @xmath125 . from our previous analysis , we expect the fidelity to be maximal when @xmath126 , and when @xmath127 approach zero to fulfill the requirement @xmath128 . this behavior is totally captured by the numerical results in fig . [ fig : fidelity_n3]*(a ) * , displaying the asymptotic fidelity . the maximum fidelity , highlighted with a white dot , is indeed found in this limit and is equal to 1 within the numerical errors . we remark that , even when the above conditions are only partially fulfilled , the dynamics can provide the desired state with fidelities well above @xmath129 . therefore , we can claim that our dissipative protocol is considerably robust against parameter imperfections . nevertheless , the time required to approach the desired target state might be prohibitively long . therefore , we also characterize the fidelity for a fixed protocol duration . in fig . [ fig : fidelity_n3]*(b ) * , we show the value of the fidelity at a finite fixed time @xmath130 . as can be seen in this figure , the optimal choice for the oscillator detuning is still @xmath123 , while the optimal value for @xmath131 has moved away from zero . in fact , the best choice now results from the interplay between the condition @xmath132 , and the required convergence speed . we emphasize that it is still possible to find parameters such that the achieved fidelities remain very high , @xmath133 . we now address the dissipative generation of w - like states by adding boundary transverse fields . for the @xmath121 spin chain under consideration , the target state is @xmath134 in fig . [ fig : fidelity_n3]*(c)-(d ) * , we show the asymptotic and finite - time fidelities for the dissipative generation of such a tripartite entangled state with homogeneously - distributed excitations . we observe high fidelities comparable to the inhomogeneous spin - wave state , and moreover , an analogous robustness with respect to a non - optimal choice of the system parameters . let us note that , for @xmath121 , there are two different transitions in the spectrum of these w - type spin - waves : one is resonant at a frequency of @xmath135 , and the remaining one occurs at @xmath136 . therefore , it is impossible to set an oscillator detuning at resonance with both transitions simultaneously , and thus the conditions are not completely fulfilled . this explains the fact that the w - type fidelities achieved for finite times are lower than those corresponding to the inhomogeneous spin waves in fig . [ fig : fidelity_n3]*(b ) * , where the two transitions have the same frequency @xmath137 . in this section , we have analyzed numerically the validity of the scheme for the generation of the simplest case of multipartite entanglement , namely tripartite entangled states . a question that should be carefully addressed is whether the same scheme can be used to generate @xmath8-partite entangled states , and how large can the attained fidelities be as @xmath8 is increased . this is the topic of the following section . we start by assessing the fulfillment of the necessary conditions - , which rely on the energetic difference between the dissipative processes that climb up / down the spin - wave ladder , as the number of spins @xmath8 is increased . for very large spin chains @xmath138 , the energy differences between neighboring spin waves @xmath139 . this implies that the energetic argument selecting only processes going up or down the ladder can no longer hold . indeed , @xmath140 , so that the protocol ceases to be operative in the thermodynamic limit of infinitely long chains . however , for mesoscopic spin chains , the ratio @xmath141 can be controlled to an acceptable degree . in fig . [ fig : nsites]*(a ) * , we show that for positive detunings and @xmath142 , @xmath143 for the most of the spin - wave excitations , whereas slightly higher values are attained for the extremal spin excitations ( where @xmath144 ) . for ratios @xmath141 on this order , we expect that the dissipative preparation of entangled states still works with acceptable fidelities . to be more specific , we note that the presence of different transition frequencies @xmath63 is generic for @xmath145 , which represents an obstacle for the efficiency of the procedure . as the number of spins is increased , the convergence also becomes slower because of the larger number of steps down / up the ladder towards the target state , and the lower transition frequencies which require a decrease in @xmath49 and @xmath72 . in fig . [ fig : nsites]*(b ) * , we show numerical results for the dependence of the protocol error @xmath146 achieved for a finite time @xmath147 as a function of the number of sites and optimizing the values of @xmath49 , @xmath72 and @xmath15 . we observe that the errors obtained for the dissipative state preparation of the inhomogeneous spin wave @xmath148 are below 10% for chains of up to ten sites . in comparison , the creation of the w - like states is worse , and such high fidelities can only be achieved for short chains of up to five sites . * the dissipative protocol in mesoscopic chains : * * ( a ) * ratio @xmath149 of the relative strength of the dissipative processes climbing up and going down the ladder as a function of quasimomentum index @xmath69 for different number of spins @xmath8 . the values @xmath150 correspond to negative oscillator detunings @xmath151 which select the processes up the ladder , whereas @xmath152 corresponds to positive oscillator detunings @xmath153 selecting the cooling processes down the ladder . * ( b ) * error @xmath154 for the dissipative generation of the target state in a finite time @xmath147 as a function of the number of spins @xmath8 in the chain . the blue circles represent the errors corresponding to the inhomogeneous spin waves @xmath155 . the red diamonds represent the error for the w - like states @xmath156 . for the numerical calculation , the hilbert space of the oscillator was truncated to four levels . ] so far , our analysis has been restricted to an @xmath8-dimensional subspace of the spin chain , since the initial spin state @xmath26 contains a single excitation , and the number of spin excitations is preserved by the complete dsbc liouvillian . in the following we explore the full @xmath157-dimensional hilbert space of the spin chain . in order to treat the full hilbert space of the spins , we fermionize the spin chain via the so - called jordan - wigner transformation @xcite , namely @xmath158 where @xmath159 are fermionic creation - annihilation operators . the spin - chain and spin - boson hamiltonians can be expressed in terms of the jordan - wigner fermions as follows @xmath160 where we recall that the spin - boson couplings are @xmath36 with @xmath37 . the next step is to express these hamiltonian terms in the spin - wave basis introduced in eq . , which leads to the following expressions @xmath161 where the fermionic operators in momentum space are @xmath162 . in combination with the dissipative part given by eq . , the spin - boson system is analogous to a _ damped single - mode holstein model _ , a dissipative version of the familiar holstein model describing electron - phonon interactions @xcite . we can obtain the same formal expressions as in the single - excitation problem by rewriting the ladder operators in second - quantized form @xmath163 accordingly , in the regime where the boson degrees of freedom can be integrated out , we obtain a purely fermionic master equation which coincides with eqs . - , but with the collective jump operators now expressed in terms of the jordan - wigner ladder operators @xmath164 as a result , the dissipative dynamics restricted to the single - excitation sector can be generalized directly to the full hilbert space with arbitrary numbers of spin excitations . let us now consider an initial state with an arbitrary number of spin excitations @xmath165 distributed along the chain @xmath166 we would like to determine the steady - state of the spin - boson system if the conditions are fulfilled . in this limit , the lindblad operators are only of the form @xmath167 such a jump operator has the effect of lowering the energy of the fermionic quasiparticles . due to the pauli exclusion principle , the initial @xmath168 excitations can not all occupy the lowest energy level , with quasimomentum @xmath169 . instead , the stationary state must be of the following form @xmath170 this is precisely the ground state of the original isotropic xy model , if supplemented with a homogeneous transverse field @xmath171 . in this hamiltonian , the transverse field plays the role of an effective chemical potential @xmath172 , which is determined by the initial number of spin excitations . due to the to the dissipative process , the excitations are distributed in the lowest available single - particle states ( see fig . [ fig_6]*(a ) * ) . this means , for each number of spins up in the initial state prepared , there is a value of the transverse field @xmath173 such that the asymptotic state corresponds to the ground state of @xmath174 . conversely , a given choice of @xmath175 determines the number of spins that should be up in the initial state so that the dissipative dynamics take the system into the desired ground state . spin excitations to the lowest non - occupied state . the stationary state corresponds to the ground state of the isotropic xy hamiltonian , with the particular filling determined by an additional transverse field acting as a chemical potential . * ( b ) * the same as in * ( a ) * but selecting the processes that climb up the ladder . the stationary state would correspond to the zero - temperature state of the same xy model in a transverse field , but with reversed coupling strengths . ] we could also explore the steady state when the conditions are fulfilled . in this limit , the only lindblad operators are of the form : @xmath176 and pump all the excitations to the highest - energy available single - particle states @xmath177 corresponding to the ground - state of the xy hamiltonian @xmath178 with a ferromagnetic spin - spin coupling , and an inverted chemical potential ( see fig . [ fig_6]*(b ) * ) . * scaling of the ground - state - cooling protocol : * error @xmath179 in producing the desired ground - state @xmath180 after a fixed time @xmath147 and for a number of spins ranging from @xmath181 to @xmath182 . for each number of sites , we plot in blue circles the average over the different possible numbers of excitations ( @xmath183 $ ] ) , and with yellow and red squares the best and worst cases , respectively . for the numerical calculation , the hilbert space of the oscillator was truncated to three levels . ] let us now confirm the validity of the above argument by numerical analysis for small spin chains with different numbers of initial excitations . in fig . [ fig : xx ] , we display the error in producing the ground - state taking a fixed protocol time of @xmath147 . the highest errors occur when the number of excitations is 1 or @xmath184 , since the transition frequencies are lowest at these points ( this is the worst case , with the errors plotted in fig . [ fig : xx ] by means of red squares ) . with yellow squares , we represent the lowest attained errors . note that the average errors of the protocol ( blue circles , averaged over all the possible excitation numbers ) are as low as @xmath185 for chains up to @xmath182 spins , which would thus allow us to prepare dissipatively the ground state of the spin model with fidelities of @xmath186 . let us close this section by noting that , in the general case with an arbitrary number of spin excitations , a different choice for the site - dependence of the coupling coefficients @xmath18 may allow for jump operators that go down / up the spin ladder in larger steps , and therefore favor a faster convergence to the asymptotic state ( this appears to be specially well - suited for chains around the half - filling condition ) . however , for up to @xmath182 spins , we have not found any sizable advantage , and the highest fidelities were always achieved for transitions between neighboring spin waves @xmath187 . let us describe a specific platform where the dsbc model can be realized . we consider a system of @xmath188 ions confined in a linear radio - frequency trap @xcite , and arranged forming a one - dimensional chain ( see fig . [ fig_ion_scheme]*(a ) * ) . the vibrational degrees of freedom around the equilibrium positions can be expressed in terms of the so - called normal modes @xcite , namely @xmath189 where @xmath190 refers to the trap axis , @xmath191 labels the different modes with frequencies @xmath192 , and @xmath193 are bosonic operators that annihilate / create phonons in a particular mode . in a linear chain of ions with equal mass , the phonon branches have the structure shown in fig . [ fig_ion_scheme]*(b ) * , such that the different modes spread around the trap frequencies lying in the range @xmath1941 - 10mhz , @xmath1951 - 10mhz , and @xmath1960.1 - 1mhz . a property of the longitudinal branch that will be used in this work is the existence of an energy gap between the lowest - energy mode ( i.e. the center - of - mass mode ) and the following one . as shown in fig . [ fig_ion_scheme]*(a ) * , @xmath8 ions of the chain correspond to a particular atomic species with hyperfine structure , whereas the remaining @xmath197 ions , that will be used for cooling , do not necessarily have hyperfine splitting ( we note that though we focus on hyperfine qubits , our proposal could be generalized to optical or zeeman qubits ) . for simplicity , in the following we describe the case @xmath198 . for the first @xmath8 ions , we select two hyperfine levels of the ground - state manifold , which are referred to as spin states @xmath199 , and have a transition frequency @xmath200 in the microwave range with a negligible linewidth @xmath201 ( fig . [ fig_ion_scheme]*(c ) * ) . for the remaining @xmath202-th ion , we select two internal states @xmath203 from a certain transition with frequency @xmath204 , and linewidth @xmath20510 - 100@xmath206khz ( fig . [ fig_ion_scheme]*(d ) * ) . this is the so - called resolved sideband limit @xmath207 , which will allow for an efficient laser cooling close to the vibrational ground state for the longitudinal center - of - mass mode @xcite . let us note that these small linewidths can be obtained by a variety of methods depending on the particular ion species ( e.g. for weakly - allowed dipole transitions , for quadrupole - allowed transitions with an additional laser that admixes the excited state with that of a dipole - allowed transition , or by a raman configuration ) @xcite . the complete degrees of freedom are thus described by the following master equation @xmath208+\tilde{\mathcal{d}}(\rho),\ ] ] where we have introduced the hamiltonians @xmath209 such that @xmath210 , and @xmath211 ( we use @xmath212 instead of @xmath213 to emphasize the pseudo - spin character of these internal states ) . additionally , we have to take into account the dissipation from the dipole - allowed transition considering recoil effects @xcite . for trapped ions , this can be described by the following super - operator @xmath214 , where @xmath215 where @xmath216 . this is the usual dissipator of a two - level atom in an electromagnetic environment . the spontaneous emission also has effects on the vibrational degrees of freedom due to the photon recoil . to leading order in the lamb - dicke parameter @xmath217 for each normal mode , the jump operators create and annihilate phonons as given by @xcite : @xmath218\tilde{\tau}_{n+1}^++\text{h.c . } , \end{split}\ ] ] where we have introduced @xmath219 , with @xmath220 the coefficients describing the displacement of the @xmath221 ion in mode @xmath69 along direction @xmath222 . once the vibrational and atomic degrees of freedom have been introduced , let us summarize the role that each of them will play in the realization of the damped spin - boson chain : _ ( i ) _ the hyperfine levels of the @xmath8 ions will simulate the spins in the dsbc model . _ ( ii ) _ the @xmath223 vibrational modes along the @xmath224 and @xmath225 axes will be used to mediate the spin - spin interactions leading to the isotropic xy model in eq . . _ ( iii ) _ the longitudinal center - of - mass mode will play the role of the harmonic oscillator in the dsbc model . _ ( iv ) _ the atomic levels of the @xmath202-th ion will be used to laser - cool the longitudinal center - of - mass mode . we note that the position of this auxiliary ion is irrelevant for the purpose of cooling , but it modifies the spin couplings in the chain . moreover , it is also possible to use more than one auxiliary ion as an additional knob to control the damping rate . in the following sections , we will detail the different elements required for the ion - trap implementation of the dsbc . we emphasize that all of them can be realized with state - of - the - art technology . we start by addressing how to implement the isotropic xy model in the ion chain by means of a variation of the mlmer - srensen gate @xcite . the idea is to use the so - called spin - dependent forces , a tool that has already been implemented in several laboratories for different purposes @xcite . by combining a pair of raman laser beams ( see fig . [ fig_ion_scheme]*(c ) * ) , each of which is tuned to the red / blue first vibrational sideband of the hyperfine transition @xcite , it is possible to obtain the following state - dependent force @xmath226 where the `` hat '' indicates the interaction picture with respect to @xmath227 . in the above expression , we have assumed that the effective wavevector of the raman beams @xmath228 points along the @xmath229 axis of the trap ( see fig . [ fig_ion_scheme]*(a ) * ) . additionally , we assume that the strength of both raman beams is equal , such that they share a common rabi frequency @xmath230 , and have opposite detunings , such that we can define a common @xmath231 . we have introduced the laser lamb - dicke parameter @xmath232 . the constraints on these parameters are @xmath233 to neglect other terms in addition to such a state - dependent force . finally , we have defined the sum and difference of the raman beam phases @xmath234 and the pauli spin operator @xmath235 . in order to obtain the desired isotropic xy model , we apply two orthogonal state - dependent forces with the following phases and directions @xmath236 altogether , the hamiltonian of the ion chain in the interaction picture becomes @xmath237 performing a magnus expansion @xcite , the evolution turns out to be given by the unitary operator @xmath238 , where @xmath239 displays the following form to second order @xmath240.\ ] ] we want to derive an effective hamiltonian from this expression @xmath241 ( where the superscript `` ti '' stands for the trapped - ions realization ) . the first - order contribution leads to a couple of orthogonal state - dependent displacements , which can only be neglected in the limit @xmath242 @xcite . in addition , from the non - commutativity of the @xmath243 and @xmath244 forces , we get a residual spin - phonon coupling in the second - order term of the magnus expansion . in order to neglect it , we have to impose a further constraint , namely @xmath245 , which can be fulfilled if the trap frequencies along the @xmath246 axes are sufficiently different @xmath247 . under these constraints , an interacting quantum spin chain is obtained @xmath248 therefore , by adjusting the strengths of the raman beams , and their detunings , it is possible to find a regime where @xmath249 , and the above spin hamiltonian corresponds to the desired isotropic xy model since @xmath250 . let us remark that the resulting xy model has the peculiarity of displaying long - range couplings . in fact , when @xmath251 , and the raman lasers are far - detuned from the whole vibrational branch , it can be shown that these couplings decay with a dipolar law . the trapped - ion hamiltonian becomes a realization of the xy interaction in the dsbc model after the identifications @xmath252 , and @xmath253 . for typical nearest - neighbor distances of @xmath2541 - 10@xmath173 m , the spin - spin couplings attain strengths in the @xmath2551 - 10 khz . let us note that modifications of the scheme of state - dependent forces have been proposed to yield other quantum spin models @xcite , some of which have also been realized experimentally @xcite . [ ion_spin_boson ] let us now address how to implement the spin - boson coupling in the ion chain . the idea is again to use a @xmath256-beam configuration ( see fig . [ fig_ion_scheme]*(c ) * ) , but in a different regime . rather than tuning the two - photon frequencies @xmath257 to the vibrational sidebands of the hyperfine transition , we impose that @xmath258 . moreover , we consider that the laser beams form a standing wave along the trap @xmath259-axis ( see fig . [ fig_ion_scheme]*(a ) * ) . under these constraints , the laser - ion interaction leads to a crossed - beam differential ac - stark shift , which can be interpreted as another state - dependent force in the @xmath260 basis @xmath261 where the `` hat '' refers to the interaction picture with respect to @xmath262 . here , the parameters are defined in analogy to those in eq . , with three important differences : _ ( i ) _ @xmath263 is not the two - photon rabi frequency of the hyperfine transition , but rather a differential ac - stark shift coming from processes where a photon is exchanged between the pair of laser beams in the @xmath256 configuration . _ ( ii ) _ the detunings are changed to @xmath264 . _ ( iii ) _ since the ion chain lies along the @xmath259 axis , there is an additional site - dependent phase when shining the lasers such that @xmath265 . when adjusting the difference of the laser phases to be @xmath266 , then we obtain @xmath267 . let us highlight that the standing - wave nature of this spin - dependent force is not essential for the dissipative protocol . however , it makes the connection with the dsbc studied in sec . [ dissipative_protocol ] more transparent . we will show in sec . [ numerical ] that essentially the same fidelities can be achieved for a traveling - wave configuration , which has also been experimentally demonstrated @xcite . we now exploit the particular properties of the longitudinal phonon branch ( see fig . [ fig_ion_scheme]*(b ) * ) . more precisely , we use the presence of a gap between the axial center - of - mass mode and the rest of the modes along the same direction . for the case of equal ions , @xmath268 , while for different ion species or isotopes , the exact value of the gap will depend on the mass ratio and the position of the cooling ions . we assume that @xmath269 is red - detuned with respect to the axial center - of - mass mode @xmath270 , such that @xmath271 . since the next vibrational modes are separated by a large energy gap , the laser coupling to the remaining phonon branch is highly off - resonant and can be thus neglected . accordingly , this state - dependent force gives _ l_z=_i_i1^z_i^zb_1ze^-i_1zt+ h.c .. we can finally move to a different picture where the above hamiltonian becomes time - independent @xmath272 , where [ spin_boson_ion ] h_b^ti&=_i_1zb_1z^b_1z , + h_sb^ti&=_i_i1^z_i^z(b_1z+b_1z^ ) . therefore , it is clear that the center - of - mass mode plays the role of the harmonic oscillator @xmath273 in the dsbc , with the laser detuning corresponding to the oscillator detuning @xmath274 in the dsbc model . in addition , the strength of the state - dependent force determines the spin - boson coupling @xmath275 in the dsbc model . let us now comment on realistic values for the trapped - ion parameters . since the laser detuning should only fulfill @xmath2761 - 10@xmath206mhz , it will be easy to reach the required condition of @xmath2771 - 10@xmath206khz . on the other hand , we know from the previous section that @xmath278 is necessary to have an accurate pumping to the desired entangled state . nevertheless , @xmath49 should not be too small that the total preparation time becomes prohibitively long . a suitable choice could be @xmath279 0.1 - 0.5 khz . since @xmath280 , it will suffice to set the rabi frequency in the @xmath2811 - 5@xmath282khz . finally , we should adjust the laser wavevector such that @xmath283 . once the implementation of the xy model and the spin - boson coupling has been described , let us turn into the last required ingredient : an effective damping of the bosonic mode . as discussed previously , the idea is to exploit the atomic levels of the @xmath284-th ion to laser - cool the longitudinal center - of - mass mode . since the resonance frequencies are very different , @xmath285 , the cooling lasers do not affect the spin dynamics of the other @xmath8 ions . however , since the vibrational modes are collective , it is possible to sympathetically cool the vibrations of the crystal by only acting on the @xmath284-th ion . this sympathetic laser cooling @xcite has been already realized experimentally in small crystals for quantum computation @xcite . our starting point is the master equation with the dissipation superoperators - . to control the effective damping , we introduce a laser beam red detuned from the atomic transition @xmath286 , such that @xmath287 ( see fig . [ fig_ion_scheme]*(d ) * ) . let us note that the particular laser - beam configuration will depend on the particular ion species , and the scheme to attain the resolved - sideband limit @xcite . we consider that the laser is in a traveling - wave configuration , such that its wavevector is aligned parallel to the trap axis @xmath288 ( see fig . [ fig_ion_scheme]*(a ) * ) . after expanding in series of the lamb - dicke parameter @xmath289 , the laser - ion interaction leads to two different terms , the so - called carrier term @xmath290 and the red and blue sideband terms @xmath291 where we introduced @xmath292 , and the corresponding rabi frequency @xmath293 . by controlling appropriately these sideband terms , and their interplay with the atomic spontaneous emission , it is possible to tailor the damping of the longitudinal modes . let us rearrange the full liouvillian as a sum of two terms @xmath294 , where @xmath295+\tilde{\mathcal{d}}^0(\bullet),\\ \mathcal{l}_1(\bullet)&=-{{\rm i}}[\tilde{h}_{\tilde{\rm l}_z}^1,\bullet]+\tilde{\mathcal{d}}^1(\bullet ) , \end{split}\ ] ] which is justified for small lamb - dicke parameters . to obtain the effective damping of the longitudinal center - of - mass mode , we use a similar formalism as in @xcite . in this case , the fastest timescale in the problem is given by the decay rate @xmath296 of the atomic states of the @xmath202-th ion . therefore , we must `` integrate out '' these atomic degrees of freedom , which can be accomplished by projecting the density matrix of the full system into @xmath297 , where @xmath298 fulfills @xmath299 , and @xmath300 is the reduced density matrix for the @xmath8 spins and the chain vibrational modes . for this we use the expression @xmath301 where the `` hats '' refer to the interaction picture with respect to @xmath262 . after some algebra , one gets the effective damping of the longitudinal modes @xmath302+{\mathcal{d}}_{\rm b}(\rho_{n}),\ ] ] where we have introduced the bosonic dissipator @xmath303 together with the effective rates for laser cooling and heating @xmath304 here , the so - called diffusion coefficient accounting for the recoil heating is @xmath305 , and the spectral functions of the sideband terms are @xmath306 which can be obtained by means of the quantum regression theorem . the formalism and cooling rates coincide with those of a single trapped ion @xcite , with the difference that in the diffusion coefficient @xmath307 and forces @xmath308 one must consider the normal vibrational modes at the position of the cooled ion . the idea is to work in the resolved sideband regime @xmath207 , and set the laser - cooling parameters to optimize the cooling of the longitudinal center - of - mass mode @xmath309 . in this regime , the cooling dominates in , and we are left with the desired damping of the bosonic mode @xmath310 once more , after the identifications @xmath273 , and @xmath311 , we obtain an analogous dsbc damping . let us finally comment on the required cooling rates . ideally , we should achieve the regime @xmath3120.1 - 0.5 khz . since such cooling is not particularly fast , it is possible to find the right detuning @xmath287 such that @xmath3130.1 - 0.5 khz . besides , since we are in the resolved sideband limit , the heating can be made much smaller , @xmath314 . the stationary state of a harmonic oscillator under the action of @xmath315 in eq . is a thermal state with mean number of quanta @xmath316 . in the limit @xmath317 , such that the cooling is the dominant effect and the heating only presents a small correction , the equilibrium state of the center - of - mass mode is close to the vacuum state . in this last section , we will explore numerically how the dissipative protocol described in the part [ dissipative_protocol ] of this manuscript can be implemented with the trapped - ion dscb model in eqs . , and , namely @xmath318+\mathcal{d}_{\rm b}^{\rm ti}(\rho).\ ] ] therefore , the results discussed below correspond to the actual spin - spin couplings @xmath319 in an ion trap , which are not nearest - neighbors couplings , but display a dipolar decay with the cube of the distance . furthermore , we have taken into account that in harmonic ion traps , the interparticle distance is not constant over the chain . more importantly , we have considered the effects of the always - present heating term in the trapped - ion effective damping . * heating in the trapped - ion dissipative protocol : * * ( a ) * fidelity with the desired target state for a fixed time @xmath147 and as a function of the ratio @xmath320 between the sympathetic heating and cooling rates in a chain of six ions in which the central four represent sites of the isotropic xy spin chain . the blue dots correspond to the case of @xmath321 initial spin excitations , and the orange squares to the case of @xmath322 excitations . the oscillator was truncated to five levels . * ( b ) * same as in * ( a ) * , but considering a shorter time @xmath323 , and a traveling - wave configuration of the spin - boson coupling in eq . . the achieved fidelities for this experimentally simpler configuration are similar to the ones in * ( a)*. ] in fig . [ fig:6 ion chain]*(a ) * , we show the optimal fidelity for the dissipative generation of the ideal ground state at different fillings , by integrating numerically the dipolar and inhomogeneous trapped - ion dsbc model as a function of the ratio between heating and cooling rate @xmath324 . we study a chain of @xmath325 ions , in which @xmath326 have a hyperfine structure and play the role of the spins in the xy chain , while the two peripheral ions are assumed to be used for sympathetic cooling ( this choice of auxiliary cooling ions at the ends of the chain makes the interparticle spacing slightly more homogeneous for the `` system '' ions ) . as can be seen in the figure , when @xmath327 , the fidelities with the target ground state approach 100@xmath328 , which allows us to conclude that the differences due to the inhomogeneity of the chain and the dipolar range of interactions with respect to the ideal homogeneous and nearest - neighbor dsbc do not compromise the fidelities severely . the situation is different as finite heating rates @xmath329 are considered . in fig . [ fig:6 ion chain]*(a ) * , we observe that the fidelity with the target ground state decays essentially linear with @xmath324 . in the worst case considered , where @xmath330 leads to a mean phonon number of @xmath331 , the fidelities obtained are reduced to roughly @xmath332 . we note that resolved sideband cooling of single vibrational modes has been experimentally achieved reaching @xmath333 below 0.1 for single trapped ions @xcite . sympathetic cooling of crystals up to @xmath326 ions has also been achieved @xcite , reaching values as low as @xmath334 with pulsed techniques . we note that our scheme does not demand such an accurate ground - state cooling ( see fig . [ fig:6 ion chain]*(a ) * ) , and we expect that the required mean phonon numbers @xmath335 can also be achieved in a continuous cooling scheme . another source of experimental imperfections is given by slow drifts of the trap frequencies which will modify the detuning @xmath336 in different experimental runs , such that the conditions will not be perfectly fulfilled . nonetheless , in fig . [ fig : fidelity_n3 ] we have shown that the dissipative character of the protocol endows it with a natural robustness with respect to non - optimal choices of the parameters . therefore , one can expect large fidelities even in the presence of slow drifts of the trap frequency within the khz - range . in any case , to alleviate imperfections caused by both magnetic - field dephasing and trap frequency drifts , we have considered a faster protocol where the required time is @xmath3371 - 10@xmath206ms . in fig . [ fig:6 ion chain]*(b ) * , we show that the achieved fidelities in this case are still above 0.8 ( for heating / cooling ratios below 0.2 ) . moreover , in this figure we have also considered substituting the standing - wave force in eq . by a traveling - wave configuration , which is experimentally less demanding . the need of the @xmath260 state - dependent force forbids the use of magnetic - field insensitive states ( i.e. clock states ) . therefore , the hyperfine spins will be subject to fluctuations induced by non - shielded external magnetic fields , typically leading to dephasing in a timescale of @xmath3381 - 10@xmath206ms . these timescales are comparable to the protocol times @xmath3391 - 10@xmath206ms , where we have taken @xmath3401 - 10@xmath206khz . nevertheless , these magnetic - field fluctuations act globally for standard radio - frequency traps @xmath341 , where @xmath342 is a fluctuation of the resonance frequency due to the zeeman shift ( note that this might not be the case for micro - fabricated surface traps , where fluctuating magnetic - field gradients can also arise ) . since the dynamics of the dsbc conserves the number of spin excitations , this magnetic - field noise only introduces a global fluctuating phase , and thus does not decohere the state of the system . therefore , as far as the induced spin dynamics occurs within any subspace with a conserved number of excitations , the dissipative protocol is robust to global magnetic - field noise . * anisotropy in the trapped - ion dissipative protocol : * * ( a ) * fidelity with the desired target state for a fixed time @xmath343 and as a function of the anisotropy ratio @xmath344 for a chain of six ions in which the central four ions represent sites of the slightly - anisotropic xy spin chain . the blue dots correspond to the case of @xmath321 initial spin excitations , and the orange squares to the case of @xmath322 excitations . the oscillator was truncated to five levels . ] it is , however , important to stabilize the intensity of the state - dependent forces to the sweet spot where @xmath249 , and make sure that @xmath345 to neglect any term that does not conserve the number of excitations . the effect of off - resonant terms in the scheme leading to the xy interactions , that could give rise to non - conservation of the number of spin excitations , should be much smaller than in stroboscopic proposals @xcite since our protocol does not require fast gates @xcite . in fig . [ fig:7 ion chain ] , we show the achieved fidelities for anisotropies in @xmath346 . the results have been obtained numerically for the same regime as in fig . [ fig:6 ion chain]*(b ) * , and indicate that if @xmath347 differ by 1 part in @xmath348 , the fidelities are not severely reduced . the results in fig . [ fig:7 ion chain ] also show a marked difference in the robustness depending on the number of spin excitations in the initial state . this behaviour , at first surprising , can be understood by looking at the small undesired hamiltonian terms responsible for the anisotropy in an interaction picture with respect to the ideal isotropic hamiltonian . the anisotropy term creates or destroys fermionic quasiparticles in pairs , and rotates with a frequency that is the sum of the energies of the two quasiparticles . for the case with only one spin up , it is possible to create from the ideal target state pairs of fermions with energy adding up to zero , so that these error terms do not rotate . in the case with two spin excitations , on the contrary , the lower half of the fermionic spectrum is filled , so that any hamiltonian term creating or destroying a pair of fermions from the ideal target state oscillates in time , and therefore its effect is strongly suppressed . we have presented a method to dissipatively generate multipartite - entangled states corresponding to the ground states of small spin chains with isotropic xy hamiltonian in a transverse field . the protocol for dissipative state preparation is interesting from a fundamental point of view , as it illustrates how local and even purely markovian noise can assist entanglement production . indeed , the jump operators required are sums of terms acting on individual sites , in contrast with the few - body ( quasi - local ) nature of the operators in @xcite . when the noise deviates from exact markovianity , the required time for achieving the steady state is reduced , which illustrates the value of non - markovian effects for practical purposes . we would like to emphasize that the general idea underlying the procedure presented in this work is not specific to the model considered , and might be generalized to other spin hamiltonians . as an example , we introduced a variation that can be used to generate states locally equivalent to w - states . it would also be interesting to modify our dissipative protocol to prepare the ground state of a gapped spin model . for longer ion chains , the combination of more state - dependent forces with different wavevectors could improve the scalability of the protocol . we have also explained in detail how to implement this method in small chains of trapped ions . in our proposal , two internal levels of the ions embody the spin system and a collective motional mode represents the damped oscillator . the preparation procedure requires the implementation of spin - spin interactions using the so - called mlmer - srensen scheme , the action of a state - dependent force , and sympathetic cooling , all ingredients within the capabilities of present ion - trap technology . numerical simulations including different sources of errors indicate that the protocol can produce the target states with fidelities comparable to the more standard coherent protocols @xcite . the method presented is indeed robust against a number of experimental imperfections , and the implementation is simplified since only global addressing of the ions is required . our results , however , may find a realization in different experimental setups . for instance , the application of our ideas to arrays of superconducting qubits in stripline resonators seems feasible . morever , since this model can also be understood as a quadratic fermonic model with an additional chemical potential fixing the number of particles , the results may also be interesting for fermionic atoms in optical lattices . the realization of this kind of dissipative system paves the way for the implementation of more complex scenarios to study the interplay of coherent and incoherent dynamics giving rise to noise - induced criticality @xcite . indeed , recent work in the area of non - equilibrium quantum phase transitions @xcite has shown a number of fascinating results that could be demonstrated in systems of trapped ions using tools similar to the ones in our protocol . d. e. browne , m. b. plenio , and s. f. huelga , http://prl.aps.org/abstract/prl/v91/i6/e067901[phys . * 91 , * 067901 ( 2003 ) ] ; l .- m . duan and h. j. kimble , http://prl.aps.org/abstract/prl/v90/i25/e253601[phys . rev . lett . * 90 , * 253601 ( 2003 ) . ] a. beige , d. braun , b. tregenna , and p. l. knight , http://prl.aps.org/abstract/prl/v85/i8/p1762_1[phys . * 85 , * 1762 ( 2000 ) ] ; j. pachos and h. walther , http://prl.aps.org/abstract/prl/v89/i18/e187903[phys . rev . lett . * 89 , * 187903 ( 2002 ) ] ; a. beige , d. braun , s. bose , s.f . huelga , p.l . knight , m.b . plenio and v. vedral , http://www.tandfonline.com/doi/abs/10.1080/09500340008232183[j . mod . opt . * 47 * , 2583- 2598 ( 2000 ) ] . j. f. poyatos , j. i. cirac , and p. zoller , http://prl.aps.org/abstract/prl/v77/i23/p4728_1[phys . * 77 , * 4728 ( 1996 ) ] ; m. b. plenio and s. f. huelga , http://prl.aps.org/abstract/prl/v88/i19/e197901[phys . rev . lett . * 88 , * 197901 ( 2002 ) ] . a. gonzalez - tudela , d. martin - cano , e. moreno , l. martin - moreno , c. tejedor , and f.j . garcia - vidal , http://prl.aps.org/abstract/prl/v106/i2/e020501[phys . * 106 , * 020501 ( 2011 ) ] ; a. bermudez , t. schaetz , and m. b. plenio , http://prl.aps.org/abstract/prl/v110/i11/e110502[phys . lett . * 110 * , 110502 ( 2013 ) ] ; m. j. kastoryano , f. reiter , and a. s. srensen , http://prl.aps.org/abstract/prl/v106/i9/e090502[phys . rev . * 106 * , 090502 ( 2011 ) ] . j. t. barreiro , m. mller , p. schindler , d. nigg , t. monz , m. chwalla , m. hennrich , c. f. roos , p. zoller , and r. blatt , http://www.nature.com/nature/journal/v470/n7335/full/nature09801.html[nature * 470 , * 486 ( 2011 ) ] ; p. schindler , m. mller , d. nigg , j. t. barreiro , e. a. martinez , m. hennrich , t. monz , s. diehl , p. zoller , and r. blatt , http://www.nature.com/nphys/journal/v9/n6/abs/nphys2630.html[nature phys . * 9 * , 361 ( 2013 ) ] . c. monroe , d.m . meekhof , b.e . king , and d.j . wineland , http://www.sciencemag.org/content/272/5265/1131.abstract?sid=cc4c0d27-8270-4f9d-bd28-4e7af11fb993[science * 272 , * 1131 ( 1996 ) ] ; p. c. haljan , k .- a . brickman , l. deslauriers , p. j. lee , and c. monroe , http://prl.aps.org/abstract/prl/v94/i15/e153602[phys . * 94 , * 153602 ( 2005 ) ] ; m.j . mcdonnell , j.p . home , d.m . lucas , g. imreh , b.c . keitch , d.j . szwer , n.r . thomas , s.c . webster , d.n . stacey , a.m. steane , http://prl.aps.org/abstract/prl/v98/i6/e063603[phys . lett . , * 98 , * 063603 ( 2007 ) ] ; u. poschinger , a. walther , k. singer , and f. schmidt - kaler , http://prl.aps.org/abstract/prl/v105/i26/e263602[phys . rev . lett . * 105 , * 263602 ( 2010 ) . ] sackett , d. kielpinski , b.e . king , c. langer , v. meyer , c.j . myatt , m. rowe , q.a . turchette , w.m . itano , d.j . wineland , and c. monroe , http://www.nature.com/nature/journal/v404/n6775/abs/404256a0.html[nature * 404 , * 256 ( 2000 ) ] ; d. leibfried , b. demarco , v. meyer , d. lucas , m. barrett , j. britton , w. m. itano , b. jelenkovic , c. langer , t. rosenband , and d. j. wineland , http://www.nature.com/nature/journal/v422/n6930/abs/nature01492.html [ nature * 422 , * 412 ( 2003 ) ] ; j. benhelm , g. kirchmair , c. roos , and r. blatt , http://www.nature.com/nphys/journal/v4/n6/full/nphys961.html [ nat . phys . * 4 , * 463 ( 2008 ) ] ; a. noguchi , s. haze , k. toyoda , and s. urabe , http://prl.aps.org/abstract/prl/v108/i6/e060503[phys . * 108 , * 060503 ( 2011 ) ] . c. ospelkaus , u. warring , y. colombe , k. r. brown , j. m. amini , d. leibfried , and d. j. wineland , http://www.nature.com/nature/journal/v476/n7359/full/nature10290.html[nature * 476 , * 181 ( 2011 ) ] ; a. khromova , ch . piltz , b. scharfenberger , t. f. gloger , m. johanning , a. f. varon , ch . wunderlich , http://prl.aps.org/abstract/prl/v108/i22/e220502[phys . * 108 , * 220502 ( 2012 ) . ] a. friedenauer , h. schmitz , j. t. glueckert , d. porras , and t. schaetz , http://www.nature.com/nphys/journal/v4/n10/full/nphys1032.html[nat . phys . * 4 , * 757 ( 2008 ) ] ; k. kim , m .- s . chang , s. korenblit , r. islam , e. e. edwards , j. k. freericks , g .- d . lin , l .- m . duan , and c. monroe , http://www.nature.com/nature/journal/v465/n7298/full/nature09071.html[nature * 465 , * 590 ( 2010 ) ] ; r. islam , e. e. edwards , k. kim , s. korenblit , c. noh , h. carmichael , g .- d.lin , l .- m . duan , c .- c . joseph wang , j. k. freericks , and c. monroe , http://www.nature.com/ncomms/journal/v2/n7/full/ncomms1374.html[nat . * 2 , * 377 ( 2011 ) ] ; b. p. lanyon , c. hempel , d. nigg , m. mller , r. gerritsma , f. zhringer , p. schindler , j. t. barreiro , m. rambach , g. kirchmair , m. hennrich , p. zoller , r. blatt , and c. f. roos , http://www.sciencemag.org/content/334/6052/57[science * 334 , * 57 ( 2011 ) ] ; j. w. britton , b. c. sawyer , a. c. keith , c .- c . j. wang , j. k. freericks , h. uys , m. j. biercuk and j. j. bollinger , http://www.nature.com/nature/journal/v484/n7395/full/nature10981.html[nature * 484 , * 489 ( 2012 ) . ] h. schmitz , r. matjeschk , ch . schneider , j. glueckert , m. enderlein , t. huber , and t. schaetz , http://prl.aps.org/abstract/prl/v103/i9/e090504[phys . * 103 , * 090504 ( 2009 ) ] ; f. zhringer , g. kirchmair , r. gerritsma , e. solano , r. blatt , and c. f. roos , http://prl.aps.org/abstract/prl/v104/i10/e100503[phys . rev . lett . * 104 , * 100503 ( 2010 ) . ] h. rohde , s. t. gulde , c. f. roos , p. a. barton , d. leibfried , j. eschner , f. schmidt - kaler , r. blatt , http://iopscience.iop.org/1464-4266/3/1/357/[j . b * 3 , * s3 ( 2001 ) ] ; m. d. barrett , b. demarco , t. schaetz , v. meyer , d. leibfried , j. britton , j. chiaverini , w. m. itano , b. jelenkovic , j. d. jost , c. langer , t. rosenband , and d. j. wineland , http://pra.aps.org/abstract/pra/v68/i4/e042302[phys . a * 68 , * 042302 ( 2003 ) ] . j. p. home , m. j. mcdonnell , d. j. szwer , b. c. keitch , d. m. lucas , d. n. stacey , and a. m. steane , http://pra.aps.org/abstract/pra/v79/i5/e050305[phys . a * 79 , * 050305(r ) ( 2009 ) ] ; j. p. home , d. hanneke , j. d. jost , j. m. amini , d. leibfried , and d. j. wineland , http://www.sciencemag.org/content/325/5945/1227[science * 325 , * 1227 ( 2009 ) ] . a. imamoglu , http://pra.aps.org/abstract/pra/v50/i5/p3650_1[phys . a * 50 , * 3650 ( 1994 ) ] ; b. m. garraway , http://pra.aps.org/abstract/pra/v55/i3/p2290_1[phys . a * 55 , * 2290 ( 1997 ) . ] chin , a. rivas , s.f . huelga and m.b . * 51 * , 092109 ( 2010 ) ] . t. holstein , http://www.sciencedirect.com/science?_ob=articlelisturl&_method=list&_articlelistid=-67535988&_sort=r&_st=13&view=c&_acc t = c000057787&_version=1&_urlversion=0&_userid=2532525&md5=e2f1d3979b0d8f1c0b747b3c2c93f84e&searchtype = a[ann . * 8 , * 325 ( 1959 ) ] ; ( ibid ) http://www.sciencedirect.com/science/article/pii/000349165990003x[*8 , * 343 ( 1959 ) ] d. j. wineland , c. monroe , w. m. itano , d. leibfried , b. e. king , and d. m. meekhof , http://nvlpubs.nist.gov/nistpubs/jres/103/3/cnt103-3.htm[j . res . * 103 , * 259 ( 1998 ) ] , _ and references therein_. j. dalibard and c. cohen - tannoudji , http://iopscience.iop.org/0022-3700/18/8/019[j . b : at . mol . * 18 , * 166 ( 1985 ) ] ; y. castin , h. wallis , and j. dalibard , http://www.opticsinfobase.org/josab/abstract.cfm?uri=josab-6-11-2046[j . b. * 6 , * 2046 ( 1989 ) . ] w. magnus , http://onlinelibrary.wiley.com/doi/10.1002/cpa.3160070404/abstract[commun . pure appl . * 7 , * 649 ( 1954 ) ] ; s. blanes , f. casas , j. a. oteo and j. ros , http://iopscience.iop.org/0143-0807/31/4/020[eur . * 31 , * 907 ( 2010 ) . ] m. pons , v , ahufinger , ch . wunderlich , a. sanpera , s. braungardt , a. sen , u. sen , and m. lewenstein , http://prl.aps.org/abstract/prl/v98/i2/e023003[phys . * 98 , * 023003 ( 2007 ) ] ; a. bermudez , d. porras , and m. a. martin - delgado , http://pra.aps.org/abstract/pra/v79/i6/e060303[phys . a * 79 , * 060303 ( r ) ( 2009 ) ] ; p. hauke , f. m. cucchietti , a. mller - hermes , m.c . bauls , j. i. cirac , and m. lewenstein , http://iopscience.iop.org/1367-2630/12/11/113037/[new j. phys . * 12 , * 113037 ( 2010 ) ] ; p. a. ivanov and f. schmidt - kaler , http://iopscience.iop.org/1367-2630/13/12/125008[new j. phys . * 13 , * 125008 ( 2011 ) ] ; a. bermudez , j. almeida , f. schmidt - kaler , a. retzker , and m. b. plenio , http://prl.aps.org/abstract/prl/v107/i20/e207209[phys . * 107 , * 207209 ( 2011 ) ] ; r. schmied , j. h. wesenberg , and d. leibfried , http://iopscience.iop.org/1367-2630/13/11/115011/[new j. phys . * 13 , * 115011 ( 2011 ) ] ; g .- d . lin , c. monroe , and l .- m . duan , http://prl.aps.org/abstract/prl/v106/i23/e230402[phys . * 106 , * 230402 ( 2011 ) ] ; a. bermudez , j. almeida , k. ott , h. kaufmann , s. ulm , u. poschinger , f. schmidt - kaler , a. retzker , and m. b. plenio , http://iopscience.iop.org/1367-2630/14/9/093042/[new j. phys . * 14 , * 093042 ( 2012 ) . ] d. kielpinski , b. e. king , c. j. myatt , c. a. sackett , q. a. turchette , w. m. itano , c. monroe , d. j. wineland , and w. h. zurek , http://pra.aps.org/abstract/pra/v61/i3/e032310[phys . rev . a * 61 , * 032310 ( 2000 ) ] ; g. morigi and h. walther , http://epjd.epj.org/index.php?option=com_article&access=standard&itemid=129&url=/articles/epjd/abs/2001/02/d0131/d0131.h tml[eur . j. d * 13 , * 261 ( 2001 ) ] . g. schn and a.d . zaikin , http://www.sciencedirect.com/science/article/pii/037015739090156v[phys . rep . * 198 * , 237 ( 1990 ) ] . l. capriotti , a. cuccoli , a. fubini , v. tognetti , and r. vaia , http://prl.aps.org/abstract/prl/v94/i15/e157001[phys . * 94 * , 157001 ( 2005 ) ] . m. a. cazalilla , f. sols , and f. guinea , http://prl.aps.org/abstract/prl/v97/i7/e076401[phys . * 97 * , 076401 ( 2006 ) ] . f. verstraete , m. m. wolf , and j. i. cirac , http://www.nature.com/nphys/journal/v5/n9/abs/nphys1342.html[nature phys . * 5 * , 633 ( 2009 ) ] . j. eisert and t. prosen , e - print http://arxiv.org/abs/1012.5013[arxiv:1012.5013 ( 2012 ) ] . s. morrison , a. s. parkins , http://prl.aps.org/abstract/prl/v100/i4/e040403[phys . lett . * 100 * , 040403 ( 2008 ) ] . t. prosen and e. ilievski , http://prl.aps.org/abstract/prl/v107/i6/e060403[phys . * 107 * , 060403 ( 2011 ) ] ; m. j. bhaseen , j. mayoh , b. d. simons , and j. keeling , http://pra.aps.org/abstract/pra/v85/i1/e013817[phys . a * 85 * , 013817 ( 2012 ) ] ; v. m. bastidas , c. emary , g. schaller , and t. brandes , http://pra.aps.org/abstract/pra/v86/i6/e063627[phys . a * 86 * , 063627 ( 2012 ) . ]

since the discovery that quantum coherences might have a functional role in biological systems even at room temperature @xcite , there has been great interest in understanding how coherences can be maintained and used under the influence of different environments with competing effects . in particular , much of recent research focused on one - dimensional nanostructures , due to their relevance to molecular aggregates , such as the j - aggregates @xcite , natural photosynthetic systems @xcite , bio - engineered devices for photon sensing @xcite and light - harvesting systems @xcite . here we focus on a ring - like structure of two - level systems coupled with nearest neighbor tunneling amplitudes which has been recently considered in literature @xcite . usually , under low light intensity , in many natural photosynthetic systems or in ultra - precise photon sensors the single - excitation approximation can be used . in this case the system is equivalent to a tight binding model where one excitation can hop from site to site , see fig . [ ring ] . many photosynthetic organisms contain ring - like chlorophyll molecular aggregates in their light - harvesting complexes , which are called lhi and lhii @xcite . these complexes have the purpose to absorb light and to transfer the excitations to other structures or to a central core absorber , the reaction center , where charge separation , necessary in the next steps of photosynthesis , occurs . these complexes are subject to the effects of different environments : @xmath0 dissipative , where the excitation can be lost ; @xmath1 proteic , which induces static or dynamical disorder . the efficiency of excitation transfer can be determined only through a comprehensive analysis of the effects due to the interplay of all those environments . here , in particular , we consider a system subject to the influence of both a common decay channel where the excitation can be lost , and a static disorder . the first environment can be thought of as a model for the coupling of a molecular aggregate to the electromagnetic field @xcite ( loss of excitation by recombination ) or for the coupling of the molecular aggregate to a central core absorber ( loss of excitation by trapping ) . for many molecular aggregates , the single channel approximation is appropriate to describe the coupling with the electromagnetic field , since the wavelength of the absorbed light is much larger than the system size ( natural complexes such as lhi , lhii typically span few tens of nanometers , while the wavelength of the involved photon is hundreds of nanometers ) . moreover , it can also be considered as a good approximation for the coupling to a central core absorber , modeled for instance by a semi - infinite one - dimensional lead @xcite . the second environment consists of a protein scaffold , in which photosynthetic complexes are embedded , that induces fluctuations in the sites energies . the fluctuations which occur on a time scale much larger than the time scale of the dynamics are usually described as static disorder . by static disorder we mean position dependent , but time - independent , fluctuations of the site energies . the case of time - dependent fluctuations of site energies has been considered in a separate paper @xcite . it is well known that , when many sites are all coupled to the same channel , we can have a superradiant behavior @xcite . superradiance implies the existence of some states with a cooperatively enhanced decay rate ( i.e. proportional to the number of sites ) . superradiance comes always together with subradiance , that is the existence of states with a cooperatively suppressed decay rate ( i.e. smaller than the single - site decay rate ) . though originally discovered in the context of atomic clouds interacting with the electromagnetic field @xcite , in presence of many excitations , superradiance was soon recognized to be a general phenomenon in open quantum systems @xcite under the conditions of coherent coupling with a common decay channel . most importantly , it can also occur in presence of a single excitation ( the super of superradiance @xcite ) , entailing a purely quantum effect . the functional role that superradiance might have in natural photosynthetic systems has been discussed in many publications @xcite , and experimentally observed in molecular aggregates @xcite . superradiance ( or supertransfer ) is also thought to play an important role w.r.t . the transfer of excitation to the central core absorber @xcite , and its effects on the efficiency of energy transport in photosynthetic molecular aggregates have been recently analyzed @xcite . the origin of superradiance lies in the fact that the excitation can be coherently spread over several sites , thus inducing a cooperative effect . on the other hand , static disorder is expected to destroy superradiance , since it induces localization @xcite , which implies that excitons are localized on one site only , thus hindering cooperativity . the main question we want to address here is whether a critical disorder exists at which superradiance and , thus , cooperativity are destroyed . the relation between superradiance and localization have been already analyzed in literature in different contexts @xcite . in particular , in @xcite the case of weak coupling to the continuum ( weak opening ) has been analyzed for one - dimensional systems . it has been already analyzed also by some of the authors of the present paper : in @xcite the case of open 1-d and 3-d anderson models in the strong opening regime was considered . it was pointed out there that the sensitivity to disorder can be very different for superradiant and subradiant states : while the latter localize at the same critical disorder of the closed system ( i.e. a system with no coupling to the continuum of states ) , the former localize at the critical disorder for which superradiance is quenched . interestingly , though subradiant states essentially localize at the localization threshold associated with the closed system , they display some peculiar features due to opening , being neither fully localized nor extended ( hybrid states ) @xcite . in this paper we aim to study both the regimes of weak and strong opening and their effects on localization in one - dimensional nanostructures . even if it is easy to imagine that opening and disorder have competing effects on the efficiency of energy absorption and transfer , a deeper analysis is necessary to fully understand their action . for instance , disorder decreases the efficiency of the superradiant states in absorbing light or in transferring excitations , but , at the same time , it can allow for energy absorption and transfer from the subradiant states . thus , for these states , disorder is useful to enhance efficiency . the latter effect is strongly related to the enhancement of efficiency due to noise : the so called noise - assisted transport , discussed in @xcite . noise - assisted transport constitutes a general phenomenon in quantum networks , even if its relation with subradiance has never been stressed up to now , to the best of our knowledge . the plan of the paper is the following : in sec . [ sec:2 ] we introduce the model , in sec . [ sec:3 ] we derive analytically the critical disorder strength needed to quench superradiance , identifying the different regimes of weak and strong opening . in sec . [ sec:4 ] we analyze in detail the relation with localization , while sec . [ sec:5 ] is devoted to study the consequences of the previous findings on the system dynamics . a brief discussion about the relevance to photosynthetic complexes is given at the end of each section . sites coupled with nearest - neighbors tunneling transition amplitude @xmath2 . all sites are connected to a common decay channel , where the excitation can escape , with an equal coupling strength @xmath3 . , width=302 ] we considered a 1-d chain of sites with periodic boundary conditions , arranged to form a ring - like structure , as shown in fig . [ ring ] , where the excitation can hop from site to site . the model is characterized by the following tight binding hamiltonian : @xmath4 where the summation index @xmath5 runs over the pairs of nearest - neighbor sites and @xmath6 is the tunneling transition amplitude . here @xmath7 represents a state in which the excitation is at the site @xmath8 , while all the other sites are unoccupied . in terms of two - level system states ( @xmath9 the eigenvalues @xmath10 and the eigenstates @xmath11 of the system can be computed exactly . concerning the components of the eigenstate @xmath11 on the site basis @xmath12 , one has @xmath13 for @xmath14 , and @xmath15 for @xmath16 . the ground state , corresponding to @xmath17 and energy @xmath18 , is fully symmetric and extended in the site basis : @xmath19 the 1-d anderson model can be `` opened '' by allowing the excitation to escape the system from any site into the same continuum channel . this situation of `` coherent dissipation '' can be met in many systems and it has been recently considered in @xcite , where it has been shown to give rise to the following effective non - hermitian hamiltonian ( see also @xcite ) : @xmath20 where @xmath21 are the transition amplitudes from the discrete state @xmath22 to the continuum channel @xmath23 . in our case , we have a single decay channel , @xmath24 , and equal couplings , @xmath25 , so that @xmath26 @xmath27 . the quantum evolution is given by the operator @xmath28 which is non - unitary , and gives rise to a loss of probability in the decay channel . the complex eigenvalues of @xmath29 can be written as @xmath30 , where @xmath31 represent the decay widths of the eigenstates . usually , in molecular aggregates , energy is measured in units of @xmath32 , corresponding to energy divided by @xmath33 . in these units , time is measured in @xmath34 which corresponds to the mapping @xmath35 ( @xmath36 is the speed of light ) . in the following all units of energy will be given in @xmath32 and in order to have time in @xmath37 we need to divide it by @xmath38 . due to its specific structure , the operator @xmath39 has only one eigenstate with a non - zero eigenvalue : this is the fully extended state with eigenvalue equal to @xmath40 . this eigenstate also corresponds to the ground state of @xmath41 , given in eq . ( [ sr ] ) . all the other eigenstates of @xmath39 are degenerate with null eigenvalue and , since @xmath42 = 0 $ ] , they can be chosen to match the eigenstates @xmath11 , @xmath43 , of @xmath41 . this implies that only the state @xmath44 , eq . ( [ sr ] ) , has a non - vanishing decay width equal to the total decay width of the system : @xmath45 . this is the superradiant state . note that the dependence on @xmath40 of that decay width is the hallmark of the cooperative nature of superradiance . all the other states with zero decay width are called subradiant . the full expression for the complex eigenstates of the non - hermitian hamiltonian , eq.([amef ] ) , is given in appendix @xmath46 , see eq.([ceig ] ) . importantly , the superradiant effect might explain the strong dependence on the initial state of the efficiency of energy transfer to a central core absorber discussed in ref . @xcite . several features of the model above in absence of disorder are quite atypical . indeed , superradiance , as discussed in many papers @xcite , is usually reached only above a critical coupling strength with the continuum ( in the overlapping resonance regime ) when @xmath47 where @xmath48 is the average decay width and @xmath49 is the mean level spacing of the closed system . on the other hand , _ we are in a superradiant regime for any @xmath50 _ , even if the overlapping resonance condition is not satisfied . moreover , the widths of the subradiant states are usually small , but not zero as in this case . this is also due to the particular symmetric configuration chosen , from which it follows that @xmath41 and @xmath39 commute . note that such geometrically - induced subradiant subspaces with zero decay width are equivalent to the trapping - free subspaces discussed in literature @xcite . finally , let us notice that the presence of a superradiant regime for any coupling strength to the continuum might indicate a relation between structure and function in natural complexes , and it might also suggest the use of ring - like structures to exploit the superradiant behavior . the peculiar features discussed above disappear when we introduce the diagonal disorder , described by adding to the hamiltonian , eq . ( [ amef ] ) , the term @xmath51 where the random diagonal energies @xmath52 are taken uniformly distributed in @xmath53 $ ] , @xmath54 being the disorder strength . with the addition of this term our model becomes equivalent to a 1-d open anderson model as considered in ref . @xcite . the presence of static disorder on the site energies breaks the symmetry of the system under rotations , inducing the width of the superradiant state to decrease and the widths of the subradiant states to increase ( the total decay width @xmath55 is a constant that does not depend on the degree of disorder @xmath54 ) , so that all of the eigenstates can decay into the continuum channel . the effect of static disorder on the decay widths has been analyzed in fig . [ gw ] , where the width of the superradiant state and the average width of all subradiant states are shown as a function of the disorder strength @xmath54 , for different parameters value . as one can see , for small disorder , the effect on subradiant states is much more evident than that on the superradiant state . for large disorder , all widths approach the value @xmath3 , corresponding to the decay width of an isolated site . in this regime there is no collective behavior anymore and superradiance is completely quenched . _ vs _ the disorder strength , @xmath54 , for @xmath56 ( circles ) , @xmath57 ( squares ) , and @xmath58 ( crosses ) . black symbols ( upper curve ) refer to the average over disorder of the largest width ( the superradiant state ) . red symbols ( lower curve ) refer to the average over disorder of the mean subradiant width ( mean taken over the @xmath59 smallest widths ) @xmath60 . as a dashed line we plot the perturbative result , eq . ( [ gsub ] ) , for @xmath61 . the horizontal dot - dashed line indicates the value @xmath62 . the critical disorder strength , eq . ( [ wlimit ] ) , is indicated as the intersection between the line given by perturbation theory and the horizontal line , see text for details . , width=302 ] for small disorder , it is possible to use perturbation theory ( see appendix [ app : a ] ) to obtain the mean decay width of the @xmath59 smallest widths : @xmath63^{-1}\ , . \end{aligned } \label{gsub}\ ] ] the sum in eq . ( [ gsub ] ) can be well approximated in different parameter regimes to give ( see appendix [ app : b ] ) @xmath64 let us name _ weak opening _ the regime characterized by @xmath65 and _ strong opening _ the one in which @xmath66 . the different regimes shown above can be understood if we consider that @xmath67 is the minimal nearest neighbor energy distance , see eq . ( [ eigq ] ) . in ref . @xcite a perturbative result was obtained in the regime of weak opening , @xmath68 , which agrees with our findings . as one can see from eq . ( [ gsub2 ] ) , the average subradiant width in any regime increases as @xmath69 , but the dependence on the system size , @xmath40 , and on the degree of opening , @xmath3 , changes : in the weak opening regime , the widths increase with @xmath40 and @xmath3 , whereas , for very strong opening , they decrease with @xmath40 and @xmath3 . in fig . [ gw ] the perturbative expression is shown as a dashed line and agrees very well with numerical data . from eq . ( [ gsub ] ) one can define a critical disorder strength , @xmath70 , at which superradiance is quenched , given by the condition @xmath71 from which one gets @xmath72^{-1 } } } \ , . \label{wcr}\ ] ] for @xmath73 , all of the widths become essentially the same and equal to @xmath3 , while below @xmath70 they strongly depend on the chosen state . usually , the transition between these two regimes , which corresponds also to a transition from a non - cooperative to a cooperative regime , is referred to as superradiance transition ( st ) in literature @xcite . even if the validity of eq . ( [ wcr ] ) has been shown in fig . [ gw ] only in the weak opening regime , we checked that it gives an excellent estimate of the disorder at which superradiance is quenched also for strong opening . from eq . ( [ wcr ] ) it is possible to get an approximate expression ( see appendix [ app : b ] ) for the critical disorder strength , @xmath70 , in the different regimes : @xmath74 the results contained in eq . ( [ wlimit ] ) are very interesting , since they show that in some region of parameters ( typically small system size and weak opening ) the critical disorder at which superradiance is quenched is independent of @xmath3 ( a quantity often difficult to be experimentally determined ) , while it decreases with the system size as @xmath75 . this independence from @xmath3 is also shown in fig . [ gw ] where we plotted data obtained with different values of @xmath3 , for which @xmath76 . in particular , for the largest value of @xmath3 considered in fig . [ gw ] , @xmath77 , we have @xmath78 . the existence of a regime ( weak opening ) in which the critical disorder strength is independent of @xmath3 could be surprising . indeed , applying the overlapping resonance criterion , eq . ( [ orc ] ) , one would obtain @xmath79 , since @xmath80 and @xmath81 . an explanation of this effect , due to localization , will be given in the next section . a second remarkable result is the linear dependence of @xmath70 on @xmath40 and @xmath3 in the strong opening regime . since , on increasing @xmath40 , one always enters the strong opening regime , it is possible to preserve the cooperative nature of superradiant states up to arbitrarily large disorder . we may thus say that the opening induces a cooperative robustness to disorder , as was also recently found by some of the authors of this paper @xcite . it is interesting to observe that also in the case of dynamical disorder it was found @xcite that the critical dephasing necessary to destroy the cooperative superradiant effects is proportional to both @xmath40 and @xmath3 . note that this regime was not analyzed in ref . @xcite , where it was stated that the critical disorder needed to quench superradiance does not depend on the superradiant decay rate . from eq . ( [ wlimit ] ) we can infer that the dependence of @xmath70 on the system size @xmath40 is non - monotone . setting @xmath82 , we can roughly estimate the @xmath40 value at which @xmath70 has a minimum : @xmath83 of sites in the ring for different values of the coupling strength @xmath3 , as indicated in the legend . the numerical data ( symbols ) are compared with the analytical expression ( curves ) for the critical disorder strength , eq . ( [ wcr ] ) . arrows indicate the size @xmath84 at which critical disorder is minimal , see eq . ( [ ncr ] ) . , width=302 ] to confirm the validity of the critical disorder strength , @xmath70 , computed above , as the value at which the st occurs , we computed the variance of the decay widths . indeed , it is well known @xcite that , at the st , the variance of the widths has a maximum . the results of such a comparison are presented in fig . [ varg ] , showing a good agreement between the two estimates of the st . finally , it is interesting to estimate the value of @xmath70 for the photosynthetic complexes lhi and lhii . in this case @xmath85 and @xmath86 , respectively @xcite . the coupling with the electromagnetic field can be estimated from the radiative decay time @xmath87 of a single molecule , which is of the order of few nanoseconds @xcite . hence we get @xmath88 . on the other side , for the lhi complex , the common decay channel can also represent the reaction center . this coupling can be estimated to be @xmath89 from the mean transfer time to the reaction center of the lhi complex @xcite , as discussed at the end of sec . [ sec:5 ] . both these couplings are very weak if compared to the energy scale of @xmath2 , so that we can assume that we are in a weak opening regime , @xmath90 , where @xmath70 does not depend on @xmath3 and it will then be the same for both environments . we can thus use @xmath91 , see eq . ( [ wlimit ] ) , getting @xmath92 for lhi and @xmath93 for lhii . these values of disorder are in agreement with the experimental observation that static disorder in lhii complexes is @xmath94@xmath95 times larger than the value of disorder in lhi complexes @xcite . those values are also quantitatively compatible with the estimated ranges of static disorder strength in natural photosynthetic complexes ( @xmath96@xmath97 for lhi complexes @xcite , @xmath98@xmath99 for lhii complexes @xcite . to make a comparison with the data contained in these references , one should take into account that they considered gaussian static disorder with a standard deviation @xmath100 , so that @xmath101 ) . these estimates might suggest that natural photosynthetic complexes operate close to the st . in the previous section we analyzed how diagonal disorder modifies the decay widths of the states . on the other hand , it is well known that disorder in isolated tight binding models induces anderson localization @xcite . in 1-d systems any disorder strength induces localized eigenstates , @xmath102 , where @xmath8 labels the position of the sites on the lattice and @xmath103 is the localization length , measured in units of intersite distance . the localization length is , in general , a function of the disorder strength @xmath54 and of the energy @xmath104 . in particular , it is well known that , for weak disorder and away from the edges of the energy band , @xmath105 . therefore , it is possible to define a critical disorder @xmath106 for the localization effect to be important by the simple equation @xmath107 indeed , while any increase of the disorder strength will produce eigenstates with a localization length smaller than the sample size , decreasing @xmath54 gives rise to eigenstates with a localization length larger than the system size , i.e. effectively delocalized . _ vs _ the disorder strength @xmath54 . upper panel represents the weak opening regime , namely @xmath108 , while the lower panel depicts the strong opening regime , @xmath109 . in both panels the blue open circles represent the behavior of the superradiant state as a function of the disorder strength , while the red full circles stand for the average @xmath110 of the subradiant states . vertical dashed lines represent in both panels the localization condition given by eq . ( [ loc88 ] ) ( red ) , and @xmath70 ( blue , eq . ( [ wlimit ] ) ) . , width=302 ] the interplay of disorder and opening can be studied by means of the participation ratio @xmath111 of the eigenstates @xmath112 of @xmath113 , given in eq . ( [ amef ] ) , where @xmath114 stands for the ensemble average over different realizations of the static disorder . the @xmath110 is widely used to characterize localization properties @xcite and it clearly satisfies the bounds @xmath115 . for extended states , it increases proportionally to the system size @xmath40 , while , for localized states , it is independent of @xmath40 . our aim is to compare the disorder strength at which superradiance is quenched , @xmath70 , see eq . ( [ wcr ] ) , with the disorder strength at which the states localize , @xmath106 , see eq . ( [ loc88 ] ) . to do that , we analyze separately the @xmath110 of the superradiant state ( the state with the maximum decay width ) and the average @xmath110 of the other @xmath59 states as a function of @xmath54 . the typical behavior of the @xmath110 as a function of the disorder strength @xmath54 has been analyzed in fig . [ prring ] in two different situations : for weak opening ( @xmath116 , upper panel ) , and for strong opening ( @xmath117 , lower panel ) . in both cases , the @xmath110 of the superradiant state decreases roughly at the st , as given by @xmath70 , while the @xmath110 of subradiant states decreases roughly at @xmath106 . decreases its value by @xmath118 with respect to the value at @xmath119 , as a function of the system size , @xmath40 , for the superradiant state ( upper panel ) and for the average @xmath110 over all the other states ( lower panel ) . circles stand for the open system ( @xmath120 ) , crosses for the closed system ( @xmath121 ) . here is @xmath122 . in the upper panel the full curve stands for the analytical @xmath70 , given by eq . ( [ wlimit ] ) , rescaled by a factor @xmath123 to fit numerical data . in the lower panel a curve proportional to @xmath124 has been drawn to guide the eye ( for the explanation see text ) . vertical dashed lines mark @xmath84 , see eq . ( [ ncr ] ) , and separate the weak opening regime ( left ) from the strong opening regime ( right ) . , width=321 ] to be more quantitative , we numerically computed , for the superradiant state and for the subradiant states , the disorder strength @xmath125 at which their @xmath110 decreases by 3% w.r.t . the value at zero disorder . to highlight the peculiar effects due to opening , these results should be compared with those for the closed system ( @xmath121 ) . for the closed system we can not define superradiant and subradiant states , but , since the localization length depends on the energy level , we can compare states of the open system with states of the closed system having the same real energy . in particular , the superradiant state is compared with the ground state of the closed system . results are shown in fig . [ wcrpr1 ] for the superradiant state ( upper panel ) and for the subradiant ones ( lower panel ) as a function of the system size @xmath40 . in this figure we fix @xmath3 and , by varying @xmath40 , we switch from the weak opening regime ( for small @xmath40 values ) to the strong opening regime ( for large @xmath40 ) . the @xmath84 value which separates the two different regimes can be estimated from eq . ( [ ncr ] ) , and has been indicated as a dashed vertical line in both panels . the opening does not modify the behavior of the subradiant states if compared with the behavior of the closed system , compare circles with crosses in the lower panel of fig . [ wcrpr1 ] . in particular , for non - edge states of the closed system @xcite @xmath126 and , from eq . ( [ loc88 ] ) , one gets that the disorder strength at which states localize scales as @xmath127 the same dependence on @xmath40 is found in presence of opening and it has been indicated for the sake of comparison in fig . [ wcrpr1 ] , lower panel . let us now analyze the behavior of superradiant states , fig . [ wcrpr1 ] upper panel . in the weak opening regime , @xmath128 , the open and the closed model display the same behavior , while in the strong opening regime , @xmath129 , the behavior is very different : in this regime @xmath125 decreases with @xmath40 for the closed model , while it increases with @xmath40 for the open one . even if the behavior of the superradiant state of the open system in the weak and strong opening regimes is very different , it is always captured by the disorder strength at which superradiance is quenched , @xmath70 , see eq . ( [ wlimit ] ) . indeed , the disorder strength at which the superradiant state starts lo localize ( phenomenologically described by @xmath125 ) scales with the parameters as @xmath70 ( compare full line with symbols in upper panel of fig . [ wcrpr1 ] ) . this fact allows us to understand the scaling of @xmath125 with @xmath40 in both regimes : @xmath130 in the weak opening regime , while @xmath131 in the strong opening regime . note that the dependence @xmath130 is the same as that of the disorder strength necessary to localize the edge states of the closed system @xcite , for which we have @xmath132 and , from eq . ( [ loc88 ] ) , we obtain a disorder strength scaling as @xmath75 . the different sensitivity of super and subradiant states to disorder is far from being trivial . due to the fact that the @xmath39 matrix in eq.([amef ] ) is a full matrix , the opening induces a long - range hopping which contrasts localization , and one might expect such long range to affect all states equally . on the other hand , the correlated nature of the long range hopping implies that only superradiant states are affected , leaving the subradiant states effectively decoupled from the environment and thus behaving more similarly to the states of the closed system . for more details see ref . @xcite . summarizing we can conclude that : * the disorder strength necessary to localize the subradiant states is the same of the corresponding value for the closed system ; * the disorder strength necessary to localize the superradiant states is proportional to the disorder strength necessary to quench superradiance , @xmath70 ; * in the weak opening regime , the quenching of superradiance is determined only by the localization properties of the closed model , resulting in a @xmath70 independent of @xmath3 ; * in the strong opening regime , superradiance is quenched at a critical disorder proportional to @xmath55 , as in the case of time - dependent disorder @xcite , thus showing its cooperative robusteness to disorder ; * for the realistic parameters of natural photosynthetic complexes , such as lhi and lhii ( see end of sec . [ sec:3 ] ) , we are in the weak opening regime , so that it is possible to determine @xmath70 only by analyzing the localization properties of the closed system . this fact can be very useful since the exact value of @xmath3 is not easy to be determined experimentally . , @xmath133 , @xmath122 , for different disorder strengths as indicated in the legend . numerical results are shown with symbols , while full curves show the analytical expression , eq . ( [ pth ] ) . the arrow shows the value of @xmath134 obtained analytically by eq . ( [ ts ] ) for @xmath135 . for comparison , both the exponential decay of the superradiant state for zero disorder ( dashed ) and the decay at large disorder ( dotted ) are shown . , width=302 ] in this section we aim at studying how the time evolution of the survival probability @xmath136 ( that is the probability of finding the excitation in the system , initially prepared in some state @xmath137 ) is modified by the presence of static disorder . that quantity is given by @xmath138 let us choose @xmath139 , the fully extended state of eq . ( [ sr ] ) , for our first analysis . for @xmath119 we clearly have @xmath140 , since the fully extended state is the only one with a decay width , see eq.([ceig ] ) . for @xmath141 the fully extended state does not coincide with the superradiant state anymore , and it should be written as a superposition of superradiant and subradiant states . using first order perturbation theory ( in the disorder strength @xmath54 ) we can derive an approximate expression for @xmath136 valid for small time , see appendix [ app : a ] , eq . ( [ c1 ] ) : @xmath142 eq . ( [ pth ] ) takes into account only the superradiant decay and the fastest subradiant decay , @xmath143 , which can be computed from eq . ( [ gq ] ) given in appendix [ app : a ] , setting @xmath144 . in fig . [ ptth ] we compared numerical data for @xmath136 with the analytical expression given in eq . ( [ pth ] ) : the agreement is excellent for disorder strength @xmath145 , where the decay is well approximated by a bi - exponential function . from eq . ( [ pth ] ) it is also possible to compute the time at which a change in the decay occurs , @xmath146 , by equating the two terms on the r.h.s . of eq . ( [ pth ] ) . dividing by @xmath147 in order to have @xmath146 in picoseconds , we obtain @xmath148 such a time , for one value of the disorder strength , is shown with an arrow in fig . [ ptth ] . note that @xmath146 can be considered as the time up to which the decay is superradiant . as the disorder increases , @xmath146 goes to zero and the decay of the extended state becomes similar to the decay of independent sites , i.e. @xmath149 . , rescaled to the individual decay width @xmath3 , as a function of the rescaled disorder strength @xmath150 . different sets of values have been considered : full circles are for @xmath151 , @xmath152 , @xmath153 , @xmath154 ; open circles are for @xmath155 , @xmath133 , @xmath156 , @xmath157 . the dashed red line represents @xmath158 . , width=302 ] the generality of our results can be assessed by observing that the critical disorder at which superradiance is quenched is an important threshold for the whole system dynamics and not only for the superradiant state . to this end , let us consider as initial state a random superposition of site states @xmath159 @xmath160 being random complex coefficients such that @xmath161 . for such initial state , we compute the survival probability @xmath136 , for one realization of disorder . by changing the random initial state and the random diagonal disorder , we can consider the average survival probability @xmath162 and define the decay time @xmath163 as @xmath164 these @xmath165 values are reported in fig . [ dyn ] as a function of @xmath150 for different parameter values . for the sake of comparison we show , in the same figure , the analytical expression for the average decay width of the subradiant states , see eq . ( [ wcr ] ) . as one can see , up to a numerical constant , the agreement is very good . in other words , the disorder strength necessary to quench superradiance ( obtained analytically imposing the average decay width of the subradiant states , @xmath60 , to be equal to the single site decay @xmath3 ) is also a valid tool in estimating the decay time of the survival probability associated with generic random initial conditions . the problem of computing the survival probability of the superradiant state in presence of inhomogeneous broadening was also considered in @xcite for @xmath40 two - level systems . a bi - exponential behavior was numerically found for any excitation number . for the case where only one excitation is present , our results are compatible with the bi - exponential behaviour of the survival probability found in @xcite , and we also give an approximate analytical expression for the survival probability @xmath136 . finally , in order to stress the relevance of superradiant energy transfer in natural photosynthetic complexes , let us consider the excitation transfer from the lhi complex to the reaction center @xcite . first of all , we point out that all models used to study the dynamics of this complex are characterized by a large inhomogeneity in the transfer time of different energy eigenstates @xcite , which is , for instance , typical of the superradiant regime . thus , we reasonably assume superradiance in transfer to be relevant in natural complexes . we can estimate @xmath3 , representing the coupling to the reaction center , from the following considerations . using realistic parameters as was done at the end of sec . [ sec:3 ] , @xmath85 , @xmath166 , we computed @xmath136 for the fully extended state of eq . ( [ sr ] ) , in presence of a realistic value of the static disorder @xmath167 ( a large disorder corresponding to the critical disorder at which superradiance starts to be quenched ) . we choose @xmath168 , so that we have a transfer time ( time at which @xmath169 ) starting from the fully extended state of @xmath170 , in agreement with experimental data @xcite . note that a single occupied site would give a transfer time of @xmath171 , showing that , even in presence of strong and realistic static disorder , superradiance is able to strongly enhance energy transfer . we analyzed the interplay of superradiance , induced by the coupling to a common decay channel , and localization , induced by static disorder , in 1-d ring - like structures , usually used to model some natural light - harvesting complexes . the common decay channel can represent both the coupling to the electromagnetic field or to a central core absorber , such as the reaction center in natural photosynthetic complexes . we have shown that , for zero disorder , these structures are in a superradiant regime for any value of the coupling strength to a common decay channel . above a critical disorder strength superradiant effects decrease until , for very large disorder , all of the states decay independently with the common width @xmath3 , and cooperativity is completely lost . our main purpose was to determine the critical disorder at which superradiance is hindered . using a perturbative approach we determined analytically such critical disorder and we related it firstly with the localization properties of superradiant and subradiant states and then to the system dynamics . we found that superradiance can be quenched by disorder in different ways , depending on the regime entailing either weak or strong coupling to the continuum . these regimes are triggered by the parameter @xmath172 , which represents the ratio between the coupling strength to the continuum , @xmath3 , and the unperturbed mean level spacing in absence of disorder , @xmath173 . when this ratio is small , i.e. @xmath174 , the critical disorder is independent of the coupling strength with the external environment and it is determined only by the parameters of the molecular chain , since the opening is unable to affect the disorder - induced localization . in this regime , the critical disorder decreases with the size of the system , but , for large system size @xmath175 , such a regime becomes less and less feasible ( to be in the weak opening regime implies the condition @xmath176 ) . on the other hand , for strong opening , @xmath177 , the critical disorder increases with both the size of the system and the coupling strength with the external environment this is in agreement with the results recently found in ref . @xcite , where the same ring structure has been analyzed in presence of dephasing ( dynamical disorder ) and the strength necessary to destroy superradiance was found to be proportional to both @xmath3 and @xmath40 . we also demonstrated that the critical disorder at which superradiance is suppressed is close to the disorder at which superradiant states localize @xcite . specifically , we found that , in the weak opening regime , @xmath178 , superradiance is quenched at the same disorder at which the edge state of the closed system , with real energy equal to that of the superradiant state , localizes . we have also found that , in the strong opening regime , @xmath177 , superradiance is a manifestation of cooperative robustness to disorder , in that the superradiant state localizes at a disorder strength ( proportional to the system size ) much larger than the one needed to localize the corresponding edge state of the closed system . as for subradiant states , in any regime , they begin their process of localization at the same disorder strength at which the states of the closed system do . finally , we have shown the relevance of our findings to natural photosynthetic complexes : @xmath0 for the realistic parameters of natural complexes , superradiance is quenched at a disorder strength which is independent of the coupling to the external environment ( electromagnetic field or reaction center ) , which is difficult to determine experimentally . thus our findings allow to determine the critical disorder from the localization properties of the closed system alone ; @xmath1 the critical disorder thus obtained is compatible with experimental estimates , suggesting that natural systems operate close to the superradiance transition ; @xmath179 even in presence of large and realistic static disorder , superradiance can strongly enhance energy transfer to the reaction center and light absorbtion . _ _ we would like to thank r. kaiser and b. sterzi for providing many useful discussions . perturbation theory is applied to the _ symmetric unperturbed hamiltonian , _ @xmath180 in order to find the critical disorder strength , @xmath70 , at which superradiance is destroyed . let us rewrite the hamiltonian given in eq . ( [ amef ] ) as @xmath181 where @xmath182 is the tight binding hamiltonian , eq . ( [ am ] ) , in absence of disorder and @xmath183 , see eq . ( [ dam ] ) , is a diagonal matrix which contains the disordered site energies @xmath184 . it is necessary to define the non - hermitean `` bra '' as the transposed of a ket @xmath185 while the standard bra is the adjoint @xmath186 indeed , given the right eigenvectors of a symmetric hamiltonian @xmath187 , the left eigenvectors are @xmath188 , that is @xmath189 and we have the _ biorthogonality _ condition @xmath190 clearly , for real eigenstates we have @xmath191 . matrix elements of the operators defined above , in the site basis @xmath192 , are given by , @xmath193 with @xmath194 . ( [ eq : defh0 ] ) @xmath6 , and @xmath195 , are independent identically distributed random variables with mean @xmath196 and variance @xmath197 . since @xmath198=0,$ ] it is convenient to study the whole system on the basis of eigenstates of @xmath41 , which are given by @xmath13 for @xmath14 , and @xmath15 for @xmath16 , with eigenvalues @xmath199 the eigenvalues of the hamiltonian , @xmath200 , are thus given by , @xmath201 that is , only the ground state acquires a decay width @xmath55 . such a state is called _ superradiant _ , and the others are _ subradiant_. notice that @xmath202 and @xmath203 are non - degenerate , while , for any @xmath204 , @xmath205 and @xmath206 span a two - dimensional degenerate eigenspace . when the disorder strength is turned on every state will get an eigenenergy with a negative imaginary part ( decay width ) . perturbation theory up to second order can be applied , for sufficiently small disorder strength , to give @xmath207 for degenerate energy levels the first order correction is given by the eigenvalues of the 2 by 2 symmetric matrices @xmath208 while the second order correction is @xmath209 we are interested in the imaginary part of the perturbed eigenvalues , and , since the eigenstates , @xmath205 , are _ real _ , first order corrections never contribute to those terms . considering now averages over disorder and writing @xmath210 , with @xmath211 real , we obtain the average decay widths for the superradiant state @xmath212 and , for the subradiant ones , @xmath213 , @xmath214 the maximum decay widths of the subradiant states are clearly @xmath215 and @xmath216 , while the average of the subradiant widths is @xmath217 we finally define the critical disorder @xmath70 as the one at which @xmath218 i.e. equals the single - site decay width @xmath3 . let us now apply first order perturbation theory to the superradiant state . for @xmath119 , the superradiant state is given by the extended state , eq . ( [ sr ] ) , while , for @xmath141 , we can write @xmath219 from this expression we can compute the probability to be in the superradiant state when starting from the extended state as @xmath220 let us rewrite eq . ( [ pte ] ) in the form @xmath221 where we have defined @xmath222 and @xmath223 eq . ( [ essea ] ) can be put in integral form for sufficiently large @xmath40 , that is @xmath224 , as @xmath225 it is easy to show that eq . ( [ esseaint ] ) has two different limits , namely @xmath226 nevertheless , substituting a sum with an integral works only for very large @xmath40 . for small @xmath227 , or , in general for a sufficiently small @xmath228 value , it is more convenient approximating the sum with only two terms , namely those for which the denominator in eq . ( [ essea ] ) is small . in detail , one has @xmath229 this implies that , in this regime , @xmath230 on the other hand , for @xmath231 ( for all those @xmath40 values for which @xmath232 ) , one can approximate the sum with the integral and use the asymptotic behavior given in eq . ( [ asymp ] ) . to summarize we have the following behavior : @xmath233 x. hu , t. ritz , a. damjanovic , and k. schulten , j. phys . b * 101 * , 3854 ( 1997 ) ; x. hu , a. damjanovic , t. ritz , and k. schulten , pnas * 95 * , 5935 ( 1998 ) ; t. ritz , s. park , and k. schulten , j. phys . b * 105 * , 8259 ( 2001 ) . v. v. sokolov and v. g. zelevinsky , nucl . a504 * , 562 ( 1989 ) ; phys . b * 202 * , 10 ( 1988 ) ; i. rotter , rep . 54 * , 635 ( 1991 ) ; v. v. sokolov and v. g. zelevinsky , ann . ( n.y . ) * 216 * , 323 ( 1992 ) . g. l.celardo , f. borgonovi , v. i. tsifrinovich , m. merkli , and g. p. berman , j. phys . c * 116 * , 22105 ( 2012 ) . d. ferrari , g. l. celardo , g. p. berman , r. t. sayre , f. borgonovi , j. phys . c * 118 * , 20 ( 2014 ) . g. l. celardo , f. m. izrailev , v. g. zelevinsky , and g. p. berman , phys . lett b * 659 * , 170 ( 2008 ) ; g. l. celardo , f. m. izrailev , v. g. zelevinsky , and g. p. berman , phys . e , * 76 * , 031119 ( 2007 ) ; g. l. celardo , s. sorathia , f. m. izrailev , v. g. zelevinsky , and g. p. berman , cp995 , nuclei and mesoscopic physics - wnmp 2007 , ed . p. danielewicz , p. piecuch , and v. zelevinsky .

there is a mounting evidence that the galactic bulge is a triaxial structure , or a bar . this was first postulated by de vaucouleurs ( 1964 ) , based on similarities between the kinematics of the gas observed towards the galactic center and in other barred galaxies . however , the hypothesis that our galaxy is a barred galaxy was for a long time overshadowed by `` @xmath4 expanding arm '' hypothesis , despite a number of papers arguing for bar s presence ( e.g. peters 1975 ; liszt & burton 1980 ; gerhard & vietri 1986 ) . only recently has the view of our galaxy as a barred spiral gained momentum , mostly due to the work of blitz & spergel ( 1991 ) . they analyzed @xmath5 observations of the galactic center and showed convincingly that the observed asymmetry in the galactic longitude distribution of surface brightness is naturally explained by the bar with the near side in the first galactic quadrant . binney et al . ( 1991 ) have constructed a dynamical model for gas in the inner galaxy , and their resulting bar has the same orientation as that suggested by blitz & spergel ( 1991 ) in the sense that the closer part of the bar is at positive galactic longitudes . cobe - dirbe multiwavelength observations of the galactic center ( weiland et al . 1994 ) confirmed the existence of the longitudinal asymmetry discussed by blitz & spergel ( 1991 ) . this data was used by dwek et al . ( 1995 ) to constrain a number of analytical bar models existing in the literature . star counts have also shown evidence for triaxial structure in the center of the galaxy . nakada et al . ( 1991 ) analyzed the distribution of iras galactic bulge stars and found asymmetry in the same sense as blitz & spergel ( 1991 ) . whitelock & catchpole ( 1992 ) analyzed the number distribution of mira variables in the bulge as a function of distance modulus and found that the half of the bulge which is at positive galactic longitude is closer to us than the other half . the observed stellar distribution could be modelled with a bar inclined at roughly @xmath6 to the line of sight . weinberg ( 1992 ) used agb stars as star tracers and mapped the galaxy inside the solar circle . he found evidence for a large stellar bar with semimajor axis of @xmath7 and inclination placing the nearer side of the bar at positive galactic longitudes . a third thread of evidence comes from the gravitational microlensing towards the galactic bulge ( udalski et al . 1994 ; alcock et al . 1995 ; alard et al . 1995 ) . observed high microlensing rate can be accounted for by stars placed in the near part of the galactic bar microlensing the stars from the far side of the bar ( kiraga & paczyski 1994 ; paczyski et al . 1994b ; zhao , spergel & rich 1995 ) . there are many implications if this scenario is correct , discussed latter in this paper . for recent reviews on the galactic bar see gerhard ( 1996 ) and kuijken ( 1996 ) . in this paper we use a color - magnitude data obtained by the optical gravitational lensing experiment collaboration for 12 fields scattered across the galactic bulge to construct the three - dimensional model of the mass distribution in the galactic bar . the optical gravitational lensing experiment ( ogle , udalski et al . 1993 ; 1994 ) is an extensive photometric search for the rare cases of gravitational microlensing of galactic bulge stars by foreground stars , brown dwarfs and planets . it provides a huge database ( szymaski & udalski 1993 ) , from which color - magnitude diagrams have been compiled ( udalski et al . stanek et al . ( 1994 ; 1996 ) used the well - defined population of bulge red clump stars to investigate the presence of the bar in our galaxy . comparing extinction - adjusted apparent magnitudes of the red clump giants observed at fields lying at @xmath8 , we found that the red clump stars lying at the positive galactic longitudes are systematically brighter by @xmath9 than the stars lying at negative @xmath1 . a bar - shaped bulge whose nearer side is at the positive galactic longitude most easily explains this offset . this agrees with blitz & spergel ( 1991 ) and other works . the paper is organized as follows . in section 2 we discuss the basis for using the red clump stars as a distance indicator . in section 3 we discuss the data used for the bar modelling . in section 4 we discuss various analytical models of the bar . in section 5 we constrain these models using our data . in section 6 we discuss various factors which might affect the fits . in section 7 we discuss some astrophysical implications of the constructed models , among the others for the gravitational microlensing optical depth . before we model the galactic bar using red clump stars , we shall discuss the evidence that the red clump stars would be a good distance indicator for the galactic bulge population . red clump stars are the equivalent of the horizontal branch stars for a metal rich population , i.e. relatively low mass stars burning helium in their cores . from stellar evolution theory ( e.g. seidel , demarque & weinberg 1987 ; castellani , chieffi & straniero 1992 ) we expect the red clump stars to have a narrow luminosity distribution with weak dependence on age and the metallicity . for given metallicity and helium abundance , the total difference in mean clump luminosity for ages between 1 and @xmath10 is @xmath11 ( seidel et al . the effect of metallicity on red clump giants luminosity and color is also relatively weak ( see sarajedini , lee & lee 1995 , their figs.4,5 ) . therefore , red clump stars form a suitable population with which to investigate the properties of high - metallicity systems , like the galactic bulge . because of high and variable interstellar extinction towards the galactic bulge , we use extinction - adjusted parameter @xmath12 in our analysis ( see eq.1 ) . we would like to be able to say something about the intrinsic width of the red clump population . for that , we use @xmath13 color - magnitude diagrams ( cmds ) of the globular cluster 47 tuc obtained by one of the authors ( jk ) using 2.5-meter dupont telescope at the las campanas observatory . these cmds are described in more detail by kauny & wysocka ( 1996 ) . 47 tuc is a metal rich globular cluster with @xmath14\approx-0.7 $ ] ( zinn & west 1984 ) and age of @xmath15 ( sarajedini et al . 1995 ) . we first construct a histogram of @xmath12 values for stars in the red clump dominated region of the 47 tuc cmd ( fig.[fig1 ] ) . the definition of the @xmath12 parameter causes that , along with the red clumps giants we are interested in , also stars from the red giant branch which happen to have the same values of @xmath12 are included into histogram . the resulting peak in the luminosity function is quite narrow for comparison we plotted a gaussian with @xmath16 . we want also to be able to estimate what are the effects of possible metallicity spread ( mcwilliam & rich 1994 ; minniti et al . 1995 ) on the red clump width . to do so , we selected one of baade s window fields observed by ogle , bw4 , which has relatively uniform extinction ( see stanek 1996 ) . we than shifted the cmd of 47 tuc to match the location of the red clumps giants in the bw4 . this is shown in the upper part of fig.[fig2 ] . we then plotted the histogram of @xmath17 color for red giant branch stars from both the cluster and the bw4 field ( fig.[fig2 ] , lower part ) . the red giant branch of 47 tuc is very narrow , @xmath18 , and it is much broader for the bw4 field , with @xmath19 . this combined with the @xmath16 measured in the previous paragraph gives us the intrinsic width of the red clump giants population in the galactic bulge @xmath20 . we will use this number later in this paper . we use a color - magnitude data obtained by the optical gravitational lensing experiment collaboration for 12 fields scattered across the galactic bulge to construct the 3-d model of the mass distribution in the galactic bar . the schematic map of the fields location in the galactic coordinates is shown in fig.[fig3 ] and the coordinates of the fields are given in table 1 . we use 10 color - magnitude diagrams ( cmds ) from udalski et al . ( 1993 ) ( fields bw1 - 6 , mm5-b , mm7-ab , tp8 ) and two cmds obtained last year specifically for the current project ( fields gb11 , gb13 ) . all the data were obtained using the 1-meter swope telescope at the las campanas observatory , operated by the carnegie institution of washington , and @xmath21 ford / loral ccd detector with the pixel size 0.44 arcsec covering @xmath22 field of view . we did not use fields bw7 and bwc of udalski et al . ( 1993 ) because they each contain a globular cluster , which could potentially contaminate our star counts . we also excluded fields bw8 and mm5-a because of very strong ccd bleeding columns caused by bright stars . most of each cmd is dominated by bulge stars , with a distinct red clump , red giant , and turn - off point stars . the part of the diagram dominated by disk stars for nine bw fields was analyzed by paczyski et al . ( 1994a ) . stanek et al . ( 1994 ) used well - defined population of bulge red clump giants in nine bw fields and four mm fields to find an evidence for the galactic bar . this paper is an extension of this earlier work . lrrrrr bw1 & 18:02:19.9 & @xmath2329:49:28 & 1.0764 & @xmath233.5894 & 5,496 bw2 & 18:02:20.6 & @xmath2330:15:18 & 0.7017 & @xmath233.8028 & 4,380 bw3 & 18:04:19.5 & @xmath2330:15:22 & 0.9103 & @xmath234.1769 & 3,865 bw4 & 18:04:18.8 & @xmath2329:49:12 & 1.2908 & @xmath233.9625 & 4,549 bw5 & 18:02:20.1 & @xmath2330:02:27 & 0.8878 & @xmath233.6961 & 5,382 bw6 & 18:03:20.8 & @xmath2330:15:10 & 0.8099 & @xmath233.9909 & 5,045 mm5b & 17:47:24.3 & @xmath2334:57:38 & @xmath234.9577 & @xmath233.4479 & 3,844 mm7a & 18:10:52.1 & @xmath2325:54:16 & 5.4272 & @xmath233.3385 & 3,572 mm7b & 18:11:39.4 & @xmath2325:55:08 & 5.4996 & @xmath233.5008 & 3,493 tp8 & 18:17:55.0 & @xmath2332:54:03 & @xmath230.0583 & @xmath237.9773 & 855 gb11 & 18:20:34.6 & @xmath2323:58:45 & 8.1703 & @xmath234.3672 & 2,339 gb13 & 17:46:58.5 & @xmath2323:01:14 & 5.2157 & 2.8113 & 5,144 to analyze the distribution of bulge red clump stars in a quantitative manner , we use the extinction - insensitive @xmath12 parameter ( paczyski et al . 1994a ; stanek et al . 1994 ) @xmath24 where we use reddening law @xmath25 , following woniak & stanek ( 1996 ) and stanek ( 1996 ) . the parameter @xmath12 has been defined so that if @xmath26 is independent of location then for any particular star its value is not affected by the unknown extinction ( woniak & stanek 1996 ) . then we consider only the region of the cmd clearly dominated by the bulge red clump stars : @xmath27 for one of the fields , tp8 , the average interstellar extinction is clearly lower than for the other fields , so for tp8 the selected region is : @xmath28 on the other hand , for one of the new fields , gb13 , the average interstellar extinction is clearly higher than for the other 11 fields , so for gb13 the selected regions is : @xmath29 please note that in eqs.([eq : select][eq : select1 ] ) the range of @xmath12 is always the same . if the value of the coefficient of the selective extinction is indeed constant throughout the galactic bulge , as found by woniak & stanek ( 1996 ) , then these different selections for tp8 and gb13 fields should not affect our calculations . all stars that satisfy the inequalities ( [ eq : select ] ) and ( [ eq : select2],[eq : select1 ] ) were counted in bins of @xmath30 . the result appears in fig.[fig4 ] , where we see the number of stars as a function of @xmath12 for all the fields , ordered by decreasing galactic longitude . please note the systematic shift between the red clump giants distribution as the @xmath1 decreases , as already seen by stanek et al . ( 1994 ; 1996 ) . please also note that the @xmath12 histogram of the field gb13 is not shifted comparing to the mm7 fields , which indicates that the coefficient of the selective extinction is indeed constant throughout the galactic bulge . we seek to reproduce the observed luminosity function in each of the 12 fields ( fig.[fig4 ] ) assuming some analytical function for the galactic bar density distribution and fitting some simple form of the intrinsic luminosity function , the same in all the fields , for the stars in the red clump region . for the galactic bar density distributions we follow dwek et al . ( 1995 ) approach and select three families of the analytical functions ( for detailed description see dwek et al . ) : 1 . _ gaussian type functions _ ( g ) , described by @xmath31 2 . _ exponential - type functions _ ( e ) , described by @xmath32 3 . _ power - law type functions _ ( p ) , described by @xmath33 where : @xmath34^{1/2 } \\ r_e & \equiv & \left(\frac{\left|x\right|}{x_0}+\frac{\left|y\right| } { y_0}+\frac{\left|z\right|}{z_0}\right ) \\ r_s & \equiv & \left\{\left[\left(\frac{x}{x_0}\right)^2+\left(\frac{y}{y_0}\right)^2 \right]^2+\left(\frac{z}{z_0}\right)^4\right\}^{1/4}\end{aligned}\ ] ] all these functions are allowed to be rotated in the plane of the galaxy by the angle @xmath35 , related to dwek s et al . angle @xmath36 by @xmath37 ( @xmath38 corresponds to the major axis of the bar pointing at us ) . we also allow for the tilt out of the galactic plane @xmath39 , corresponding to dwek s et al . angle @xmath39 . we assume the distance to the center of the galaxy @xmath40 . we parametrize the fitted luminosity function in the form @xmath41~\ ; \left[l^{-1}_{\odot}\right ] . \label{eq : lumfun}\ ] ] the power - law with index @xmath42 is intended to represent the underlying broad population of stars , and the gaussian is intended to represent red clump stars . for a given @xmath12 bin , the number of stars in that bin is then given by @xmath43 where we take @xmath44 is a constant , @xmath45 is a distance from the observer to the source , and @xmath46 is given by @xmath47 @xmath48 being another constant . in equation ( [ eq : integ ] ) we assume that the number of observable stars is everywhere proportional to the density of matter and that the intrinsic luminosity function is independent of location , which gives a constant @xmath49 throughout the bar . we now have a model of the mass distribution with four or five free parameters ( three axis plus one or two angles , eqs.[eq : gauss]-[eq : power ] ) and the intrinsic luminosity function with four free parameters ( two for the underlying power law plus two for the gaussian red clump , eq.[eq : lumfun ] the intrinsic width of the gaussian is not fitted , see below ) , total of 8 or 9 free parameters . we do not fit the intrinsic width of the red clump distribution @xmath50 we pre - set it to three values corresponding to @xmath51 and @xmath52 , with @xmath53 being the preferred value based on analysis in section 2 . we should mention here that the luminosity function from eq.[eq : lumfun ] is only approximately gaussian when transferred into magnitudes . on the other hand we have 12 fields with 55 bins in each of the fields , a total of 660 bins . this gives us number of degrees of freedom @xmath54 , which should be adequate to fit the model of the galactic bar . please note that , unlike the case of cobe - dirbe data , we have the information about all three dimensions of the bar , but we lack cobe - dirbe spatial coverage . table 2 presents some of the fitted parameters in the case with the value of the assumed intrinsic width of the red clump distribution @xmath55 . as we mentioned above , we also investigated cases with @xmath56 and we will discuss some aspects of this below . in all three cases we have two sub - cases : one when we allow only eight free parameters ( with @xmath57 ) and one in which we allow for the tilt out of the galactic plane ( @xmath58 ) . ccrrrrc g1 & 24.6 & & 1514 & 691 & 463 & 2.60 & 24.5 & 0.5 & 1529 & 676 & 460 & 2.60 g2 & 24.9 & & 1239 & 609 & 438 & 2.90 & 23.9 & 2.4 & 1293 & 565 & 423 & 2.84 g3 & 24.8 & & 3851 & 1698 & 1114 & 2.43 & 24.7 & @xmath230.1 & 3857 & 1705 & 1115 & 2.43 e1 & 20.7 & & 1665 & 691 & 297 & 2.47 & 20.1 & @xmath230.1 & 1680 & 704 & 296 & 2.47 e2 & 23.8 & & 899 & 386 & 250 & 2.37 & 23.8 & 0.0 & 897 & 387 & 250 & 2.37 e3 & 23.9 & & 854 & 381 & 278 & 2.59 & 24.0 & 0.4 & 857 & 376 & 276 & 2.59 p1 & 22.7 & & 724 & 299 & 189 & 2.27 & 22.6 & @xmath230.2 & 738 & 308 & 194 & 2.27 p2 & 22.6 & & 1056 & 435 & 275 & 2.27 & 22.4 & @xmath230.2 & 1061 & 443 & 278 & 2.27 p3 & 22.8 & & 1400 & 580 & 368 & 2.27 & 22.7 & @xmath230.2 & 1389 & 582 & 368 & 2.27 we use @xmath59 statistics as both a fitting tool , and also as a relative measure of the quality of each of the models . as we discuss later in this paper , our best fit models have values of the @xmath60 , which indicates that @xmath59 should not be used as a measure of absolute goodness of the fit . we therefore do not give formal errors of the fitted parameters in table 2 based on usual @xmath59 confidence levels . before we start discussing triaxial bar models , we should mention that we also tried to fit axially - symmetric set of models , described by the same functions as the triaxial bars , but with @xmath61 set to be the same as @xmath62 . not surprisingly , axially symmetric models fitted our data substantially worse than triaxial models , with the best fit model achieving @xmath63 compared to @xmath60 for the best triaxial models . from table 2 we can immediately see that none of the models fits the data perfectly , i.e. @xmath64 is relatively high . we discuss possible reasons for that in section 6 . still , some analytical functions fit better then others . we find that the whole class of power - law models ( p1 , p2 , p3 ) gives us the best fits to the data , for all three values of the @xmath50 investigated . however , the values of @xmath64 for other analytical functions , especially for the g3 and e2 models , are not much worse then for the power - law models . what is somewhat surprising is that the g2 model is our `` worst fit '' model , given that this was dwek et al . best fit model . we will discuss agreement of our work with the analysis of cobe - dirbe data in section 7 . to illustrate how significant the differences are between the best and the worst fit models , in fig.[fig5 ] we show the same histograms as in fig.[fig4 ] , but with the best ( p1 continues line ) and worst ( g2 dashed line ) model fits shown . for some of the fields , like gb13 or mm5-b , these differences are easily noticeable . in fig.[fig6 ] we show a cut through the e2 ( our `` recommended '' model of the galactic bar , see section 7.1 ) model at @xmath65 below the galactic plane , i.e. at the galactic latitude @xmath66 at the distance of @xmath67 . with dashed lines we show the lines of sight for the 12 fields used in this paper . contours start at 10% of the central value and increase by 10% . as we discuss below , this figure would be similar for other analytical functions used . the characteristic angles of the fitted models ( @xmath68 ) are defined in the same way for each model , so there is possible direct comparison between the models . as can be seen from table 2 , the tilt angle @xmath39 is never important in reducing the @xmath59 values significantly . also , except for the g2 model , angle @xmath39 is always smaller than @xmath69 . we will therefore set @xmath70 in further discussion . to analyze the behavior of the angle @xmath35 we plot in fig.[fig7 ] the value of this angle for various models as the function of the assumed @xmath50 . the horizontal spread along the @xmath71-axis for each @xmath50 was done for the clarity of presentation . it is immediately clear from fig.[fig7 ] that for a given @xmath50 the values of @xmath35 are very similar between the models ( see also table 2 ) . even for the e1 model , which has the most discrepant value of @xmath35 comparing to other models , this value is still only few degrees below the other values . also , there is visible in fig.[fig7 ] clear trend in the sense that for larger @xmath50 given analytical model has larger inclination angle @xmath35 . this trend was to some extent expected : to fit the observed data , the models with larger @xmath50 have to have smaller depth along the line of sight , which for triaxial objects decreases with increasing inclination angle . for the preferred value of @xmath55 ( section 2 ) the inclination angle is between @xmath72 . even taking the most extreme values of @xmath35 from fig.[fig7 ] we find that the inclination angle is between @xmath73 . this is a very good constraint comparing to the values of @xmath35 found in the literature ( for discussion see dwek et al . 1995 ; gerhard 1996 ; kuijken 1996 ) . our result agrees roughly with dwek s et al . range of @xmath74 , with an average of @xmath75 . in our case , however , constraints on the value of @xmath35 are much better due to the nature of the data used , which , unlike the cobe - dirbe data , contains the third dimension information in the form of the shift between the red clump giants distributions ( fig.[fig4 ] ) . unlike the characteristic angles , the scale lengths for different functional forms can not be compared directly . we can however compare the axis ratios for various models . in fig.[fig8 ] we show the axis ratios @xmath76 for the various models at the assumed width of the red clump @xmath77 . the horizontal spread along the @xmath71-axis for each @xmath50 was done for the clarity of the presentation , as in fig.[fig7 ] . let us first discuss the ratio @xmath78 , i.e. the ratio of the bar major axis to the minor axis lying in the plane of the galaxy . this ratio for all the models does not change significantly with @xmath50 and is between @xmath79 , except for the e1 model with @xmath80 , for which @xmath81 . this is a very tight constraint , again much tighter than @xmath82 obtained by dwek et al . from cobe - dirbe data ( see also their fig.5 ) . on the other hand , there is a clear trend between @xmath83 and @xmath50 in the sense that for larger @xmath50 given analytical model has smaller axis ratio @xmath83 . this was somewhat expected : as we increase the @xmath50 , @xmath61 must decrease to fit the data , but we expect that @xmath84 would remain approximately the same , since @xmath84 is the scale length perpendicular to the plane of the galaxy and it is not strongly affected by `` rotating '' the bar . also , the scatter of @xmath83 for different models is bigger than in case of @xmath78 . for the preferred value of @xmath55 , @xmath83 is between @xmath85 , again with the exception of e1 model , for which it is @xmath86 . judging from the map of the observed fields ( fig.[fig3 ] ) , one could be worried that most of the constraint we have on the @xmath84 comes from the tp8 field , in which there are very few red clump stars ( fig.[fig4 ] and table 1 ) . this is however not the case : even within baade s window there is a clear gradient , in the number of the red clump stars , with the distance from the galactic center . the bw1 field has about 40% more red clump stars than the bw3 field . in fact , we run a series of fits without the tp8 field and the resulting models were very close to what we present in this paper . there is evidence from the gas kinematics pointing to the fact that the centroid of the gas emission from the galactic bar is offset at @xmath87 ( blitz 1995 ; weinberg 1996 ) . we decided to add to our models one additional parameter @xmath88 which allows for the offset of the centroid of the galactic bar in the @xmath1 direction . in table 3 we list the best fit parameters of the resulting models , and also compare the @xmath64 for models with the offset and without it . cccrrrcc g1 & 24.1 & 0.60 & 1575 & 663 & 477 & 2.46 & 2.60 g2 & 24.1 & 0.82 & 1325 & 560 & 452 & 2.64 & 2.90 g3 & 23.9 & 0.53 & 4028 & 1656 & 1155 & 2.33 & 2.43 e1 & 19.1 & 0.24 & 1745 & 723 & 303 & 2.47 & 2.47 e2 & 23.5 & 0.46 & 935 & 385 & 262 & 2.29 & 2.37 e3 & 23.2 & 0.51 & 904 & 376 & 293 & 2.44 & 2.59 p1 & 22.5 & 0.18 & 784 & 321 & 207 & 2.26 & 2.27 p2 & 22.4 & 0.17 & 1095 & 448 & 288 & 2.26 & 2.27 p3 & 22.7 & 0.27 & 1486 & 606 & 396 & 2.25 & 2.27 ccrccc e2 & 20.0 & 1000 & 1:2.00 & 1:2.50 & 1.11 g1 & 18.1 & 1478 & 1:1.96 & 1:2.42 & 1.19 g2 & 17.8 & 1298 & 1:1.87 & 1:2.27 & 1.28 g3 & 19.2 & 4537 & 1:1.97 & 1:2.63 & 1.09 e1 & 13.8 & 1650 & 1:1.90 & 1:3.48 & 1.40 e2 & 18.3 & 1000 & 1:1.98 & 1:2.58 & 1.07 e3 & 18.0 & 1075 & 1:1.91 & 1:2.43 & 1.12 p1 & 19.2 & 1565 & 1:2.00 & 1:2.69 & 1.14 p2 & 19.4 & 2653 & 1:2.01 & 1:2.72 & 1.16 p3 & 18.6 & 1869 & 1:1.99 & 1:2.62 & 1.08 it is interesting to note that all the models prefer to have positive @xmath89 , in agreement with blitz ( 1995 ) . also , for some models , most noticeably the g2 model , introducing this additional parameter reduced @xmath64 considerably this is not simply an effect of having one additional parameter the tilt angle @xmath39 had hardly any effect on our models . we conclude by saying that our results are suggestive of a small ( @xmath90 ) centroid offset of the galactic bar . one could in principle introduce another additional parameter describing the centroid offset in the @xmath91 direction . with current data , however , we have only one field at positive galactic latitude , which provides very little lever arm to constrain an offset in @xmath91 . we thus make no attempt to include it in our models . in the previous section we fitted various analytical models to the luminosity function of the bulge red clumps stars and obtained the parameters of the 3-d mass distribution of the galactic bar . this procedure is subject to possible various systematic errors , which we discuss in some detail below . one obvious disadvantage of the whole procedure described in the previous section is that we fit only a number of selected analytical functions describing the 3-d light ( mass ) distribution of the galactic bar . indeed , high value of the formal @xmath59 we obtained even for our best - fit model p1 with @xmath55 and 8 free parameters has @xmath92 suggests that we either underestimate our errors ( which were taken to be poissonian @xmath93 ) or the galactic bar 3-d mass distribution is described by some other triaxial function or most probably is only roughly described by an analytical function . nevertheless , by following dwek et al . approach and fitting a whole set of different analytical functions with different properties we showed we can retrieve successfully at least some of the galactic bar properties . for example , it is clear from fig.[fig7 ] that the inclination angle of the bar to the line of sight @xmath35 does not depend strongly on the model we fit and is affected stronger by the assumed value of the intrinsic width of the red clump @xmath50 . to investigate to some extent how strong the effect of fitting an improper analytical function is on the fitted parameters , we generated artificial random histograms of the @xmath12 distributions using the e2 model . we then fitted this artificial data using all nine different analytical functions used in this paper . in table 4 we show both the parameters of the input model ( first row ) and the fitted parameters of all nine models . for reasons discussed in the previous section , instead of @xmath94 scale lengths we show the axis ratios @xmath95 , as these are more relevant for comparing the agreement between the models . as can be seen in table 4 , agreement in the fitted inclination angle is very good . the e1 model deviates most from the input @xmath96 , but this is also true when fitting to the real data . this only reconfirms our strong conviction that we can constrain the inclination angle @xmath35 of the bar very well . the @xmath97 axis ratio is also recovered remarkably well . as to the @xmath98 axis ratio , except for the e1 model , it is also well recovered by all other models . we conclude by saying that all the models we use , except for the e1 model , can be used to recover the parameters of the galactic bar with similar fidelity . it is possible that we detect red clump giants with different efficiency among the 12 fields we use in this paper . this could affect some of the parameters we fit , some stronger than others , for example inclination angle of the bar probably would not depend very strongly on this effect , but the axis ratios probably would . red clump stars we use in our analysis are always well above the detection threshold , @xmath99 in @xmath100 , so we do not expect a strong variation in efficiency among the fields . looking at fields which are adjacent ( like bw1 and bw5 or the two mm7 fields ) we generally find that the histograms of @xmath12 values are similar for these fields ( fig.[fig4 ] and table 1 ) . there are two levels on which stars can be lost : the ccd detection and the construction of a cmd from @xmath101 and @xmath100 databases . udalski et al . ( 1993 ) addressed the question of ccd star detection efficiency by placing artificial stars on the ccd and then retrieving them through the standard ogle data pipeline . they found that the ccd detection efficiency in the interesting us magnitude range is roughly constant and is about @xmath102 , which would not change the parameters of the fit we are interested in . we therefore make no correction for the ccd detection efficiency . the other reason that some stars are not included in the cmd is that the star photometry is not good enough to pass a corresponding criterion . udalski et al . ( 1993 ) used rather stringent photometric quality constrains on what stars can qualify to be included in the cmd , so only about 60% of stars from the @xmath103 databases are finally included . these constrains are however the same for all the fields , so we neglect that in our fitting procedure . one can imagine that a large scale gradient in the reddening law could explain the systematic shift of the red clump giants in magnitude with the galactic longitude , which we discuss in this and earlier works ( stanek et al . 1994 ; 1996 ) . this would cause @xmath12 defined by eq.[eq : free ] to be invalid globally . however , woniak and stanek ( 1996 ) found no such gradient . what they could not exclude were smaller variations of this coefficient among various fields , but again , inspection of fig.[fig4 ] shows a good agreement in red clump peak position for fields located at the approximately same position , like the bw fields . even fields mm7 and gb13 , located at approximately the same galactic @xmath1 , but on opposite sides of the galactic plane ( table 1 ) , have practically the same red clump peak position ( fig.[fig4 ] ) , even that the gb13 field has somewhat higher extinction then the mm7 fields . as we select a region of the cmd and then use all stars in this region to construct a luminosity function , we may expect some contamination from stars located in the galactic disk . stanek et al . ( 1994 ) considered this problem using bahcall & soneira ( 1980 ) model of the galactic disk and found the contamination to be negligible ( see stanek s et al . this is because in the region corresponding to red clump giants we expect only either bright and nearby , or evolved disk stars , of which there are very few . a related problem : our star counts could also be contaminated by stars from the metal poor stellar spheroid . we do not account for this population , having already 89 free parameters in our fit . however , this should be negligible , just as was contamination by disk stars . red clump becomes weak quite rapidly , as seen from @xmath104 histogram of the fields gb11 and tp8 , i.e. metal rich galactic bar star counts , in the region of the cmd we select for our analysis , strongly dominate . there seems to be a small metallicity gradient throughout the galactic bulge ( minniti et al . 1995 ; but see mcwilliam & rich 1994 ) , which could shift a little intrinsic color and luminosity of the red clump stars across the bulge . for all the models we took that into account by assuming @xmath105 where the choice of @xmath106 or @xmath107 depends on the density model fitted . we found that this additional parameter did not improve significantly the quality of the fits , we therefore did not present in table 2 the models with @xmath108 . the analytical models we used in this work were identical to those used by dwek et al . , so we should be able to compare our results directly ( with the remainder that the angle @xmath35 we use differs from dwek et al . @xmath36 by @xmath109 ) . for comparison we use dwek et al . best fitting models at @xmath110 with cutoff at @xmath111 , the same cutoff we use in this paper ( eq.[eq : integ ] ) . the dwek et al . fitting procedure ( their table 1 ) does not like power - law models ( our best fit models ) , so here we will only compare six other models we both used . in fig.[fig9 ] we plot the values of @xmath112 from dwek et al . ( @xmath71-axis ) and from this work ( @xmath113-axis ) . the dashed lines correspond to these values being equal . there is a very good agreement between the values of @xmath84 found by dwek et al . and those we find : they can be compared directly using dwek et al . table 1 and our table 2 . this is extremely reassuring because @xmath84 is not affected by `` rotating '' the bar , so the fact that we both get essentially the same values for @xmath84 means that , although here we are using data of entirely different nature from cobe - dirbe data , we probe in our analysis the same stellar populations . the correlations between @xmath114 values are not nearly as good as for @xmath84 , but still rather clear . on the other hand , there is no correlation between the inclination angles @xmath35 obtained by dwek et al . and our values . as we discussed earlier in this paper , regardless of the fitted model we obtain almost the same value for this angle , but dwek et al . obtained a wider range of values for this angle . we are confident that the nature of the data we use in this work allows for much better constraint of this parameter than the 2-d cobe - dirbe infrared intensity maps . as we mentioned in section 5 , the g2 model which was dwek et al . best fit model fits our data worst . there is however several models which seem to fit both cobe - dirbe and our data reasonably well . in fact , for the g3 model the values of the bar parameters obtained by dwek et al . ( their table 1 ) are remarkably similar to ours ( table 2 ) . the e2 model also fits reasonably well in both cases , and we suspect that if the value of @xmath35 for this model was fixed by dwek et al . to our value of @xmath115 , the agreement between @xmath116 would be much better . according to the trends we discussed in section 5 , lowering @xmath35 from dwek et al . @xmath117 to our @xmath118 would increase both @xmath61 and @xmath62 to values closer to ours . the e2 model might be preferred over the g3 model as it fits better the cobe - dirbe data in the inner region of the galaxy , where we do not have the cmd data from a low extinction fields . to summarize , for those who want to use a galactic bar model which should not be grossly wrong over large range of galactocentric distances , we recommend the e2 exponential model with _ our _ values for the bar parameters : @xmath119 . in fig.[fig10 ] we show the projected light from this model for the inner @xmath120 region of the galaxy , with our fields also shown . contours start at 10% of the central value and increase by 10% . the asymmetry between the positive and negative @xmath1 sides of the galactic bar is visible , but not striking . microlensing can provide us with additional information on the properties of the galactic bar . as noticed first by kiraga & paczyski ( 1994 ) , stars in the galactic bulge lensing other stars in the bulge produce significant contribution to the microlensing optical depth , comparable to that produced by the galactic disk stars lensing bulge stars . this effect is strongly enhanced when the galactic bulge is a bar with the major axis inclined to the line of sight by less the @xmath6 ( paczyski et al . 1994b ; zhao et al . recently zhao & mao ( 1996 ) investigated in detail how different models of the galactic bar would differ in the optical depth for microlensing they produce , so here we want to show only some of the aspects of the microlensing by the galactic bar . in fig.[fig11 ] we show a map of the microlensing optical depth @xmath121 produced by our `` recommended '' e2 model in the inner @xmath120 region of the galaxy , with our fields also shown . the mass of the galactic bar was assumed to be @xmath122 . contours start at @xmath123 and increase also by @xmath123 to the maximum of @xmath124 . the contribution to the optical depth from the galactic disk was not included . details of this microlensing map would be different if we used one of other models discussed in this paper , but the overall effect of the optical depth dropping off rapidly with the distance from the galactic center , faster along the minor axis , would remain . as discussed by zhao & mao ( 1996 ) , some few hundred microlensing events are needed before the microlensing can really make the difference for the existing galactic bar models . fortunately , this is very feasible judging from the existing microlensing experiments observing the galactic bar ( udalski et al . 1994 ; alcock et al . 1995 ; alard et al . 1995 ; alcock et al . 1996 ) . another aspect of the microlensing by the galactic bar was first noticed by stanek ( 1995 ) . as can be deduced from fig.[fig6 ] , the stars from the far side of the galactic bar will be preferentially lensed if most of the lensing comes from the bar stars ( paczyski et al . 1994b ; zhao et al . this leads to the offset in the brightness of the lensed stars , as compared to all observed stars , discussed in detail by stanek ( 1995 ) . the value of this offset correlates very well with the depth of the galactic bar along the line of sight ( stanek 1995 , his fig.3 ) . about a hundred of microlensing events in the red clump region of the cmd are needed to measure this effect with desired accuracy . in this paper we have used color - magnitude data obtained by the ogle collaboration for 12 fields scattered across the galactic bulge ( fig.[fig3 ] ) to construct the three - dimensional model of the mass distribution in the galactic bar . we model the galactic bar by fitting for all fields the observed luminosity functions in the red clump region of the color - magnitude diagram ( fig.[fig4 ] , [ fig5 ] ) . we find that almost regardless of the analytical function used to describe the 3-d stars distribution of the galactic bar , the resulting models are inclined to the line of sight by @xmath2 ( fig.[fig7 ] ) , with axis ratios corresponding to @xmath3 ( fig.[fig8 ] ) . this puts a strong constraint on the possible range of the galactic bar models . comparing our result with the results of dwek et al . ( 1995 ) , we find a good agreement between the derived parameters of the models ( fig.[fig9 ] ) , but we constrain the bar parameters much tighter . we recommend as a model which fits well both cobe - dirbe and our data the e2 model ( eq.9 ) with the values for the bar parameters : @xmath119 . we show various properties of this model in fig.reffig6 , [ fig10 ] , [ fig11 ] . gravitational microlensing can provide us with additional constrains on the structure of the galactic bar .

asking for the time evolution of a system means solving the schrdinger equation @xmath0 . restricted to the case of closed systems - which is done from now on - this equation can formally be solved by @xmath1 with @xmath2 for the sake of simplicity and @xmath3 . the whole dynamics is thus governed by the unitary propagator @xmath4 which maps any given `` initial state '' @xmath5 at time @xmath6 to its propagated counterpart @xmath7 at time @xmath8 . quantum mechanical determinism is expressed by the fact that @xmath5 can also be regarded as a `` final state '' into which a state @xmath9 evolves by satisfying @xmath10 ( see fig . [ fig : evol1 ] ) . @xmath11{images / evol1.eps}\ ] ] since equation ( [ prop1 ] ) is valid for all times @xmath12 , giving @xmath13 , one is led to the operator identity @xmath14 in other words : the set of evolution operators @xmath15 forms a unitary group @xmath16 with the group property ( [ group1 ] ) , giving any propagator @xmath17 its inverse element @xmath18 . the physical meaning of this mathematical statement is profound : in quantum mechanics there is no microscopic arrow of time . any state @xmath19 can be propagated to the future as well as to the past , so that there is a complete time symmetry . in order to explain an arrow of time despite these facts , there are several ideas a few of them are : 1 . any measurement on the system interrupts the deterministic evolution of the state by applying random projections ( plus normalization ) instead of unitary operations ( the `` collapse of the wave function '' ) . this process can not be reversed . real physical systems can never be regarded as closed . there is always an interacting environment with many unobservable degrees of freedom , inducing non - unitary time evolution of the observed system . 3 . time symmetry in classical mechanics is only broken with respect to macroscopic properties of a system ( e.g. temperature or entropy ) . considering macroscopic observables in quantum mechanics , this effect also shows up ( coarse graining argument ) . all these ideas have been worked out to elaborate physical models with strong experimental support and all of them break time symmetry ( see e.g. @xcite ) . so one could be content with the situation and regard those models as fairly good explanations of an arrow of time , which is indeed an undeniable part of human experience . a group of scientists around _ ilja prigogine _ ( nobel prize in chemistry 1977 ) , referred to as the `` brussels school '' , is not satisfied with the situation and has worked out a quantum mechanical formalism especially applying to unstable systems ( i.e. systems involving decay ) , where time symmetry is claimed to be broken at the microscopic level ( see e.g. @xcite ) . @xmath11{images / evol2.eps}\ ] ] this fact is expressed by the splitting of the unitary group @xmath16 into two semigroups @xmath20 and @xmath21 of propagators for negative and positive times , such that past and future states can no longer be regarded as equivalent counterparts . this leads to a so - called `` intrinsic irreversibility '' , implying a microscopic arrow of time ( see fig . [ fig : evol2 ] ) . this paper is separated into two different parts . the first part is dedicated to a review of the rigged hilbert space formalism worked out by the brussels school . it is shown how it works , how it applies to physical systems and what its advantages over standard quantum mechanics are . the second part is devoted to the role of time symmetry within the rigged hilbert space formalism . apart from its practical advantages , the brussels school points out that a microscopic arrow of time can be implemented here . it will be shown that such an implementation is possible but not mathematically necessary . writing down the formal solution ( [ prop1 ] ) of a closed system s schrdinger equation is not yet really satisfying . how can expectation values of observables be numerically calculated ? to this aim one has to find a spectral decomposition of the system s hamiltonian @xmath22 , thus one has to solve the time - independent schrdinger equation @xmath23 . since @xmath22 is a self - adjoint operator , its spectral decomposition is complete and orthogonal and may therefore be used as a spectral decomposition of the unity operator @xmath24 . furthermore the spectrum of @xmath22 is real and consits in general of a discrete part @xmath25 and a continuous part @xmath26 . in special cases there are other parts , e.g. singular continuous and dense spectra , but these cases are excluded here . with this restriction , using the dirac bra - ket notation , the spectral decomposition of the unity operator reads in general @xmath27 where @xmath28 , @xmath29 and @xmath30 is the energy density . as soon as the energy decomposition of the initial state @xmath31 , generally given by @xmath32 is known , the propagated state is also known and reads @xmath33 the expectation value of an observable @xmath34 can be calculated for any time @xmath8 using @xmath35 so the fundamental key to the dynamics of a system is the spectral decomposition of its hamiltonian . all dynamical properties are monitored by this decomposition and here is where the new formalism enters . there is a class of systems where the brussels schools s spectral decomposition of the hamiltonian differs from the usual decomposition : systems envolving decay , called _ unstable systems_. the basic concept of the brussels school s formalism is to find a different spectral decomposition of the hamiltonian . in contrast to the method of _ complex scaling _ ( see @xcite ) , the hamiltonian is not modified and remains self - adjoint ; it only acquires a different representation involving generalized eigenvectors contained in an extended distribution space . the modified decomposition can most easily be achieved by using the hamiltonian s _ resolvent _ @xmath36 the spectrum of @xmath22 is defined by the irregularities of its resolvent . the discrete eigenvalues are the singularities of @xmath37 , i.e. those points in the complex plane where the operator @xmath38 is not everywhere defined on the hilbert space @xmath39 . the continuous eigenvalues are those points where @xmath38 is an unbound ( and thus non - continuous ) operator on @xmath39 . anywhere else on the complex plane the resolvent is an analytic function mapping complex numbers onto the set of linear bounded operators . as @xmath22 is self - adjoint , any irregularities of @xmath38 take place on the real axis , where the discrete eigenvalues form the point spectrum and the continuous eigenvalues define a cut in the complex plane ( see fig . [ fig : spectrum ] ) . @xmath40{images / spectrum.eps}\ ] ] it is possible to find an analytic continuation of @xmath38 from both sides of the cut , each leading to a different operator @xmath41 with @xmath42 understood here and in the following . note that this continuation is not possible at singularities embedded in the continuous spectrum . note further that on the real axis outside the continuous spectrum both operators @xmath43 and @xmath44 coincide with @xmath38 . the two continued resolvents are known as the _ retarded _ and the _ advanced _ _ green operator _ , respectively . the resolvent can be used to construct eigenvectors . the residuum of the resolvent at a discrete eigenvalue yields the dyadic product of the corresponding eigenvector : @xmath45 note that both dyadic products project onto the same subspace corresponding to @xmath46 . if the discrete eigenvalue is isolated , i.e. not embedded in the continuous spectrum , both limits from above and below the real axis coincide . in the following we will assume that all discrete eigenvalues are isolated and we therefore have @xmath47 a discrete eigenvector @xmath48 of @xmath22 can be obtained by applying the residual resolvent to any hilbert vector @xmath31 being not orthogonal to the eigenspace of @xmath46 . : @xmath49 where @xmath50 is an adequate normalization constant , such that @xmath51 . in analogy to the discrete case the residuum of the resolvent at one of the continuous eigenvalues yields the dyadic product of the corresponding continuous eigenvector : @xmath52 note that both dyadic products are equivalent representations of the @xmath53-operator @xmath54 , defined by @xmath55 where @xmath56 is the continuous part of the total hamiltonian @xmath22 . since the continuous eigenvalues are not isolated , there are always pairs of eigenvectors . they correspond to retarded and advanced solutions of the continuous eigenvalue equation and are known as _ incoming _ and _ outgoing waves_. both solutions can equivalently be used in the spectral decomposition and they can be constructed by @xmath57 where @xmath58 is a suitable normalization constant , such that @xmath59 . using dyadic products , the retarded and advanced green operators can be written as @xmath60 where @xmath61 is any of the two dyadic products @xmath62 and the principal value distribution @xmath63 is defined by @xmath64 the difference of the two green operators gives the dyadic product of the continuous eigenvector : @xmath65 altogether , the spectral decomposition of the unity operator may be expressed by the exclusive use of the resolvent : @xmath66 this unity decomposition will be used as the starting point of the brussels schools s unity spectral decomposition . the resolvent can be analytically continued to the complex plane beyond the real axis entering the _ second riemann sheet _ : @xmath67 on the one half of the complex plane where the continuation started ( the first riemann sheet ) , both operators coincide with the original resolvent . on the other half ( the second riemann sheet ) it might happen that new singularities appear , and that in a pairwise manner : if @xmath68 is a singularity of @xmath69 in the lower half of the complex plane , then @xmath70 is a singularity of @xmath71 in the upper half ( see fig . [ fig : riemann ] ) . are exactly those with a resolvent having complex poles on the second riemann sheet . @xmath72{images / riemann.eps}\ ] ] the second riemann sheet poles act as complex resonances and cause the decay of certain states . consider the continuous part of the unitary decomposition ( [ unity ] ) : @xmath73 the following manipulations of @xmath74 are reducing its domain . as a consequence one is forced to use a smaller _ test space _ , enlarging at the same time the distribution space and inducing this way a certain _ gelfand tripel _ or _ rigged hilbert space _ where the modified completeness relation remains valid . we delay this discussion to section [ rhs ] . if one analytically deforms the integration path along the continuous spectrum @xmath26 to a path @xmath75 leading through the lower half of the complex plane , one enters the second riemann sheet and crosses the singularities of @xmath69 . let the set of these singularities be defined by @xmath76 and be enclosed by @xmath77 . in the course of the deformation circular integrations around the singularities are split off , each leading to a residuum of @xmath69 at these points ( see fig . [ fig : spectrum2 ] ) . @xmath40{images / spectrum2.eps}\ ] ] the modified operator now reads @xmath78 introducing a _ complex distribution _ @xmath79 , defined by @xmath80 where @xmath81 and @xmath82 being the analytic continuation of @xmath83 to respectively the upper and lower half of the complex plane , one can rewrite @xmath84 as @xmath85 in analogy to the standard case one defines dyadic products of _ generalized eigenvectors _ of @xmath22 by @xmath86 for the discrete case and , using ( [ dyad2 ] ) , @xmath87 for the continuous case , where @xmath61 is any of the two dyadic products @xmath62 . hence we have @xmath88 the generalized eigenvectors appear as pairs of right and left eigenvectors . the discrete eigenvectors are called _ gamov vectors _ and can be constructed by : @xmath89 where @xmath31 is a suitable vector and @xmath90 is a normalization constant , such that @xmath91 . by adjunction one sees that @xmath92 . furthermore we have @xmath93 . these objects are left and right eigenvectors of the hamiltonian @xmath22 corresponding to complex eigenvalues : @xmath94 since the gamov vectors have zero norm , they all coincide in hilbert space with the null vector , which is not allowed to be an eigenvector . also , since @xmath22 is self - adjoint , it can not have any eigenvectors corresponding to complex eigenvalues . concluding , the gamov vectors can not be elements of the hilbert space . in fact they are distributions in a suitably chosen distribution space ( see section [ intrinsic ] ) . the continuous eigenvectors can be constructed in two ways : @xmath95 by application to test vectors with analytic extensions to the complex plane one sees that @xmath96 hence there are two equivalent bases given by @xmath97 both bases can equally be used to span the hilbert space @xmath39 . arbitrarily we select @xmath98 and obtain a _ retarded _ unity decomposition @xmath99 in complete analogy one can modify the spectral decomposition by deformation of the integration path to @xmath81 in the _ upper _ half of the complex plane , entering the domain of @xmath71 with singularities at @xmath100 . one then obtains the two bases @xmath101 such that @xmath102 arbitrarily choosing @xmath103 one obtains an _ advanced _ unity decomposition : @xmath104 which is just the adjoint of the retarded unity decomposition , so @xmath105 . note , however , that both decompositions are equivalent representations of the unity operator , thus in the sense of an operator identity we have @xmath106 . the unity decompositions obtained in the preceding section can be used for different representations of vectors and operators . using the retarded decomposition the hamiltonian reads @xmath107 the adjoint of the above expression yields the hamiltonian in the _ advanced _ decomposition . since both decompositions are equivalent representations , one sees that @xmath108 , i.e. the hamiltonian remains self - adjoint . the propagator in the retarded decomposition reads @xmath109 since the complex numbers @xmath68 are located in the lower half of the complex plane , the contribution from the gamov vectors decays in the future . if an initial hilbert state @xmath31 has gamov components , they will disappear for @xmath110 . so if the state has no rotating components , i.e. @xmath111 , the whole state will disappear in the future with the decay rates of the gamov contributions given by the imaginary part of the contributing @xmath68 : @xmath112 with @xmath113 , @xmath114 and @xmath115 . each contribution from a gamov state also induces a certain energy shift , given by the real part of the @xmath68 . the advantage of the extended spectral decomposition is getting clear now : not only the stable frequencies can immediately be read from the spectrum , but the decay rates and the energy shifts , too . by additionally checking the amplitudes @xmath116 between the initial state and the gamov vectors one has immediate access to the dissipative dynamics of the system . apart from the contribution of the gamov vectors one has to consider the contribution from the generalized continuous eigenvectors . let us call it the _ background contribution_. since the decomposition involves the complex distribution @xmath117 , a test vector @xmath118 has to be introduced . the background contribution then reads : @xmath119 where @xmath120 and @xmath121 . due to @xmath122 the integration path is transformed into the curve @xmath75 below the complex eigenvalues @xmath68 . @xmath123 where @xmath124 and @xmath125 are the analytic continuations of respectively @xmath126 and @xmath127 to the lower half of the complex plane ( provided their existence ) . the curve @xmath75 is arbitrary as long as @xmath128 surrounds @xmath129 , i.e. , @xmath75 is placed _ below _ the poles @xmath68 , which all have negative imaginary parts . for @xmath130 , due to the exponential factor @xmath131 , the background contribution thus becomes more and more neglectable in comparison with the pole contributions . for initial states with high decay rates , i.e. pole contributions with small imaginary parts , the background contribution may be neglected already for small times @xmath130 . this approximation is called the _ pole approximation _ and it is essentially equivalent to the _ wigner - weisskopf approximation _ ( see @xcite ) . the time evolution of the initial state may be approximated by an exponential decay and its energy spectrum by a breit - wigner distribution . this approximation is very popular , e.g. in quantum optics , and indeed in most cases it is a very good one . obviously the above argument does not hold for negative times . here , the contributions from the gamov vectors diverge . in several publications of the brussels school ( see e.g. @xcite or _ a. bohm _ in @xcite ) this is an argument in favour of a broken time symmetry , since for negative times the _ advanced _ decomposition ( [ unity_ad ] ) is the better choice . however , as will be shown below , this is a rather intuitive reasoning and does not enforce a break of time symmetry . let us see what happens with the retarded decomposition of a hilbert state for negative times . of course the gamov contributions now diverge , but they are compensated by the background contribution . for @xmath132 the background integral ( [ bg2 ] ) gains influence , since the integration path @xmath75 is placed _ below _ the complex eigenvalues @xmath68 . the same argument , that for increasing @xmath130 the background contribution can be neglected in comparison with the pole contributions , is turned upside down for @xmath133 . the smaller imaginary part of @xmath75 induces a higher divergence than the contribution of the poles . by construction their divergences cancel each other , since the sum of both yields the contribution of the standard continuous eigenvectors ( see eq . ( [ unity_2rep ] ) ) , so a divergence of an initial hilbert state backwards in time is prevented . the divergence of the gamov contributions is even more than compensated : if a hilbert state decays in the future , it will also decay in the past . to see this one can use the advanced decomposition ( [ unity_ad ] ) to investigate the system s evolution to the past and one would find a completely symmetric situation with majorizing pole contributions decaying backwards in time . the situation is different if a gamov vector @xmath134 is taken as the initial state . now there is only one pole contribution @xmath68 decaying in the future and diverging in the past . but gamov vectors are only `` existing '' in the spectral representation of hilbert states and they themselves do not refer to elements of physical reality , similiar to all other `` generalized eigenstates '' appearing in spectral decompositions . as constituted by the axiomatic foundations of quantum mechanics a physically realizable state is represented by a hilbert vector ( actually a family of hilbert vectors normalized to unity and differing only in phase ) . gamov states are no hilbert vectors , so their properties are not expected to be physically meaningful . concluding , both decompositions are equivalent , but the use of the retarded decomposition is more _ intuitive _ for @xmath130 . however , this should not be taken as the reason for a broken time symmetry . a more convincing argument in favour of a broken time symmetry is the splitting of _ test spaces_. as we have seen in ( [ bg2 ] ) , analytic continuations of the test functions are needed to have the decompositions well - defined . in section [ rhs ] we will get back to this point . besides the advantages pointed out in section [ advantages ] we now face another important advantage of the brussels school s formalism . perturbation theory can be applied even in cases where standard quantum mechanics fails . those cases are featured by the phenomenon of instability , which is strongly connected to the sensitivity of a system towards small perturbations . let the system s hamiltonian be of the form @xmath135 where @xmath136 is the perturbation operator , switched on by the perturbation parameter @xmath137 $ ] , and @xmath138 is the unperturbed part of the hamiltonian , whose spectral decomposition is known and reads @xmath139 to simplify the discussion let the spectrum of @xmath138 be non - degenerate , so @xmath140 for @xmath141 . as long as the perturbation strength @xmath142 is not explicetely needed , it is set to 1 and @xmath143 is written as @xmath22 . the spectral decomposition of @xmath22 is yet unknown and reads in general @xmath144 where the sum is taken over the discrete spectrum @xmath25 and the integration is performed over the continuous spectrum @xmath26 of @xmath22 . the continuous spectrum @xmath26 is not affected by the perturbation ( provided the perturbation operator @xmath136 is @xmath138-compact , which is assumed here ) and there is an identity mapping of perturbed and unperturbed continuous eigenvalues , so in both cases the integration is performed over @xmath26 . the discrete case is different . switching to _ brioullin - wigner _ perturbation theory let us define pairs of orthogonal projection operators @xmath145 furthermore let the perturbed eigenvector @xmath146 of @xmath22 be not normalized to 1 but rather let @xmath147 , such that @xmath148 . applying these projections to the perturbed eigenvalue equation @xmath149 one obtains the _ brioullin - wigner formulas _ for the shifted eigenstate @xmath146 and the shifted energy @xmath46 : @xmath150 iteration leads to the corresponding perturbation series @xmath151 take a closer look at the above formulas ( [ briou1 ] ) and ( [ briou2 ] ) . they are only valid if the expression @xmath152 is well - defined . here _ resonance _ appears as a major obstacle . there are two different types of resonance making a perturbation expansion of eigenstates and eigenvalues impossible : 1 . * discrete resonance * : the eigenvalue @xmath153 may be shifted in the course of the perturbation to a neighboured eigenvalue of @xmath138 ( see fig . [ fig : shift]a ) . expression ( [ proj1 ] ) is no longer defined and the corresponding perturbation series diverges on account of the summation term . * continuous resonance * : if the eigenvalue @xmath153 lies next to the continuous spectrum of @xmath138 , it may be shifted there ( see fig . [ fig : shift]b ) . now the integral term in ( [ proj1 ] ) causes a divergence of the perturbation series . in both cases the perturbation expansion is limited to a finite convergence radius of the perturbation parameter @xmath142 . outside its range the shifted eigenvalue disappears from the spectrum of @xmath154 and the mapping from the unperturbed to the perturbed spectrum is no longer one - to - one ( see fig . [ fig : vanish ] ) . @xmath11{images / shift.eps}\ ] ] now regard the case of discrete unperturbed eigenvalues _ embedded _ in the continuous spectrum . here the convergence radius of @xmath142 reduces to zero , i.e. the perturbation series are no longer analytic in @xmath142 . an arbitralily small perturbation induces continuous resonance . as a consequence embedded states get unstable and start to decay . decaying states can not be part of the spectral decomposition of @xmath154 , they disappear from there and can not be reconstructed for @xmath155 , i.e. @xmath156 . hence perturbation theory is not aplicable . @xmath157{images / vanish.eps}\ ] ] there is a way out : one has to include unstable states , i.e. gamov vectors , into the spectral decomposition of @xmath154 . in terms of perturbation theory the gamov vectors can be constructed by modified brioullin - wigner equations : @xmath158 where @xmath159 are the singularities of @xmath43 on the lower half of the complex plane and @xmath160 are those of @xmath44 on the upper half . there is a one - to - one mapping between the disappearing discrete eigenvalues @xmath161 and the complex eigenvalues @xmath68 . iteration leads to the corresponding perurbation series @xmath162 the generalized continuous eigenstates are @xmath163 with the complex distribution @xmath164 defined in ( [ compl_dist ] ) . the standard continous eigenvectors can be constructed using the _ lippmann - schwinger equation _ @xmath165 or its iteration , the _ born series _ : @xmath166 take a closer look at the construction of gamov states in eq . ( [ briou3 ] ) . the divergences due to continuous resonance are prevented by introducing a small complex shift in the denominator and analytic continuation to the second riemann sheet . as a consequence , the energy eigenvalues are shifted to respectively the lower and upper complex plane , inducing exponential decay and breit - wigner energy distribution of the corresponding gamov states . the expressions are still analytic in @xmath142 , so perturbation theory can be applied . this is a major advantage of the brussels school s formalism . the friedrichs model ( k. friedrichs 1948 , see @xcite ) is the simplest model involving instability . here one single discrete state is coupled to a continuum of states by an interaction hamiltonian @xmath136 . a physical analogon to this model is e.g. the _ wigner - weisskopf model _ ( wigner and weisskopf 1930 , see @xcite ) of a twolevel atom coupled to the vacuum field . another example is the _ auger effect _ of autoionisation ( see e.g. @xcite ) . the total hamiltonian is of the form @xmath167 using the _ second resolvent identity _ @xmath168 with the resolvent @xmath169 of the free hamiltonian @xmath138 and the resolvent @xmath170 of the total hamiltonian . the resolvent @xmath37 can be obtained by applying it to the free states @xmath171 and @xmath172 and doing some rearrangement : @xmath173\nonumber \\ & & \left[\langle1|+\int_0^\infty d\omega\,\frac{w^*(\omega ) } { z-\omega}\langle\omega|\right]\nonumber \\ & & + \int_0^\infty d\omega\frac1{z-\omega}|\omega\rangle\langle\omega|\quad.\end{aligned}\ ] ] the complex function @xmath174 has analytic extensions @xmath175 from above and below to the real axis . the zeros of @xmath176 on the second riemann sheet define the complex poles of the resolvent : @xmath177 the pole @xmath178 can be obtained by numerically calculating the complex zero of @xmath179 . in first order one gets @xmath180 the gamov vectors can be obtained using eqns . ( [ g_eigenvector1a ] ) and ( [ g_eigenvector1b ] ) : @xmath181_+}|\omega\rangle\right]\\ \langle f_1 ^ -|&=&\frac1{c_1}\lim_{e\rightarrow z_1}(e - z_1)\langle1|r_+(e)\\ & = & \frac1{\sqrt{\eta_+{}'(z_1)}}\left[\langle1|+\int_0^\infty d\omega\ , \frac{w^*(\omega)}{[z_1-\omega]_+}\langle\omega|\right]\quad,\end{aligned}\ ] ] where the complex distribution is defined by @xmath182_+ } : = \lim_{e\rightarrow z_1}\int_0^\infty d\omega\ , \frac{\varphi(\omega)}{e-\omega+i\epsilon}\quad.\ ] ] the standard outgoing states can be constructed using eq . ( [ eigenvector2 ] ) : @xmath183\quad.\end{aligned}\ ] ] using eq . ( [ g_eigenvector2 ] ) the generalized continuous eigenvectors read @xmath184\quad,\\ \langle f^-(\omega)|&=&\langle u^+(\omega)|\\ & = & \langle\omega|+\frac{w(\omega)}{\eta_-(\omega)}\times\nonumber \\ & & \times\left[\langle1|+\int_0^\infty d\omega'\,\frac{w^*(\omega')}{\omega-\omega'-i\epsilon } \langle\omega|\right]\quad,\end{aligned}\ ] ] where the complex distribution @xmath164 is defined by ( [ compl_dist ] ) with the curve @xmath75 leading from @xmath185 to @xmath186 below @xmath178 . in the pole approximation , i.e. neglecting the background contribution from the continuous eigenvectors , the initial state @xmath171 decays with a decay rate @xmath187 given by two times the imaginary part of @xmath178 , @xmath188 which is in agreement with fermi s golden rule . when dealing with infinite dimensions , the hilbert space appears to be too small in the following sense : position and momentum operator have no hilbert eigenvectors and dirac s bra - ket formalism is mathematically not justified . to correct this problem , the hilbert space must be enlarged to include _ improper states _ like @xmath53-functions and plane waves . the framework to enable these features is given by the theory of the _ gelfand triplet _ , also called the _ rigged hilbert space _ ( rhs ) . @xmath11{images / leaving.eps}\ ] ] the concept is the following ( comp . e.g. @xcite ) . take the hilbert space @xmath189 of squared integrable functions and find a suitable test space @xmath190 with the properties 1 . @xmath191 is a locally convex topological vector space , 2 . @xmath191 is complete with respect to its own topology , 3 . @xmath191 is dense in @xmath192 , such that the topological dual @xmath193 of @xmath191 , consisting of all linear continuous functionals on @xmath191 , called _ distributions _ , includes the desired improper states . the hilbert space itself is his own topological dual , i.e. @xmath194 . for any test space @xmath191 being smaller than @xmath39 , the gelfand triplet @xmath195 makes it possible to extend the bracket @xmath196 to all pairs @xmath197 , @xmath198 by @xmath199 on pairs of hilbert vectors the bracket conicides with the scalar product in @xmath192 . the `` kets '' and `` bras '' are now identified by linear functionals on the opposite dual space , i.e. @xmath200 where @xmath201 or @xmath193 . an operator @xmath34 can be extended to the distribution space @xmath193 using @xmath202 hence @xmath34 can only be extended to @xmath193 if its adjoint @xmath203 is defined on and does not lead out of the test space @xmath191 : @xmath204 the test space of standard quantum mechanics is the schwartz space @xmath205 of rapidly decreasing functions , defined by @xmath206 the schwartz space is a physically convenient test space , since here position and momentum operator are everywhere defined , they fulfill ( [ extend_op ] ) , they are self - adjoint and have their eigenvectors among the _ tempered distributions _ in @xmath207 . furthermore , they keep their real spectrum , which makes the rhs @xmath208 be called a _ tight rigged _ hilbert space . another popular test space is the space @xmath209 of @xmath210-functions with compact support . but @xmath209 is not tight rigged , since the momentum operator obtains a complex spectrum with complex plane waves @xmath211 as eigenvectors . now we turn to the rhs of the brussels school . as already stated in section [ deform ] the definition of generalized eigenvectors is only possible if the test functions can be analytically continued to whether the lower or the upper half of the complex plane . the retarded right eigenvectors @xmath134 and @xmath212 involve complex distributions continuating the test functions @xmath213 to the lower half of the complex plane , denoted by @xmath214 . there is a class of functions fulfilling this condition : the hardy class @xmath215 from below ( for detailed information on hardy classes see @xcite ) . if @xmath216 is in @xmath215 then its complex conjugate @xmath217 is in the hardy class @xmath218 from above , whose members can be analytically continuated to the upper half of the complex plane , denoted by @xmath219 . in detail : the function @xmath220 is in @xmath221 if and only if it has an analytic continuation to @xmath222 , such that there is a @xmath223 and @xmath224 thus , restricted to the real line , hardy functions are @xmath225-functions . an equivalent definition can be given using the _ paley - wiener theorem _ : the hardy spaces @xmath221 are formed by the inverse fourier transforms of @xmath225-functions with support on respectively the positive and negative semiaxis @xmath226 : @xmath227 for physical convenience it is necessary , that the test functions are also in the schwartz class @xmath228 . furthermore , their energy representation can be restricted to the positive semiaxis , since energy is bounded from below by zero . altogether , the test spaces of the brussels school are defined by @xmath229 the topological duals @xmath230 then include the complex distributions used in the generalized eigenvectors . the formalism of the brussels school can be applied in the rhs @xmath231 since there are _ two _ rhs s , one has to investigate how to deal with them . the retarded unity decomposition ( [ unity_ret ] ) contains the dyadic products @xmath232 . applying test vectors @xmath233 and @xmath234 to both sides one obtains @xmath235 the above expression only makes sense , if @xmath127 and @xmath236 are both in @xmath237 , thus @xmath238 must be in @xmath239 . hence the instructions for computing brackets while using the retarded unity decomposition are : on the left hand you have to use @xmath239-vectors and on the right hand @xmath237-vectors . however , since both test spaces are dense in @xmath39 , you would not have to care too much , if there was not a significant subtlety : time symmetry is broken . to see this we investigate the time evolution of a test vector @xmath240 : @xmath241 since @xmath220 is in @xmath215 , there is an analytic continuation @xmath242 to @xmath214 and so there is one for its propagated counterpart , given by @xmath243 obviously , condition ( [ hardy_def ] ) is violated for @xmath244 and @xmath133 , thus time evolution leads out of the test space for @xmath133 . as a consequence test functions in @xmath237 can only be propagated to the future . analogically , test functions in @xmath239 can only be propagated to the past . using condition ( [ extend_op ] ) one finds in addition that the extension of the unitary propagator @xmath17 to the distribution spaces @xmath245 and @xmath246 is only possible for respectively @xmath130 and @xmath133 . so retarded gamov vectors @xmath134 in @xmath245 `` propagate '' ( as part of the spectral decomposition of hilbert vectors ) only to the future and advanced gamov vectors @xmath247 in @xmath246 propagate only to the past . the unitary propagator group @xmath248 hence splits into a semigroup @xmath249 propagating to the future ( and applying to vectors in @xmath237 and @xmath245 ) and a second one @xmath250 propagating to the past ( and applying to vectors in @xmath239 and @xmath246 ) . this feature is called by the brussels school the _ intrinsic irreversibility _ of quantum mechanics and is identified with a microscopic arrow of time . is the special choice of the two rhs s _ necessary _ for the correct mathematical implementation of gamov states ? if so , one could regard the splitting of the unitary time evolution group into two semigroups as enforced by the mathematical concept . this point of view is taken by _ a. bohm _ and _ n.l . harshman _ in @xcite : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ `` philosophizing alone would not be enough to take a semigroup instead . to arrive at the semigroup , we start from the empirically desirable properties of gamov resonance states and let mathematics determine the path . '' _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the above mentioned properties are given by @xmath251 with @xmath252 . starting from these conditions , following the reasoning of section [ intrinsic ] , one is led to the two rhs s with broken time symmetry . however , the argumentation is not valid if a rigged hilbert space can be found , where gamov vectors are included and time symmetry is preserved . let us start with the space @xmath209 of @xmath210-functions with compact support . as already mentioned , this space is a perfect test space fulfilling the requirements given in section [ rhs2 ] . furthermore we have @xmath253 . now regard the space @xmath254 of inversely fourier transformed @xmath209-functions : @xmath255 since the inverse fourier transform is a one - to - one mapping on the schwartz space @xmath205 , we have @xmath256 let @xmath257 be a @xmath209-function . its inverse fourier transform @xmath258 reads @xmath259 where @xmath260 is the compact support of @xmath257 . the analytic continuation of @xmath220 to the complex plane exists and is an entire function , given by @xmath261 for all @xmath262 ( see @xcite , theorem 7.22 ) . since the inverse fourier transform is a linear continuous one - to - one mapping , the transformed topology @xmath263 of @xmath209 , @xmath264 transfers all topological properties from @xmath209 to @xmath254 . thus , @xmath265 is a topological vector space fulfilling all requirements of a test space in @xmath266 . since @xmath254 contains functions being analytic in the whole complex plane , the topological dual @xmath267 contains the retarded gamov vectors defined by ( [ req1 ] ) and ( [ req2 ] ) . furthermore , it contains also the _ advanced _ gamov vectors @xmath247 with eigenvalues @xmath70 in the upper half of the complex plane , as well as any other generalized eigenvector used in the extended spectral decompositions ( [ unity_ret ] ) and ( [ unity_ad ] ) . because of relation ( [ z_schwartz ] ) all @xmath254-functions are @xmath205-functions , so @xmath254 is already a physical convenient test space and standard quantum mechanics is applicable , too . since @xmath268 the space @xmath254 does neither include any of the test spaces @xmath269 nor is it part of it , i.e. @xmath270 but of course @xmath271 . so @xmath254 is a different test space , whose topological duals include any of the complex distributions needed for a comfortable treatment of decaying systems . however , there is no intrinsic irreversibility . let the @xmath209-function @xmath257 have a compact support bounded by the finite interval @xmath272 $ ] . its inverse fourier transform @xmath220 can be propagated in time yielding @xmath273 , whose fourier transform , @xmath274 has a compact support bounded by the finite interval @xmath275 $ ] , thus it remains compact for all @xmath276 and @xmath277 stays in @xmath209 , so @xmath278 stays in @xmath254 . hence the propagator @xmath17 does not lead out of the test space for any @xmath276 . consequently , the unitary propagator group @xmath279 can be extended to the distribution space @xmath267 without any splitting . the support of @xmath280-functions can be restricted to the positive semiaxis , due to the positivity of energy , and the rhs @xmath281 is thus suitable for the presented concepts . any desired unity decomposition including gamov vectors , retarded or advanced , can equivalently be used . as far as hilbert states are concerned , no unphysical behaviour is predicted and any calculated physical quantity is numerically identical to the one calculated by standard methods . apart from pure mathematics there is a physical reasoning presented by members of the brussels school who try to derive the special choice of the two hardy classes from the assumption of _ causality _ ( see e.g. schulte and twarock in @xcite or antoine , bohm and harshman in @xcite , from where the citations below are taken ) . this is done the following way . starting point is a special assumption of _ causality _ which is called the `` preparation@xmath282registration arrow of time '' : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ `` time translation of the registration apparatus relative to the preparation apparatus makes sense only by an amount of @xmath283 . '' _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ this is translated into the language of quantum mechanics to a so - called `` quantum mechanical arrow of time '' : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ `` a state @xmath284 must be prepared before an observable @xmath285 can be measured in that state , i.e. @xmath286 must be prepared during @xmath287 . '' _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ interpreting measurement formally as a scattering experiment , the prepared state @xmath288 is identified with an incoming state and the measured state @xmath289 with an outgoing state ( see fig . [ fig : measurement ] ) . survival probabilities of the initial ( i.e. prepared ) state are thus calculated by transition probabilities of the incoming and outgoing state : @xmath290 using the energy decomposition this implies @xmath291 using the paley - wiener theorem ( [ paley - wiener ] ) one is led to the desired conclusion @xmath292 hence , founding on the given formulation of causality the only correct choice of a rigged hilbert space is the choice of the two rhs s of the brussels school . in other words : standard quantum mechanics has not yet found a microscopic arrow of time because it has been incomplete in its axiomatic foundation . @xmath293{images / measurement.eps}}\ ] ] the above argumentation is correct , but it can not be taken as a _ derivation _ of the arrow of time . as is obvious even by the choice of words ( `` before '' , `` after '' , etc . ) , the given `` causality '' assumption is already based on the arrow of time , hence it is logically impossible to derive the latter from the former . causality itself does not require an arrow of time , since neither newton s law nor schrdinger s law contradict causality , although both theories are perfectly time - symmetric . yet , the `` preparation@xmath282registration arrow of time '' refers explicitely to _ open subsystems _ , namely the preparation apparatus and the measuring apparatus , which are coupled together to form a composite system . though one can not generally infer from statements about the time evolution of open subsystems to statements about the evoution of closed composite systems , especially not when the subsystems are involved in a measurement process . hence postulating a `` preparation@xmath282registration arrow of time '' for _ all _ systems equals introducing a microscopic arrow of time _ by hand _ into quantum mechanics . once the arrow of time has been put into the mathematics , it is not surprising that it will reappear eventually . the situation can be visualized by fig . [ fig : circular ] . @xmath294{images / circular.eps}\ ] ] thus , the special choice of the two rhs s must be regarded as being _ equivalent to a postulation of a global microscopic arrow of time_. the split rhs s are a mathematical mirror image of the empirically motivated splitting between past and future . as already stated in the introduction , it is possible to explain an arrow of time due to time - asymmetric evolution of _ macroscopic _ observables and due to _ local _ observations on subsystems . but respecting a _ global microscopic arrow of time _ , there is a major difference between standard and rigged hilbert space quantum mechanics . in standard quantum mechanics there is no way to implement a global microscopic arrow of time except perhaps by applying brute force and splitting the unitary evolution group axiomatically into two semigroups . in rigged hilbert space quantum mechanics one has the choice of either time - symmetric or time - asymmetric microscopic evolution , due to the choice of the test spaces . hence , in the rhs formulation it is possible to implement a global microscopic arrow of time in contrast to standard quantum mechanics . so one could argue that nature _ obviously _ has taken the time - asymmetric choice of two test spaces instead of one . but what does `` obviously '' mean ? one might mention our everyday experiences or global time - asymmetric boundary conditions like the big bang or the second law of thermodynamics . though these experiences and boundary conditions are all related to subsystems or far from being microscopic . the `` universe '' is a construction from observations on subsystems on a _ very _ macroscopic scale . altogether , these time - asymmetric boundary conditions all base on _ local macroscopic observations _ and hence , as already stated , they do not give rise to a real problem with quantum mechanics , which is a set of microscopic laws . so the choice of global microscopic time - symmetry or time - asymmetry remains merely axiomatic and can not be derived from other physical postulates . the quantum mechanical formalism , developped by the brussels school , offers a comfortable treatment of decaying systems , where standard quantum mechanics requires a lot more effort . the generalized spectral decompositions involve gamov states and can be interpreted in a physically intuitive manner . they directly yield important physical quantities like decay rates and energy level shifts . perturbation expansions become possible in cases , where standard quantum mechanics fails . in contrast to standard quantum mechanics , where the implementation of a microscopic arrow of time is not possible , the rigged hilbert space formalism offers the possibility to incorporate such an arrow by the special choice of two different gelfand triplets . due to this choice , the unitary time evolution group splits into two semigroups , each one for respectively the future and the past . however , it is shown that the use of gamov states does not require this splitting . as a counter example , one single gelfand triplet is constructed , where the formalism of the brussels school can be applied and where time symmetry is preserved . furthermore , it is pointed out that the physical arguments in favour of the special choice of two gelfand triplets are not independent from the arrow of time and thus can not be used for a derivation of the latter . hence even within the framework of the rigged hilbert space formalism , the global microscopic arrow of time remains a postulate . i would like to thank f. petruccione , h .- breuer , m. bordemann , and m. wilkens for stimulating discussions and editorial advice and the referee for valuable comments . i also thank all other people talking with me about the topic of this paper and contributing in many ways to its completion .

quantum parameter estimation , originally motivated by optical metrology @xcite , considers how to estimate parameters associated with the evolution of quantum systems by physical means . a theory of quantum estimation , derived from the laws of quantum physics and statistics , has been developed and applied to various evolution processes yielding important results for various metrology situations @xcite . we consider the single qubit depolarizing channel , which maps a qubit , initially in the state @xmath0 , via @xmath1 } \ ; \hat{i } + \lambda { \hat{\rho}}\label{eq : depol}\ ] ] where @xmath2 the aim of a physical estimation protocol would be to determine @xmath3 as accurately as possible by subjecting a collection of qubits to this evolution . the depolarizing channel is interesting for various reasons . first , depolarization is a standard model of certain noise processes and is of general interest for quantum information processing @xcite . specific examples appear in nuclear magnetic resonance ( nmr ) @xcite and optical quantum information processing @xcite . second , quantum parameter estimation reveals fundamental and quantifiable differences with classical approaches when manifestly quantum resources such as entanglement are used . for example , in the estimation of the phase associated with a unitary parameter , classical approaches yield an uncertainty that scales as @xmath4 where @xmath5 is the number of times the channel is invoked , whereas using entangled states can result in a lower uncertainty that scales as @xmath6 , thus pointing to a measurable gain provided by entanglement @xcite . depolarizing channel parameter estimation has been investigated previously in the context of finding optimal estimation protocols @xcite where any possible initial state is assumed to be available . it emerges that the optimal estimation protocol requires qubit pairs , each initially in an entangled pure state , and that states displaying entanglement or correlations amongst more than two qubits will not give any advantages to estimation accuracy @xcite . we consider various estimation protocols for the constrained situation where _ when the available initial state of each qubit is mixed . _ this situation arises in room temperature solution state nmr , where the initial state for any nuclear spin is highly mixed @xcite . we ask whether , for mixed initial state situations , protocols involving correlated states can yield gains in estimation accuracy compared to protocols that use product states and whether increasing the number of correlated qubits beyond two is advantageous . while we will consider these questions for situations involving initial states with all possible ranges of mixedness , we will focus on the special case where the initial state of each qubit is highly mixed . our approach is motivated by similar work for other types of channels where gains were found when using correlated states @xcite and is also inspired by various other approaches to investigating parameter estimation with mixed initial states @xcite . this article is organized as follows . section [ sec : genestimation ] offers a review of quantum estimation theory . section [ sec : indest ] applies this to the simplest depolarizing channel parameter channel estimation protocol , establishes a benchmark against which other schemes are to be compared and reviews existing results for parameter estimation using pure initial states . section [ sec : indchanneluseest ] considers mixed initial state estimation protocols where there are never any correlations between qubits . section [ sec : correst ] , which contains the key general results of our article , considers an estimation protocol that employs correlated states . section [ sec : weakpol ] applies these to the special case where the available initial states are highly mixed . finally section [ sec : correlation ] briefly discusses the roles of entanglement and quantum discord in the protocol that uses correlated states . physical parameter estimation requires that suitably prepared quantum systems be subjected to the operation @xmath7 which is dependent on a real parameter @xmath3 that is to be estimated . the generic scheme for doing this , illustrated in fig . [ fig : generalscheme ] , assumes that @xmath8 copies of the operation , or channel , are available . these may be invoked on one or more of @xmath5 systems with invocations surrounded by a series of parameter independent unitary transformations . quantum systems , each represented by a horizontal line . this procedure uses @xmath8 invocations of the operation @xmath7 and parameter - independent unitary transformations @xmath9 the initial state of each quantum system is @xmath10 and the pre - measurement state of the collection of systems is @xmath11 measurements on individual systems follow @xmath12 [ fig : generalscheme ] ] each quantum system is prepared in a known initial state @xmath10 . after the final unitary , a measurement is chosen and performed on each system and the measurement outcomes are recorded . the final pre - measurement state @xmath13 may depend on the parameter and thus so will the probabilities with which the measurement outcomes occur . thus information about @xmath3 might be inferred from the measurement outcomes by classical data processing , thus yielding an estimate for the parameter . the fact that measurement outcomes in quantum physics are usually statistically distributed implies that different runs of the same estimation procedure will yield estimates that fluctuate about a mean value . the aim of quantum parameter estimation is to tailor the process by choosing the initial state @xmath10 , the parameter independent unitaries @xmath14 and the final measurements so that the estimates will display minimal fluctuations about the actual parameter value . the accuracy of the estimation is quantified via these statistical fluctuations as follows . given a set of measurement outcomes , one on each qubit and denoted @xmath15 , an estimate for the parameter @xmath16 is attained via @xmath17 where @xmath18 is an estimator function . fluctuations in these measurement outcomes from one run of the estimation procedure to another generate fluctuations in @xmath16 . the extent to which this occurs depends on the construction of the estimator function . key requirements for an estimation protocol will be that the estimator is unbiased , i.e. @xmath19 , and that it minimizes the variance , @xmath20 where the angle brackets indicate the mean over all possible measurement outcomes . the variance will depend on the choice of estimator . however , a key result @xcite is that _ the variance on any estimator is bounded from below via the cramr - rao bound _ @xmath21 where @xmath22 ^ 2\ ; \ ; \mathrm{d}x_1 \ldots \mathrm{d}x_n . \label{eq : classicalfisher}\ ] ] is the ( classical ) fisher information associated with the probability distribution for the measurement outcomes @xmath23 . this is independent of the choice of estimator . quite generally , there is always an estimator @xcite , which asymptotically saturates the bound of eq . . thus in classical parameter estimation the fisher information quantifies the accuracy of any possible estimate ; the larger @xmath24 , the better the estimate . in classical estimation , it is easily shown @xcite that if the same estimation trial is repeated @xmath8 times independently , the the fisher information becomes @xmath25 where @xmath24 is the fisher information for a single trial . thus the cramr - rao bound implies a reduction in the variance by a factor of @xmath8 , yielding the typical classical reduction in uncertainty of the estimate by a factor of @xmath26 in quantum evolution processes the probability distribution for the measurement outcomes depends on the type of measurement . thus , given many possible choices of measurement , there will be many possible values for the classical fisher information for the same evolution process . quantum estimation aims for a measurement which results in the largest classical fisher information . this is constrained and simplified by the _ quantum cramr - rao bound , _ which states that , for any measurement the resulting classical fisher information satisfies @xmath27 where @xmath28 is the quantum fisher information and is _ independent of the choice of measurement _ @xcite . the quantum fisher information depends only on @xmath13 and is calculated via @xcite @xmath29 } \label{eq : quantumfisher}\ ] ] where the symmetric logarithmic derivative ( sld ) or score operator , @xmath30 is defined implicitly via @xmath31 . \label{eq : slddefintion}\ ] ] in general the sld is not immediately apparent from @xmath13 ; sometimes it can be computed using relatively simple algebra @xcite . however , the sld and quantum fisher information can always be computed whenever an eigenvalue and eigenstate decomposition for @xmath13 can be found . specifically , if @xmath32 then @xcite @xmath33 quite generally , there exists a measurement procedure which asymptotically attains the quantum cramr - rao bound @xcite . thus the quantum fisher information quantifies the accuracy of the quantum estimation process . the main task in quantum parameter estimation is to provide a protocol giving a final pre - measurement state @xmath13 with the maximum quantum fisher information ; this approach has been widely adopted elsewhere @xcite . before continuing , we remark that a useful computational alternative to eq . is @xcite @xmath34}. \label{eq : quantumfishertwo}\ ] ] the quantum fisher information will depend on the resources , such as the numbers of qubits , channel invocations , single qubit unitaries or entangling unitaries that are used and various competing estimation protocols can only be compared in terms of their resource use . in this article we take a common approach , which regards the number of channel invocations @xmath8 in the circuit of fig . [ fig : generalscheme ] as the only important resource cost @xcite . in this context , the goal of any quantum estimation protocol is to produce a final state with the maximum quantum fisher information per channel invocation . the initial state for the entire system can be assumed to be a product of states for the individual systems . if necessary , any of @xmath14 can be used to generate entangled or otherwise correlated states as the procedure unfolds . we can then consider two broad classes of estimation schemes . in the first the unitaries @xmath14 never produce any correlation between the individual systems and the entire system evolves through a series of product states . various classical approaches to estimation , including repetition and averaging , fall into this realm . in the second , some of the unitaries @xmath14 produce states where there is some form of quantum correlation , such as entanglement , between the individual systems . frequently quantum estimation aims to show that using such correlated states results in a parameter estimate with a greater quantum fisher information per channel invocation @xcite than is possible with non - correlated states ; indirectly this indicates the utility of quantum resources for information processing . within the framework of fig . [ fig : generalscheme ] , the typical task is to find the initial state @xmath10 and the unitaries @xmath14 that produce a final state @xmath13 that yields the maximum quantum fisher information per channel invocation . the convexity of the quantum fisher information guarantees that this will be attained by using a pure initial state @xcite , and most investigations of depolarizing channel parameter estimation have considered this case @xcite . however , it may be that case that the available initial states are not pure . our aim is to consider various estimation protocols when the available initial states are limited to those that are mixed . the general initial state of a single qubit is @xmath35 where @xmath36 and @xmath37 are the components of a three dimensional real vector , @xmath38 . the magnitude of this vector @xmath39 is called the _ polarization or purity _ of the state and it satisfies @xmath40 when @xmath41 the state is pure and as @xmath42 , which we use to quantify the mixedness or purity of the state , decreases the state becomes more mixed . thus the principal issue of such mixed state quantum estimation is : _ given that the available qubits have initial states each with polarization @xmath42 , how do the accuracies , quantified by the quantum fisher information , of various parameter estimation protocols compare ? _ note that , in this approach , the orientation of the polarization vector @xmath38 is irrelevant , as this can be modified at no cost with a single qubit unitary without changing the polarization @xmath42 . the simplest parameter estimation scheme , which also forms the baseline against which all other schemes may be compared , is that where the channel is invoked once on a single qubit and is described in fig . [ fig : sqsc ] . this will be termed the single qubit , single channel ( sqsc ) protocol . . a parameter independent unitary transformation @xmath43 precedes the single channel invocation , producing a channel input state , @xmath44 . a measurement follows the channel invocation . [ fig : sqsc ] ] the unitary @xmath43 of fig . [ fig : sqsc ] only changes the orientation of @xmath38 , leaving the polarization unaltered and thus can be omitted . then eq . yields that the state after one channel invocation is @xmath45 . the quantum fisher information can be computed by first determining the sld using prop . 1 of @xcite and then substituting into eq . . this proposition ( appropriately corrected ) states that if @xmath46 then @xmath47/\operatorname{tr}{{\hat{\rho}_\mathrm{f}}}$ ] while if @xmath48 then @xmath49/\operatorname{tr}{{\hat{\rho}_\mathrm{f } } } + \frac{\partial}{\partial \lambda } \ln{(\lvert \alpha/\operatorname{tr}{{\hat{\rho}_\mathrm{f } } } \rvert ) } { \hat{i}}.$ ] straightforward calculation gives that @xmath50 and @xmath51 the only circumstances under which @xmath52 are when @xmath41 _ and _ @xmath53 we henceforth only consider @xmath54 it follows that , _ for a single channel invocation on the state of eq . _ the sld is @xmath55 substituting into eq . then gives that _ the quantum fisher information for the sqsc protocol using an initial state with polarization @xmath42 _ is @xmath56 this is the baseline quantum fisher information against which all other estimation schemes are to be compared . we briefly digress to discuss the previously investigated @xcite extreme case , where the initial states are pure . if it were possible to modify the initial state polarization , then eq . , shows that , in accordance with general results @xcite , the largest quantum fisher information would be attained for a pure input state , i.e. when @xmath57 giving @xmath58 usually the key issue of interest in quantum parameter estimation is whether a greater quantum fisher information can be attained when entangled multiple qubits are available as input for the channel . for the depolarizing channel it has been shown that entanglement can yield a greater quantum fisher information . specifically if the channel is applied to one of two qubits then the optimal quantum fisher information is attained @xcite when the two qubits are prepared in a maximally entangled pure state , such as @xmath59 . the resulting quantum fisher information is @xmath60 it is straightforward to show that @xmath61 whenever @xmath62 the remaining question is whether any additional qubits and channel invocations can yield a greater quantum fisher information per channel invocation . remarkably the result is @xcite that the optimal depolarizing channel parameter estimation scheme uses pairs of qubits , each in a maximally entangled pure input state . the channel is applied once to only one of each entangled pair and the resulting quantum fisher information per channel use is that of eq . . thus _ when qubits are available in pure initial states then entangling more than two qubits can not yield any advantages for depolarizing channel parameter estimation . _ however , we stress that this does not necessarily pertain to the situation that we address , in which the available input states are not pure . the previous discussion motivates the central questions of our work : _ given a collection of qubits , each initially in a state with the same polarization @xmath42 , is it possible to enhance the parameter estimation accuracy by using multiple channel invocations or entanglement or correlations amongst more than two qubits ? if so to what extent can the accuracy , as measured by the quantum fisher information per channel use , be enhanced ? _ a first strategy for enhancing estimation accuracy is one where the same process is repeated independently , as is done for classical repeat and average schemes . specifically we consider the protocol of fig . [ fig : indchanneluse ] , called the _ independent channel use protocol . _ here there are @xmath8 qubits , each in the initial state @xmath10 and subject to a parameter independent single qubit preparatory unitary . the channel is subsequently invoked exactly once on each qubit . in this protocol there are never any correlations between the qubits during the entire process ; in this sense it resembles a classical approach . single qubits , each initially in the state @xmath10 . qubit @xmath63 is subjected to a single qubit unitary , @xmath64 , which is independent of the channel parameter , producing the input state @xmath65 thereafter a depolarizing channel is applied once to each qubit . [ fig : indchanneluse ] ] the quantum fisher information is additive in such cases @xcite and thus the quantum fisher information for the independent channel use protocol with @xmath8 channel invocations is @xmath66 although this is an improvement over the sqsc protocol , clearly the quantum fisher information _ per channel use _ in the independent channel use protocol is identical to that of the sqsc protocol . thus _ the independent channel use protocol attains the same estimation accuracy , measured in terms of quantum fisher information per channel invocation , as the sqsc protocol . _ a second strategy , illustrated in fig . [ fig : seqchanneluse ] , is that where _ only one qubit is available and the channel is invoked sequentially a total of @xmath8 times . _ successive channel invocations are interspersed with parameter - independent unitaries . is subjected to @xmath8 channel invocations . each channel invocation is preceded a single qubit unitary , @xmath67 , all of which are independent of the channel parameter . [ fig : seqchanneluse ] ] assuming the standard initial state , @xmath10 , the effect of @xmath43 is to rotate the polarization vector , @xmath68 without altering its magnitude . . implies that the state after the first channel invocation is @xmath69 . using the same argument , the state after the second channel invocation is @xmath70 where @xmath71 rotates @xmath72 continuing , the state prior to measurement is @xmath73 where @xmath74 a calculation similar to that giving eq . yields that the quantum fisher information for @xmath8 sequential channel invocations on a single qubit is @xmath75 thus the quantum fisher information per channel invocation is @xmath76 to compare this to the sqsc protocol we define the _ sequential channel use gain _ as @xmath77 figure [ fig : gainseqplot ] illustrates the typical gain , plotted here for the @xmath78 case . this shows that the sequential channel use protocol is advantageous for certain ranges of parameter values and polarizations but disadvantageous for others . . [ fig : gainseqplot ] ] the sequential channel use gain decreases monotonically with @xmath42 and increases monotonically with @xmath3 ( these are proved in appendix [ app : gainseq ] ) . a more detailed analysis of this separates the pure initial state case , i.e. @xmath41 , from mixed initial state cases , i.e. @xmath79 . for the _ pure initial state case , _ it is shown in appendix [ app : gainseq ] that @xmath80 for the _ mixed initial state case , _ it is also shown appendix [ app : gainseq ] that @xmath81 in both cases the upper bound is attained when @xmath82 it follows that _ if the available initial state is pure there is no advantage to the sequential channel use protocol . if the available initial state is mixed then , for certain initial polarizations and the parameter values , there can be an advantage to the sequential channel use protocol . _ it remains to assess whether for mixed initial states , there is any advantage to increasing the number of channel invocations . to do so , we compare the quantum fisher information per channel use of eq . for @xmath83 channel invocations versus that for @xmath8 channel invocations . an advantage is attained when the ratio of that for @xmath83 to that for @xmath8 exceeds @xmath84 , and straightforward algebra shows that this is equivalent to @xmath85/\lambda^{2m+2}$ ] . the right hand side of this increases monotonically with @xmath3 and attains a maximum of @xmath84 when @xmath86 . also it is negative when @xmath87 , implying that when @xmath87 there is no polarization for which an additional channel use will give an advantage . thus for any polarization there will be some parameter values for which additional invocations in the sequential channel use protocol will be advantageous . however , the range of possible parameter values for which this occurs , bounded from below by at least @xmath88 , shrinks with increasing channel invocations . an additional resource available to multiple qubit systems , is the existence of states in which there are correlations between the individual qubits and we now consider whether these can enhance the accuracy of parameter estimation for the depolarizing channel . we consider the particular correlated state protocol procedure of fig . [ fig : preparedscheme ] ; this was previously investigated for other channels @xcite . we assume that each qubit is initially in the particular state @xmath89 with polarization @xmath42 and that a preparatory unitary @xmath90 , illustrated in fig . [ fig : uprep ] , is subsequently applied jointly to the qubits . single qubits , each initially in the state @xmath10 . these are subjected to a joint preparatory unitary , @xmath90 , producing the channel input state @xmath65 thereafter a depolarizing channel is applied once to each of the uppermost @xmath8 qubits . the preparatory unitary is illustrated in fig . [ fig : uprep ] . [ fig : preparedscheme ] ] denote the density operator for the system after the preparatory gate by @xmath91 consists of a collection of controlled-@xmath92 operations applied to each distinct pair of qubits followed by a hadamard transformation applied to each qubit . [ fig : uprep ] ] it can be shown @xcite that @xmath93 where @xmath94 and @xmath95 \\ & + i { c_{j(x)}}\ ; \biggl [ { \lvert x \rangle}{\langle n - x \lvert } - { \lvert n - x \rangle } { \langle x \lvert } \biggr ] . \end{split } \label{eq : rhoprepped}\ ] ] the counter - diagonal terms of the resulting matrix of eq . are constructed by first computing the number of zeroes @xmath96 in the binary representation @xmath97 for @xmath98 and then using @xmath99 . \label{eq : prepcounterdiag } \end{aligned}\ ] ] the diagonal terms are @xmath100 . \label{eq : prepdiag } \end{aligned}\ ] ] as described previously @xcite , this scheme results in a density operator which has supports consisting of mutually orthogonal two dimensional subspaces ; in terms of matrix representations it can be expressed as a sum of @xmath101 matrices acting on distinct vector spaces . when the initial states are pure , i.e. @xmath57 the diagonal and counter - diagonal terms are all zero except @xmath102 and @xmath103 and @xmath104 is the density operator corresponding to the ghz - type state @xmath105 suppose that the channel acts _ once on each of @xmath8 qubits _ corresponding to the least significant ( or rightmost ) digits in the computational basis representation . the action of the channel on multiple qubits is readily described by noting that eq . implies that , for a single qubit , [ eq : depoloncompbasis ] @xmath106 where @xmath107 and @xmath108 are computational basis states such that @xmath109 and @xmath110 here @xmath111 and @xmath112 . a general state of multiple qubits can be expressed in terms of linear combinations and tensor products of basic operators such as @xmath113 and thus eqs . can be used to describe the action of any number of channels acting on multiple qubits . eq implies the counter - diagonal terms of the matrix representing @xmath13 are attained from those of @xmath104 by multiplication by @xmath114 eq . implies that the diagonal terms of the matrix representing @xmath13 result from rearrangements of those of @xmath104 multiplied by various factors of @xmath115 and @xmath116 . the remaining terms are again all @xmath117 . an additional simplification , demonstrated in appendix [ app : rhofdiag ] is that @xmath118 thus the final density operator has a form similar to that for @xmath104 , i.e.@xmath119 where @xmath120 \\ & + i { c_{\textrm{f}\ ; x}}\ ; \biggl [ { \lvert x \rangle}{\langle n - x \lvert } - { \lvert n - x \rangle } { \langle x \lvert } \biggr ] . \end{split } \label{eq : rhofinal}\ ] ] eq . yields @xmath121 for the counter - diagonal terms where @xmath96 is the number of zeroes in the bit string for @xmath98 . determining the diagonal terms is further simplified by noting that the preparatory unitary is invariant under interchange of any two qubits and that this is also true of the channel operation provided that the channel either acts on each qubit that is interchanged or that it acts on neither of the qubits that are interchanged . the invariance of the preparatory unitary is readily apparent from its structure , as shown in fig . [ fig : uprep ] . more precisely , suppose that @xmath122 is the unitary operation which swaps the qubits labeled @xmath63 and @xmath123 . then if @xmath124 ( recall that the channel acts on the rightmost @xmath8 bits ) , @xmath125 with an identical result when @xmath126 . thus if the bit string for @xmath98 is @xmath127 , then @xmath128 is invariant under the interchange of any two of @xmath129 and is also invariant under the interchange any two of @xmath130 ( it is not in general invariant under interchanges between one element from one of these sets and another from the other set ) . for example , if the channel is applied the two least significant qubits out of a total of five qubits then @xmath131 ( or , using a binary representation for the index subscript , @xmath132 ) and @xmath133 ( or @xmath134 ) . a consequence of this is that @xmath135 only depends on the number of zeroes in the leftmost @xmath136 bits of the binary representation of @xmath98 and also the number of zeroes in the rightmost @xmath8 bits of the binary representation of @xmath98 . this consideration and a derivation , provided in appendix [ app : rhofdiagterms ] , yields @xmath137 where @xmath138 is the number of zeroes in the leftmost @xmath136 bits ( i.e. @xmath139 ) of the binary representation of @xmath98 , @xmath140 is the number of zeroes in the rightmost @xmath8 bits ( i.e. @xmath141 ) of the binary representation of @xmath98 , @xmath142 and @xmath143 . clearly @xmath135 depends on @xmath144 and @xmath8 and @xmath3 , via @xmath145 and @xmath146 . since @xmath3 is implicit throughout this work and @xmath8 is fixed by the number of channel invocations we will find it convenient to express @xmath135 as @xmath147 according to eqs . and , @xmath148 is a sum of operators , each acting on a two dimensional space , which have mutually orthogonal supports . thus prop . 2 of @xcite implies that the quantum fisher information is @xmath149 where @xmath150 is the quantum fisher information associated with @xmath151 a derivation ( see appendix [ app : qfifirst ] ) based on this and eq . yields @xmath152 \label{eq : qfidepolone}\end{aligned}\ ] ] where the dot indicates differentiation with respect to @xmath153 although the sum in eq . is over @xmath98 , the terms indexed as such depend on @xmath154 and @xmath155 and it follows that the quantum fisher information can be expressed as a sum over these . in appendix [ app : qfifirst ] it is shown that _ if @xmath156 then _ @xmath157 \label{eq : qfidepoltwomlessthann}\end{aligned}\ ] ] where @xmath158 is given by eq . . _ if @xmath159 _ then @xmath160 \label{eq : qfidepoltwomequalsn}\end{aligned}\ ] ] where @xmath161 is again given by eq . . in both cases @xmath162 is given by eq . . these represent the simplest expressions that we have been able to obtain for the quantum fisher information for the correlated state protocol . the quantum fisher information per channel invocation is @xmath163 we aim to compare the performance of the correlated state protocol to that of the sqsc protocol _ in terms of the quantum fisher information per channel use and given that the initial qubit polarizations are the same . _ we define the _ correlated state protocol quantum fisher information gain _ as @xmath164 and whenever this exceeds @xmath84 , the correlated state protocol outperforms the sqsc protocol . the expression for the quantum fisher information is a rational polynomial in @xmath42 and @xmath3 and even with as few as two qubits , this is too complicated to yield insight into the estimation accuracy for the correlated state protocol for all values of polarization and the parameter . while the same applies to the gain , aspects of its behavior can be illustrated graphically . first , suppose that there is only one channel invocation ( @xmath165 ) . [ fig : gainplotssingle ] illustrates the gain for the examples @xmath166 and @xmath167 . and @xmath40 [ fig : gainplotssingle ] ] these suggest that when the channel is invoked once only , the correlated state protocol will be more accurate than the sqsc protocol , assuming the same initial state polarization for both . however , a general proof of this is elusive ; the methods of @xcite used to prove a similar fact for the phase - flip channel are not immediately applicable here . second , suppose that there are multiple channel invocations ( @xmath168 ) . [ fig : gainplotsmultiple ] illustrates the correlated state gain of eq . for various numbers of channel invocations on @xmath169 qubits . all plots are over the range @xmath170 . [ fig : gainplotsmultiple ] ] these show that when the channel is invoked multiple times , there are polarizations and parameter values for which the the correlated state protocol will be more accurate than the sqsc protocol but there are also parameter values for which it is inferior . given that the channel is invoked multiple times , the correlated state protocol can also be compared to the sequential channel protocol that uses the same number of channel invocations . we define the correlated vs sequential channel gain as @xmath171 where the number of channel invocations is the same for the two protocols . qubits . plots on the left are over the range @xmath172 . plots on the right provide details for @xmath173 [ fig : gaincorrmultvsseq ] ] figure [ fig : gaincorrmultvsseq ] suggests that when the channel is invoked a fixed number of times , the correlated state protocol is more accurate than the sequential protocol , using the same number of channel invocations , over all parameter values and for all polarizations . in some situations the initial state polarization is very small , i.e. @xmath174 this is true of room - temperature solution state nmr @xcite , where typically @xmath175 when @xmath176 the sqsc protocol quantum fisher information becomes @xmath177 and the sequential channel use protocol quantum fisher information per channel use becomes @xmath178 each to lowest order in @xmath179 in appendix [ app : smallrqfi ] we show that the correlated state quantum fisher information per channel use becomes @xmath180 to lowest order in @xmath42 . here the sequential channel use gain approximates to @xmath181 eq . shows that the gain exceeds @xmath84 when the depolarizing channel parameter satisfies @xmath182 where @xmath183 is the characteristic parameter cutoff . the characteristic parameter cutoff , plotted in fig . [ fig : lambdacutoff ] , is a monotonically increasing function of @xmath8 , with a minimum value of @xmath184 attained as @xmath185 and a maximum value of @xmath84 attained as @xmath186 . this indicates that for any given parameter value , there will be a maximum number of channel invocations beneath which the sequential channel use protocol offers enhanced parameter estimation accuracy and beyond which it will not do so . figure [ fig : lambdacutoff ] shows that as @xmath3 increases , accuracy gains will persist with increasing number of channel invocations . note that the larger @xmath3 is , the weaker the depolarization is , and for such weak depolarization cases enhanced estimation accuracies can be attained with multiple sequential channel invocations . . [ fig : lambdacutoff ] ] thus , _ for low polarizations , the sequential channel use protocol can yield a greater estimation accuracy per channel invocation than the sqsc protocol . _ however , this requires that the parameter value exceed the characteristic parameter cutoff and fig . [ fig : lambdacutoff ] indicates that such advantages will be absent for low enough parameter values and , even when there is an advantage , increasing the number of channel invocations will eventually eliminate this advantage . the correlated state protocol gain ( versus the sqsc protocol ) approximates to @xmath187 we consider two cases separately . first , with _ only one channel invocation the correlated state protocol offers an @xmath5-fold gain in accuracy regardless of the parameter value , for low initial polarization states . _ this is strikingly different from the pure state case where the optimal accuracy is attained @xcite with a single channel invocation on one of only two qubits ( @xmath166 and @xmath165 ) ; by contrast when the initial state polarization is low and the channel is only invoked on one qubit , then adding more qubits always increases the estimation accuracy . second , with multiple channel invocations , eq . indicates that enhanced estimation accuracy will occur if @xmath188 the right hand side forms a new cutoff which is again an increasing function of the number of channel invocations but a decreasing function of the total number of qubits . thus increasing the number of spectator qubits while leaving the number of channel invocations fixed extends the range of parameter values over which the correlated state protocol is advantageous . furthermore eq . shows that adding more spectator qubits while keeping the number of channel invocations fixed will increase the quantum fisher information per channel use ; the increase can be arbitrarily large by adding sufficiently many spectator qubits . _ unlike the pure initial state case , for low polarization initial states , there will always be an advantage to adding additional spectator qubits . _ separately , when @xmath156 we can assess whether one additional channel invocation yields any advantage . . implies that starting with @xmath8 channel invocations and adding one more invocation increases the quantum fisher information if @xmath189 , decreases it if @xmath190 and makes no difference if @xmath191 thus the optimal number of channel invocations in the correlated state protocol is @xmath192 unless @xmath193 is an integer , in which case , @xmath194 examples of this and the associated gains are provided in table [ tab : optinvocations ] . ccd @xmath3 & @xmath195 & + @xmath196 & @xmath84 & 1.00 n + @xmath197 & @xmath198 & 1.28 n + @xmath199 & @xmath200 & 2.15 n + @xmath201 & @xmath202 & 3.97 n + @xmath203 & @xmath204 & 18.67 n + @xmath205 & @xmath206 & 37.07 n when @xmath3 is close enough to @xmath84 , this offers considerable opportunity for improved accuracy . again _ unlike the pure initial state case , for low initial polarization , increasing the number of channel invocations in the correlated state protocol can result in enhanced estimation accuracy , per channel use . _ in many physical situations every qubit is subjected to the depolarizing channel . when the channel is invoked on all qubits of the correlated state protocol ( @xmath207 ) , the gain becomes @xmath208 and this implies that the correlated state protocol provides enhanced accuracy whenever @xmath209 this lower bound is plotted in fig . [ fig : lambdacutoff ] . this is again an increasing function of @xmath8 , approaching @xmath210 as @xmath211 . the slightly lower cutoff implies that there is somewhat greater scope for enhanced parameter estimation when all qubits are subjected to the channel . however , for a given parameter value , this cutoff establishes an upper bound on the number of qubits / channel invocations for which the correlated state protocol provides any gain . we can again ask when adding one more qubit and channel invocation gives any advantage . equation with @xmath207 implies that if there are originally @xmath8 channel invocations then adding one more gives a greater quantum fisher information per channel use if @xmath212 where @xmath213 unless @xmath214 is an integer , in which case , @xmath215 examples of this and the associated optimal gains are provided in table [ tab : optinvocationsall ] . ccd @xmath3 & @xmath195 & + @xmath216 & @xmath84 & 1.00 + @xmath196 & @xmath217 & 2.16 + @xmath199 & @xmath218 & 15.01 + @xmath201 & @xmath219 & 56.96 + @xmath220 & @xmath221 & 155.03 + @xmath203 & @xmath206 & 1367.00 clearly , for relatively weak depolarizing channels , i.e. @xmath222 substantial enhancements of estimation accuracy are possible with the correlated state protocol . finally , the correlated state protocol and sequential channel protocol can be compared in the low initial polarization limit , given the same number of channel invocations . here the correlated vs sequential channel gain as @xmath223 and thus when @xmath224 the correlated state protocol will always offer a better estimation accuracy than the sequential channel protocol provided that the number of channel invocation for each is the same . the presence of quantum correlations such as entanglement or discord in the correlated state protocol may be assessed for the @xmath166 along the same lines as was done for the pauli channel parameter estimation of @xcite . then eqs . , and yield that with @xmath8 channel invocations , @xmath225 in the computational basis . the state of the system prior to channel invocation can be obtained from this by setting @xmath53 the presence of entanglement can be assessed by considering partial transpose of @xmath148 and this is @xmath226 the state is separable if @xmath227 is positive @xcite . the eigenvalues of this will only be positive if and only if @xmath228 thus the state prior to channel invocation will be separable if and only if @xmath229 the correlated state protocol can give estimation gains as @xmath230 and this clearly corresponds to a state which is separable throughout the protocol . thus the correlated state protocol gains can not always be ascribed to the presence of entanglement . quantum discord describes correlations between quantum systems in terms of mutual information and measures the extent to which these exceed any possible classical correlations between the systems @xcite . it is possible to compute the quantum discord for states with the form @xcite @xmath231 where @xmath232 are all real and @xmath233 defining [ eq : evalsforq ] @xmath234 and @xmath235 yields that the quantum discord is @xcite @xmath236 following the scheme of @xcite , the final density operator can be brought into the form of eq . by applying the single qubit unitary @xmath237 to each qubit and this will not alter the discord . then @xmath238 and @xmath239 since @xmath40 again the discord for the correlated state protocol prior to channel invocation can be determined by setting @xmath86 and this gives @xmath240 then @xmath241 whenever @xmath242 and @xmath243 when @xmath244 implies that the state prior to channel invocation has non - zero discord whenever the polarization of each qubit is non - zero . in this sense , we can regard the correlated state protocol as using states that have quantum correlations . we have considered estimation of the depolarizing channel parameter for various situations where the available initial state of each qubit is mixed and we quantified the accuracy of each in terms of the quantum fisher information after channel invocation . we showed that for certain ranges of initial state polarization and channel parameter , a protocol that uses correlated states prior to channel invocation attains a larger quantum fisher information per channel invocation than either a protocol which uses product states throughout or another protocol which uses a sequence of channel invocations on a single qubit . for qubits with small initial polarizations we showed that using more than two qubits and more than two channel invocations can be advantageous ; this is in contrast to the situation where qubits in pure initial states are available . our results suggest that , as with pauli channel parameter estimation , a single invocation of the channel on many correlated qubits gives gains for all parameter values and initial polarizations . we were only able to prove this for the depolarizing channel in the limit as the initial polarization approaches zero , but we were also unable to generate a counterexample . this was not central to our results but it does merit further investigation . the source of the gain can not always be attributed to the presence of entangled or non - separable states prior to channel invocation ; we demonstrated this for the two qubit example . we also showed that there is non - zero discord present prior to channel invocation in this example . a full analysis of the evolution of the discord and its possible relationship to gains in the quantum fisher information might provide insight into the source of these accuracy gains ; this is beyond the scope of our article and remains to be investigated . finally , we chose one particular preparatory unitary transformation for the analysis of the correlated state protocol ; that this yields gains in some cases is interesting enough . however it is unclear whether this is optimal or whether there exists another parameter independent unitary that offers greater gains . the behavior of @xmath245 with respect to @xmath42 can be assessed by first setting @xmath246 . then @xmath247 gives that @xmath248 since @xmath42 increases as @xmath98 increases this proves the fact that @xmath245 is a monotonically decreasing function of @xmath42 . to assess the behavior of @xmath245 with respect to @xmath3 , let @xmath249 and note that @xmath250 then @xmath251 and @xmath252 thus @xmath253 the term @xmath254 is negative since @xmath255 and @xmath256 and thus decreases as @xmath257 increases . the remaining term @xmath258 increases as @xmath257 increases . thus @xmath245 decreases as @xmath257 increases . however , @xmath257 decreases as @xmath3 increases and this proves the fact that @xmath245 is a monotonically increasing function of @xmath3 . for the pure initial state case , @xmath41 and eq . gives @xmath259 where @xmath260 . then @xmath261 gives @xmath262 this is a monotonically decreasing function of @xmath263 and thus attains a maximum of @xmath84 at @xmath264 thus @xmath265 for a pure initial state , where the lower bound stems from the fact that the gain is a ration of positive quantities . for the mixed initial state case , @xmath245 is a monotonically increasing function of @xmath3 and attains a maximum of @xmath8 when @xmath82 this shows that @xmath266 when @xmath267 eqs . give that the diagonal terms of @xmath148 only arise from the action of the channel on @xmath268 \label{eq : rhoprepdiag}\ ] ] and that this will be mapped to the diagonal component of the final density operator , denoted @xmath269 and defined via @xmath270 a useful expression for @xmath269 involves the following representation of the channel on any diagonal operator . eq . gives @xmath271 thus if the channel acts on the rightmost qubit , labeled @xmath84 , then its action on the diagonal component of the density operator is @xmath272 where @xmath273 and the superscript on @xmath274 indicates action on qubit @xmath84 . suppose that the channel acts on qubit @xmath84 and then on qubit @xmath198 . this maps the diagonal component of the prepared density operator as @xmath275 \nonumber \\ & & + q^2 { \hat{x}}_2{\hat{x}}_1 { \hat{\rho}_\textrm{prep diag}}{\hat{x}}_2{\hat{x}}_1 \end{aligned}\ ] ] where @xmath276 this extends to action of the channel once on each of qubits @xmath277 and gives @xmath278 where the inner sum is over all @xmath279 that satisfy @xmath280 and range over the entire possibilities starting with @xmath281 to @xmath282 if @xmath283 the inner sum is taken to mean @xmath284 . for example , if @xmath285 and @xmath286 there are three indices @xmath287 and the summation range yields the following possibilities for @xmath288 : @xmath289 this consists of the set of all flips on three out of the four qubits with least significant label . a more symmetrical alternative is @xmath290 where the sum is over all @xmath291 where no two indices are equal . we introduce the following notation to render this more compact . let @xmath292 where , @xmath291 and no two elements in this list are equal . then define @xmath293 it follows that @xmath294 we use this to prove that @xmath295 to do so , note that @xmath296 and let @xmath297 thus , using @xmath298 @xmath299 eq . gives @xmath300 since @xmath301 commutes with @xmath302 . . gives @xmath303 this and eq . let the binary representation of @xmath98 be @xmath305 then @xmath306 and eq . give @xmath307 with the same constraints for the inner sum as for eq . and where @xmath308 . we shall evaluate the inner sum for any value of @xmath309 , which represents the number of flips on the bits within the binary representation of @xmath98 . the fact that the diagonal terms of @xmath148 only depend on the number of ( and not their order ) zeroes , @xmath154 , in the @xmath136 most significant bits and also the number of zeroes , @xmath155 , in the @xmath8 least significant bits implies that it is sufficient to consider @xmath135 where the binary representation of @xmath98 is @xmath310 here eqs . and give @xmath311 with the same limits for the inner sum . these limits can be modified by observing that @xmath312 only depends on the number of zeroes and ones in the binary representation of @xmath98 together with the number and arrangement of flips . suppose that , of the @xmath309 flips , @xmath313 occur on the rightmost cluster of @xmath155 bits . thus @xmath314 occur on the second to right cluster of @xmath315 bits . then the bit representation , @xmath316 , contains @xmath154 zeroes from the leftmost cluster , @xmath117 from the next cluster to the right , @xmath317 from the second to right cluster and @xmath318 from the rightmost cluster . thus the total number of zeroes in this bit string are @xmath319 and @xmath320 the exact location of the flips within each cluster is irrelevant . thus within the inner sum of eq . there are @xmath321 ways of attaining @xmath322 , consisting of @xmath323 ways of flipping @xmath313 bits in the rightmost cluster together with @xmath324 ways of flipping @xmath314 in the second to right cluster . finally not every value of @xmath313 is possible and these are constrained by @xmath325 ( at most @xmath155 flips in the rightmost cluster ) and @xmath326 ( at most @xmath315 flips in the second to right cluster ) . taken together these give @xmath327 combining these yields eq . . we shall show that @xmath328\end{aligned}\ ] ] and the result will follow from eq . . the expression for @xmath329 comes from the eigenvalue decomposition of @xmath151 the eigenvalues are of @xmath330 @xmath331 and the associated eigenvectors are @xmath332 the fact that the eigenvalues are independent of @xmath3 means that the second term on the right hand side of eq . is zero and the first term then yields @xmath333 straightforward algebra yields the desired result . the expression for the quantum fisher information can be rewritten as a sum over @xmath154 and @xmath155 . the only relevance that each value of @xmath98 has for the summand is that it determines @xmath334 and @xmath335 typically many values of @xmath98 will yield the same combination of values of @xmath154 and @xmath155 . to account for this , note that the sum in eq . runs from @xmath336 to @xmath337 ( in binary notation ) . the fact that the leftmost bit is always zero gives two distinct cases . in the first , where @xmath338 this bit structure reveals that @xmath339 and @xmath340 . for any choice of @xmath154 , there are @xmath341 ways in which the zeroes can appear in the leftmost @xmath136 bits , given that the leftmost single bit must be zero . there are independently @xmath342 ways in which the zeroes can appear in the rightmost @xmath8 bits . thus there are @xmath343 values of @xmath98 in the range of the sum which result in exactly the same pair of values @xmath154 and @xmath155 . for each the contribution of the summand is identical . this means that the terms within the summand can be expressed in terms of @xmath154 and @xmath155 ( rather than @xmath98 ) and the result sum reduces to that of eq . . in the second , where @xmath207 , clearly @xmath344 and @xmath345 a similar counting argument reveals that there are @xmath346 ways in which the zeroes can appear in the rightmost @xmath347 bits , since the rightmost bit must be zero . a similar argument then results in eq . . the correlated state protocol quantum fisher information , eqs . and , requires @xmath162 and @xmath348 which in turn requires @xmath349 eqs , and yield , to second order in @xmath42 , @xmath350}{2^{n+1}}.\end{aligned}\ ] ] the latter of these gives @xmath351 where @xmath352 and @xmath353}{2^{n+1}}.\end{aligned}\ ] ] both @xmath354 and @xmath355 are independent of @xmath42 . then @xmath356 which can be demonstrated indirectly by noting that , when @xmath357 each diagonal term of the final density operator equal @xmath358 however , when @xmath357 @xmath359 and thus @xmath360 . the channel does not alter this and thus , when @xmath244 , @xmath361 . it follows that @xmath362 this fact can independently be verified by explicitly evaluating the three summations in @xmath358 note that these imply @xmath363 then eqs , and , yield that for the terms in eqs . and @xmath364 thus if @xmath156 , eqs . becomes @xmath365 the two summations in eq give @xmath366 and this yields the result of eq . . a similar derivation gives the same result when @xmath367 48ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty link:\doibase 10.1103/revmodphys.52.341 [ * * , ( ) ] link:\doibase 10.1103/physrevd.23.1693 [ * * , ( ) ] link:\doibase 10.1103/physreva.43.3795 [ * * , ( ) ] @noop _ _ ( , , ) link:\doibase 10.1103/physrevlett.72.3439 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.79.3865 [ * * , ( ) ] http://stacks.iop.org/0305-4470/39/i=26/a=015 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.96.010401 [ * * , ( ) ] link:\doibase 10.1142/s0219749909004839 [ * * , ( ) ] @noop _ _ ( , , ) \doibase doi:10.1038/nphoton.2011.35 [ * * , ( ) ] link:\doibase 10.1103/physreva.54.3824 [ * * , ( ) ] link:\doibase 10.1103/physreva.56.3470 [ * * , ( ) ] @noop _ _ ( , , ) link:\doibase 10.1103/physreva.82.012306 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.86.5811 [ * * , ( ) ] http://stacks.iop.org/1367-2630/11/i=1/a=013034 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.92.257901 [ * * , ( ) ] link:\doibase 10.1002/mop.21210 [ * * , ( ) ] link:\doibase 10.1364/josab.25.000668 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.105.180402 [ * * , ( ) ] link:\doibase 10.1103/physreva.85.042112 [ * * , ( ) ] link:\doibase 10.1103/physreva.63.042304 [ * * , ( ) ] link:\doibase 10.1103/physreva.66.022308 [ * * , ( ) ] http://stacks.iop.org/0305-4470/36/i=29/a=314 [ * * , ( ) ] http://stacks.iop.org/1751-8121/44/i=20/a=205306 [ * * , ( ) ] link:\doibase 10.1103/physrevx.1.021022 [ * * , ( ) ] link:\doibase 10.1103/physreva.87.032301 [ * * , ( ) ] link:\doibase 10.1103/physreva.72.042327 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.100.100503 [ * * , ( ) ] link:\doibase 10.1140/epjd / e2010 - 00195 - 3 [ * * , ( ) ] link:\doibase 10.1103/physreva.83.063836 [ * * , ( ) ] link:\doibase 10.1103/physreva.88.040102 [ * * , ( ) ] @noop _ _ ( , , ) @noop ( ) , in link:\doibase 10.1007/978 - 4 - 431 - 54493 - 7_4 [ _ _ ] , ( , ) pp . http://stacks.iop.org/0305-4470/33/i=24/a=306 [ * * , ( ) ] link:\doibase 10.1103/physreva.69.022303 [ * * , ( ) ] link:\doibase 10.1103/physreva.70.012317 [ * * , ( ) ] link:\doibase 10.1103/physreva.72.052334 [ * * , ( ) ] \doibase http://dx.doi.org/10.1109/tit.2008.929940 [ * * , ( ) ] link:\doibase 10.1038/nphys1958 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.98.090501 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.77.1413 [ * * , ( ) ] link:\doibase 10.1016/s0375 - 9601(96)00706 - 2 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.88.017901 [ * * , ( ) ] link:\doibase 10.1103/physreva.77.042303 [ * * , ( ) ] link:\doibase 10.1103/physreva.84.042313 [ * * , ( ) ]

in our planetary system , most objects go around the sun following counterclockwise ( prograde ) orbits as viewed from above the north pole of our star . however , a large number of comets orbit the sun clockwise ( retrograde ) and a substantial percentage of these objects have highly inclined orbits . they are believed to have their origin in the oort cloud ( see e.g. duncan 2008 ) , a vast and remote spherical reservoir of cometary material tens of thousands of astronomical units in diameter that completely surrounds the solar system . out of 3265 objects currently classified by the jet propulsion laboratory ( jpl ) small - body database as comets , 1918 move in retrograde orbits ( 58.7% ) . for long - period comets the retrograde fraction is 49.7% , for halley - type comets is 35.7% and for jupiter - family comets is 16.7% . in contrast , only 53 of the 641502 currently catalogued asteroids orbit the sun clockwise ( 0.008% ) . for asteroids in the outer solar system ( catalogued as centaurs or trans - neptunian objects ) the fraction of retrograde objects is 2.7% ( 48 out of 1804 ) . therefore , most retrograde asteroids have been found in the region of the giant planets and beyond . at 11.0% ( 30 out of 272 catalogued objects in the jpl small - body database ) , centaurs have the highest retrograde fraction among all the dynamical groups or asteroid families . centaurs are a transient asteroidal population whose orbits cross those of the outer planets ( see e.g. di sisto and brunini 2007 ) . most centaurs may have their origin in the trans - neptunian belt , some may have come from the oort cloud . the main source of centaurs remains to be discovered . they are probably an heterogeneous population with multiple sources ( see e.g. di sisto et al . 2010 ) , perhaps none being dominant ( but see horner and lykawka 2010 ) . the most intriguing of the known centaurs are those moving in retrograde orbits . the study of retrograde centaurs , especially those following high inclination trajectories , may represent a rare opportunity to look at objects that have suffered little surface alteration over the age of the solar system ( sheppard 2010 ) . the three largest known ( in terms of absolute magnitude , @xmath0 ) retrograde centaurs are ( 342842 ) 2008 yb@xmath3 ( @xmath0 = 9.3 mag ) , 2011 mm@xmath4 ( @xmath0 = 9.3 mag ) and 2013 lu@xmath5 ( @xmath0 = 8.1 mag ) ; they all move in near - polar orbits . minor bodies following eccentric , high - inclination paths spend most of the time well beyond the plane of the solar system and away from the destabilizing influence of the giant planets , although close encounters with these planets are still possible at the nodes . here , we show that , far from being just dynamical curiosities , the three largest retrograde centaurs may be the key to identify the size threshold for catastrophic collisions in the early solar system . on the other hand , they are also submitted to retrograde mean motion resonances that make them dynamically unstable , emphasizing the fact that these objects have not remained in their current orbits for long . this paper is organized as follows . in sect . 2 , we present a comparative statistical analysis of the properties of known prograde and retrograde centaurs that reveals compelling reasons to single out the largest retrograde objects . the current dynamical status of these large retrograde centaurs is studied in sect . 3 , where the numerical model used in our @xmath6-body simulations is also briefly discussed . in sect . 4 , we consider the implications of our results and in sect . 5 we summarize our conclusions . following alexandersen et al . ( 2013 ) , let us consider the group of minor bodies whose semi - major axes are between 6 and 34 au . there are 266 such asteroids currently ( as of 2014 may 5 ) in the jpl small - body database . this number is slightly less than the total number of known centaurs quoted above . out of them , 236 ( 88.7% ) follow prograde orbits and 30 ( 11.3% ) follow retrograde orbits . figure [ pro - ret ] shows the distribution in orbital parameter space and @xmath0 of prograde ( left - hand panels ) and retrograde ( right - hand panels ) centaurs . even if the number of known retrograde centaurs is small , it is highly improbable that the two centaur populations have a common origin . the distribution in absolute magnitude , which is a proxy for size , clearly shows that retrograde centaurs are unlikely to be the result of gravitational scattering of originally prograde centaurs , unless we assume that smaller objects are more efficiently scattered to become retrograde than larger ones ( see the @xmath7- and @xmath8-panels ) or that smaller retrograde objects are easier to identify than their prograde counterparts . most prograde orbits have eccentricities in the range 0.20.6 and relatively low inclinations , @xmath920 . the small but noticeable group of prograde orbits with inclination @xmath1063 in fig . [ pro - ret ] ( left - hand panel ) is consistent with predictions made by e.g. gallardo ( 2006 ) on the existence of objects submitted to the kozai resonance at that critical inclination . the relative excess of objects with argument of perihelion close to 0or 180could be due to observational bias because only objects with declinations @xmath1124are observed ( see de la fuente marcos and de la fuente marcos 2014 ) . the raw distribution in @xmath0 of prograde orbits shows evidence of the so called divot or sudden decrease in numbers for @xmath12 9 mag ( or a diameter of nearly 100 km ) as found by e.g. shankman et al . this has been interpreted by those authors as the relic signature of the size distribution of the primordial planetesimal population , with the larger objects being the remnants . a divot in the size distribution of trans - neptunian objects was predicted by fraser ( 2009 ) using the results of collisional evolution calculations . a lack of objects in the size range 120 km was also anticipated by charnoz and morbidelli ( 2007 ) for minor bodies in the scattered disk and the oort cloud . this feature had been previously identified and confirmed observationally for the size range 2560 km by bernstein et al . ( 2004 ) , fuentes and holman ( 2008 ) , and fraser and kavelaars ( 2009 ) . this break in the size distribution may signal the onset of collisional evolution . schlichting et al . ( 2013 ) consider that , for trans - neptunian objects , the observed size distribution above a radius of @xmath1030 km is primordial . they also found that there is a significant deficit of bodies with size @xmath1010 km and a strong excess of bodies of nearly 2 km in radius ; this excess leaves a permanent signature in the size distribution of these objects that is still clearly observable after 4.5 gyr of collisional evolution . fraser et al . ( 2014 ) have found that the observed size distributions are difficult to explain within the currently accepted standard model of planetesimal formation . the distribution in @xmath0 for retrograde orbits in fig . [ pro - ret ] ( left - hand panel ) exhibits a most unusual behaviour . there are only three objects ( 10% ) with @xmath1 10 mag : ( 342842 ) 2008 yb@xmath3 ( @xmath0 = 9.3 mag ) , 2011 mm@xmath4 ( @xmath0 = 9.3 mag ) and 2013 lu@xmath5 ( @xmath0 = 8.1 mag ) . the rest have @xmath2 12 mag ( 73.3% have @xmath2 14 mag ) . although the number of known retrograde centaurs is small , this result can be seen as a dramatic confirmation of the existence of a break in the size distribution as predicted by e.g. fraser ( 2009 ) . there is an obvious deficit of objects with @xmath0 in the range 1014 mag although they must be significantly easier to detect , if they do indeed exist , than fainter objects of a size below 10 km . although the available sample is far from complete and likely biased , it should not be biased in favour of 1-km objects against those in the size range 10100 km . the scarcity of intermediate - size objects must be real and it may have a collisional , dynamical or even primordial origin ( or a combination of the three ) . in principle , and following fraser ( 2009 ) , schlichting et al . ( 2013 ) or shankman et al . ( 2013 ) , some or all the large retrograde centaurs may represent a remnant population of primordial planetesimals . in addition , the distribution in @xmath0 of retrograde centaurs suggests that at least two distinct populations of rather different dynamical origin are present . in support of this interpretation , fig . [ aecen ] ( see also fig . [ real ] ) shows the distribution in the ( @xmath13 , @xmath7 ) plane of both groups and comets in the same semi - major axis range : the three largest retrograde centaurs clearly stand out ( bottom panel ) . the brightest ( largest ) objects also appear to be dynamically different . the distribution of semi - major axes and eccentricities of retrograde objects with @xmath2 12 resembles that of members of the detached or scattered disk populations ( levison and duncan 1997 ; gomes 2011 ) . in terms of the size of their members , these trans - neptunian populations appear to be of collisional origin and , dynamically , they may have reached their current state after suffering strong gravitational perturbations . the surprising abundance of retrograde minor bodies at @xmath14 mag may signal a high density of objects with size @xmath15 km as predicted by fraser ( 2009 ) or , more recently , belton ( 2014 ) . , @xmath7 ) plane of prograde ( top panel ) and retrograde ( bottom panel ) centaurs and comets with semi - major axes in the range 634 au . there are 145 prograde and 20 retrograde comets in that semi - major axis range . see also fig . the three largest retrograde centaurs , namely ( 342842 ) 2008 yb@xmath3 , 2011 mm@xmath4 and 2013 lu@xmath5 , are clear dynamical outliers ; they have the largest perihelia . ] , @xmath13 ) plane . the dark gray area represents the eccentricity / semi - major axis combination with periapsis between the perihelion and aphelion of jupiter , the light gray area shows the equivalent parameter domain if saturn is considered instead of jupiter . the orange area corresponds to the ( @xmath7 , @xmath13 ) combination with apoapsis between the perihelion and aphelion of uranus and the brown area shows the counterpart for neptune . ] it may be argued that we unfoundedly invoked the observed decrease in the number of prograde centaurs at magnitude @xmath12 9 to implicitly justify the division into two groups ( @xmath1 10 mag and @xmath2 10 mag ) in fig . [ aecen ] . however , such concern is completely unjustified as fig . [ aecen ] clearly shows . it is rather obvious from the figure that the three largest retrograde centaurs are clear dynamical outliers because they have the largest perihelia . out of the three largest retrograde centaurs , only 342842 is a relatively well - studied object ( sheppard 2010 ; pinilla - alonso et al . 2013 ; bauer et al . but , what is special or unique about these objects ? both 342842 and 2013 lu@xmath5 can suffer close encounters with jupiter ; 2011 mm@xmath4 may be considerably more stable . brasser et al . ( 2012 ) and volk and malhotra ( 2013 ) have found that retrograde or very high inclination centaurs can not have evolved from the trans - neptunian belt due to close encounters with the jovian planets . instead , they favour a source in the oort cloud for these objects . in the solar system , near - polar orbits are almost dynamically frozen in time and , potentially , they could be stable for hundreds of myr . this is particularly true if the nodes of the orbits of the objects remain relatively far from the paths of the planets . however , if the objects following these unusual trajectories are also submitted to mean motion resonances , their dynamical stability could be compromised even if they do not undergo close encounters with planetary bodies because they may experience chaotic diffusion . morais and namouni ( 2013a ) have brought to the attention of the astronomical community that minor bodies following retrograde orbits around the sun may also be trapped in a mean motion resonance with a planet . the same authors have identified several asteroids currently moving in retrograde resonance with jupiter and saturn ( morais and namouni 2013b ) . this discovery is of considerable practical interest because of its potential impact on our understanding of the early stages of the evolution of the solar system and also on the dynamics of transient populations like the centaurs . are the largest retrograde centaurs perhaps submitted to retrograde mean motion resonances with the giant planets ? the dynamics of resonant retrograde minor bodies has been studied by morais and namouni ( 2013a , 2013b ) . for an object in retrograde resonance with a planet that moves in a prograde orbit , the resonant argument can be written as @xmath16 , where @xmath17 and @xmath18 are the mean longitudes of the retrograde object and the prograde planet , respectively , @xmath19 is the longitude of perihelion of the retrograde object , @xmath20 , @xmath21 is the longitude of the ascending node and @xmath22 is the argument of perihelion . the mean longitude of a prograde object is given by @xmath23 , where @xmath24 is the mean anomaly , but the one for a retrograde object is written as @xmath25 . the numbers @xmath26 , @xmath27 and @xmath28 are integers with @xmath29 . if the two objects are in resonance , @xmath30 , where @xmath31 and @xmath32 are the orbital periods of the retrograde object and the prograde planet , respectively . if , over a given period of time , the resonant argument , @xmath33 , can take any value in the range 0360 , then it circulates but if it exhibits oscillations ( symmetric or asymmetric ) around a certain value ( usually 0or 180 ) , then we say that it librates . if the resonant argument librates , then the object is trapped in a @xmath26:@xmath27 retrograde ( or @xmath26:@xmath34 ) mean motion resonance with the planet . retrograde resonances are weaker than their prograde counterparts ( morais and namouni 2013a , 2013b ) . 0.35truecm [ cols="<,<,<,<,<",options="header " , ] [ elements ] in this work , the orbital evolution of the three largest retrograde centaurs is computed using the hermite integration scheme described by makino ( 1991 ) and implemented by aarseth ( 2003 ) . the standard version of this @xmath6-body sequential code is publicly available from the ioa web sitesverre / web / pages / nbody.htm ] . our numerical simulations include the gravitational perturbations of the eight major planets , the moon , the barycentre of the pluto - charon system and the three largest asteroids ( dwarf planet ( 1 ) ceres , ( 2 ) pallas and ( 4 ) vesta ) . initial positions and velocities are based on the de405 planetary orbital ephemerides ( standish 1998 ) referred to the barycentre of the solar system . together with the orbital elements , they have been obtained from the jpl on - line solar system data service ( giorgini et al . 1996 ) . in the integrations , the relative error in the total energy was always as low as @xmath35 or lower after a simulated time of 1 myr . the corresponding error in the total angular momentum is several orders of magnitude smaller . further details on the physical model and the validation of the code can be found in de la fuente marcos and de la fuente marcos ( 2012a ) . besides the calculations performed using the nominal orbits as provided by the jpl solar system data service , we have followed the evolution of 50 additional control orbits obtained from the nominal ones and the quoted uncertainties ( see table [ elements ] ) . their orbital elements have been sampled from a six - dimensional gaussian distribution around the nominal ones ( within 3@xmath33 with the exception of 2013 lu@xmath5 ) . in all the figures , @xmath36 = 0 coincides with the jd2456800.5 epoch unless explicitly stated . the dynamical evolution of the objects is followed for @xmath37500 kyr . , 2011 mm@xmath4 and 2013 lu@xmath5 in the time interval ( -500 , 500 ) kyr . relatively long episodes of horizontal ( resonant ) oscillations are visible . for 2011 mm@xmath4 , no large changes in the orbital elements are observed over the studied time interval . the dark gray area represents the ( @xmath7 , @xmath13 ) combination with periapsis between the perihelion and aphelion of jupiter , the light gray area shows the equivalent parameter domain if saturn is considered instead of jupiter . the orange area corresponds to the ( @xmath7 , @xmath13 ) combination with apoapsis between the perihelion and aphelion of uranus and the brown area shows that of neptune . these results correspond to the nominal orbits in table [ elements ] . ] but for the ( @xmath7 , @xmath8 ) evolution . ] figures [ ea ] , [ ei ] and [ all ] evidence that the overall orbital evolution of the three objects is rather different although all of them experience horizontal oscillations in the ( @xmath7 , @xmath13 ) plane as secular resonances modify @xmath7 at constant @xmath13 . these figures also show that , when @xmath22 is close to 0or 180 , ( 342842 ) 2008 yb@xmath3 can experience close encounters with jupiter at perihelion and with uranus at aphelion , 2011 mm@xmath4 can encounter saturn at perihelion and neptune at aphelion , and 2013 lu@xmath5 can find jupiter at perihelion and neptune at aphelion ( see also fig . [ real ] ) . encounters can only take place at the nodes of the orbit . for a retrograde orbit , the distance between the sun and the nodes ( fig . [ all ] , f - panels ) is given by @xmath38 , where the ` + ' sign denotes the descending node and the ` - ' sign the ascending node . in the following , we explore the dynamics of these unusual objects in more detail . asteroid ( 342842 ) 2008 yb@xmath3 was discovered on 2008 december 18 by r. h. mcnaught observing for the siding spring survey with the 0.5-m uppsala schmidt ( mcnaught et al . 2008 ) . when first observed , it had a @xmath39 magnitude of 18.1 . it was soon obvious that it follows a near - polar retrograde orbit characterized by @xmath13 = 11.62 au , @xmath7 = 0.44 and @xmath8 = 105(see table [ elements ] ) . its orbit is well determined with 466 observations spanning a data - arc of 1856 d. it is relatively large with @xmath0 = 9.3 mag or a diameter of 67 km for an albedo of 6.2% , although it is probably larger . sheppard ( 2010 ) found moderately red colours for this minor body , consistent with those of the neutral group of the centaurs , and concluded that due to its relatively short perihelion , the object may have undergone surface sublimation and interior recrystallization . pinilla - alonso et al . ( 2013 ) used photometry , spectroscopy and numerical simulations to show that this asteroid has a rotational period of 16.4 hr and , when compared with solar colours , is moderately red with low albedo ( perhaps depleted of ices ) , and it came from the trans - neptunian region . asteroid 342842 is currently trapped in an accidental 3:10 retrograde resonance with jupiter ( libration around 0 , see fig . [ yb3 ] , top panel ) . we call it accidental because the resonant argument circulates with a superimposed short - period oscillation . this behaviour is similar to the one found for plutino ( 15810 ) 1994 jr@xmath40 , a quasi - satellite of pluto ( de la fuente marcos and de la fuente marcos 2012b ) . this resonant episode with jupiter started nearly 30 kyr ago and it will cease in about 10 kyr . just after the episode , this minor body will be submitted to a 3:4 retrograde resonance with saturn ( asymmetric libration around 0 , see fig . [ yb3 ] , bottom panel ) for nearly 10 kyr . close encounters with jupiter and more distant flybys with saturn trigger the ejection from the retrograde resonances with jupiter and saturn . all the control orbits exhibit similar behaviour during the time - span plotted in fig . [ yb3 ] but the dynamical evolution after the close encounters differs significantly between control orbits . in about half the cases , the 3:4 retrograde resonance with saturn is observed prior and after the 3:10 retrograde resonance with jupiter . the time - scale necessary for two initially infinitesimally close trajectories associated with this object to separate appreciably , or @xmath7-folding time , is @xmath103 kyr . the object will be ejected towards the trans - neptunian region within the next few hundred kyr . our results indicate that this object is less dynamically stable than implied in pinilla - alonso et al . ( 2013 ) . . ] asteroid 2011 mm@xmath4 was discovered on 2011 june 24 by n. primak , a. schultz , s. watters , j. thiel and t. goggia observing for the pan - starrs 1 project with the 1.8-m ritchey - chretien telescope from haleakala ( woodworth et al . when discovered , its @xmath41 magnitude was 21.1 . it follows a path characterized by @xmath13 = 21.21 au , @xmath7 = 0.47 and @xmath8 = 100(see table [ elements ] ) . its orbit is relatively well determined with 92 observations spanning a data - arc of 396 d. with @xmath0 = 9.3 mag , it is very similar in size to ( 342842 ) 2008 yb@xmath3 although probably larger ( bauer et al . 2013 ) . asteroid 2011 mm@xmath4 is currently trapped in the 3:10 retrograde resonance with saturn ( asymmetric libration around 90 , see fig . [ mm4 ] , top panel ) . this resonant episode started nearly 14 kyr ago and it will end in about 3 kyr . prior to this episode , we observe that the critical angle alternates between circulation and asymmetric libration , indicating that the motion is chaotic . before and after the resonant interaction with saturn , this minor body was submitted to the 5:3 retrograde resonance with neptune ( asymmetric libration around 180 , see fig . [ mm4 ] , bottom panel ) . relatively close encounters with saturn and neptune are responsible for the ejection from the resonances . during the entire resonant phase , the argument of perihelion remains close to 0 , suggesting a behaviour similar to that of a kozai secular resonance ( kozai 1962 ) although both @xmath7 and @xmath8 remain relatively constant ( see fig . [ all ] ) . as in the previous case , all the control orbits exhibit similar behaviour during the time - span plotted in fig . [ mm4 ] . in some cases , the 3:10 retrograde resonance with saturn results in asymmetric libration around 180instead of 90 . our integrations indicate that this object can remain dynamically stable for several myr although this issue will be studied in more detail later . its @xmath7-folding time is @xmath1010 kyr though . it is not submitted to any typical secular resonance ( see fig . [ rellon ] ) . the cause of the relative stability of this object in comparison with the others , is connected with the position of its nodes ( see fig . [ all ] , panel f , middle column ) . for a minor body moving in a near - polar orbit , close encounters with major planets are only possible in the vicinity of the nodes . when @xmath22 is close to 0or 180 , the perihelion takes place within the ecliptic plane with one node situated at aphelion and the other at perihelion ; however , if @xmath22 is close to 90 or 270 , the perihelion occurs well away from the plane . in the case of 2011 mm@xmath4 , when @xmath22 is 0or 180 , encounters with saturn and neptune take place at perihelion and aphelion ; if @xmath22 is 90or 270 , encounters with uranus are possible at both nodes in a manner similar to what is observed for 83982 crantor ( 2002 go@xmath42 ) , a well - studied uranian co - orbital ( de la fuente marcos and de la fuente marcos 2013 ) . the period of this oscillation of the nodes is nearly 700 kyr . at present time , the closest approach to saturn is at 1.08 au but the hill radius of saturn is 0.41 au ; for neptune , the closest approach is at 0.94 au but its hill radius is 0.77 au . this explains why its motion is rather stable , the planetary perturbations exerted on 2011 mm@xmath4 are currently not strong enough to change the orbit appreciably . . ] , of 2011 mm@xmath4 with respect to the giant planets : referred to jupiter ( @xmath43 ) , to saturn ( @xmath44 ) , to uranus ( @xmath45 ) , and to neptune ( @xmath46 ) . in all cases , the resonant argument @xmath47 circulates over the entire simulated time interval although it is nearly in secular resonance with jupiter . these results are for the nominal orbit in table [ elements ] . ] asteroid 2013 lu@xmath5 was discovered on 2013 june 8 by a. boattini , r. e. hill and j. a. johnson observing with the 1.5-m telescope of the mt . lemmon survey ( bressi et al . 2013 ) . it had a @xmath39 magnitude of 21.7 when first observed . its orbit is very poorly determined with just 27 observations acquired during 7 d and it is mainly included here to encourage follow - up observations . its orbit is , in principle , the most eccentric of the ones discussed here . this object is also the largest , with @xmath0 = 8.1 mag or 50150 km . due to the large uncertainty associated to the orbit of 2013 lu@xmath5 , our discussion here is more speculative . assuming an object moving in an orbit like the nominal one ( @xmath13 = 19.25 au , @xmath7 = 0.63 , @xmath8 = 11700 , @xmath21 = 25457 , @xmath22 = 17905 and @xmath24 = 29787 ) , we observe that it is trapped in the 1:1 retrograde resonance with uranus ( asymmetric quasi - satellite behaviour , see fig . [ lu28 ] , middle panel ) . the object has remained in the neighbourhood of this resonance for nearly 200 kyr but its dynamical status is rather unstable because it is submitted to additional resonances with neptune and saturn ( see fig . [ lu28 ] , bottom and top panels ) , 2:1 and 4:11 retrograde , respectively . our results are consistent with previous studies of the 1:1 resonance at high inclination ( see e.g. brasser et al . all the control orbits suggest that it is a recent visitor from the trans - neptunian region . in this case , and given its very uncertain parameters , we have restricted our exploration of the orbital domain to orbits very close to the nominal one . its @xmath7-folding time is @xmath103 kyr . . the distribution in absolute magnitude of known retrograde centaurs with semi - major axis in the range 634 au exhibits a clear deficit of objects within 1014 mag and a significant excess at @xmath1014 mag . this is consistent with theoretical expectations based on collisional models ( fraser 2009 ) . objects with @xmath1 10 mag form a distinct group in dynamical terms . the largest known retrograde centaurs move in near - polar , comet - like orbits but no cometary activity has been detected yet on any of them . the analysis of the short - term orbital evolution of these objects shows that the group is heterogeneous in origin . all of them are currently trapped in transient retrograde resonances with the jovian planets . compared with the objects analysed by morais and namouni ( 2013b ) , the duration of the resonant episodes of the objects studied here is longer as they are globally more stable . asteroid ( 342842 ) 2008 yb@xmath3 will escape towards the trans - neptunian region in the near future and 2013 lu@xmath5 , a recent visitor from the trans - neptunian region , is the first known uranian retrograde co - orbital candidate ( a quasi - satellite ) but further observations are necessary to absolutely confirm that this object is indeed a retrograde companion to uranus . another interesting feature of the dynamics discussed above is that high - order resonances seem to be preferred by these initial conditions characterized by very high inclinations . this is , however , not surprising as pointed out by , for example , robutel and laskar ( 2001 ) or gallardo ( 2006 ) . the strength of high - order resonances in general is greater for high - inclination orbits . at high inclination , high - order resonances are strong enough to capture objects for relatively long periods of time . in our case , the high eccentricities are also a contributing factor to produce stronger dynamical effects ( see e.g. gallardo 2006 ) . gallardo ( 2006 ) has found that , for a given eccentricity and due to short - period planetary perturbations , high - inclination resonant orbits have a higher probability of survival than low - inclination ones . this is also consistent with the comparatively weaker stability observed for the retrograde resonances analysed by morais and namouni ( 2013b ) and pointed out above ; the orbits followed by the objects studied in their paper are not almost perpendicular to the ecliptic as in our case and their resonances are low order . in general , chaotic diffusion is enhanced for high - inclination , high - eccentricity orbits submitted to high - order resonances . asteroids 342842 and 2013 lu@xmath5 are not long - term stable but 2011 mm@xmath4 may be considerably more stable . in order to study the long - term stability of 2011 mm@xmath4 we use the regularized mixed variable symplectic ( rmvs ) integrator ( _ swift - rmvs3 _ ) which is part of the swift package ( levison and duncan 1994 ) . the swift package is available from h. levison web site.hal/swift.html ] the physical model here is similar to the one described in sect . 3 but now the earth and the moon are replaced by the barycentre of the system . the calculations do not consider mercury or any asteroids but they include the barycentre of the pluto - charon system , use a timestep of 50 d and run for @xmath375.0 gyr . relative errors in the total energy are @xmath48 . an ejection distance of 10000 au from the sun was used ; any test particle that reached that distance was removed from the calculations . as in the previous simulations , we have followed the evolution of 50 control orbits obtained from the nominal orbital elements of 2011 mm@xmath4 and their quoted uncertainties ( see table [ elements ] ) . these orbital elements have been sampled within 3@xmath33 from a six - dimensional gaussian distribution around the nominal ones . results from rmvs and hermite are consistent within the applicable timeframe . horner et al . ( 2004a , 2004b ) studied the dynamical half - lives of prograde centaurs and found values in the range 0.5432 myr . the mean half - life of their sample of centaurs was 2.7 myr . for a given orbit , a large number of control orbits , @xmath49 , with parameters close to those of the nominal one were integrated . these orbits were followed until they resulted in a collision with the sun or the ejection of the test particle from the solar system . the number of control orbits , @xmath6 , that still remain within the boundaries of the solar system ( in our case , within 10000 au from the sun ) after a time @xmath36 , is given by @xmath50 , where @xmath51 , @xmath17 is the decay constant and @xmath52 is the dynamical half - life of the objects . using _ swift - rmvs3 _ , we have applied the same approach to 2011 mm@xmath4 in an attempt to study its long - term stability . asteroid 2011 mm@xmath4 is clearly far more dynamically stable than most centaurs ; its mean dynamical half - life is 23.12 myr . however , the range in dynamical half - lives forward in time is 5.26310.32 myr and for the past injection into its present orbit is 3.561795.35 myr . in other words , this object probably arrived to the region of the giant planets nearly 23 myr ago and certainly not earlier than nearly 2 gyr ago . therefore , it is not a primordial minor body that has remained around its present location since the formation of the solar system like jovian or martian trojans . it will be ejected also within 23 myr and not later than 310 myr into the future . among the control orbits , we have found one that is particularly interesting ( @xmath13 = 21.1623 au , @xmath7 = 0.47301 , @xmath8 = 100.446 , @xmath21 = 282.604and @xmath22 = 7.224 , see fig . [ control ] ) . it was very long lived into the past ( 1795.35 myr ) and its associated dynamics may signal the presence of a reservoir of asteroids moving in high - inclination prograde orbits that are submitted to a kozai resonance as described by gallardo ( 2006 ) . in fig . [ control ] , the control orbit begins at a critical inclination of @xmath1063and very high eccentricity . from there , the eccentricity decreases and the initially prograde orbit turns into a retrograde one . if , somehow , a reservoir of objects moving with orbital inclinations @xmath1063and high eccentricities was created ( and may still survive ) at 200500 au in the remote past , it may be a viable secondary source for retrograde centaurs . the dynamical evolutionary pathway uncovered by our calculations , even if it has a statistically low probability , shows that the oort cloud is not the only possible source region for retrograde centaurs . the existence of additional stable reservoirs of rocky and icy bodies at 100300 au can not be simply casted aside . this interpretation is , however , tentative and should be treated with caution , as there is recent circumstantial evidence on the existence of a massive planet at 250 au from the sun ( trujillo and sheppard 2014 ) . if such planet does indeed exist , the architecture of the entire trans - plutonian region could be very different from the one assumed here , i.e. no truly massive bodies beyond neptune . a massive perturber in the region 100300 au may affect the dynamical evolution presented in figure [ control ] very significantly . brasser et al . ( 2012 ) , and volk and malhotra ( 2013 ) argued that retrograde or very high inclination centaurs have their origin in the oort cloud . fouchard et al . ( 2014 ) also pointed out that the oort cloud , and in particular its inner part , is a potential source of high inclination centaurs . long - period comets are believed to have their origin in the oort cloud ( see e.g. duncan 2008 ) . halley - type comets may come from the outer edge of the scattered disc ( levison et al . 2006 ) or the oort cloud ( emelyanenko et al . 2013 ) . jupiter - family comets appear to have their origin in the scattered disk ( levison and duncan 1997 ) . for the group of minor bodies whose semi - major axes are between 6 and 34 au , the fraction of retrograde orbits is 11.3% . there are 165 known comets with semi - major axes in the range 634 au . out of them , 20 objects or 12.1% move in retrograde orbits . the fractions of retrograde orbits in both groups are more similar than expected by chance alone . in principle , this fact argues strongly in favour of a dominant source in the oort cloud for the retrograde centaurs but , again , the possible existence of trans - plutonian planets may change the entire dynamical scenario . : semi - major axis , @xmath13 , eccentricity , @xmath7 , inclination , @xmath8 , longitude of the perihelion , @xmath21 , and , argument of perihelion , @xmath22 . the output time - step for this plot is 0.045 myr . these results correspond to the control orbit defined by @xmath13 = 21.1623 au , @xmath7 = 0.47301 , @xmath8 = 100.446 , @xmath21 = 282.604and @xmath22 = 7.224 . ] asteroid 2011 mm@xmath4 is a promising candidate to be a representative of a remnant population of primordial planetesimals and it is a good candidate to be submitted to the nodal libration mechanism ( verrier and evans 2008 , 2009 ; farago and laskar 2010 ; doolin and blundell 2011 ) , at least temporarily . this mechanism , when active , can significantly enhance the stability of near - polar orbits . minor bodies like 2011 mm@xmath4 may signal the size threshold for catastrophic collisions in the early solar system . returning to the issue of its origin , and as alternative to the oort cloud or the hypothetical high - inclination reservoir pointed out above , we speculate that this and other similar objects may be ancient ejected uranian satellites ( regular or irregular ) . the uranian satellite system is the least massive among those of the jovian planets even if the largest one , titania , is the eighth largest moon in the solar system . uranus rotates nearly on its side with the axis lying nearly in its orbital plane . uranus large moons follow regular paths in equatorial - plane orbits . any object originally bound to uranus and tracing an orbit lying nearly in its equatorial plane will follow an almost polar orbit when ejected . in this research , we have used a comparative statistical analysis of the properties of known prograde and retrograde centaurs to reveal compelling reasons to single out the largest retrograde objects . the dynamical evolution of these objects has been subsequently studied using @xmath6-body simulations . the main conclusions of our work can be summarized as follows : * the distribution in absolute magnitude of known retrograde centaurs with semi - major axis in the range 634 au exhibits a clear deficit of objects in the range 10 - 14 mag and a significant excess at @xmath1014 mag . * the largest retrograde centaurs move in almost polar , very eccentric paths with their nodal points currently located near perihelion and aphelion . within the retrograde centaur population , they are clear outliers both in terms of dynamics and size . * the largest retrograde centaurs are currently trapped in transient retrograde , mostly high - order , resonances with the jovian planets . these resonances are strong enough to induce substantial chaotic diffusion and make their orbits dynamically unstable . * asteroid ( 342842 ) 2008 yb@xmath3 appears to be significantly less dynamically stable than usually assumed . * asteroid 2011 mm@xmath4 may be a visitor from the oort cloud but an origin in a relatively stable closer reservoir can not be ruled out . * asteroid 2013 lu@xmath5 is the first known uranian retrograde co - orbital candidate ( a quasi - satellite ) . * this population may have its origin in the oort cloud but alternative or concurrent sources could be ancient ejected uranian satellites or a hypothetical reservoir of minor bodies moving in high - inclination , high - eccentricity orbits at 200500 au and submitted to the kozai resonance . our dynamical analysis has been restricted to just three objects and their associated orbital parameter space . however , a number of distinctive properties that should be common to any retrograde centaurs have been identified . in particular , the role of high - order resonances at high inclination as anticipated by gallardo ( 2006 ) seems to be determinant on the short - term evolution of this population . we would like to thank s. j. aarseth , h. f. levison and m. j. duncan for providing the codes used in this research , and the referee for his / her quick and helpful report . this work was partially supported by the spanish ` comunidad de madrid ' under grant cam s2009/esp-1496 . we thank m. j. fernndez - figueroa , m. rego fernndez and the department of astrophysics of the universidad complutense de madrid ( ucm ) for providing computing facilities . most of the calculations and part of the data analysis were completed on the ` servidor central de clculo ' of the ucm and we thank s. cano alsa for his help during this stage . in preparation of this paper , we made use of the nasa astrophysics data system , the astro - ph e - print server and the mpc data server . aarseth , s.j . : gravitational n - body simulations . cambridge university press , cambridge , p. 27 ( 2003 ) alexandersen , m. , gladman , b. , greenstreet , s. , kavelaars , j.j . , petit , j .- m . , gwyn , s. : science * 341 * , 994 ( 2013 ) bauer , j.m . , et al . : * 773 * , 22 ( 2013 ) belton , m.j.s . : * 231 * , 168 ( 2014 ) bernstein , g.m . , trilling , d.e . , allen , r.l . , brown , m.e . , holman , m. , malhotra , r. : * 128 * , 1364 ( 2004 ) brasser , r. , heggie , d.c . , mikkola , s. : celest . * 88 * , 123 ( 2004 ) brasser , r. , schwamb , m.e . , lykawka , p.s . , gomes , r.s . : * 420 * , 3396 ( 2012 ) bressi , t.h . , et al . : mpec 2013-l69 ( 2013 ) charnoz , s. , morbidelli , a. : * 188 * , 468 ( 2007 ) de la fuente marcos , c. , de la fuente marcos , r. : * 427 * , 728 ( 2012a ) de la fuente marcos , c. , de la fuente marcos , r. : * 427 * , l85 ( 2012b ) de la fuente marcos , c. , de la fuente marcos , r. : * 551 * , a114 ( 2013 ) de la fuente marcos , c. , de la fuente marcos , r. : * 441 * , 2280 ( 2014 ) di sisto , r.p . , brunini , a. : * 190 * , 224 ( 2007 ) di sisto , r.p . , brunini , a. , de ela , g.c . : * 519 * , a112 ( 2010 ) doolin , s. , blundell , k. m. : * 418 * , 2656 ( 2011 ) duncan , m. j. : * 138 * , 109 ( 2008 ) emelyanenko , v.v . , asher , d.j . , bailey , m.e . : earth , moon , and planets * 110 * , 105 ( 2013 ) farago , f. , laskar , j. : * 401 * , 1189 ( 2010 ) fouchard , m. , rickman , h. , froeschl , c. , valsecchi , g. b. : * 231 * , 99 ( 2014 ) fraser , w.c . : * 706 * , 119 ( 2009 ) fraser , w.c . , kavelaars , j.j . : * 137 * , 72 ( 2009 ) fraser , w.c . , brown , m.e . , morbidelli , a. , parker , a. , batygin , k. : * 782 * , 100 ( 2014 ) fuentes , c.i . , holman , m.j . : * 136 * , 83 ( 2008 ) gallardo , t. : * 181 * , 205 ( 2006 ) giorgini , j.d . , et al . : * 28 * , 1158 ( 1996 ) gomes , r.s . : * 215 * , 661 ( 2011 ) horner , j. , lykawka , p.s . : * 402 * , 13 ( 2010 ) horner , j. , evans , n.w . , bailey , m.e . : * 354 * , 798 ( 2004a ) horner , j. , evans , n.w . , bailey , m.e . : * 355 * , 321 ( 2004b ) kozai , y. : * 67 * , 591 ( 1962 ) levison , h. , duncan , m.j . : * 108 * , 18 ( 1994 ) levison , h.f . , duncan , m.j . : * 127 * , 13 ( 1997 ) levison , h.f . , duncan , m.j . , dones , l. , gladman , b.j . : * 184 * , 619 ( 2006 ) makino , j. : * 369 * , 200 ( 1991 ) mcnaught , r.h . , et al . : mpec 2008-y38 ( 2008 ) morais , m.h.m . , namouni , f. : celest . * 117 * , 405 ( 2013a ) morais , m.h.m . , namouni , f. : * 436 * , l30 ( 2013b ) pinilla - alonso , n. , et al . ; * 550 * , a13 ( 2013 ) robutel , p. , laskar , j. : * 152 * , 4 ( 2001 ) schlichting , h.e . , fuentes , c.i . , trilling , d.e . : * 146 * , 36 ( 2013 ) shankman , c. , gladman , b.j . , kaib , n. , kavelaars , j.j . , petit , j.m . : * 764 * , l2 ( 2013 ) sheppard , s.s . : * 139 * , 1394 ( 2010 ) standish , e.m . : jpl planetary and lunar ephemerides , de405/le405 , interoffice memo . 312.f-98 - 048 , jet propulsion laboratory , pasadena , ca , usa ( 1998 ) trujillo , c.a . , sheppard , s.s . : * 507 * , 471 ( 2014 ) verrier , p.e . , evans , n.w . : * 390 * , 1377 ( 2008 ) verrier , p.e . , evans , n.w . : * 394 * , 1721 ( 2009 ) volk , k. , malhotra , r. : * 224 * , 66 ( 2013 ) woodworth , d. , et al . : mpec 2013-n02 ( 2011 )

rare decays stand among the most powerful probes of physics beyond the standard model ( sm ) . indeed , most new physics ( np ) scenarios suffer from severe constraints due to their potentially large contributions to flavor observables . this has motivated an intense experimental search , with special focus on observables which are strongly suppressed in the sm . in 2013 , the lhcb collaboration reported on the measurement of several observables in processes involving @xmath3 transitions . a significant tension with the sm was found in some cases . these include angular observables in @xmath4 @xcite , particularly large in case of the popular @xmath5 @xcite , as well as a decrease , with respect to the sm expectation , in several branching ratios @xcite . these anomalies received immediate attention in the flavor community , and soon several independent global fits @xcite showed that the tension could be alleviated in the presence of new physics contributions . interestingly , in 2014 the lhcb collaboration also found an indication of lepton universality violation in the theoretically rather clean ratio @xcite @xmath6 measured in the low dilepton invariant mass regime , which departs from the sm result @xmath7 by @xmath8 @xcite . again , this led to some excitement in the community , in particular after it was found that this hint is compatible with the previous anomalies in @xmath3 transitions : they can be explained by the same type of np contributions to muonic operators @xcite . so far , the discussion has focused on results obtained by lhcb with an integrated luminosity of @xmath9 fb@xmath10 . the latest chapter of this story is the recent announcement of the lhcb collaboration of new results using the full lhc run i dataset @xcite , with an integrated luminosity of @xmath11 fb@xmath10 , which confirm the robustness of the lhcb data . indeed , the new results are compatible with those found with @xmath9 fb@xmath10 , and several theorists have used the new data to update their analyses . according to @xcite , the hypothesis of non - zero np contributions is preferred over the sm by @xmath12 , @xmath13 if the 2014 measurement of @xmath14 is included , whereas the analysis of @xcite finds a slightly larger statistical significance for np ( slightly larger than @xmath15 ) even in the absence of @xmath14 . in any case , assuming that hadronic effects @xcite are not behind these anomalies ( something impossible in case of @xmath14 ) , there is clear evidence of new physics in @xmath16 meson decays . for a complete review of the subject see @xcite and references therein . several approaches have been considered in order to explain the @xmath17 anomalies . some papers @xcite consider generic @xmath2 bosons with flavor violating couplings . these have been shown to provide a simple way to reconcile theory predictions with experimental data . there are also some works that were built to address the first anomalies but fail to address the lepton universality violating @xmath14 measurement , see for example refs . @xcite in the context of 331 scenarios or ref . @xcite , where 4-quark scalar interactions were considered . finally , a few recent models can also account for the lepton universality violating @xmath14 measurement . one finds three types of working models : models with scalar or vector leptoquarks ( or similarly , r - parity violating supersymmetry , where the squarks can play the role of the leptoquarks ) @xcite , composite higgs models @xcite and models with a @xmath2 . the latter is generally considered the easiest approach , since the anomalies can be solved with new contributions to vectorial operators . however , to the best of our knowledge , the only ( complete ) models that have been put forward in order to account for these anomalies using a @xmath2 are : ( 1 ) the one introduced in ref . @xcite ( and the two - higgs - doublet version in @xcite ) , which makes use of a @xmath18 gauge symmetry , and ( 2 ) the one proposed in the recent paper @xcite , which extends the horizontal gauge symmetry to the quark sector . the existence of dark matter ( dm ) is a well established evidence of new physics , supported by a plethora of astrophysical and cosmological observations . this motivates further extensions of the sm of particle physics that include valid dm candidates . in dm models , the current planck @xmath19 limits for the dm relic density , @xmath20 @xcite , typically imply strong constraints on the mass and couplings of the dm candidate . this , combined with additional constraints ( such as those from flavor physics ) , sometimes leads to very predictive scenarios where the parameter space of the dm model shrinks to small regions compatible with all observations . we propose a simple model which captures the main ingredients required to explain the @xmath3 anomalies and , simultaneously , provides a dm candidate . in order to do so , we extend the sm with a dark sector charged under an additional @xmath1 gauge symmetry . the spontaneous breaking of this symmetry gives rise to a massive @xmath2 boson , which communicates the sm particles with the dark matter particle and solves the @xmath3 anomalies with contributions to the relevant flavor observables . the interplay between dm and flavor physics leads to a very constrained scenario . in particular , achieving the measured dm relic density while providing the required np contributions to @xmath16 meson decays turns out to be very restrictive on the model parameters . the rest of the paper is organized as follows : in sec . [ sec : general ] we discuss some general aspects about the @xmath0 anomalies and introduce the basic language to be used throughout the paper . in sec . [ sec : model ] we define our model , show how it addresses the @xmath3 anomalies while providing a dm candidate and discuss extensions to generate non - zero neutrino masses . [ sec : dm ] is devoted to the dm phenomenology of our model and the @xmath2 portal responsible for its production in the early universe . [ sec : constraints ] is a review of the most relevant constraints in our model , to be considered in sec . [ sec : results ] , where we present our numerical results . finally , in sec . [ sec : conclusions ] we summarize our results , discuss some related aspects and derive some general conclusions . in this section we discuss some general aspects about the @xmath3 anomalies . for a complete review of the subject , we refer the reader to @xcite . the effective hamiltonian for @xmath3 transitions is usually written as @xmath22 where @xmath23 is the fermi constant , @xmath24 the electric charge and @xmath25 the ckm matrix . @xmath26 and @xmath27 are the effective operators that contribute to @xmath3 transitions , and @xmath28 and @xmath29 their wilson coefficients . since the most important anomalies have been found in semileptonic @xmath16 meson decays , we will consider the following set of operators , @xmath30 here @xmath31 and @xmath32 , for muons . ] . it is also customary to split the wilson coefficients in two parts : the sm contributions and the np contributions . since the primed operators , @xmath33 and @xmath34 , do not receive significant sm contributions , this is usually applied only to the unprimed wilson coefficients , which can be written as @xmath35 the sm contributions , @xmath36 and @xmath37 , have been computed by different groups . assuming that these are the only contributions to the wilson coefficients , several independent global fits have found a sizable tension with experimental data on @xmath3 transitions . this motivates the addition of np contributions . when this is done , the global fits are clearly improved . according to @xcite ( we will mainly consider the results of this global fit ) , the best improvements are found in two cases : * * scenario 1 : * np provides a negative contribution to @xmath38 , with @xmath39 , leading to a wilson coefficient @xmath40 significantly smaller than the one in the sm . * * scenario 2 : * np enters in the @xmath41 invariant direction @xmath42 , with @xmath43 . in both cases , the rest of operators involving muons are perfectly compatible with the sm expectations . similarly , no np is required for operators involving electrons or tau leptons . once determined the type of contributions to @xmath44 and @xmath45 a np model has to induce , one can figure out the main ingredients of a simple working model . arguably , the simplest one contains the following elements : * a massive @xmath2 boson , responsible for the vectorial operators @xmath46 and @xmath47 * the @xmath2 must have flavor violating couplings to quarks * the @xmath2 must couple differently to electrons and muons this setup can be easily parameterized by the lagrangian @xcite @xmath48 in order to account for the anomalies , one requires @xmath49 and either ( 1 ) @xmath50 , or ( 2 ) @xmath51 and @xmath52 , depending on the scenario one wants to consider . the rest of the @xmath2 couplings to sm fermions can be set to zero . as we will see in sec . [ sec : model ] , the model we are going to consider belongs to scenario ( 2 ) . in this case , the quark and lepton currents contributing to @xmath46 and @xmath47 are both left - handed and one finds at tree - level @xcite @xmath53 with @xmath54 where @xmath55 is the fine structure constant . note that @xmath56 and the ckm elements appear in eq . in order to normalize the wilson coefficients as defined in eqs . and . + we now introduce a complete renormalizable model with these properties . we extend the sm gauge group with a new dark @xmath57 factor , under which all the sm particles are assumed to be singlets . the only particles charged under the @xmath57 group are the following vector - like fermions , @xmath58 as well as the complex scalar fields @xmath59 where we denote the gauge charges under @xmath60 and the @xmath41 doublets can be decomposed as @xmath61 and @xmath62 . besides canonical kinetic terms , the new vector - like fermions have dirac mass terms , @xmath63 as well as yukawa couplings with the sm fermions @xmath64 where @xmath65 and @xmath66 are @xmath11 component vectors . the scalar potential takes the form @xmath67 here @xmath68 is the sm higgs doublet and latexmath:[$\mathcal v_{\text{sm } } = m_h^2 involving the @xmath57 charged scalars are @xmath70 and @xmath71 we will assume that the scalar potential is such that only the standard higgs boson and the @xmath72 field acquire non - zero vacuum expectation values ( vevs ) , @xmath73 therefore , the @xmath72 field will be responsible for the spontaneous breaking of @xmath57 , giving a mass to the @xmath2 , @xmath74 , where @xmath75 is the @xmath57 gauge coupling , and inducing mixings between the vector - like fermions and their sm counterparts thanks to the yukawa interactions in eq . . furthermore , after spontaneous symmetry breaking , the resulting lagrangian contains a remnant @xmath76 symmetry , under which @xmath77 is odd and all the other fields are even . therefore , @xmath77 is a stable neutral scalar , and thus a potentially valid dm candidate . it is worth noting that the mechanism to stabilize the dm particle does not introduce additional ad - hoc symmetries , but simply makes use of the same @xmath57 symmetry that is required in order to give an explanation to the lhcb observations . this goal has been achieved by breaking the continuous @xmath57 symmetry to a remnant @xmath76 , something that can be easily accomplished with a proper choice of @xmath57 charges @xcite . before concluding this section we must comment on @xmath1 mixing . it is well known that nothing prevents @xmath1 factors from mixing . in the model under consideration , this would be given by the lagrangian term @xcite @xmath78 where @xmath79 are the usual field strength tensors for the @xmath80 groups . in the presence of a non - zero @xmath81 coupling , kinetic mixing between the @xmath57 and @xmath82 gauge bosons is induced . as a consequence of this , the physical @xmath2 boson would couple to all particles that carry hypercharge , this is , to all the sm fermions . this would lead to phenomenological problems since couplings to the first generation are strongly constrained . therefore , we will assume that the tree - level @xmath81 coupling vanishes . this is easily justified in our model because this term is not induced via renormalization group running if it is zero at some high - energy scale ( where one may speculate about a ultraviolet completion ) . the reason is our choice of the @xmath57 charges @xcite . in addition , we must keep the 1-loop induced @xmath81 coupling , generated in loops including heavy vector - like quarks and leptons , under control . we find @xmath83 therefore , @xmath84 would ensure small @xmath81 couplings , while allowing large @xmath75 . something required by dm constraints ( see sec . [ sec : dm ] ) . and @xmath47 in our model . ] this model solves the @xmath3 anomalies in a similar fashion as the one in ref . the @xmath2 couplings to the sm fermions are generated after their mixing with the corresponding vector - like quarks and leptons , as shown in fig . [ fig : couplings ] . neglecting @xmath85 and @xmath86 , the resulting @xmath2 couplings are found to be @xmath87 in case @xmath88 , the @xmath2 coupling to a pair of sm quarks can be further approximated to @xmath89 . these results can be combined with eq . in order to determine the allowed ranges for the model parameters that explain the @xmath3 anomalies found by lhcb . is a new scalar field with @xmath57 charge @xmath90 , necessary in order to make the operator gauge invariant . ] we can further extend the model to get non - zero masses for the sm neutrinos . this can be done trivially by adding new particles , singlets under @xmath57 , which mediate the standard mechanisms . for example , the addition of right - handed neutrino singlets , @xmath91 , allows for the usual type - i seesaw mechanism . it is , however , more interesting to consider a mechanism that involves the @xmath57 sector of the model . this can be done by means of the effective operator @xmath92 as shown in fig . [ fig : numass ] . here @xmath93 is a new scalar field with with @xmath57 charge @xmath94 , necessary in order to make the operator gauge invariant . an example model that can serve as ultraviolet completion of @xmath95 is obtained with the addition of the scalar @xmath96 , together with a vector - like ( dirac ) fermion @xmath97 . this allows for the yukawa couplings @xmath98 and @xmath99 , which lead to @xmath100 after integrating out @xmath101 and @xmath102 . although @xmath93 must get a non - zero vev in order to break lepton number and generate neutrino masses , we note that our choice @xmath94 guarantees that the remnant @xmath76 symmetry that stabilizes the dm particle @xmath77 is preserved . at high temperatures @xmath77 attains thermal equilibrium in the heat bath through its reactions with the different degrees of freedom to which it couples . at decoupling , the dominant processes determining the dm yield are the following @xmath103 reactions : , @xmath104 plays a subdominant role . ] * @xmath105 ( higgs portal ) : processes enabled by the @xmath106 coupling . * @xmath107 , with @xmath101 standing for the sm and the new vector - like quarks and leptons ( @xmath2 portal ) : the former enabled by @xmath108 mixing and @xmath57 coupling , while the latter solely by @xmath75 . for very heavy dm with @xmath109 , also @xmath110 can take place . depending on the relative size of these processes ( pure scalar and @xmath2-mediated ) one can then distinguish several scenarios . among them those solely involving @xmath2-mediated processes , are arguably expected to be dominant ( they are driven by a gauge coupling ) . interestingly enough , in that case a clear correlation with flavor physics must exist . note that these processes match those in fig . [ fig : couplings ] if one trades one of the fermion pairs for @xmath111 . therefore , under the fairly reasonable assumption that the @xmath2-mediated processes play a dominant role , an interplay between flavor and dm physics is possible establishing . thus , in turn , further constraining the flavor - transition parameters through the restrictions imposed by the condition of generating the correct dm relic density ( see the useful ref . @xcite ) . although our results rely on micromegas @xcite , a simple analytical discussion is worth doing . the cross section for the @xmath107 processes can be estimated to be @xmath112 where the @xmath113 and @xmath114 is a kinematic function . depending on the relative size of @xmath115 and @xmath116 ( @xmath117 ) , one can distinguish two regimes . for @xmath118 , dm annihilation processes involve dominantly sm quarks and leptons ( @xmath119 and @xmath120 ) . in that case , however , the corresponding cross sections are suppressed by chiral / vector - like mixing ( see eq . ( [ eq : delta - parameters ] ) ) , annihilation is rather inefficient and therefore leads to an overpopulation of @xmath77 scalars . for @xmath121 , processes involving the vector - like fields are also possible , and since they are not suppressed by mixing parameters they can lead to the appropriate dm relic abundance . note that if the dm and vector - like fermion mass splitting is small , then the cross section will be phase space suppressed : @xmath122 . in this case , resonant annihilation is needed to efficiently deplete the scalar @xmath77 population ( see fig . [ fig : c9 dm ] ) . for large mass splittings , in contrast , annihilation is very efficient , and so the correct relic density can be readily obtained . however , reconciling this `` scenario '' with flavor constraints turns out to be tricky . avoiding the resonance requires @xmath123 , which means @xmath124 , and a large @xmath75 close to 1 . in that limit , @xmath125 which for typical vector - like quark masses , @xmath126 tev , implies large chiral / vector - like quark mixing ( order one @xmath127 couplings ) in order to be compatible with the lhcb observations . this , however , is forbidden by quark flavor constraints ( see next section ) . in summary , the correct dm relic density can be easily produced within our setup provided @xmath128 and dm annihilation proceeds resonantly . resonant annihilation can be avoided for @xmath129 , but finding spots in parameter space consistent with both , dm and flavor constraints seems challenging . the right relic density might as well be produced when @xmath118 , but probably this would require being sharply at the @xmath2 resonance . there are constraints on the mass of additional gauge bosons coupled to sm states . the ballpark of these limits is about 2.53.0 tev for @xmath1 extensions which predict a coupling of the @xmath130 to light quarks of @xmath131 @xcite . however , in our model these couplings are suppressed by the small mixing between the sm quarks and the vector - like states . thus , we nearly get any constraint on the mass of the @xmath130 from lhc searches . in addition , we are going to assume in the lepton sector only a sizable mixing for muons with the new states . therefore , we get also hardly any limit on @xmath132 from lep searches . thus , we can safely assume in the following that the @xmath2 couplings to the first generation of sm fermions can be neglected and evade all lep and lhc limits . similarly , @xmath2 contributions to flavor observables related to the first generation can be safely neglected . let us now review several relevant constraints in our model . the masses of the vector - like quarks have strong bounds from the lhc . being colored particles , vector - like quarks can be efficiently produced in @xmath133 collisions , which typically pushes their masses towards the tev scale , see @xcite and references therein . since our setup works with vector - like quarks with masses at or above the tev scale , these bounds are easily satisfied . vector - like leptons can be searched for at the lhc in the standard multi - lepton channels . using the cms analysis in @xcite , based on searches for final states including @xmath11 or more leptons with an integrated luminosity of about @xmath134 fb@xmath10 at @xmath135 tev , ref . @xcite obtained a lower limit on the mass of doublet vector - like leptons of about @xmath136 gev , in case they decay to electrons or muons , and about @xmath137 gev , in case they decay to tau leptons . more recently , ref . @xcite considered the analogous atlas multi - lepton search @xcite , with @xmath138 fb@xmath10 at @xmath135 tev , also looking for final states with @xmath11 or more leptons . the results were similar to the ones derived from the cms analysis , with lower limits on the mass of doublet vector - like leptons of about @xmath139 gev . finally , one can also derive limits from @xmath140 searches at the lhc . however , according to @xcite , these are rather mild , only relevant for @xmath2 masses below @xmath141 gev . contrary to the model in @xcite , where the muonic current contributing to the @xmath3 observables is purely vectorial ( @xmath142 ) , in our model the current is left - handed and thus contains an axial vector contribution as well ( @xmath143 ) . this is relevant for the @xmath144 decay , especially sensitive to axial - vector currents . currently , there is a little tension between the sm prediction for the average time - integrated branching ratio , @xmath145 , and the cms and lhcb measurements . while the sm prediction is found to be @xmath146 @xcite , the combination of the cms and lhcb measurements leads to @xmath147 @xcite . in view of this little deficit , the new @xmath2 mediated contribution might be potentially welcome . the @xmath144 decay is known for its reach in probing new physics scenarios ( see for example the recent papers @xcite ) . in the sm , the amplitude is dominated by the axial vector leptonic current , and thus by the @xmath148 wilson coefficient ( @xmath149 ) . similarly , in the model under consideration , the main np contribution is given by a modification of @xmath148 . we note that vector contributions vanish for lepton flavor conserving channels . therefore , our prediction for the average time - integrated branching ratio is simply given by @xmath150 nevertheless , the large experimental error in @xmath145 precludes from obtaining useful bounds . using the sm and the experimental average time - integrated branching ratios given above , one finds @xmath151 ( at the @xmath152 level ) . by combining @xmath153 with the sm values for @xmath40 and @xmath154 , one finds that in our setup @xmath155 , which is slightly above the @xmath156 limit , and thus perfectly compatible with the experimental measurement of @xmath145 at the @xmath157 level . another relevant flavor constraint comes from @xmath158 mixing , induced at tree - level by @xmath2 exchange as soon as a non - zero @xmath159 coupling is considered . allowing for a @xmath160 deviation from the sm expectation in the mixing amplitude , @xmath161 , one finds @xcite @xmath162 in principle , the mixings with the vector - like leptons can induce lepton flavor violating ( lfv ) processes of the type @xmath163 , mediated by the @xmath2 boson . however , one can suppress them with a proper parameter choice . since all @xmath3 anomalies can be simultaneously explained with ( only ) new contributions to @xmath32 , @xmath164 is required , but one can choose @xmath165 . this would eliminate the coupling of the @xmath2 boson to electrons and tau leptons , and thus all lfv processes mediated by the @xmath2 boson . this includes lfv in @xmath16 decays , recently suggested in @xcite and further studied in @xcite . besides the anomalies in @xmath3 transitions , the lhc might have found additional hints for new physics . recently , the cms collaboration found a 2.4@xmath166 excess in the @xmath167 channel which translates into br@xmath168% @xcite . for this result , the collaboration made use of the 2012 dataset taken at @xmath169 tev with an integrated luminosity of 19.7 fb@xmath10 . this large higgs lfv branching ratio can not be accommodated in our setup , since vector - like leptons are known to be unable to reach such lfv rates in higgs decays due to limits from the radiative @xmath170 decay @xcite . however , as suggested in @xcite and recently confirmed in several works @xcite , a simple extension with a second higgs doublet suffices to explain the cms hint . finally , we comment on limits from the ( former ) non - observation of lepton universality volation . indeed , very strong bounds have been derived in many scenarios due to the non - observation of lepton universality violating effects in , for example , pion and kaon decays ( see for example @xcite ) . however , these are absent in our model due to the suppression of the vector - like and first generation sm quark mixing . there are several precision measurements in the lepton sector which might be potentially sensitive to the @xmath2 interactions considered in this model . first of all , there is the so - called _ neutrino trident production _ @xcite . this is the production of a muon anti - muon pair by scattering of muon neutrinos in the coulomb field of a target nucleus . the cross - section gets additional contributions in the presence of a new @xmath2 boson that couples to the muons , and these can be used to constrain @xmath171 and @xmath132 . the impact of this bound on @xmath2 models designed to solve the @xmath3 transitions was studied in @xcite . using the ccrf measurement of the trident cross - section @xcite , one can derive bounds on the mass of the @xmath2 boson and its coupling to muons . for scenarios with left - handed @xmath2 coupling to muons , one finds @xcite @xmath172 finally , there are two other constraints related to precision measurements in leptonic observables : @xmath173 and the @xmath174 boson couplings to the muon ( which are modified at the 1-loop level due to the @xmath2 interactions ) . however , in @xcite these are found to be relatively mild . for this reason we will not consider them in our numerical analysis . for the numerical analysis we have implemented the model into the mathematica package sarah@xcite and generated fortran code to link this model to spheno@xcite . the big advantage of this setup is that we can cross - check our tree - level approximations with a full numerical calculation . moreover , based on the flavorkit interface @xcite , the generated sphenoversion does not only calculate the tree - level contributions to the wilson coefficients but also all 1-loop corrections . this way , we can see in what parameter range the tree - level approximation works well and where it breaks down . in general , one can expect loop effects to become important when the mixing between sm quarks and the vector - like states is sizable : this leads to significant changes in the sm - like box and penguin contributions to @xmath44 and @xmath45 . this alone would not be necessarily a problem . however , the same mixing influences also loop contributions to the other wilson coefficients , in particular @xmath175 and @xmath176 . these effects have to be small in order to satisfy the results of global fits to @xmath3 data . + as second step , we used the calchep @xcite output of sarahto implement the model in micromegas to calculate the relic density . interfacing the parameter values between sphenoand micromegas is done via the slha+ functionality of calchep @xcite . and @xmath177 ( dashed black ) in the @xmath178 plane . for @xmath179 we show the full 1-loop results via the black lines , while the dotted grey lines give the values using the tree - level approximation only . ] ( full black lines ) and @xmath180 ( dotted grey lines ) in the @xmath178 plane . ] our main numerical results are summarized in figs . [ fig : c9 dm ] and [ fig : c7c9 ] . in fig . [ fig : c9 dm ] we show the dark matter relic density in the @xmath181 plane , together with the ratio @xmath179 . the other important parameters entering this calculation have been fixed to @xmath182 first of all , we see that with these parameters the model is perfectly compatible with the constraints discussed in section [ sec : constraints ] . furthermore , there is a region for moderately large @xmath183 where the dm constraint can be satisfied and @xmath184 holds . to reduce the relic density to @xmath185 one has to be rather close the the @xmath130 resonance around 2 tev . in the same figure we show also @xmath186 when using the tree - level approximation only . in the interesting region where the flavor anomalies can be explained , this approximation is working quite well . however , for decreasing @xmath75 one would expect smaller values for @xmath186 at tree - level , while we find a large increment . the reason is that for constant @xmath132 and smaller @xmath75 , @xmath187 is increasing , leading to a larger mixing in the quark sector with the vector - like states . thus , loop effects become important and quickly dominate the behavior . these effects are not only expected to show up in @xmath188 but also in the other wilson coefficients as stated above . to demonstrate this , we plot in fig . [ fig : c7c9 ] the contours for @xmath189 and @xmath190 in the same plane . obviously , we find a similar enhancement for small @xmath75 for both coefficients . however , for the most interesting region where tree - level contributions to @xmath188 dominates and can explain the anomalies , the change in @xmath175 is very moderate : @xmath191 . we have presented a dm model that successfully addresses the anomalies in @xmath3 transitions recently found by the lhcb collaboration . in our setup , @xmath16 mesons decay through the @xmath2 into the dark sector , which mixes with the sm via yukawa couplings . an interesting connection between dm and flavor physics emerges , leading to a very constrained scenario where the np masses and parameters are restricted to lie in thin regions of the full parameter space . the model offers several interesting phenomenological possibilities . due to the mixing between the sm muons and the vector - like lepton doublet , our model predicts a reduction of the @xmath192 coupling , @xmath193 , with respect to the sm prediction . the most direct probe of this coupling is , of course , the higgs boson decay into a pair of muons , @xmath194 , experimentally very challenging in a hadron collider due to the expected low branching ratio ( @xmath195 in the sm ) . in 2013 , the atlas collaboration presented results based on an integrated luminosity of @xmath196 fb@xmath10 in collisions at @xmath135 tev , without any evidence of a signal @xcite . for @xmath197 gev , they obtained an upper limit on the @xmath198 signal strength , @xmath199 ( at @xmath200 cl ) . assuming a sm - like higgs boson production , this limit translates into @xmath201 . obviously , our model is well within the limit , since it predicts a reduction of the coupling . nevertheless , it is worth pointing out that a precise determination of this coupling in a future linear collider would be a strong test of the model coupling with an uncertainty of about @xmath202 . ] . let us also comment on the mixings between the vector - like fermions and the sm ones . these are determined by free yukawa parameters which , in principle , are naturally expected to be of the same order for all sm generations . however , in order to avoid conflict with existing experimental data , one is forced to strongly suppress the mixings with the first generation , implying @xmath203 . this might seems like an ad - hoc assumption . however , it just reflects the _ constant _ need for a theory of flavor , also required to understand the yukawa structure of the sm . finally , one can envisage several phenomenological directions in which this model can be further explored . by allowing for non - zero couplings to first generation quarks and leptons , the impact of the model gets extended to many additional observables of interest . in particular , the lhc would be able to probe the parameter space of the model via the production of the @xmath2 boson . furthermore , since the scalar dm carries , at early times , a conserved @xmath57 charge it will develop an asymmetry , provided chemical equilibrium with the thermal bath is guaranteed . in such a case ( asymmetric dm scenario @xcite ) , a link between the baryon and dm asymmetries is something worth exploring . an update of the global fit @xcite was recently presented in @xcite . while this would change slightly the numerical output of our analysis , our conclusions would remain unchanged . we are grateful to javier virto and jorge martin camalich for enlightening discussions . das wants to thank diego restrepo and carlos yaguna for very useful conversations on dark matter . das is supported by a `` charg de recherches '' contract funded by the belgian fnrs agency . av acknowledges partial support from the expl / fis - nuc/0460/2013 project financed by the portuguese fct . r. aaij _ et al . _ [ lhcb collaboration ] , phys . * 111 * ( 2013 ) 19 , 191801 [ arxiv:1308.1707 [ hep - ex ] ] . j. matias , f. mescia , m. ramon and j. virto , jhep * 1204 * ( 2012 ) 104 [ arxiv:1202.4266 [ hep - ph ] ] . s. descotes - genon , j. matias , m. ramon and j. virto , jhep * 1301 * ( 2013 ) 048 [ arxiv:1207.2753 [ hep - ph ] ] . s. descotes - genon , t. hurth , j. matias and j. virto , jhep * 1305 * ( 2013 ) 137 [ arxiv:1303.5794 [ hep - ph ] ] . r. aaij _ et al . _ [ lhcb collaboration ] , jhep * 1406 * ( 2014 ) 133 [ arxiv:1403.8044 [ hep - ex ] ] . r. aaij _ et al . _ [ lhcb collaboration ] , jhep * 1307 * ( 2013 ) 084 [ arxiv:1305.2168 [ hep - ex ] ] . s. descotes - genon , j. matias and j. virto , phys . d * 88 * ( 2013 ) 7 , 074002 [ arxiv:1307.5683 [ hep - ph ] ] . w. altmannshofer and d. m. straub , eur . j. c * 73 * ( 2013 ) 12 , 2646 [ arxiv:1308.1501 [ hep - ph ] ] . f. beaujean , c. bobeth and d. van dyk , eur . j. c * 74 * ( 2014 ) 6 , 2897 [ erratum - ibid . c * 74 * ( 2014 ) 12 , 3179 ] [ arxiv:1310.2478 [ hep - ph ] ] . t. hurth and f. mahmoudi , jhep * 1404 * ( 2014 ) 097 [ arxiv:1312.5267 [ hep - ph ] ] . r. aaij _ et al . _ [ lhcb collaboration ] , phys . * 113 * ( 2014 ) 15 , 151601 [ arxiv:1406.6482 [ hep - ex ] ] . g. hiller and f. kruger , d * 69 * ( 2004 ) 074020 [ hep - ph/0310219 ] . r. alonso , b. grinstein and j. martin camalich , phys . * 113 * ( 2014 ) 24 , 241802 [ arxiv:1407.7044 [ hep - ph ] ] . g. hiller and m. schmaltz , phys . d * 90 * ( 2014 ) 5 , 054014 [ arxiv:1408.1627 [ hep - ph ] ] . d. ghosh , m. nardecchia and s. a. renner , jhep * 1412 * ( 2014 ) 131 [ arxiv:1408.4097 [ hep - ph ] ] . t. hurth , f. mahmoudi and s. neshatpour , jhep * 1412 * ( 2014 ) 053 [ arxiv:1410.4545 [ hep - ph ] ] . w. altmannshofer and d. m. straub , arxiv:1411.3161 [ hep - ph ] . c. langenbruch , talk presented in the ew session of the _ 50th rencontres de moriond _ , march 20th , 2015 . s. jger and j. martin camalich , jhep * 1305 * ( 2013 ) 043 [ arxiv:1212.2263 [ hep - ph ] ] . s. jger and j. martin camalich , arxiv:1412.3183 [ hep - ph ] . r. gauld , f. goertz and u. haisch , phys . rev . d * 89 * ( 2014 ) 1 , 015005 [ arxiv:1308.1959 [ hep - ph ] ] . a. j. buras and j. girrbach , jhep * 1312 * ( 2013 ) 009 [ arxiv:1309.2466 [ hep - ph ] ] . a. j. buras , j. girrbach - noe , c. niehoff and d. m. straub , jhep * 1502 * ( 2015 ) 184 [ arxiv:1409.4557 [ hep - ph ] ] . r. gauld , f. goertz and u. haisch , jhep * 1401 * ( 2014 ) 069 [ arxiv:1310.1082 [ hep - ph ] ] . a. j. buras , f. de fazio and j. girrbach , jhep * 1402 * ( 2014 ) 112 [ arxiv:1311.6729 [ hep - ph ] , arxiv:1311.6729 ] . a. datta , m. duraisamy and d. ghosh , phys . d * 89 * ( 2014 ) 7 , 071501 [ arxiv:1310.1937 [ hep - ph ] ] . s. biswas , d. chowdhury , s. han and s. j. lee , jhep * 1502 * ( 2015 ) 142 [ arxiv:1409.0882 [ hep - ph ] ] . b. gripaios , m. nardecchia and s. a. renner , arxiv:1412.1791 [ hep - ph ] . i. d. m. varzielas and g. hiller , arxiv:1503.01084 [ hep - ph ] . c. niehoff , p. stangl and d. m. straub , arxiv:1503.03865 [ hep - ph ] . w. altmannshofer , s. gori , m. pospelov and i. yavin , phys . rev . d * 89 * ( 2014 ) 9 , 095033 [ arxiv:1403.1269 [ hep - ph ] ] . a. crivellin , g. dambrosio and j. heeck , arxiv:1501.00993 [ hep - ph ] . a. crivellin , g. dambrosio and j. heeck , arxiv:1503.03477 [ hep - ph ] . p. a. r. ade _ et al . _ [ planck collaboration ] , astron . astrophys . * 571 * ( 2014 ) a16 [ arxiv:1303.5076 [ astro-ph.co ] ] . a. j. buras , f. de fazio and j. girrbach , jhep * 1302 * ( 2013 ) 116 [ arxiv:1211.1896 [ hep - ph ] ] . l. m. krauss and f. wilczek , phys . * 62 * ( 1989 ) 1221 . b. petersen , m. ratz and r. schieren , jhep * 0908 * ( 2009 ) 111 [ arxiv:0907.4049 [ hep - ph ] ] . d. a. sierra , m. dhen , c. s. fong and a. vicente , arxiv:1412.5600 [ hep - ph ] . b. holdom , phys . b * 166 * ( 1986 ) 196 . r. m. fonseca , m. malinsk and f. staub , phys . b * 726 * ( 2013 ) 882 [ arxiv:1308.1674 [ hep - ph ] ] . x. chu , y. mambrini , j. quevillon and b. zaldivar , jcap * 1401 * ( 2014 ) 01 , 034 [ arxiv:1306.4677 [ hep - ph ] ] . a. alves , s. profumo and f. s. queiroz , jhep * 1404 * ( 2014 ) 063 [ arxiv:1312.5281 [ hep - ph ] ] . a. alves , a. berlin , s. profumo and f. s. queiroz , arxiv:1501.03490 [ hep - ph ] . g. belanger , f. boudjema , a. pukhov and a. semenov , comput . phys . commun . * 185 * , 960 ( 2014 ) [ arxiv:1305.0237 [ hep - ph ] ] . e. accomando , a. belyaev , j. fiaschi , k. mimasu , s. moretti and c. shepherd - themistocleous , arxiv:1503.02672 [ hep - ph ] . d. barducci , a. belyaev , m. buchkremer , g. cacciapaglia , a. deandrea , s. de curtis , j. marrouche and s. moretti _ et al . _ , jhep * 1412 * ( 2014 ) 080 [ arxiv:1405.0737 [ hep - ph ] ] . collaboration , _ a search for anomalous production of events with three or more leptons using 19.5/fb of sqrt(s)=8 tev lhc data _ , tech . cms - pas - sus-13 - 002 , cern , geneva , 2013 . a. falkowski , d. m. straub and a. vicente , jhep * 1405 * ( 2014 ) 092 [ arxiv:1312.5329 [ hep - ph ] ] . r. dermek , j. p. hall , e. lunghi and s. shin , jhep * 1412 * ( 2014 ) 013 [ arxiv:1408.3123 [ hep - ph ] ] . collaboration , _ search for new phenomena in events with three charged leptons at a center - of - mass energy of 8 tev with the atlas detector _ , atlas - conf-2013 - 070 , cern , geneva , 2013 . c. bobeth , m. gorbahn , t. hermann , m. misiak , e. stamou and m. steinhauser , phys . * 112 * ( 2014 ) 101801 [ arxiv:1311.0903 [ hep - ph ] ] . and * lhcb * collaborations , eps - hep 2013 european physical society conference on high energy physics , stockholm , sweden , 2013 , conference report no . cms - pas - bph-13 - 007 , lhcb - conf-2013- 012 , http://cds.cern.ch/record/1564324 . k. de bruyn , r. fleischer , r. knegjens , p. koppenburg , m. merk , a. pellegrino and n. tuning , phys . * 109 * ( 2012 ) 041801 [ arxiv:1204.1737 [ hep - ph ] ] . a. j. buras , r. fleischer , j. girrbach and r. knegjens , jhep * 1307 * ( 2013 ) 77 [ arxiv:1303.3820 [ hep - ph ] ] . g. yeghiyan , phys . d * 88 * ( 2013 ) 3 , 034020 [ arxiv:1305.0852 [ hep - ph ] ] . d. becirevic , n. kosnik , f. mescia and e. schneider , phys . d * 86 * ( 2012 ) 034034 [ arxiv:1205.5811 [ hep - ph ] ] . h. k. dreiner , k. nickel and f. staub , phys . d * 88 * ( 2013 ) 11 , 115001 [ arxiv:1309.1735 [ hep - ph ] ] . w. altmannshofer , s. gori , m. pospelov and i. yavin , phys . * 113 * ( 2014 ) 091801 [ arxiv:1406.2332 [ hep - ph ] ] . s. r. mishra _ et al . _ [ ccfr collaboration ] , phys . rev . * 66 * ( 1991 ) 3117 . s. l. glashow , d. guadagnoli and k. lane , phys . lett . * 114 * ( 2015 ) 9 , 091801 [ arxiv:1411.0565 [ hep - ph ] ] . b. bhattacharya , a. datta , d. london and s. shivashankara , phys . b * 742 * ( 2015 ) 370 [ arxiv:1412.7164 [ hep - ph ] ] . s. sahoo and r. mohanta , arxiv:1501.05193 [ hep - ph ] . s. m. boucenna , j. w. f. valle and a. vicente , arxiv:1503.07099 [ hep - ph ] . v. khachatryan _ et al . _ [ cms collaboration ] , arxiv:1502.07400 [ hep - ex ] . s. davidson and g. j. grenier , phys . d * 81 * ( 2010 ) 095016 [ arxiv:1001.0434 [ hep - ph ] ] . j. kopp and m. nardecchia , jhep * 1410 * ( 2014 ) 156 [ arxiv:1406.5303 [ hep - ph ] ] . d. aristizabal sierra and a. vicente , phys . d * 90 * ( 2014 ) 11 , 115004 [ arxiv:1409.7690 [ hep - ph ] ] . j. heeck , m. holthausen , w. rodejohann and y. shimizu , arxiv:1412.3671 [ hep - ph ] . i. dorsner , s. fajfer , a. greljo , j. f. kamenik , n. kosnik and i. nisandzic , arxiv:1502.07784 [ hep - ph ] . a. abada , d. das , a. m. teixeira , a. vicente and c. weiland , jhep * 1302 * ( 2013 ) 048 [ arxiv:1211.3052 [ hep - ph ] ] . a. abada , a. m. teixeira , a. vicente and c. weiland , jhep * 1402 * ( 2014 ) 091 [ arxiv:1311.2830 [ hep - ph ] ] . t. asaka , s. eijima and k. takeda , phys . b * 742 * ( 2015 ) 303 [ arxiv:1410.0432 [ hep - ph ] ] . [ atlas collaboration ] , atlas - conf-2013 - 010 , atlas - com - conf-2013 - 003 . f. staub , arxiv:0806.0538 [ hep - ph ] . f. staub , comput . commun . * 181 * ( 2010 ) 1077 [ arxiv:0909.2863 [ hep - ph ] ] . f. staub , comput . commun . * 182 * ( 2011 ) 808 [ arxiv:1002.0840 [ hep - ph ] ] . f. staub , comput . commun . * 185 * ( 2014 ) 1773 [ arxiv:1309.7223 [ hep - ph ] ] . w. porod , comput . commun . * 153 * ( 2003 ) 275 [ hep - ph/0301101 ] . w. porod and f. staub , comput . commun . * 183 * ( 2012 ) 2458 [ arxiv:1104.1573 [ hep - ph ] ] . w. porod , f. staub and a. vicente , eur . j. c * 74 * ( 2014 ) 8 , 2992 [ arxiv:1405.1434 [ hep - ph ] ] . e. e. boos , m. n. dubinin , v. a. ilyin , a. e. pukhov and v. i. savrin , hep - ph/9503280 . a. pukhov , hep - ph/0412191 . g. belanger , n. d. christensen , a. pukhov and a. semenov , comput . commun . * 182 * ( 2011 ) 763 [ arxiv:1008.0181 [ hep - ph ] ] . g. milutinovi - dumbelovi , i. boovi - jelisavi , c. grefe , s. luki , m. pandurovi and p. roloff , arxiv:1412.5791 [ hep - ex ] . s. m. boucenna and s. morisi , front . * 1 * ( 2014 ) 33 [ arxiv:1310.1904 [ hep - ph ] ] . w. altmannshofer and d. m. straub , arxiv:1503.06199 [ hep - ph ] .

observations indicate that the escape of ly@xmath0 photons is facilitated enormously by the presence of outflowing interstellar gas ( e.g. * ? ? ? * ) . scattering off these outflows effectively doppler shifts ly@xmath0 photons out of resonance into the red wing of the absorption line profile . in models of this process the outflow is often represented with a geometrically thin shell of gas ( e.g. * ? ? ? * ; * ? ? ? * ) . in spite of its simplicity , this so - called ` shell - model ' has been very successful at reproducing observed spectra line shapes @xcite . however , the shell - model has recently been shown to have difficulties , especially in reproducing the strength of ` blue bumps ' in a subset of observed spectra @xcite . @xcite studied what may be viewed as an extreme example of blueshifted ly@xmath0 emission in which the ly@xmath0 line of spatially extended ly@xmath0 emission around a radio galaxy as a whole is blueshifted relative to the observed 21-cm absorption , with emission extending beyond 1500 km s@xmath2 into the blue wing of the line profile . blueshifted ly@xmath0 radiation arises naturally when the photons scatter through optically thick inflowing gas @xcite . indeed , @xcite show that their data can be reproduced if more than @xmath3 of cold inflowing gas is present . this large mass of cold neutral gas inside a massive dark matter halo ( @xmath4 ) though , is at odds with expectations from theory , which predicts that the gas should be predominantly accreted in the hot - mode ( e.g. @xcite ; fig . 7 of @xcite ) . an alternative process which gives rise to blueshifted ly@xmath0 emission is described by neufeld & mckee ( 1988 ; henceforth nm88 ) , who show that scattering across a shock front can give rise to such blueshift . nm88 presented analytical calculations , and therefore were forced to adopt simplifying assumptions , namely that the shock - crossing acceleration process is modelled by ly@xmath0photons bouncing between two partially transparent , frequency - preserving mirrors . in reality , however , the frequency of a ly@xmath0photon is not preserved at each ` reflection ' . instead , the frequency of the photon diffuses through frequency - space as the photon is scattered by the material on either side of the shock front , in turn changing the optical depth of the gas to the photon . in addition , they assume that photons cross the shock front isotropically . finally , the analysis presented by nm88 was restricted to a simplified geometry of two adjacent semi - infinite slabs . in spite of an increasing prevalence of ly@xmath0 radiative transfer monte - carlo codes @xcite , they have so far not been used to investigate this fermi - like acceleration mechanism ( while acknowledging the inaccuracy of doing so , we henceforth refer to this as ` fermi acceleration ' for brevity ) . the goals of this paper are two - fold : ( _ i _ ) numerically study fermi acceleration of ly@xmath0 photons across a shock front without the simplifying assumptions that were required in the analytical treatment , and ( _ ii _ ) investigate whether this mechanism can help to reproduce observed blue bumps in spectra that have been difficult to explain with conventional shell - models . the outline of this paper is as follows : [ mechanism ] describes the basic acceleration mechanism , [ simulations ] details the simulations which we use to investigate the physical mechanism , [ simresults ] presents the output of our numerical experiments , [ applications ] discusses the applications of our findings , and in [ conclusions ] we summarise the conclusions . ly@xmath0photons are resonantly scattered by hi , making the ly@xmath0radiative transfer complex . the propagation of ly@xmath0photons is affected by both bulk gas motion and , in detail , microscopic motion of individual hydrogen atoms . the result is that ly@xmath0photons undergo a random - walk like motion in both physical and frequency - space ( see * ? ? ? * for a detailed review ) . as ly@xmath0 photons diffuse away from the resonance frequency , their mean free path - and hence their escape probability - increases . as a result , the spectrum of ly@xmath0 photons emerging from static optically thick media consists of two peaks , which are distributed symmetrically around the resonance frequency ( as frequency diffusion can occur to lower and higher energies with equal probability k ) the energy deposited in the recoiling scattering atom becomes important , and the red peak is enhanced . ] ) . when the scattering medium is contracting [ expanding ] however , the frequency diffusion preferentially occurs towards higher [ lower ] frequencies . as a result , the spectrum of ly@xmath0 photons emerging from contracting optically thick media is blueshifted . an alternative way to see this is that the converging flow of contracting gas is doing work on the ly@xmath0 photons as they scatter outwards , which increases their mean energy @xcite . the fermi acceleration mechanism described in nm88 invokes similar converging flows of gas . the mechanism is illustrated with a simplified geometry in figure [ setupsfig ] , which shows two semi - infinite adjacent slabs of neutral gas . the slab on the left is moving to the right with velocity @xmath5 into a stationary slab . if @xmath6 exceeds the sound speed in the left slab , then the two slabs are separated by a shock front . when a ly@xmath0source is in the vicinity of this shock front we expect some ly@xmath0photons to diffuse through space and cross the front . when a photon traverses the shock front , the doppler boost will impart a blueshift in the local gas frame . partially coherent scattering off atoms with thermal velocity @xmath7 mostly preserves the blueshift of this photon : for a photon with frequency @xmath8 each scattering event pushes back the photon to the line centre by an average amount @xmath9 @xcite . when the photon scatters back across the shock front before this ` restoring force ' has returned the photon to line center , the blueshift of the photon increases with each shell crossing . this is what nm88 referred to as ` fermi acceleration ' . in order for fermi acceleration of a ly@xmath0photon to actually occur we require it to scatter after crossing the shock front . the optical depth to a ly@xmath0 photon that is blueshifted to @xmath10 through a slab of gas with column density @xmath11 equals @xmath12 . the condition @xmath13 thus translates to @xmath14 @xmath15 . inversely , a photon can be fermi accelerated to a maximum blueshift of @xmath16 km s@xmath2 , _ in the frame of the scattering medium . _ after scattering , the ly@xmath0 photon can get an additional doppler shift @xmath17 depending on whether the scattering medium is moving or not . in this section we describe the setup and results of the numerical simulations which we use to study the fermi acceleration mechanism briefly summarised in [ mechanism ] . and [ simulations ] ) . the ly@xmath0source is denoted by the cross in the left slab . the left slab has hi column density @xmath18 , and the right slab has hi column density @xmath19 . the gas in the left slab moves with velocity @xmath20 , photons cross the shock front with velocity @xmath21 , and the angle between @xmath20 and @xmath21 is labelled @xmath22 . @xmath23 is then the projection of @xmath21 onto the unit direction vector of the shock front , @xmath24 . , width=302 ] we use modified versions of the slaf 3d monte - carlo ly@xmath0radiative transfer code ( chung et al . 2015 , in prep ) , as well as the code described in @xcite . the vanilla version of slaf is a grid - based code , able to handle arbitrary gas and source distributions within a finite volume of 3d space . for this work we have modified it , removing the dependence on an underlying grid - structure , so that the code can handle the specific setup described in nm88 . we choose to focus on the geometry discussed in nm88 to facilitate a straightforward comparison . explicitly , there are two distinct regions of gas , represented by slabs of infinite width ( y ) and height ( z ) . the slab on the left side represents outflowing gas , and has a bulk velocity of @xmath25 km s@xmath2 from left to right . the hi column density across the outflowing slab is @xmath26@xmath15 . the slab on the right side represents static gas with an hi column density @xmath27 @xmath15 @xmath28 . a shock front exists at the interface of the two slabs . ly@xmath0photons are emitted at line centre in the rest frame of the outflowing slab on the left , which represents ly@xmath0 emission powered by star formation triggered by the passing shockwave ( see nm88 ) . we ignore dust throughout the analysis in [ simresults ] , which simplifies the interpretation of our results and does not affect our main results . we discuss the impact of this assumption separately when we discuss applications of our results in [ bluewingsrg ] and [ bluebumpspectra ] . figure [ setupsfig ] shows the described experimental setup . figure 2 was produced from a run tracking @xmath29 crossing events , while figures 3 and 4 were produced from runs of @xmath29 photons . in all cases the results were checked for convergence , and found to already be converged with much lower photon counts . table [ crossingproportions ] shows the fraction of photons escaping the simulation volume , @xmath30 , as a function of the total number of shock crossings , @xmath31 . the second and third columns indicate the @xmath32 found in our simulations and the analytic estimate , respectively . the following paragraphs explain how we obtained these analytic estimates . photons emitted in the centre of the outflowing slab are equally likely to leave this slab on the left or right hand side . we therefore expect @xmath33 of all photons not to cross the shock front , which is in agreement with the simulated fraction . photons that enter the static slab for the first time are doppler boosted by an average amount @xmath34 ( see 3.2.2 ) . the probability of these photons being transmitted all the way through the slab is given by @xmath35 , where @xmath36 denotes the optical depth of the entire static slab to photons that enter at frequency @xmath37 ( in the frame of the slab , see @xcite ; note that this transmission probability only applies when the optical depth of the slab to incoming photons is @xmath38 ) . substituting numbers yields @xmath39 , where @xmath40 denotes the hi column density of the slab in units of @xmath41 @xmath15 , and @xmath42 denotes @xmath10 in units of 100 km s@xmath2 ( notation adopted from nm88 ) . we then get @xmath43 , which compares reasonably to the fraction @xmath44 we found in our simulation ( exact agreement is not expected as photons cross the slab over a range of frequencies ) . photons that cross the front twice are back in the low column density outflowing slab , and appear doppler shifted by an average amount @xmath45 in the slab frame ( see 3.2.2 ) . the slab will appear optically thin to these photons , and we estimate the transmission probability from @xmath46 $ ] . here , @xmath47 denotes the optical depth through the outflowing shell for the incoming photons , where @xmath48 is the ly@xmath0 absorption cross - section at @xmath10 . the factor of @xmath49 in the exponent accounts for the fact that the photons enter the slab under an angle @xmath22 with probability @xmath50 ( see 3.2.2 ) . the transmission probability is thus the sum of the probability ( _ i _ ) that photons are transmitted directly through the slab ( the @xmath51 term ) , and ( _ ii _ ) that photons that do scatter eventually escape the slab without crossing the shock again . for the latter , we assume that photons are equally likely to escape on the left and right hand side after scattering , which seems reasonable given that the slab appears optically thin on average . substituting numbers we obtain @xmath52 , which agrees with the simulation result . with @xmath35 and @xmath53 , we estimate @xmath54 , 48.5% larger than what we obtain from the simulation . we attribute this discrepancy to the approximation @xmath35 breaking down at larger shifts from line centre . the previous analysis allows us to understand quantitatively the simulation results . more photons escape after undergoing an even number of shock crossings than odd . this is simply a reflection of the fact that in our experimental setup we have specified that @xmath55 . as we pointed out in [ introduction ] , it is theoretically possible that a ly@xmath0photon that has been fermi accelerated into the line wing can subsequently diffuse back into the core of the line prior to escaping or crossing the shock front again . when this happens , any memory of previous fermi acceleration is erased . however , our numerical simulations indicate that in practice this almost never occurs , and therefore we can identify odd [ even ] numbered contiguous shock crossings with photons which exit on the right [ left ] side of the simulation . .fraction of photons undergoing _ n _ successive shock crossings before exiting the simulation ( middle column ) . the right column refers to the same fraction calculated with an analytic method . [ cols="^,^,^",options="header " , ] we denote the angle at which a photon crosses the shock front with @xmath22 ( see fig [ setupsfig ] ) . for an isotropic distribution of shock crossing directions we have @xmath56 , where @xmath57 . the average doppler shift experienced by a photon as it crosses the shock front is @xmath58 . after @xmath59 crossings the photon experiences an average blueshift of @xmath60 ( nm88 ) . with our code we are able to track individual photons as they cross the shock front , which allows us to directly measure @xmath61 . figure [ crossingdistr ] shows the number count , @xmath62 , for photons that cross the @xmath63 time for various @xmath59 . @xmath61 only differs from @xmath62 by a normalisation factor . here , @xmath59 refers to the current ( rather than final ) contiguous crossing count of a particular photon as it crosses the shock front . a photon which finally exits the simulation after @xmath64 crossings is represented @xmath64 times ( once for each crossing ) . figure [ crossingdistr ] shows clearly that the distribution of photon crossing projections is not isotropic @xmath65 [ @xmath66 for photons that cross the shock front from left - to - right [ right - to - left ] . in figure [ crossingdistr ] only photons on an odd [ even ] shock crossing contribute to @xmath62 for positive [ negative ] values of @xmath67 . ] . instead , we find that @xmath68 . this distribution has been found in previous analyses ( e.g. * ? ? ? * ; * ? ? ? * ) , and its origin is discussed in appendix [ app : pmu ] . for this distribution we expect @xmath69 ( see also fig . [ projvsscattering ] ) , and hence that a photon experiences a doppler boost @xmath70 after @xmath59 shock crossings , then the total hydrogen column density to the edge of the slab is @xmath71 , where @xmath72 [ @xmath73 for the slab on the left [ right ] in figure [ setupsfig ] . the angle - averaged column to the edge of the slab is given by @xmath74 , as used in 3.2.1 . ] . . for each value of @xmath59 the plotted distribution includes all photons that make an @xmath63 crossing regardless of what their final crossing count , @xmath64 , is.,width=302 ] figure [ velshiftvsscatterings ] shows the velocity shift @xmath10 of exiting photons as a function of the total number of shell crossings @xmath64 , in the frame of the gas into which the photon is crossing ( i.e. for odd [ even ] numbered shock crossings the velocity shift is given in the static [ outflowing ] frame ) . results from our monte - carlo simulations are represented by the data points . the red dashed line shows the analytic result under the assumption that photons cross the shock front isotropically ( as in * ? ? ? * ) , while the green solid line shows the analytic result for photons crossing the shock front according to @xmath68 . for @xmath75 the simulation results follow the @xmath76-relation ( where here @xmath64 is a good proxy for @xmath59 ) , after which it approaches the isotropic shock - crossing case . the reason for this transition is that as the photons get blueshifted further into the wing of the line profile their mean free path increases . as we discuss in detail in appendix [ app : pmu ] , we expect a transition to isotropic shock crossing when the mean free path becomes comparable to the thickness of the slab . + shock crossings as found in our simulations ( points ) , as measured in the frame of the gas . the error bars show the standard deviation within the distribution . the dashed red line shows the @xmath77-relation which applies to isotropic shock crossings , whereas the green line refers to the @xmath78-relation which applies to anisotropic shock - crossings of the form @xmath68 ( see text).,width=321 ] finally , figure [ spectrumsinglesource ] shows the spectrum outcoming from the slab , as measured in the lab frame . the result can be easily understood in terms of our previous analysis . more specifically , the total spectrum ( black histogram ) , can be explained as follows . * the peak at @xmath79 km s@xmath2 is composed primarily of the @xmath80 of all photons that did not cross the shock front . these photons diffused outward of the left slab , and would have emerged with a characteristic double peaked emission line profile centered around the resonance frequency ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . however , when we doppler boost these photons back into the lab frame this double peak is diluted by the fact that the photons escaping the slab do so at various angles . the doppler boost is dependent on the exit angle , with an average doppler boost of @xmath81 km s@xmath2 . the grey crosshatched histogram shows the spectrum of these photons . * the small fraction of photons that escape after a single shock crossing , @xmath82 , is indicated by the purple hatch - filled histogram and escapes with a mean blueshift of @xmath83 km s@xmath2 ( also see fig . [ velshiftvsscatterings ] , these photons escape from the static shell , and no additional doppler boost into the lab frame is required ) . * the photons that escape after two shock crossings ( green dotted line histogram ) account for @xmath84 of the total . figure [ velshiftvsscatterings ] shows that these photons have accumulated a blueshift of @xmath85 km s@xmath2 in the frame of the outflowing gas . a doppler boost back into the lab - frame transforms these photons back to a blueshift of @xmath83 km s@xmath2 . a similar reasoning applies to photons that escape after a larger number of crossings , and shows that it is easy to understand the shape of the total spectrum : the redshifted peak ( grey crosshatched histogram ) at @xmath86 is composed of photons that did not cross the shock front . the first peak blueward ( @xmath87 ) of the ly@xmath0resonance is at @xmath88 and consists of photons that crossed the shock front @xmath89 times . similarly , the second blueward peak at @xmath90 consists of photons that crossed the shock 3 - 4 times , etc . the prevalence of these peaks depends on a number of factors , including @xmath20 , @xmath18 , @xmath91 , and also the distribution of ly@xmath0 sources relative to the hi gas . finally we note that our results here are in excellent agreement with the analytic estimate of the maximum blueshift detailed in [ mechanism ] . setting @xmath92@xmath15 in the analytic formula gives @xmath93 km s@xmath2 , which lies in the blue tail of figure [ spectrumsinglesource ] . is the number of shock crossings that the photon undergoes prior to exiting the simulation ( see [ crossingstats]).,width=302 ] we now explore , by means of our monte - carlo codes and assuming some simple models , whether the fermi acceleration mechanism can provide a viable explanation for some observed systems . the analysis by nm88 was motivated by observations of 3c326.1 , a radio source with a ly@xmath0 spectral line profile with emission extending far into the blue wing of the line profile ( @xcite , @xcite ) . nm88 modelled 3c326.1 with a shock front that was propagating into a collection of dense clumps . the shock passage triggers star formation and ly@xmath0 emission inside the post - shock gas . the subsequent radiative processes in each clump proceed just as in the slab models we discussed previously . the details of the emerging spectrum depend on the distribution and covering factor of the clumps , but for simplicity and to facilitate a direct comparison to nm88 we follow their prescription and assume that the emerging spectrum is a superposition of the spectra of individual clumps . nm88 further show that dust does not affect the emerging spectra , provided that the gas - to - dust ratio @xmath94 , where @xmath95 denotes the dust - to - gas ratio relative to the local interstellar value . they argue that this limit is acceptable because the dust grains would be destroyed by the passage of the shock front . we finally assume all clumps - and therefore the radiative transfer processes inside them - to be identical , then we can simply adopt the spectrum from figure [ spectrumsinglesource ] . in reality , from our findings in 3.2.2 , we expect that a distribution of clump properties would give rise to a superposition of spectra with peaks in different locations , and therefore that the emission in the blue wing of the line profile would have less prominent features spectrum depends on the placement of ly@xmath0 sources . in particular , changing only the source position leaves the peak positions unchanged , but modulates their relative amplitudes . we have explicitly verified , however , that our spectrum barely changed if we assumed a uniform distribution of ly@xmath0 sources throughout the outflowing slab . ] . putting these reservations aside , we observe that qualitatively our simulated spectrum shows similarities to that predicted by nm88 . we should note though that in figure [ spectrumsinglesource ] we omit the direct sources at @xmath96 , which nm88 put in by hand as a delta function at @xmath96 . we note that the two largest peaks immediately redward and blueward of line centre can not be attributed to photons exiting without crossing the shock front as in nm88 ( i.e. the standard double - peak profile produced by a static , optically thick medium ) . instead , our analysis shows clearly that these peaks are primarily caused by photons which have undergone 0 and \{1 , 2 } shock crossings respectively . the smaller , secondary blueward peak is primarily composed of photons which have undergone 3 or 4 shock crossings . the primary differences between the spectra predicted by us and by nm88 are in the asymmetry of the two main peaks , and the ratio of the peaks . as we have previously mentioned , the source distribution and physical parameters of the clumps affect the outcoming spectrum , and it is likely that tuning of either or both of these will alleviate this discrepancy . the two blue peaks in our model appear to be shifted redward compared to the model predictions from nm88 . however , a comparison of the observed spectrum in nm88 and our model shows that the positions of our blue peaks are also a good fit to the data . in fact , if the amplitude of the peaks is ignored ( this is dependent on the exact source distribution ) , the position of the peaks in our model is arguably a better fit to the data than the nm88 model . we now consider the observed ly@xmath0 spectra of @xmath97 radio galaxy b2 0902 + 34 @xcite . there is broad agreement in the shape of the observed and predicted spectra , with the broad wing extending well beyond @xmath98 km s@xmath2 , and some spectra displaying prominent peaks blueshifted by @xmath98 km s@xmath2 . the spectra from b2 0902 + 34 vary widely depending on where exactly on the galaxy the fiber is placed . it is however true to say that for the majority of fibers the most prominent peak has been observed to lie either at the systemic velocity , which was determined from the observed 21-cm absorption signature , or slightly blueward of it . this would appear to be in tension with our model here which , following nm88 , predicts the most prominent peak to appear at @xmath96 ( not shown in figure [ spectrumsinglesource ] ) as a result of direct ` blister ' sources on the edge of the clump . recalling that the amplitude ratios of the peaks is determined by the source distribution , we speculate that this tension could be relieved if the most prominent peak in the b2 0902 + 34 observations is instead identified as the first blueshifted peak in our model , and the absence of a pronounced peak at systemic velocity attributed to a lower prevalence of ` blister ' sources in this system . as we mentioned in [ introduction ] , observed ly@xmath0 spectral line profiles can often be reproduced surprisingly well with shell - models . in these models , a central ly@xmath0 source is surrounded by a geometrically thin shell of gas . the shell - models contain two parameters describing the ly@xmath0 source : ( _ i _ ) the assumed fwhm of the ly@xmath0 line prior to scattering , and ( _ ii _ ) the strength of the ly@xmath0 emission line , which is quantified by the equivalent width ( ew ) . the shell itself is described by four additional parameters : ( _ i _ ) the hi column density of the shell @xmath99 , ( _ ii _ ) its outflow velocity @xmath100 , ( _ iii _ ) its dust content @xmath101 , and ( _ iv _ ) its velocity dispersion @xmath102 . some recent analyses have pointed out that shell - models have difficulties explaining blue bumps in observed spectra ( e.g. * ? ? ? * ; * ? ? ? nice examples can be seen e.g. in figure 7 of @xcite . more recently @xcite pointed out that the shell - models , when applied to blue - bump objects , consistently require the intrinsic ly@xmath0 line to have a fwhm which is too large ( an excessively high fwhm was only required for blue - bump objects ) . because fermi acceleration naturally gives rise to blueshifted emission , we investigate a minor modification of the shell - model in which we embed the outflowing shell within a static gas cloud . this modification can be interpreted as the situation in which the outflow is still propagating into the static interstellar medium ( possibly prior to breaking out of it ) . this configuration now includes a shock front as in our previous analysis . we specifically study a shell - model with parameters based loosely on those inferred by @xcite : @xmath103 @xmath15 , @xmath104 km s@xmath2 , and fwhm=200 km s@xmath2 . we further assume @xmath105 , ew@xmath106 ( i.e. pure ly@xmath0emission ) , and @xmath107 km s@xmath2 ( corresponding to gas at @xmath108k ) , but note that these assumptions do not affect our results at all . these results show clearly that dust barely affects the blue bumps . this result is not surprising : our modification only adds a small amount of additional hydrogen , and this additional gas triggers fermi - acceleration into the ( blue ) wings of the ly@xmath0 line profile , where ly@xmath0photons escape more easily . ] . finally , the key new model ingredient is the addition of a static shell of gas adjacent to the outflowing shell . this shell is characterised primarily by its column density @xmath109 . we assume that it has no dust and that it has the same temperature as the outflowing shell . the results of this analysis for @xmath110 ( i.e. the original shell - model ; solid black histogram ) , @xmath111 @xmath15 ( red dashed line ) , and @xmath112 @xmath15 ( blue dotted line ) are shown in figure [ fig : shellresults ] , spectra emerging from a ly@xmath0 source surrounded by a shell of hi gas outflowing at 200 km s@xmath2 ( the shell - model , see text ) embedded within a static neutral shell with a column density @xmath109 . the solid black histogram shows the spectrum for @xmath110 ( i.e. the standard shell - model ) , while the red dashed and blue dotted line shows the spectrum for @xmath111 @xmath15 and @xmath113 @xmath15 , respectively . the plot demonstrates that adding even a small amount of hydrogen ( 1/10@xmath114 of that in the shell ) triggers the on - set of fermi acceleration , which gives rise to a blue bump.,width=302 ] where it is clear that a small additional column of hydrogen ( @xmath115 ) dramatically affects the spectrum blueward of the systemic velocity . this large change of the spectrum can be easily understood , as the outflowing shell with @xmath103 @xmath15 directly transmits a significant fraction of ly@xmath0 photons . however , the surrounding static shell is optically thick to photons emitted near line centre , because these photons still appear close to the centre of the line in the frame of this gas . the static shell therefore effectively reflects back photons into the outflowing shell we show that this mechanism produces blue bumps for column densities as low as @xmath116 @xmath15 . this is because a static shell of gas with @xmath116 @xmath15 remains optically thick to ly@xmath0 photons emitted close to line centre , and can therefore reflect back photons into the expanding shell . ] . the reflected photons appear blueshifted by @xmath117 in the frame of the outflowing shell , where their newly acquired large blueshift makes them escape efficiently . for example , photons that are scattered by @xmath118 after being reflected back into the outflowing shell escape with a blueshift of @xmath119 , which is indeed where the new peaks in the spectrum emerge . this process is depicted schematically in figure [ fig : shell ] . the analytic estimate for the maximum blueshift in [ mechanism ] provides a decent estimate of the maximum blueshift @xmath120 km s@xmath2 ( where the @xmath121 km s@xmath2 was the maximum shift for a column density of @xmath122 @xmath15 in the frame of the scattering medium ) . spectra in the shell - model . the outflowing shell is surrounded by a static ism . , width=321 ] we have presented a detailed analysis of fermi acceleration of ly@xmath0 photons across a shock front , a process that was first studied analytically by nm88 . our monte - carlo radiative transfer simulations of the slab model ( figure [ setupsfig ] ) confirm the basic results in nm88 . in particular , we find that each time a photon crosses a shock front that is propagating at @xmath20 , it experiences a doppler boost @xmath123 in the gas frame . the precise numerical coefficient depends on the opacity of the gas to the ly@xmath0 photons . we also find the blueshift of the ly@xmath0 photons to increase in proportion with the number of shock crossings @xmath59 , i.e. @xmath124 . we discussed how our results can help to explain extended blue wings observed in spectra of radio galaxy 3c326.1 ( which was proposed previously by nm88 ) , but we note that our line as a whole appears redshifted by @xmath125 km s@xmath2 . we reached the same conclusions for the radio galaxy b2 0902 + 34 @xcite . we nevertheless consider fermi acceleration a plausible alternative to the model proposed by @xcite , which involves the collapse of @xmath126 of neutral gas . last , we show that fermi acceleration naturally gives rise to blue bumps in ly@xmath0 spectra , which are difficult to reproduce with conventional shell - models ( at least those with reasonable values for the fwhm of the intrinsic ly@xmath0 line , see @xcite ) . we presented a natural extension to shell - models in which the shell is expanding into static gas that contains a low column density of hi . these models can give rise to blue bumps in ly@xmath0 spectra without abandoning the simplicity of the conventional shell - model transfer through more complex , simulated gas distributions ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , though results from these calculations have not been compared to observed spectra in detail . note however , that the sensitivity of the ly@xmath0 transfer scattering process to the distribution and kinematics of neutral gas implies that these predictions depend on sub - grid feedback prescriptions as well as the spatial resolution of the simulation ( which ideally needs to be sub - pc , see dijkstra & kramer 2012 for a discussion ) . simulations therefore have their own uncertainties , and thus provide a complementary route to addressing ly@xmath0 scattering on interstellar scales . ] . in our model , the bump consists of photons that initially streamed through the outflowing shell , but which were reflected back into this shell by the surrounding static interstellar medium . this suggests that blue bumps in ly@xmath0 spectra are associated with outflows that are still confined to the ism of the galaxy , which may represent an earlier stage in the evolution of the galaxy . we thank koki kakiichi and michele sasdelli for enlightening discussions , and the referee for critical , helpful comments . adams , t. f. 1972 , apj , 174 , 439 j. j. , hill g. j. , macqueen p. j. , 2009 , apj , 694 , 314 ahn , s .- h . , lee , h .- w . , & lee , h. m. 2001 , apj , 554 , 604 barnes , l. a. , haehnelt , m. g. , tescari , e. , & viel , m. 2011 , mnras , 416 , 1723 behrens , c. , & braun , h. 2014 , a&a , 572 , a74 cantalupo , s. , porciani , c. , lilly , s. j. , miniati , f. , 2005 , apj , 628 , 61 - 75 chonis , t. s. , blanc , g. a. , hill , g. j. , et al . 2013 , apj , 775 , 99 dekel , a. , & birnboim , y. 2006 , mnras , 368 , 2 dijkstra , m. , haiman , z. , & spaans , m. 2006a , apj , 649 , 14 dijkstra , m. , haiman , z. , & spaans , m. 2006b , apj , 649 , 37 dijkstra , m. , & kramer , r. 2012 , mnras , 424 , 1672 dijkstra , m. 2014 , pasa , 31 , e040 s. , 1988 , in kron r. g. , renzini a. , eds , towards understanding galaxies at large redshift vol . 141 of astrophysics and space science library , lyman - alpha galaxies at large redshifts . pp 259274 garavito - camargo , j. n. , forero - romero , j. e. , & dijkstra , m. 2014 , apj , 795 , 120 harrington , j. p. 1973 , mnras , 162 , 43 hashimoto , t. , verhamme , a. , ouchi , m. , et al . 2015 , arxiv:1504.03693 kere , d. , katz , n. , weinberg , d. h. , & dav , r. 2005 , mnras , 363 , 2 kulas , k. r. , shapley , a. e. , kollmeier , j. a. , et al . 2012 , apj , 745 , 33 kunth , d. , mas - hesse , j. m. , terlevich , e. , et al . 1998 , a&a , 334 , 11 lake , e. , zheng , z. , cen , r. , et al . 2015 , apj , 806 , 46 laursen , p. , razoumov , a. o. , & sommer - larsen , j.,2009 , apj , 696 , 853 d. a. , mckee c. f. , 1988 , apjl , 331 , l87 neufeld , d. a. 1990 , apj , 350 , 216 osterbrock , d. e. 1962 , apj , 135 , 195 pierleoni , m. , maselli , a. , ciardi , b. , 2009 , mnras , 393 , 872 schaerer , d. , hayes , m. , verhamme , a. , & teyssier , r. 2011 , a&a , 531 , a12 m. a. , mccarthy p. j. , spinrad h. , djorgovski s. , van breugel w. , liebert j. , dickinson m. , 1987 , in bergeron j. , kunth d. , rocca - volmerange b. , tran thanh van j. , eds , high redshift and primeval galaxies 3c 326.1 - a forming galaxy at a redshift of 1.82 ? . pp 8592 tasitsiomi , a. , 2006 , apj , 645 , 792 verhamme , a. , schaerer , d. , & maselli , a. 2006 , a&a , 460 , 397 verhamme , a. , schaerer , d. , atek , h. , & tapken , c. 2008 , a&a , 491 , 89 verhamme , a. , dubois , y. , blaizot , j. , et al . 2012 , a&a , 546 , a111 yajima , h. , li , y. , zhu , q. , abel , t.,2012,mnras , 424 , 884 yajima , h. , li , y. , zhu , q. , abel , t.,2015,apj , 801 , 52 zheng , z. , & miralda - escud , j. 2002 , apj , 578 , 33 the goal of this appendix is to clarify why the distribution of angles at which ly@xmath0 photons cross the shock front scales as @xmath68 ( where @xmath57 ) , rather than isotropically . photons that cross the shock front will on average have traversed an optical depth @xmath127 . depending on @xmath67 , @xmath127 corresponds to a different ( frequency dependent ) physical depth , @xmath128 , away from the shock front . it is easy to see that @xmath129 , so that at fixed frequency the volume of gas that the photon is likely to have last scattered in is @xmath130 . if we further assume that the density of scattering events is homogeneously distributed in the gas , then this leads directly to the relation @xmath131 , where @xmath132 is the intensity of photons . therefore @xmath133 , which we showed in figure [ crossingdistr ] , and which was found previously by @xcite and @xcite . figure [ projvsscattering ] shows the average angle at which photons cross the shock front , @xmath134 , as a function of crossing number @xmath59 . as we mentioned in the paper , odd [ even ] number of shock crossings correspond to crossings from left - to - right [ right - to - left ] . this figure clearly shows that @xmath135 for even shock - crossings , while it approaches @xmath136 for odd - shock crossings when @xmath137 . . the solid lines represent the expected values for isotropic scattering , while the dashed lines show @xmath138 . red triangular points are for odd-@xmath59 shock crossings , blue square points are for even-@xmath59 shock crossings . , width=302 ] an explanation for this is offered in figure [ depthfromshock ] , which shows the mean distance from the shock front ( scattering depth ) of the last scattering before crossing the shock as a function of the shock crossing number @xmath59 . this distance follows an inverse exponential distribution . as the photon becomes more blueshifted with increasing @xmath59 , the mean free path of the photon also increases , so that for each successive @xmath59 the photon is coming from deeper within the slab , explaining the increasing trend in figure [ depthfromshock ] . however , as the mean free path increases , the photons which propagate the farthest actually propagate all the way through the gas slab and exit the simulation . therefore , the distribution of scattering depth is truncated for the next scattering count , resulting in a flattening in the growth of scattering depth as a function of @xmath59 . this is precisely the behaviour we observe for odd-@xmath59 in figure [ depthfromshock ] . the scattering depth distribution is truncated , but the photon blueshift is unaffected . therefore , the gas between the next scattering event and the shock front becomes effectively optically thin to the photon , and we expect photons to cross the shock front isotropically , which is why @xmath139 approaches @xmath140 in figure [ projvsscattering ] . for even-@xmath59 we do not observe this transition to isotropic crossing . this is because @xmath141 , so that at a given @xmath59 and blueshift the mean free path of a photon is much lower in the right ( static ) slab and the scattering depth distribution for even-@xmath59 ( i.e. crossings where the previous scattering was in the right , static slab ) has not yet become truncated . as discussed above it is the truncation of the scattering depth distribution that causes the transition to isotropic shock crossing , and so in its absence for even-@xmath59 the ly@xmath0 photons still cross the shock front following @xmath68 . . d is shown in simulation units , i.e. cells ( each slab is 128 cells wide ) . red triangular points and the red dotted line show the odd - l crossings , while blue square points and the blue dot - dash line show the even - l crossings.,width=302 ] finally , we note that the highest shock crossing counts that occur in our simulation are rare events , and the resultant poor statistics are responsible for the turnaround in figure [ depthfromshock ] ( recall that we are sampling from an inverse exponential distribution ) . in dusty media the increase in path length caused by scattering can allow the dust to significantly attenuate the ly@xmath0 flux that escapes . because photons of different frequencies have different scattering cross - sections the average number of scatterings they undergo will be different . therefore they will have different changes in path - length , and subsequently undergo different levels of dust attenuation . we take a direct approach to address this issue and perform numerical experiments where grey dust is added to either the inner , outer , or both shells . the results are shown in figure [ fig : b1 ] . it is clear that the differences in the emergent spectrum caused by the different dust prescriptions are very small . the addition of dust slightly enhances the blue bump . this occurs because once a photon is fermi accelerated the optical depth of the hi shells to the photon is vastly reduced and the photon easily escapes , avoiding or reducing the number of further scatterings . thus , a fermi acclerated photon has a shorter total path - length so that the presence of dust affects it less than an unaccelerated photon . this results in a slightly enhanced blue peak in the normalised spectra shown in figure [ fig : b1 ] . including three different dust prescriptions . for each histogram @xmath142 refers to the optical depth of the inner shell , @xmath143 refers to the optical depth of the outer shell . the reference dust - free spectrum is shown with a heavy black line.,width=302 ] figure [ fig : b2 ] shows that the blue bump remains visible even if we further reduce @xmath109 by order(s ) of magnitude . , but here we vary @xmath109 while keeping the other parameters fixed . the prominence of the blue peak reduces with hi column density , but a blue bump clearly remains even when @xmath144 @xmath15.,width=302 ]

numerical simulations of the bondi hoyle lyttleton ( hereafter bhl ) accretion flow , in 2d and 3d , have not only enabled the determination of the mean rate of accretion of mass , linear and angular momentum , but have also significantly modified our understanding of its dynamics . they have revealed that the accretion column foreseen by hoyle & lyttleton ( 1939 ) widens to form a detached bow shock if the adiabatic index of the accreted gas is @xmath0 ( hunt 1971 ) or @xmath3 ( hunt 1979 ) . the shock structure observed in simulations was used by eadie ( 1975 ) to interpret the x ray observations of vel x-1 . + real situations like wind accretion in x - ray binaries require taking into account the density and velocity gradients in the incoming flow ( see ishii 1993 and references therein ) , the effects of rotation ( see theuns 1996 and references therein ) as well as the intricacies of radiative transfer ( see an overview in blondin 1994 ) . for the sake of a better understanding , we restrict our present investigation to the axisymmetric meridional accretion of a gas with uniform adiabatic index onto a gravitating accretor with constant velocity . + although this configuration is highly simplified , a completely unexpected non axisymmetric instability was discovered in numerical simulations for supersonic flows ( fryxell & taam 1988 ) . this instability could explain the observed variability of the radio or x - ray luminosity in some astrophysical systems ( ruffert & melia 1994 ) . + however , its is not easy to establish the consistency of the results of the few published 3d numerical simulations which depend on the details of the numerical method : no instability appeared in the 3d sph simulations by boffin ( 1991 ) whereas it was observed in the 3d eulerian simulations by matsuda ( 1992 and references therein ) and ruffert ( 1995 and references therein ) . dome shaped structures frequently appeared with roe s method ( approximate riemann solver ) of matsuda ( 1991 ) , but not in the nested grid piecewise parabolic method ( ppm ) of ruffert ( 1991 ) . since numerical viscosity appears to play an important role , a sufficient resolution of the simulation is crucial . in this respect , numerical simulations encounter the following difficulty : the smallness of the accretor in simulations is limited by the computational power , because the shortest timestep of the simulation is usually determined by the region of high velocities in the vicinity of the accretor . the ratio of scales between the accretion radius and the radius @xmath4 of a compact star , moving with a supersonic velocity @xmath5 is typically of order : @xmath6 where @xmath7 is the schwarzschild radius . note that in the case of accretion onto a neutron star , the relevant accretor boundary is the boundary of its magnetosphere , typically a factor @xmath8 larger than the schwarzschild radius . some of the most recent 3d simulations ( ruffert 1992 , 1994a ) could allow a ratio up to @xmath9 ( the full domain of the simulation extends up to @xmath10 accretion radii ) , by using a ppm code with multiple nested grids . such an accreting sphere is still more than 100 times larger than @xmath7 , and typically @xmath11 times larger than the magnetosphere of a neutron star . one would like to be able to predict , on an analytical basis , which changes are likely to occur if the size of the accretor is further decreased . the role of the boundary conditions at the surface of the accretor is closely related to the position of the sonic surface , and we shall use analytical arguments to obtain some insights on this question . an analytical description of axisymmetric stationary flows is also a first step towards the analysis of their stability , which we postpone to a future paper . bisnovatyi - kogan ( 1979 ) found self - similar stationary solutions , and one would like to know to what extent any generic stationary solution would behave in the same way as their solution . more generality is gained by studying , in the same spirit as theuns & david ( 1992 ) for the spherical case , the first order behaviour of all quantities near the accretor for an axisymmetric flow , and in particular their departure from sphericity . some analytic formulae have been proposed to fit the mass accretion rate observed in simulations ( hunt 1979 ; ruffert 1994b ) , but none of them filled the gap between the well studied cases of axisymmetric accretion of dust ( hoyle & lyttleton 1939 ; bondi & hoyle 1944 ) on the one hand , and spherical accretion of a gas ( bondi 1952 ) on the other , for an arbitrary value of the adiabatic index . continuing in this direction , we add to these reference models the spherical accretion of a shocked gas with uniform adiabatic index , in order to see the effect of both the kinetic and the thermal energies , and extend this approach to axisymmetric meridional stationary flows . + we first recall in sect . [ sgeneral ] the general equations governing a stationary flow and formulate the axisymmetric problem with a single equation in cylindrical coordinates . the shape of the sonic surface is analysed in sect . [ ssonic ] . we establish some properties of the flow by performing local expansions in the vicinity of the accretor in sect . [ sexpansions ] . we present in sect . [ sspherical ] the analytical solution of the spherical accretion from a gas with constant kinetic and thermal energies at infinity . an interpolation formula for the mass accretion rate of axisymmetric meridional flows is proposed in sect . [ sinterp ] . we summarize our conclusions in sect . [ sconclusion ] . the continuity equation for a gas of velocity @xmath12 and density @xmath13 is : @xmath14 we restrict our study to the case of a gas of uniform adiabatic index @xmath15 . the entropy @xmath16 of a gas with pressure @xmath17 is defined as : @xmath18 in which @xmath19 is the gas constant and @xmath20 is the mean molecular weight . the entropy is constant along each flow line ( @xmath21 ) , except when matter passes through a shock . + the temperature @xmath22 and the square of the sound velocity @xmath23 are simply related through a multiplicative constant : @xmath24 for the sake of simplicity , we shall set the ratio @xmath25 everywhere in what follows . since the mean molecular weight is constant throughout our study , this convention is equivalent to changing the units of temperature and entropy , without loosing any generality . + using eq . ( [ defentro ] ) , the pressure force can be written as : @xmath26 the euler equation for a stationary flow with velocity @xmath12 around an accretor of mass @xmath27 is then : @xmath28 in which @xmath29 is the distance to the centre of the accretor and @xmath30 is the gravitational constant . + the bernoulli equation is obtained by integrating the euler equation along a flow line . it states that the quantity @xmath31 @xmath32 is constant along each flow line . the three terms of this sum are respectively the kinetic , the thermal , and the gravitational energies . + the vorticity of the flow is simply defined as @xmath33 by using the definition of the bernoulli constant ( [ bernoulli ] ) in the euler equation ( [ euler ] ) , we obtain the following equation for any stationary flow , @xmath34 this formula is particularly useful because it expresses the vorticity in terms of invariant quantities @xmath31 and @xmath16 . + the bernoulli constant @xmath31 remains constant along a flow line , even through a shock . by contrast , the entropy @xmath16 increases through a shock . + in the classic bhl case of a flow with constant and uniform velocity at infinity , the bernoulli constant is uniformly constant ( @xmath35 ) , and so does not contribute to the vorticity : @xmath36 if , in addition , the flow is isentropic at infinity , the only possible source of vorticity gradients is the shock front . note that eq . ( [ vortentro ] ) implies the conservation of entropy ( @xmath21 ) . thus the only remaining differential equations are the continuity eq . ( [ continuity ] ) and the vorticity eq . ( [ vortentro ] ) . we consider the axisymmetric stationary flow of a gas with uniform density and constant velocity @xmath37 with respect to the accretor . the geometry of the flow lines is described by a single function , the angle @xmath38 between the radial direction and the flow line , defined by : @xmath39 we define by @xmath40 the direction perpendicular to the velocity , in the plane containing the symmetry axis . denoting by @xmath41 the unit vectors in cylindrical coordinates , we can write : @xmath42 when useful throughout the paper , we denote by @xmath43 the dimensionless distance normalized to the modified accretion radius @xmath44 , defined as : @xmath45 it coincides with the classic accretion radius @xmath46 for large mach numbers , when the kinetic energy largely exceeds the thermal energy : @xmath47 we shall denote distances normalized to @xmath44 by a bar . + using eqs . ( [ defentro ] ) and ( [ bernoulli ] ) , all relevant flow quantities can be rewritten as functions of @xmath48 : @xmath49 the azimuthal component of the euler eq . ( [ euler ] ) , multiplied by @xmath29 , is : @xmath50 we denote by @xmath51 the third and only non - zero component of the vorticity vector : @xmath52 by combining the vectorial vorticity eq . ( [ vortentro ] ) and the scalar continuity eq . ( [ continuity ] ) , we obtain a third order differential system in terms of @xmath53 only , with variables @xmath54 ( see appendix [ apartial3 ] ) . this system is hyperbolic in regions where @xmath55 , and elliptic where @xmath56 . it requires _ three _ functions @xmath57 as boundary conditions on the surface of the accretor . it should be noted that if the entropy is uniform , the flow is potential ( @xmath58 ) , and the bernoulli and the continuity equations provide a second order partial differential equation on @xmath59 . in this case , the boundary conditions at the surface of the accretor consist of _ two _ functions of @xmath60 , namely @xmath61 and its radial derivative . we can simplify the system by using the modified stream function introduced by crocco ( 1936 ) , defined as follows : @xmath62 where there is no shock , @xmath63 is conserved along flow lines ( because @xmath64 ) . it is also conserved across shocks , since the derivatives of @xmath63 have only a finite discontinuity there . the entropy can be expressed as a function of @xmath63 , and thus becomes a function of a single variable @xmath65 , defined once and for all along the shock . + let us index the flow lines with their distance @xmath66 to the symmetry axis ( their impact parameter ) at infinity . only the gas within a cylinder of radius @xmath67 is accreted , so that the total mass accretion rate is @xmath68 . at infinity , the mass flux inside the cylinder of radius @xmath66 is directly proportional to @xmath69 through the following product : @xmath70 for the sake of clarity of the equations , let us define : @xmath71 because the total mass accretion rate is finite , we deduce from eq . ( [ massflux ] ) that @xmath63 is bounded in the vicinity of the accretor , and @xmath72 . + using the definition of @xmath63 and the bernoulli equation , we can formally write the mach number @xmath73 as a function of the gradient of @xmath63 through the following algebraic equation : @xmath74 at a given position ( @xmath54 ) , the maximum of the right hand side term is reached at @xmath75 . this defines a maximum value for the gradient of the stream function , for the local mass flux according to eq . ( [ massflux ] ) . for any value of @xmath76 below this maximum , two mach numbers are solutions of the eq . ( [ implicit ] ) . as for spherically symmetric flows classified by bondi ( 1952 ) , one solution is subsonic , while the other is supersonic . + equation ( [ implicit ] ) can be derived with respect to @xmath29 or @xmath60 to express the partial derivatives of @xmath73 with respect to @xmath29 or @xmath60 as function of @xmath73 and the derivatives of @xmath63 . the vorticity eq . ( [ vortentro ] ) can therefore be expressed with derivatives of @xmath63 only , as follows : @xmath77 this equation can be viewed as an explicit partial differential equation of second order on @xmath63 , with @xmath73 implicitly depending on @xmath76 through the algebraic eq . ( [ implicit ] ) . in an equivalent way , ( [ diffpsi ] ) can also be considered as a second order polynom in @xmath78 , whose two roots explicitly determine @xmath78 as a function of @xmath63 and its derivatives . inserting one of the two roots into eq . ( [ implicit ] ) , we obtain an implicit differential equation of second order containing explicitely only @xmath63 and its derivatives of first and second order . + this compact formulation will allow us to perform in sect . [ sexpansions ] a direct analysis of the behaviour of the solution in the vicinity of @xmath79 . 88.9 mm by 70.9 mm ( topology ) numerical simulations of axisymmetric accretion with different mach numbers , adiabatic indices and accretor sizes have exhibited three different topologies of the sonic surface . these three topologies are illustrated in fig . [ topologies ] . - type f ( fast accretion ) : the sonic surface is detached from the accretor . this occurs for subsonic flows for small enough accretors ( ruffert 1994b , 1995 , 1996 ) . - type sf ( mixed slow / fast accretion ) : the sonic surface reaches the surface of the accretor . it subtends an angle @xmath80 relative to the axis of symmetry at the position of the accretor . this is the case for subsonic flows with large accretors , and supersonic flows with @xmath0 ( matsuda 1992 ; ruffert & arnett 1994 ; ruffert 1994a ) . - type fsf ( mixed fast / slow / fast accretion ) : both the sonic surface and the shock surface touch the surface of the accretor . they respectively subtend the angles @xmath80 and @xmath81 relative to the axis of symmetry . this occurs for all supersonic flows when the accretor is large enough , for all the known supersonic 3d simulations with @xmath82 close to one ( ishii 1993 ; ruffert 1996 ) . more complicated topologies were observed by hunt ( 1979 ) and ruffert ( 1995 ) in simulations of supersonic flows with @xmath3 , where the bow shock is linked to the accretor by a conical shock . these can also be classified as type fsf , with the difference that the entropy in the supersonic flow ahead of the attached shock is not uniform . + the topology of the flow in numerical simulations appears to depend strongly on the size of the accretor : all the supersonic flows are of type fsf if the accretor is big enough , but some become type sf or f when the accretor size is decreased . in order to spare computing time , which increases with decreasing accretor size , one would like to use the largest accretor size leading to a physically relevant accretion process ( cf . eq . [ ratio5 ] ) . type f flows are numerically easier to handle , because unphysical effects due to a possible reflection of waves at the boundary of the too large accretor can be avoided . let us consider a flow tube of cross section @xmath83 , parametrized along the tube by the curvilinear variable @xmath84 . we denote by @xmath85 the entropy that this flow tube gains passing through a shock , and by @xmath86 and @xmath87 the density and the sound speed at large distances from the accretor , before the shock . the density @xmath13 and sound speed @xmath88 after the shock satisfy the following equation : @xmath89 using the bernoulli equation ( [ bernoulli ] ) and the entropy equation ( [ entrop ] ) , we can write the momentum as a function of the mach number @xmath73 and the position @xmath29 : @xmath90 86.7 mm by 72.3 mm ( flowtube ) the continuity equation ( [ continuity ] ) states that the mass flux is conserved along the flow tube : @xmath91 by substituting eq . ( [ momentum ] ) into eq . ( [ dmoment ] ) , we obtain , along a given flow line : @xmath92 where the derivatives are calculated along a flow line , @xmath93 . the regularity of @xmath94 at the sonic point @xmath95 implies : @xmath96 we have denoted by @xmath97 the radius of the sonic sphere in the spherical case : @xmath98 it is remarkable that the spherical sonic radius is a function of the total energy @xmath99 only , and is independent of the entropy of the flow . + the regularity condition ( [ sonic ] ) is a constraint on the shape of the flux tube at the sonic radius . note that this condition is valid for any stationary flow , with no hypothesis on its symmetry . let us illustrate it for an inflowing axisymmetric flow tube : the spherical case corresponds to radial flow lines , and thus to a conical flow tube . if the sonic radius is smaller than the spherical one , the cross section of the flow tube increases faster than the square of the distance to the accretor , thus the azimuthal component @xmath100 of the velocity at the sonic radius must be oriented towards the axis of symmetry . conversely , if the sonic radius is larger than the spherical one , the cross section of the flow tube increases more slowly than the square of the distance to the accretor , and @xmath100 must be oriented away from the axis of symmetry ( see fig . [ convexe ] ) . 86.7 mm by 72.3 mm ( property ) we prove in appendix [ aproof ] the following property @xmath101 : + _ if the sonic surface is detached from the accretor , it must intersect at least once the sonic sphere of radius @xmath97 of the corresponding spherically symmetry flow with same energy_. + @xmath97 is defined in sect . [ ssr0 ] by eq . ( [ rspherique ] ) . this property can be understood intuitively from fig . [ propriete ] , by using the geometrical interpretation of eq . ( [ sonic ] ) in sect . [ ssr0 ] , in the simple case where the azimuthal velocity is of constant sign along the sonic surface in the interval @xmath102 . if the azimuthal velocity is directed away from the axis on one side of the accretor ( for example , @xmath103 ) , it will be oriented towards the axis on the other side ( @xmath104 ) , thus implying by continuity an intersection of the sonic surface with the sonic sphere . we show in appendix [ aproof ] that this property is true for any axisymmetric flow , without any hypothesis on the distribution of azimuthal velocities . + former studies of the bhl accretion have dealt with two set of length scales , one associated to the hoyle - lyttleton accretion of dust ( the accretion radius @xmath46 ) , and the other associated to the spherical accretion ( the bondi radius @xmath105 and the corresponding sonic radius @xmath106 ) . none of these scales took into account both the kinetic energy and the thermal energy of the gas . in particular , the significance of the sonic radius @xmath106 for axisymmetric flows has always been dubious . the property @xmath101 establishes the fundamental meaning of the spherical radius @xmath107 , in relation with the sonic surface , for axisymmetric flows of type f. + the geometrical property @xmath101 is useful for numerical simulations of the bhl flow , in order to fix the maximum size of the totally absorbing accretor . as discussed in sect . [ sstypes ] , since the sonic surface must cross the sonic sphere , the accretor size @xmath4 should at least be smaller than the spherical sonic radius @xmath97 . however , although the sonic sphere is larger than the accretor size used in numerical simulations with @xmath3 and @xmath108 ( see ruffert ... ) , the sonic surface and the shock surface appear to be attached to the accretor . although necessary , the condition @xmath109 is far from being sufficient . + since @xmath110 , the sonic surface of flows with @xmath0 must always reach the surface of the accretor . following the nomenclature introduced in sect . [ sstypes ] , _ the topology of the accretion with @xmath0 is either of type sf or fsf , for any accretor size and any mach number at infinity_. the hoyle - lyttleton limit for the mass accretion rate at high mach numbers requires that the accreted matter comes from a cylinder with a radius scaling like the accretion radius ( hoyle & lyttleton 1939 ) . if @xmath82 is close to unity , the sonic spherical radius @xmath97 is @xmath111 larger than the accretion radius @xmath46 : @xmath112 if such a flow were of type f , the geometrical property @xmath101 would imply that the sonic surface goes out as far as @xmath111 times the accretion radius . this leaves only two possibilities for a stationary flow : \(i ) the accretor is able to keep material bound up to distances comparable to @xmath113 , and the mass accretion rate is infinitely larger than the hoyle - lyttleton limit . this is the case of the spherical flows we shall study in sect . [ sspherical ] . \(ii ) the accretor can not retain material at such large distances , and the flow is of type sf or fsf , as observed in numerical simulations with @xmath82 close to one by ishii ( 1993 ) and ruffert ( 1996 ) . + let us note that the artificially constructed type f geometry , where the distance of the stagnation point exceeds @xmath97 , while the accreted matter collects within a cylinder of radius @xmath114 , would be strongly unstable to the kelvin - helmholtz instability : the growth rate would scale like @xmath115 while the time needed to reach the accretor would scale like @xmath116 . + 86.7 mm by 72.3 mm ( 9/7 ) consequently , we rule out the possibility that the sonic surface is detached in the case of an isothermal stationary flow . according to eq . ( [ discrep ] ) , the spherical sonic radius is larger than the accretion radius at high mach numbers if ( see also fig . [ r0ra ] ) spherically symmetric accretion flows with @xmath15 are separated into two distinct classes , depending on whether the flow is subsonic or supersonic at the accretor boundary ( bondi 1952 ) . we learnt from eq . ( [ implicit ] ) in sect . [ sscrocco ] that these two branches of solutions also exist for axisymmetric flows , for a given value of the local mass flux at the surface of the accretor . the azimuthal parameter @xmath60 , however , introduces an additional degree of freedom , such that the solution might be subsonic in an angular sector , and supersonic in an other ( for a sonic surface attached to the accretor ) . we must also distinguish between angular directions towards which the flow lines converge ( direction of accreting " flow lines ) , and directions along which no gas is accreted ( direction of passing " flow lines ) . by conducting a local analysis , we aim to clarify the various possible configurations . for each flow line ( indexed by @xmath66 ) reaching the accretor , we denote by @xmath118 the angle between the velocity vector @xmath12 and the axis of symmetry , when @xmath119 . our notations are illustrated in fig . we define the direction @xmath120 along the axis of symmetry on the downstream side of the accretor , so that the function @xmath118 decreases monotonically in the interval @xmath121 $ ] , with @xmath122 and @xmath123 . we define @xmath124 as the largest azimuthal angle , at @xmath125 , among all the converging flow lines , apart from the axis of symmetry : @xmath1260,\varpi_0[)\}\;.\ ] ] this angle can be smaller than @xmath127 . in particular , @xmath128 for flows of type fsf . without loosing any significant physical generality , we can assume that @xmath1290,\theta_{\rm c}[\subset \{\theta_0(]0,\varpi_0[)\}$ ] in the bhl flow . + 86.7 mm by 35.2 mm ( nota ) near the surface of the accretor , let us denote by @xmath130 the local flux of mass between the azimuthal angles @xmath60 and @xmath131 : @xmath132 since the stream function is bounded in the flow ( eq . [ massflux ] ) , integrating eq . ( [ tanbeta ] ) with respect to @xmath29 implies : @xmath133\;. \label{vrvt}\ ] ] thus , if the mass accretion rate @xmath134 is locally not zero , the flow lines become locally radial when @xmath119 ( no spiraling " of the flow lines ) . on the other hand , if the flow lines are not radial , eq . ( [ mpsi ] ) implies that the mass accretion rate is zero in this direction ( passing flow lines ) . + note that along a direction of accretion , we can consider a flux tube of cross section @xmath135 around the flow line such that @xmath136 , and write eq . ( [ locflux ] ) as follows : @xmath137 the quantity @xmath134 is consequently finite only if the flow tubes are asymptotically conical ( @xmath138 ) . this leads us to call _ regular _ the accretion onto a point like accretor such that @xmath139 remains finite , and _ singular _ when it is locally infinite . + we can also express the local mass accretion rate in terms of the mass flux between the cylinders of radii @xmath66 and @xmath140 at infinity : @xmath141 thus singular accretion corresponds to a localized azimuthal angle @xmath142 where @xmath143 . an example of singular accretion is the self - similar solution found by bisnovatyi - kogan ( 1979 ) , where the whole accretion is along a single azimuthal angle @xmath1440,\varpi_0])$ ] . + for a regular flow ( @xmath145 ) , we take the limit of eq . ( [ consflux ] ) when @xmath119 and obtain , along a given flow line , @xmath146 this equation shows that for @xmath1 , the mach number along regular flow lines either tends to zero or to infinity when approaching the accretor . as in the case of spherical accretion ( bondi 1952 ) no intermediate finite mach number is possible . moreover , this equation proves , by continuity , that a mixed subsonic / supersonic accretion requires a singular accretion at the transition if @xmath1 . the sonic surface may still reach the accretor in a regular flow , by separating a region of supersonic accretion from a subsonic region of passing flow lines , as we shall see in sect . [ ssinterface ] . + we will now examine the different regions of passing flow lines , supersonic and subsonic regular accretion successively , for @xmath1 , and treat the case @xmath0 separately . we shall use the generic name @xmath147 for a function which tends to zero with @xmath29 , and the subscript @xmath148 for the limit of functions when @xmath125 . since the flux of mass is zero , @xmath151 can not vanish in the interval @xmath129\theta_{\rm c},\pi[$ ] . according to our sign convention in sect . 4.3.3 and the definition of @xmath152 in eq . ( [ defbeta ] ) , @xmath153\theta_{\rm c},\pi[\;.\ ] ] since @xmath154 , the function @xmath155 must have a maximum in the interval @xmath129\theta_{\rm c},\pi[$ ] . + \(i ) if the shock is detached , the entropy is finite in the flow . according to eq . ( [ mm ] ) , the mach number must either diverge or tend to zero in order to satisfy @xmath156 . + let us suppose that @xmath157 . we use eq . ( [ vm ] ) and obtain : @xmath158 using the vorticity equation ( [ vortentro ] ) , the entropy gradient can be expressed as : @xmath159 integrating this equation over a segment between @xmath160 and @xmath161 , for @xmath119 , contradicts the fact that the entropy is bounded for a flow with a detached shock . consequently the mach number tends to 0 when @xmath125 , and @xmath162 . _ we conclude that all the flows with a detached shock and @xmath163 are of type sf . _ + \(ii ) if the shock is attached , the incoming supersonic velocity diverges near the accretor like @xmath164 when @xmath119 . we decompose it into a parallel and a perpendicular component with respect to the shock . the mach number @xmath165 associated to the component of the velocity , after and perpendicular to the shock , is subsonic according to the rankine hugoniot jump conditions ( see landau & lifschitz 1987 ) : @xmath166 the mach number @xmath167 associated to the component parallel to the shock can be supersonic , but not arbitrarily large : @xmath168 _ the total mach number immediately after the attached shock is consequently finite when @xmath119_. this result will be important in order to study the local stability of such flows to the rayleigh taylor instability . if @xmath157 and @xmath134 is finite , eqs . ( [ vm])([pm ] ) can be expanded as follows : @xmath169 the first order azimuthal velocity can be estimated using eq . ( [ vrvt ] ) in the azimuthal euler eq . ( [ eulert ] ) . if the pressure is not spherically symmetric to first order , @xmath170 @xmath171 being negative , the azimuthal velocity is oriented towards the region of increasing pressure . the azimuthal pressure force is balanced by this dynamical term . we deduce from eqs . ( [ defbeta ] ) and ( [ vtheta ] ) an upper bound for the departure from spherical symmetry of the function @xmath152 : @xmath172 by studying the leading terms of eq . ( [ diffpsi ] ) when @xmath119 , we derive in appendix [ ainf53 ] a lower bound for the departure from spherical symmetry of the function @xmath152 , if the entropy of the flow is not uniform : @xmath173 on the other hand , the function @xmath152 may approach 0 much more rapidly than @xmath174 if the entropy is uniform . to phrase it more qualitatively , the presence of an entropy gradient sets a lower bound on the bending " of the flow lines near the accretor . if a parallel can be drawn with the simpler case of two - dimensional supersonic flow past a concave profile ( landau & lifshitz 1987 ) , this bending " could be related to the formation of long lived radial shocks in the supersonic region of flows with a non uniform entropy , as observed in numerical simulations with @xmath3 in 2d ( hunt 1979 ) and 3d ( ruffert 1995 ) . + because the radial velocity ( [ vrsup ] ) is spherically symmetric to first order , the azimuthal dependence of the accreted momentum @xmath175 comes mainly from the azimuthal dependence of the density @xmath13 , which we decompose into the contributions of pressure and entropy , @xmath176 thus we expect that for a flow with uniform entropy ( no shock ) , the mass is mainly accreted from the hemisphere where the pressure is the highest . the geometrical interpretation of eq . ( [ sonic ] ) implies that near the axis of symmetry ( for @xmath177 and @xmath178 ) , the azimuthal speed is oriented away from the axis where @xmath179 , and towards the axis where @xmath180 . if one assumes that the sign of the azimuthal velocity does not change in the supersonic region between the sonic surface and the accretor , eq . ( [ prevt ] ) indicates that the highest pressure is on the side of the accretor where @xmath179 . + for an isentropic subsonic flow , the natural orientation of the azimuthal velocity is set by the sign of the angular momentum of the fluid element at infinity . a subsonic medium would flow from the _ left _ of the accretor in fig . [ propriete ] . therefore the sonic surface is displaced opposite to the direction of the flow at infinity , and the maximum flux of mass is expected on the back hemisphere of the accretor . indeed , in the subsonic numerical simulations of a flow with @xmath3 by hunt ( 1979 ) in 2d , and by ruffert ( 1995 ) in 3d , the sign of the accreted linear momentum is negative for the smallest accretor considered . + if the flow is supersonic at infinity , the position of the shock along the axis is displaced in the direction of the flow , compared to its position for a spherical flow with same kinetic and thermal energies at infinity , because there is now a leakage of gas away from the accretor and the pressure ahead of the accretor is lower than in the corresponding spherical case . the sonic point along the axis is displaced accordingly . the sonic surface is consequently shifted in the direction of the flow , leading to a highest pressure on the front hemisphere of the accretor . the simplest configuration is the one in fig . [ propriete ] , where the supersonic flow now comes from the _ right _ of the accretor . + nevertheless , the entropy generated by the shock decreases with increasing distance from the axis of symmetry ( see appendix [ aaxis ] ) , so that entropy and pressure contribute in opposite directions to the density in eq . ( [ densentr ] ) , precluding a direct conclusion about its azimuthal distribution near the accretor . however , the accreted linear momentum is negative in the numerical simulations by ruffert ( 1996 and references therein ) indicating that the entropy effect is stronger than the pressure effect . if @xmath181 and @xmath134 is finite , eqs . ( [ vm])([pm ] ) can be written as follows : @xmath182 we use eqs . ( [ vrsub ] ) and ( [ vrvt ] ) to estimate the dynamical terms balancing the azimuthal pressure force in eq . ( [ eulert ] ) and obtain : @xmath183 comparing this to the first order of @xmath184 given by eq . ( [ csub ] ) , we conclude for @xmath1 that _ the pressure in the regular subsonic region is spherically symmetric to first order when @xmath125_. + thus eq . ( [ psub ] ) implies that the subsonic flow for @xmath1 , with its first order expansions ( [ vrsub])([machsub ] ) , is possible _ only if the entropy is uniform_. + in particular , the spherical solution proposed by theuns & david ( 1992 ) , connecting a supersonic solution to a subsonic solution through a shock , can not resist the slightest asphericity if the accretor is point like . the azimuthal pressure force due to the convergence of flow lines of different entropies is too big to be balanced by dynamical forces in the regular subsonic region close to the accretor . nevertheless , this constraint of the azimuthal euler equation is less stringent for large accretor sizes . one can also argue that the radiative processes near the surface of the star may homogenize the entropy , and allow for a spherically symmetric subsonic accretion . + the entropy in flows of types sf and fsf is a _ strictly _ decreasing quantity along the shock : \(i ) in type fsf flows , the velocity diverges to infinity when @xmath185 , thus the entropy created along the shock decreases strictly with distance from the accretor for type fsf flows , precluding a subsonic accretion with a uniform entropy . \(ii ) for type sf flows , we show in appendix [ aaxis ] that the entropy created along the shock strictly decreases with distance from the axis of symmetry . + we conclude that there can not be a region of regular subsonic accretion in shocked flows with @xmath1 , for a point like accretor . the subsonic region may reach the surface of the accretor for any finite size , but _ the mass accreted in the shocked regular subsonic region @xmath186 tends to zero when the accretor size decreases to zero_. since no regular subsonic accretion is possible with a non uniform entropy , all the regular flow lines are ultimately supersonic when they reach the point like accretor @xmath187 . note that in the singular solutions by bisnovatyi - kogan ( 1979 ) , the mach number tends to infinity along each flow line . + together with the results of sect . [ noaccretion ] , we conclude that for a flow with a detached shock , and @xmath1 , @xmath188 according to eq . ( [ sonic ] ) , the cross section of the flow tube reaching the sound speed at the surface of the accretor varies like : @xmath189 comparing this equation to eq . ( [ regular ] ) , we conclude that the azimuthal angle @xmath80 can not be a direction of regular accretion . however , @xmath80 is not necessarily a direction of singular accretion . a regular flow simply requires that @xmath1900,\varpi_0[)\}$ ] . we denote by @xmath191 the limit of the mach number along a converging flow line . for @xmath1490,\theta_{\rm c}[$ ] , eq . ( [ mm ] ) implies that the local mass accretion rate @xmath134 is always finite , its maximum value being reached for @xmath192 : @xmath193 the pressure forces for @xmath0 are strong enough to impede an arbitrarily high mass flux , even locally . thus accretion flows with @xmath0 are always regular , and the maximum mass accretion rate is the spherical bondi value . this result is particularly important to understand the instability observed in numerical simulations . since the most unstable simulations were observed with @xmath0 , we can restrict our future stability analysis to the simpler case of regular flows . + using eq . ( [ vm ] ) and ( [ vrvt ] ) in the azimuthal euler eq . ( [ eulert ] ) , we find that the azimuthal pressure force is bounded by eq . ( [ maxip ] ) . comparing it to eqs . ( [ rhom ] ) and ( [ pm ] ) , we conclude that the pressure is spherically symmetric to first order for @xmath1490,\theta_{\rm c}[$ ] . according to eq . ( [ densentr ] ) , the density is spherically symmetric to first order if there is no shock ( uniform entropy ) , but increases towards the back hemisphere if there is a shock . + with eqs . ( [ vm ] ) to ( [ pm ] ) , we conclude that : \(i ) in the absence of a shock , the pressure , density , temperature , velocity and mach number are spherically symmetric to first order when @xmath194 . the accreted momentum is therefore spherically symmetric as well . this is confirmed by the numerical simulations by ruffert ( 1994b ) , with @xmath195 and @xmath0 . the accreted linear momentum decreases to zero with the size of the accretor ( see fig . 2 , model sl " , sm " , ss " ) . moreover , the uniform mach number @xmath196 must be lower than or equal to one because of the property @xmath101 . \(ii ) if a shock is present , the pressure is spherically symmetric to first order when @xmath125 , the temperature and entropy increase towards the upstream hemisphere , and the density , the velocity and the mach number increase towards the downstream hemisphere . consequently , more mass is accreted from the downstream hemisphere than from the upstream hemisphere . this is also confirmed by the numerical simulations with @xmath0 by ruffert ( 1994b ) , at @xmath197 . this trend persists in the unstable cases , simulated by ruffert & arnett ( 1994 ) at @xmath198 , and ruffert ( 1994b ) , at @xmath199 . + thus we disagree with hunt ( 1971 ) , who chose the mach number at the surface of the accretor @xmath196 in order to maximize the mass accretion rate . as can be checked from eq . ( [ m53 ] ) , the highest mass accretion rate is obtained for @xmath192 . the mach number at the surface of the accretor is , however , not a free parameter , but is constrained by the distribution of entropy . let @xmath200 be the entropy of the flow line reaching the accretor with @xmath192 . the pressure being uniform , eq . ( [ pm ] ) implies that the final mach number of the flow line indexed by @xmath66 depends on its entropy @xmath201 according to : @xmath202 the erroneous conclusion of hunt ( 1971 ) , that the axisymmetric flow with @xmath0 never becomes supersonic close to the accretor is in contradiction with the 3d numerical simulations ( see fig . 17 in ruffert 1994b ) . it is even in contradiction with the mach number @xmath203 that one can deduce from the velocity , entropy and density displayed in figs . 5 , 6 and 8 in hunt ( 1971 ) for @xmath204 . + in a similar way as eq . ( [ sbmlies ] ) , obtained for @xmath1 , eq . ( [ machaccr ] ) shows that the functions @xmath205 also cease to be independent at the boundary of the accretor for @xmath0 if @xmath206 . for a given distribution of entropy @xmath207 , the only degree of freedom left by eqs . ( [ mpsi ] ) , ( [ m53 ] ) and ( [ machaccr ] ) is the value of @xmath200 , the amount of mass accreted subsonically . the value of @xmath208 which maximizes the total mass accretion rate satisfies an integral equation depending only on @xmath207 , which can be solved numerically . thus if the shape of the shock is known , the distribution @xmath207 is fixed by the rankine - hugoniot jump conditions along the shock ( see appendix [ aaxis ] ) , and the total mass accretion rate @xmath209 is bounded by this maximum value @xmath210 . + since the subsonic accretion of material with a non uniform entropy is possible for @xmath0 , we only have the inequality @xmath211 . nevertheless , the difference of entropies in the region of subsonic accretion is limited by the eq . ( [ machaccr ] ) to : @xmath212 an angular sector of passing flow lines may exist , and is described in sect . [ noaccretion ] . + an estimate of the first order departures from spherical symmetry is obtained by writing the differential eq . ( [ diffpsi ] ) to first order in the vicinity of @xmath79 ( see appendix [ adiff ] ) . this provides us with a lower bound on the departure from sphericity for the function @xmath152 , when the entropy is not uniform : @xmath213 as in sect . [ sssuper ] , this bending " of the flow lines enforced by the entropy gradient could be related to the formation of long lived radial shocks in the supersonic region of flows with a non uniform entropy , as observed in numerical simulations ( ruffert 1994b ) . only two situations are known to have analytical solutions : \(i ) spherically symmetric accretion from a gas at rest at infinity ( bondi 1952 ) . the mass accretion rate of the unique transsonic solution is : @xmath214 \(ii ) axisymmetric accretion from a gas with negligible temperature ( hoyle and lyttleton 1939 ) : @xmath215 other normalizations were also proposed to account for the pressure effects with more realism , a factor @xmath216 lower ( bondi & hoyle 1944 ) . + in order to bridge the analytical gap between the case of spherical accretion described by @xmath217 ( eq . [ mb ] ) and the case of supersonic accretion from a gas of negligible temperature described by @xmath218 ( eq . [ mhl ] ) , bondi ( 1952 ) proposed a simple interpolation formula , meant to give the correct order of magnitude . here we use the normalization adopted later by shima ( 1985 ) and ruffert & arnett ( 1994 ) on the basis of numerical simulations , which is a factor 2 larger : @xmath219 it has the advantage of being very simple and giving a roughly correct limiting behaviour for @xmath220 or @xmath221 . however , because it is independent of @xmath82 , it is bound to fail to reproduce quantitatively the analytical spherical accretion rate @xmath222 , which varies by a factor @xmath223 between @xmath0 and @xmath3 . moreover , numerical simulations have shown typical discrepancies of a factor 23 with this formula for various mach numbers ( see fig . [ g5/3 ] and fig . [ g4/3 ] , and also hunt 1979 ) . an intermediate situation , also analytically tractable , is the case of a spherical accretion with a constant kinetic energy at infinity : the gas is thrown " towards the accretor , from a reservoir at infinity where the flow speed is @xmath224 , the density @xmath86 and the sound speed @xmath87 are uniform and non vanishing , the kinetic energy and the thermal energies are constant at infinity . 86.7 mm by 37.0 mm ( pipes ) to this end , we imagine converging radial pipes " of constant cross section ( fig . [ pipes ] ) . they remain separate up to a minimum radius @xmath225 at which they merge . for a given distance @xmath225 , if a stationary solution exists , the mass accretion rate imposed at infinity would be : @xmath226 thus @xmath227 is also the radius of a cylinder containing the accreted material in an axisymmetric flow where the transition between the parallel flow at infinity and the spherical flow near the accretor ( and also the decrease of angular momentum along each flow line ) would be artificially driven through such pipes . this situation allows us to understand how the kinetic energy influences the mass accretion rate , for a gas with an adiabatic index @xmath82 . + for spherical flows , the mass accretion rate of the stationary solution is determined by the regularity condition at the sonic radius . we obtain exactly the same dimensionless quartic equation as in the classic spherical bondi accretion , determining , in the notations of theuns & david ( 1992 ) , @xmath228 as a function of @xmath229 ( see for example their eq . 2.9 ) , except that the velocities @xmath230 and distance @xmath29 are now normalized differently : @xmath231 for spherically symmetric flows , the regularity condition at the sonic radius fixes the mass accretion rate @xmath232 for the unique transsonic solution : by integrating the mass flux ( [ momentum ] ) over the sonic sphere ( @xmath233 ) , we obtain : @xmath234 where @xmath85 is the entropy gained when passing through a spherical shock , if the flow at infinity is supersonic . + introducing the entropy @xmath235 of the gas at infinity defined by eq . ( [ defentro ] ) , the spherical mass accretion rate can also be written : @xmath236 _ the mass accretion rate is clearly an increasing function of energy _ @xmath99 , _ and a decreasing function of the total entropy _ @xmath237 . + we now consider the two possible cases of flows with and without a shock separately . according to eq . ( [ mdotsph ] ) when @xmath238 , we expect the mass accretion rate of isentropic flows to increase with the mach number : @xmath239 this increase of the mass accretion rate with the mach number for spherical subsonic flows , is not obvious _ a priori_. an increase of the speed at infinity acts as an additional external pressure on the flow : according to eq . ( [ rspherique ] ) , the sonic radius @xmath97 decreases , and eq . ( [ momentum ] ) shows that the momentum increases . however , the decrease of the area of the sonic surface , over which the momentum must be integrated to obtain the mass accretion rate , is more than compensated by the increase of momentum , for @xmath1 . the mass accretion rate is independent of the mach number for @xmath0 . increasing the velocity of the flow at infinity increases its total energy @xmath99 , but passing through a shock also increases its entropy . according to eq . ( [ mdotsph2 ] ) , these two effects act in opposite directions on the mass accretion rate . whether the mass accretion is eventually increased or decreased depends on the position of the shock , which can be determined numerically as follows : for high mach numbers , we choose the distance of the pipes , so that the sound velocity prior to the shock @xmath240 equals the sound velocity at infinity @xmath241 . this corresponds to the position @xmath242 such that the dilution due to gravitational acceleration is balanced by the concentration of flow lines . we make this particular choice by analogy with the case of axisymmetric accretion with a detached bow shock : we show in appendix [ ahyperbolae ] that the density enhancement , along the axis of symmetry , before the shock , is smaller than a factor 2 if @xmath243 . + for each radius @xmath29 , the bernoulli equation ( [ bernoulli ] ) determines the supersonic speed @xmath244 and mach number @xmath245 prior to the shock . using the rankine hugoniot jump conditions , we can compute the entropy jump @xmath246 and the corresponding subsonic velocity @xmath247 after the shock , if the shock occurs at that radius @xmath29 : @xmath248 for a given entropy @xmath85 and bernoulli constant @xmath99 , we denote the velocity profile of the unique transsonic solution , by @xmath249(r)$ ] . the unique shock position @xmath250 corresponds to the solution of the equation @xmath251(r_{\rm sh})\;.\ ] ] 86.7 mm by 72.3 mm ( shock ) the position of the shock determined by this procedure is shown in fig . [ solnum ] , as a function of @xmath82 and @xmath252 . except for values of @xmath82 close to @xmath253 , the spherical shock distance is larger than the accretion radius , and thus larger than was observed in numerical simulations of the axisymmetric bhl flow ( typically @xmath254 ) . + the spherical mass accretion rate is given by the eq . ( [ mdotsub ] ) diminished by a factor @xmath255 defined in eq . ( [ entrorh ] ) . + 86.7 mm by 142.0 mm ( bosse ) increasing the kinetic energy of the supersonic flow always decreases ultimately the mass accretion rate below the value @xmath217 obtained for a gas at rest . between the two opposite tendencies ( entropy and energy ) , the decrease due to entropy will dominate for large mach numbers , since the effective mach number @xmath256 at the shock radius is larger than the mach number at infinity . for @xmath257 , eqs . ( [ mdotsph2 ] ) and ( [ entrorh ] ) imply : @xmath258 moreover , fig . [ solnum ] and fig . [ bosse ] show that the shock radius is proportional to the accretion radius for high mach numbers , in other words the ratio @xmath259 is a non vanishing constant for high mach numbers . apart from a multiplicative factor @xmath260 depending on @xmath82 , the mass accretion rate in this spherical geometry scales like the mass accretion rate @xmath261 derived by hoyle & lyttleton ( 1939 ) for an axisymmetric flow of negligible pressure . it is remarkable that the mass accretion rate is asymptotically independent of the value of the sound speed at infinity . we see from eq . ( [ mdotsph2 ] ) that @xmath217 is the maximum spherical mass accretion rate for @xmath0 . as we shall proceed to show , the situation is very different for @xmath82 close to 1 . the value of @xmath82 defines the strength of a pressure change in response to a density change . since the pressure gradient is the only force limiting the mass accretion rate , the higher the value of @xmath82 , the higher the pressure forces and thus the slower the accretion ( see eq . + an isothermal gas is much more easily swallowed " by the accretor : a significant increase of the mass accretion rate occurs for @xmath262 . in this range , the thermal energy dominates the kinetic energy , and eq . ( [ rspherique ] ) implies that the sonic radius scales like the bondi radius @xmath263 . more accurately , taking the rankine hugoniot jump condition ( [ entrorh ] ) in the limit @xmath264 and combining with eq . ( [ mdotsph2 ] ) gives : @xmath265 we can estimate the contribution of the exponential factor using the bernoulli equation ( [ bernouspher ] ) , for @xmath266 : @xmath267 consequently , the mass accretion rate scales like : @xmath268 using the geometrical constraint that @xmath269 , we can also write the mass accretion rate @xmath270 as the flux of mass through the shock surface , for @xmath266 . for high mach numbers , eq . ( [ bernouspher ] ) ensures that @xmath271 , and the density before the shock @xmath272 because the pipes are of constant cross section . thus , @xmath273 comparing this equation to eq . ( [ entroenerg ] ) , we conclude from @xmath274 that the shock radius is a factor @xmath275 larger than the sonic radius . inserting this result back into eq . ( [ m1minf ] ) , we further conclude that the mass accretion rate is : @xmath276 which is a factor @xmath277 larger than the spherical mass accretion rate @xmath217 from a gas at rest at infinity , and a factor @xmath278 higher than the hoyle lyttleton accretion rate @xmath261 . + we conclude that for values of @xmath82 approaching 1 , the largest spherical mass accretion rate is : @xmath279 and is reached when the kinetic and thermal energies at infinity are comparable , : @xmath280 this scaling obtained in the isothermal limit for high mach numbers is confirmed by the exact calculation of the mass accretion rate , for any value of @xmath82 , through the numerical derivation of the position of the shock , as shown by fig . [ bosse ] . + the entropy energy approach for a spherical flow has shown us that the mass accretion rate ultimately decreases like the hoyle lyttleton rate when the kinetic energy dominates the thermal energy . + nevertheless , the mass accretion rate increases considerably above its value at @xmath281 , when the kinetic energy is comparable to the thermal energy , @xmath282 . this effect is most pronounced for @xmath283 . thus the value of the adiabatic index plays a very important role in the determination of the mass accretion rate in the case of spherical accretion with a shock . similarly , we expect the mass accretion rate of an axisymmetric bhl flow with constant parallel velocities at infinity to depend on the adiabatic index . the axisymmetric mass accretion rate @xmath285 can be formally calculated from the integral of the mass flux along any surface surrounding the accretor . we can decompose the mass accretion into three components , @xmath286 where @xmath287 is the mass flux through the sonic surface , @xmath288 is the mass flux in the subsonic region ( types sf and fsf ) , and @xmath289 is the mass flux in the supersonic region ahead of the shock ( type fsf only ) . + the trajectories of the gas in the supersonic region ahead of the accretor are well approximated by hyperbolae , obtained by neglecting the temperature of the gas . @xmath289 decreases to zero when the accretor size @xmath4 decreases to zero ( eddington 1926 ) : @xmath290 according to sect . [ ssubsonic ] , @xmath288 also decreases to zero with decreasing size of the accretor , for shocked flows with @xmath1 . consequently , the mass flux through the sonic surface gives a good estimate of the total mass accretion rate for small accretors : @xmath291 using eq . ( [ momentum ] ) with @xmath75 , we can write @xmath287 as follows : @xmath292 as in the case of spherical accretion , the mass accretion rate is an increasing function of the energy and a decreasing function of the entropy . however , the geometry of the sonic surface also influences the mass accretion rate . we know from eq . ( [ sonic ] ) that the shape of the sonic surface @xmath293 is directly related to the shape of the flow lines . in contrast to the energy and the entropy , the geometrical factor depends on the shape of the flow lines at infinity , and is more difficult to estimate since it does not correspond to a conserved quantity . one can try to obtain , for the axisymmetric case , an interpolation formula for the mass accretion rate @xmath294 , based on the spherical entropy energy approach . + with some trigonometry we write : @xmath295 the direction @xmath120 being in the direction of the flow , the axisymmetric sonic surface is described by @xmath296 , with @xmath297 . for type f flows , @xmath80 must be replaced by @xmath127 in what follows : @xmath298 first note by differentiating @xmath299 with respect to @xmath43 that : @xmath300 the integral in eq . ( [ mdotfr22 ] ) can be seen as an average of @xmath301 times the entropy weighted by the geometrical function @xmath302 , thus defining an effective entropy corresponding to an effective mach number @xmath303 : @xmath304 using eq . ( [ entrorh ] ) and the definition ( [ mb ] ) of @xmath305 , eq . ( [ mdotfr22 ] ) becomes : @xmath306 one can check that for isentropic flows ( @xmath238 ) and small deviations from sphericity ( @xmath307 ) , the mass accretion rate is accurately described by the spherical model of sect . [ sspherical ] : @xmath308 in order to approximate the mass accretion rate @xmath287 , one must estimate the value of the effective mach number @xmath309 , the effective sonic radius @xmath310 and the integral of @xmath311 . if the shock is detached , we can make simple guesses if the spherical sonic radius @xmath97 is smaller than the accretion radius @xmath312 . we then use the spherical approximation for @xmath83 and @xmath313 , and use an ad - hoc expression for @xmath309 only , to account for the asymmetry . we see from fig . [ r0ra ] that @xmath314 for @xmath315 . the range @xmath316 , and especially the vicinity of @xmath317 , requires a more careful treatment because @xmath97 can be orders of magnitude larger than @xmath46 . consequently , we restrict the present approach to the range @xmath315 , and make the simple approximation @xmath318 . + since we would like to recover the hoyle lyttleton accretion rate @xmath261 for high values of the mach number , we see from eq . ( [ asymptot ] ) that this requires a constant value for the ratio @xmath319 when @xmath320 . as a first approximation , we propose the ad hoc prescription that @xmath321 when @xmath320 , and retain the same ratio @xmath319 for intermediate values of @xmath252 . neglecting the variations of the sound velocity before the shock , we can write the effective mach number in terms of the angle @xmath322 of the effective flow line " of entropy @xmath323 and the normal to the shock surface , and the distance @xmath324 between the accretor and the intersection of this flow line with the shock surface : @xmath325 a geometrical interpretation is the following . let the accretion radius @xmath46 act as a scaling factor on the topology of the accreted flow , and keep the angle @xmath322 and the ratio @xmath326 independent of the mach number . ( [ varphie ] ) then implies that the ratio @xmath327 is constant . this does not contradict the fact that the opening angle of the mach cone , well after the accretor , depends on the mach number , since our approximation deals with the region of the bow shock which is mainly ahead of the accretor . + we obtain , for @xmath328 : @xmath329 with @xmath309 defined by this relation if @xmath328 , and @xmath330 otherwise , our interpolation formula follows from eq . ( [ maxme ] ) : @xmath331 86.7 mm by 72.3 mm ( i5/3 ) 86.7 mm by 72.3 mm ( i4/3 ) we extend our rough approximation to the case @xmath0 , neglecting the mass accreted in the subsonic region . we know from sect . [ s53 ] that this corresponds to a small ( but not zero ) fraction of the total flux of mass . + the curves displayed in fig . [ g5/3 ] and fig . [ g4/3 ] show the mass accretion rates obtained with our interpolation formula @xmath294 , the bondi hoyle interpolation formula @xmath332 and the results of numerical simulations ( ruffert 1994b , 1995 ; ruffert & arnett 1994 ) . for @xmath0 our interpolation eq . ( [ interpol ] ) , in @xmath332 units , rises monotonously with mach number ( fig . [ g5/3 ] ) . the numerical results fit only marginally well . note that the numerical simulations with @xmath0 give only an upper bound for the mass accretion rate , since the accretor is always larger than the spherical sonic radius . for @xmath3 ( fig . [ g4/3 ] ) a prominent maximum is apparent both on the interpolation formula and on the numerical results . comparing fig . [ g4/3 ] with the corresponding curve in the lower part of fig . [ bosse ] gives an indication on the role played by non sphericity in reducing the mass accretion rate . + despite the considerable simplification of ignoring the geometrical term @xmath333 and keeping the ratio @xmath319 independent of @xmath252 , these curves are in reasonable agreement with the numerical simulations for @xmath3 and @xmath0 . thus we expect this analytic formula to be relevant for the whole range @xmath334 . + note that we do not solve the question of the dependence of the mass accretion rate with respect to the adiabatic index for high mach numbers . for this purpose , and following bondi & hoyle ( 1944 ) , we introduced the constant @xmath335 , assumed to be between @xmath336 and @xmath337 . our interpolation formula is nothing more than a natural way to link the bondi mass accretion rate at @xmath281 with the hoyle - lyttleton mass accretion rate at @xmath320 . + one should also keep in mind that the mass accretion rate obtained in the numerical simulations is for most models a time average over the fluctuations or quasi periodic oscillations due to the instability of the flow , and therefore differs from the exact mass accretion rate of the solution of the stationary equations . this instability is stronger for @xmath82 close to @xmath253 . by using a geometrical interpretation of the shape of the flux tube at the sonic radius , we have shown the particular significance of the radius @xmath97 ( eq . [ rspherique ] ) , which depends only on the adiabatic index and the energy of the flow . it is not only the sonic radius of all spherical flows , but is also related through the property @xmath101 ( sect . [ ssprop ] ) to the sonic surface of axisymmetric flows . + by reducing the stationary flow equations to a single partial differential equation , we have performed a local analysis near the accretor and have obtained some general results which aim at clarifying the diversity of possible configurations . we have extended the classification introduced by bondi ( 1952 ) to axisymmetric stationary flows . in addition to the subsonic and supersonic types of accretion , angular sectors without any accretion are also possible , as well as singular directions of infinitely high accretion flux . this is illustrated by the self - similar solution found by bisnovatyi - kogan ( 1979 ) , which is a very particular singular case ( @xmath338 ) . + the angular sectors without any accretion is subsonic for flows with a detached shock , so that @xmath339 in the notation introduced in sect . [ notations ] . + among some of the properties that we established for flows with @xmath1 are the following : \(i ) the sonic surface is likely to be attached to the accretor for nearly isothermal flows with high mach numbers , \(ii ) isentropic accretion proceeds mostly from the downstream hemisphere of the accretor , \(iii ) subsonic accretion must be isentropic . + we have also found a series of properties for flows with @xmath0 : \(i ) accretion is always regular , \(ii ) the sonic surface is always attached to the accretor , \(iii ) the pressure distribution is spherically symmetric to first order , \(iv ) the mach number and the entropy at a point like accretor are algebraically related to the local mass flux , \(v ) shocked matter is accreted mostly from the downstream hemisphere , where the density , velocity and mach number are maximum , and the temperature and entropy are minimum , \(vi ) isentropic accretion is spherically symmetric to first order for the velocity , temperature , density , mass flux , and mach number . + 3d numerical simulations have shown that the strongest instabilities occur with the adiabatic index @xmath0 . the fact that subsonic regions invariably reach the accretor for @xmath0 might be considered as a warning about the possible role , in the instability mechanism , of the boundary conditions at the surface of the accretor . for @xmath1 and sufficiently small accretors , it would be interesting to check for the presence of the instability in cases where the accretor is surrounded by a supersonic region ( type f in the nomenclature of sect . [ sstypes ] ) . + we have also determined the departure from sphericity , and constrained the radial variation of the angle @xmath152 for @xmath119 . in particular , we have stressed the influence of the entropy gradient on the bending of the flow lines for supersonic flows with @xmath1 and @xmath0 . + we have stressed the role of the geometrical shape of the shock in the determination of the entropy of the accreted matter , and contrasted it with the simpler case of shocked spherical flows . in this latter case , we have shown the relevance of the entropy - energy approach , and noted that the mass accretion rate scales like the hoyle lyttleton formula for high mach numbers . moreover , we have outlined the particularity of nearly isothermal flows for which the mass accretion rate can be a factor @xmath340 higher than the bondi mass accretion rate , when the kinetic energy at infinity is comparable to the thermal energy . this demonstrates the fundamental difference of the bhl flows that have @xmath82 close to 1 and those with @xmath82 close to 5/3 . + we have shown that the total mass accretion rate equals the mass accretion rate through the sonic surface , for shocked regular flows with @xmath1 . we have derived an interpolation formula for the mass accretion rate valid in the range @xmath2 . it links spherical accretion to the case of highly supersonic accretion , with reasonable agreement with numerical simulations . + analytical estimates of the following quantities , however , are still missing : \(i ) the distance of the shock as a function of @xmath82 and @xmath252 , \(ii ) the value of @xmath80 and @xmath124 as a function of @xmath82 and @xmath252 , \(iii ) the mass accretion rate in the limit of infinite mach numbers , as a function of @xmath82 . nevertheless , our classification will help perform , in a forthcoming paper , a local stability analysis of all the configurations considered . tf was supported by the ec grant erb - chrx - ct93 - 0329 , as part of the research network ` accretion onto compact objects and protostars ' . we acknowledge useful comments by dr . h. spruit and dr . the combined vorticity equation , continuity equation and entropy equation are written as functions of @xmath341 as follows : @xmath342 using the axisymmetry condition and some simple trigonometry , the variation of the cross section @xmath83 along the flow line can be written : @xmath343 where @xmath38 is defined by eq . ( [ defbeta ] ) , and its derivative is taken in the direction @xmath344 perpendicular to the flow line ( oriented towards increasing @xmath60 ) . for @xmath345 we recover the spherical case of conical flow tubes , with a cross section proportional to the square of the distance . + we can use eq . ( [ trigo ] ) along the detached sonic surface , together with eq . ( [ sonic ] ) , for @xmath346 : @xmath347 let us now prove that the sonic radius reaches the value of the spherical radius at least once , that @xmath348 has to vanish . + axisymmetry imposes @xmath349 , and by eq . ( [ deff ] ) : @xmath350 suppose , for example , that the sonic surface is inside the sphere of radius @xmath97 and detached from the accretor , @xmath351 for @xmath352 $ ] . equations ( [ f0 ] ) and ( [ fpi ] ) imply that @xmath38 is strictly positive near @xmath353 , and strictly negative near @xmath354 . a line must exist in the ( @xmath54 ) plane separating @xmath355 from @xmath356 , along which @xmath152 vanishes , when decreasing from a region of positive values to the region of negative values . we denote by @xmath357 the intersection of this line with the sonic surface : @xmath358 which contradicts the hypothesis . the same contradiction arises if we suppose that @xmath359 . consequently @xmath360 must vanish for some value of @xmath60 . by using eq . ( [ tanbeta ] ) , we can relate the second radial derivative of @xmath63 to its first radial derivative , at @xmath79 , through the following equation : @xmath362 the local mass flux is finite for regular flows , and therefore the limit when @xmath119 of ( @xmath363 ) is zero in the angular sectors where the flow is regular . + let us define the new radial variable @xmath364 . equation ( [ implicit ] ) implies to first order : @xmath365 \(i ) if eq . ( [ betamax ] ) is a strict inequality , eq . ( [ eulert ] ) implies that the pressure is spherically symmetric to first order . using eqs . ( [ mpsi ] ) and ( [ psup ] ) , the stream function @xmath366 at the surface of the accretor is directly related to the entropy @xmath367 by : @xmath368 the leading terms of the differential eq . ( [ diffpsi ] ) satisfy the following equation : @xmath369 where the six functions @xmath370 converge to zero when @xmath371 . the functions @xmath372 depend on @xmath60 only , and are defined as follows : @xmath373 assuming @xmath374 and @xmath375 contradicts eq . ( [ linearinf53 ] ) . thus we conclude that : @xmath376 by derivating eq . ( [ diffpsi ] ) with respect to @xmath29 and considering the leading terms , we find that @xmath377 satisfies the same eq . ( [ linearinf53 ] ) as @xmath378 does . using eq . ( [ tanbeta ] ) , we obtain eq . ( [ betamin ] ) for regular sectors with a non uniform entropy . \(ii ) if eq . ( [ betamax ] ) is an equality , we use eq . ( [ secondprem ] ) in order to write the first order terms of eq . ( [ diffpsi ] ) as follows : @xmath379 in both cases ( i ) and ( ii ) , @xmath380 and the radial derivative @xmath381 cease to be a set of three free functions of @xmath60 on the surface of the accretor , when the accretor size decreases to zero . let us recall that for spherical flows , one parameter ( the mass flux ) is required as boundary condition on the surface @xmath382 of the accretor , with no particular change if @xmath383 . in this respect , the differential system ( [ diffpsi ] ) is more singular at @xmath79 for axisymmetric flows than for spherical flows . because the stream function @xmath63 is finite everywhere , @xmath384 and @xmath385 tend to zero with decreasing @xmath43 . ( [ implicit ] ) implies that when @xmath0 , the mach number tends to a finite limit everywhere where the local mass flux ( @xmath386 according to eq . ( [ mpsi ] ) ) is not zero . taking the limit of eq . ( [ diffpsi ] ) when @xmath387 , we find that its last term must vanish at @xmath388 : @xmath389 where the subscript @xmath148 denotes the value of the function at @xmath79 . this equation is equivalent to the condition that the pressure is spherically symmetric to first order . + in order to find the behaviour of the quantities @xmath390 near @xmath388 , we consider only the leading terms of eqs . ( [ implicit ] ) and ( [ diffpsi ] ) . we first define @xmath256 and @xmath378 as follows : @xmath391 by linearizing eq . ( [ implicit ] ) near @xmath392 , we obtain to first order : @xmath393 introducing this into eq . ( [ diffpsi ] ) , we obtain the following leading terms : @xmath394 where the six functions @xmath395 converge to zero when @xmath387 . the functions @xmath396 depend on @xmath60 only , and are defined as follows : @xmath397 let us suppose that @xmath84 is a positive number such that @xmath398 if the entropy is not uniform ( @xmath399 ) , the first two terms of eq . ( [ linear53 ] ) are negligible compared to the rhs term . dividing the remaining equation by @xmath43 , we obtain : @xmath400 according to eq . ( [ hypolim ] ) , @xmath401 . thus eq . ( [ psi1r ] ) contradicts our hypothesis ( [ hypolim ] ) , and we conclude that if @xmath399 , @xmath402 by derivating eq . ( [ diffpsi ] ) with respect to @xmath43 , and keeping the leading terms , @xmath403 satisfies exactly the same equation ( [ linear53 ] ) as @xmath378 does . from the same reasoning , we conclude that @xmath404 this lower bound on the scaling of @xmath403 is also a lower bound on the scaling of @xmath152 : we obtain eq . ( [ dlogbeta ] ) by using eqs . ( [ tanbeta ] ) , ( [ conclulim2 ] ) and the regularity of the flow for @xmath0 . we check the validity of neglecting the pressure forces in the supersonic flow ahead of the shock , and assuming that the sound velocity is uniformly constant to its value at infinity @xmath87 . the velocity of the supersonic incoming gas is then defined by the unique hyperbolic trajectory passing trough this point , with velocity at infinity @xmath224 . + for each point ( @xmath54 ) of the shock surface , we can compute numerically the dimensionless impact parameter @xmath405 ( normalized to the accretion radius @xmath46 ) of this hyperbola : @xmath406 the half angle @xmath407 of the hyperbola is defined by @xmath408 let us calculate the variation of density along the axis of symmetry , due to both the compression by gravitational convergence of the hyperbolic trajectories and the dilation due to acceleration . this can be done using the continuity equation ( [ continuity ] ) in cylindrical coordinates , along the axis , for @xmath409 : @xmath410 the azimuthal derivative of the azimuthal component of the velocity is directly related to the impact parameter @xmath411 through the conservation of angular momentum , while the ( negative ) radial component of the velocity , along the axis , is given by the conservation of energy : @xmath412 in order to determine the azimuthal dependence of the impact parameter , let us consider the trajectories defined by eq . ( [ hyperbole ] ) for @xmath413 . since @xmath414 , we write : @xmath415 from first order expansions of eq . ( [ demiangle ] ) and eq . ( [ hyperbole ] ) , we find , for @xmath413 : @xmath416 choosing the negative solution of this equation ( @xmath405 is positive and decreases to zero when @xmath60 increases to @xmath127 ) , leads to : @xmath417 using eqs . ( [ dvtheta ] ) and ( [ enervr ] ) in eq . ( [ contax ] ) we obtain : @xmath418 this can be integrated to obtain the radial dependence of the density along the axis of symmetry : @xmath419 the compression effect due to the convergence of the flow lines is consequently stronger than the dilatation due to the acceleration of the flow . + if the shock is not too close to the accretor , gravitational focusing is negligible and the density and temperature can be approximated as constant in the upstream supersonic flow . one can check from eq . ( [ gravcomp ] ) that this density enhancement is smaller than a factor of two for @xmath420 . note that when the shock is detached , its distance is generally of order @xmath421 in numerical simulations . we deduce from eq . ( [ gravcomp ] ) that the sound velocity scales like @xmath422 when @xmath423 , and is therefore always negligible compared to the gravitational potential scaling like @xmath424 . let @xmath425 be local coordinates in a frame chosen in a way such that @xmath426 is perpendicular to the shock surface , and @xmath427 is parallel to it . let us call @xmath256 the mach number relative to the component of the velocity , ahead of and perpendicular to the shock surface . in our local system of coordinates , @xmath428 we shall use the classical convention of using the index 1 " for quantities ahead of the shock , and the index 2 " for quantities just after the shock . + we write the entropy gradient after the shock as : @xmath429 we can use the classical jump conditions ( [ entrorh ] ) to express the entropy gradient as a function of @xmath256 only : @xmath430 we obtain @xmath431 by the vector product of the velocity and the vector tangent to the shock surface . let the supersonic flow before the shock be described in cylindrical coordinates by the flow velocity @xmath432 . for the sake of clarity , we denote the velocity along the shock surface @xmath433 by @xmath434 and a dot stands for the derivative with respect to the angle @xmath60 : @xmath435 we now prove that a regular axisymmetric shock surface can not produce a local region of uniform entropy behind it , and more precisely , that the gradient of entropy created by the shock scales like the square of the azimuthal angle @xmath436 in the vicinity of the axis , for any shape of the shock which is regular and symmetric . + when not specified , the quantities are taken at @xmath160 in what follows . we also define @xmath437 and @xmath438 . the symmetry condition imposes : @xmath439 using the symmetry of the problem , we make the following expansions for small values of @xmath436 : @xmath440 using the expansions ( [ expvr ] ) , ( [ expvt ] ) , and ( [ exprs ] ) , we obtain : @xmath441 in order to evaluate this expression , we use the approximation of the supersonic flow @xmath432 by hyperbolic trajectories . derivating the bernoulli eq . ( [ bernoulli ] ) with respect to @xmath442 and @xmath60 , we obtain for @xmath160 : @xmath443 replacing these expressions , together with eqs . ( [ enervr ] ) and ( [ dvtheta ] ) into eq ( [ vnbrut ] ) , the effective mach number is : @xmath444 where the positive contribution @xmath445 depends on the second derivative @xmath446 of the shock surface on the symmetry axis , and is defined as : @xmath447 we conclude that _ for any shape of the detached shock surface _ , the effective mach number near the symmetry axis decreases at least as the square of the azimuthal angle @xmath448 . + this result is particularly useful in order to prove that the mass accretion rate in the subsonic region of type sf flows decreases to zero with the size of the accretor .

attractors for dynamical systems are traditionally viewed as being asymptotically stable invariant sets which have a neighbourhood that absorbs all sufficiently close initial conditions . as a consequence the basins of such attractors contain open sets . whereas this concept works very well for simple attractors of ordinary differential equations , for more complicated attractors the smallest asymptotically stable attractor may contain much more than the asymptotic dynamics of ` typical ' initial conditions . partly in response to this problem , milnor introduced a concept of measure attractor ( now usually called _ milnor attractor _ ) @xcite . this is a compact invariant set whose basins of attraction has positive measure in phase space . examples of riddled basins @xcite show that , even for smooth invertible dynamics , one can find milnor attractors with riddled basins ; i.e. such that any open set that intersects the basin has positive measure in the basin of a different attractor ; see @xcite for a recent review of riddled basins . in the absence of smoothness or invertibility , it is clear that more exotic attractors may appear . recent work @xcite on globally coupled networks of oscillator with delayed pulse interactions @xcite indicates that extreme cases of riddled basin attractors , _ unstable attractors _ , can appear , where there is a neighbourhood @xmath1 of the attractor such that almost all points exit from @xmath1 under the dynamics . even more surprisingly , numerical simulations @xcite seem to indicate that these attractors can appear in a robust way as long as there are sufficiently many oscillators in the system . the main aim of this paper is to give a rigorous explanation for the appearance and robustness of unstable attractors in such systems . this paper is organized as follows . section [ secunstableatts ] examines the general problem of unstable attractors in a systematic way . we give some necessary conditions for the appearance of unstable attractors and some motivating examples . subsequently , we address the relationship between networks of unstable attractors and robust heteroclinic cycles , and present a simple recipe of how to perturb a smooth flow with a robust heteroclinic cycle to a smooth semiflow with a network of unstable attractors . in section [ secpulsecos ] we consider the system of globally pulse coupled oscillators with delayed interactions studied in @xcite , including a generalization of the set of possible response functions . because the system combines continuous time evolution with discrete time events , it is strictly speaking a hybrid system @xcite with delay . the main result of this section is theorem [ thmfinite ] which shows that the dynamics reduces , after a finite time , to a finite dimensional hybrid system . it gives an explicit upper bound on the number of remaining dimensions in the system . in section [ secuasforpulsecos ] we prove that the system discussed in section [ secpulsecos ] exhibits unstable attractors ( theorem [ thm : uasexist ] ) and that these persist for an open set of parameter values ( theorem [ thm : uasrobust ] ) . the proof has two parts ; firstly we show ( by following an appropriate open set of initial conditions ) that a certain periodic orbit is a milnor attractor ; secondly we show that this orbit is a saddle . we discuss this mechanism as being an effect of ` dimension jump ' in the system . section [ secdiscuss ] discusses obstacles to proving the existence and robustness of unstable attractors in a more general setting . for convenience some of the details of the proofs have been placed in appendices : appendix [ app : k=1 ] describes the system in the case that the delay is short enough that only one delayed pulse can influence the future dynamics of the system . appendix [ app : unstable ] derives the return map on a section transverse to the unstable attracting periodic orbit . consider a dynamical system defined by a semiflow @xmath2 on a finite dimensional manifold @xmath3 for @xmath4 ( recall that a semiflow is a family of maps such that @xmath5 whenever @xmath6 , @xmath7 , and such that @xmath8 ) . in what follows we will consider continuous ( @xmath9 ) or discrete ( @xmath10 ) dynamics . our examples arise from cases where @xmath11 is not invertible for all @xmath12 : one can not extend the semiflow to a flow . for convenience we will assume that @xmath3 is compact and let @xmath13 denote lebesgue / riemann measure on @xmath3 . recall that the @xmath14-limit set of @xmath15 is defined by @xmath16 a _ milnor attractor _ for @xmath17 is a compact invariant subset @xmath18 such that ( i ) the basin of attraction @xmath19 has positive measure in @xmath3 , i.e. @xmath20 , and ( ii ) any compact invariant proper subset of @xmath0 has a basin with strictly smaller measure @xcite . it is a _ minimal milnor attractor _ if any compact invariant proper subset of @xmath0 has a basin with zero measure . all explicit examples of milnor attractors given in this paper are fixed points or periodic orbits and thus they are minimal milnor attractors because they do not contain any compact invariant proper subset . however , we will not explicitly discuss minimality in these cases . given any subset @xmath21 we define the _ lingering subset _ of @xmath1 to be @xmath22 [ def : lingeringset ] the lingering set is a ( forward ) invariant set consisting of points in @xmath1 which do not leave @xmath1 in the future ; note that if @xmath0 is an invariant set within @xmath1 then @xmath23 is non - empty as it contains @xmath0 . [ def : ua ] we say a milnor attractor @xmath0 is an _ unstable attractor _ if there is a neighbourhood @xmath1 of @xmath0 with @xmath24 requirement ( [ eq : nocontractingbasin ] ) means that almost all trajectories in a neighbourhood of the attractor must leave this neighbourhood . we distinguish between two classes of unstable attractors . [ def : uapositivelocalbasin ] an unstable attractor @xmath0 is an _ unstable attractor with positive measure local basin _ ( resp . [ def : uazerolocalbasin ] ) . ] if @xmath25 for all neighbourhoods @xmath1 of @xmath0 . [ def : uazerolocalbasin ] an unstable attractor @xmath0 is an _ unstable attractor with zero measure local basin _ if there is a neighbourhood @xmath1 of @xmath0 with @xmath26 for an unstable attractor with positive local basin , ( [ eq : nocontractingbasin ] ) means that almost all trajectories starting in a sufficiently small neighborhood @xmath1 will first leave @xmath1 . a positive measure will eventually return and be asymptotic to @xmath0 ( contrasting with with @xcite where there is a positive measure that remains in @xmath1 and is asymptotic to @xmath0 ) . for an unstable attractor @xmath0 with zero local basin almost all initial conditions in a sufficiently small neighbourhood @xmath1 of @xmath0 will leave @xmath1 and never limit to @xmath0 . has positive measure if @xmath27 ; it has full measure ( in @xmath1 ) if @xmath28 and @xmath29 . ] clearly , linearly stable fixed points or periodic orbit attractors for a smooth invertible system can not be unstable attractors ; in such cases one can construct a neighbourhood that by the hartman - grobman theorem is contained within the basin of attraction . if @xmath0 is asymptotically stable then trivially it can not be an unstable attractor . a more general observation is the following proposition characterizing a class of attractors that can not be unstable with zero local basin . suppose that @xmath0 is a milnor attractor whose basin @xmath30 contains an open set , and suppose that @xmath17 is almost everywhere a local homeomorphism . then @xmath0 can not be an unstable attractor with zero local basin . [ prop : nounsts ] * proof : * consider a point @xmath31 in the interior of the basin of @xmath0 such that @xmath11 is a local homeomorphism in a neighbourhood of @xmath31 , and choose an open neighbourhood @xmath1 of @xmath0 . there is a finite time @xmath32 after which @xmath33 . pick a small neighbourhood @xmath34 of @xmath35 in @xmath1 ; continuity of @xmath36 means that there is a neighbourhood @xmath37 of @xmath31 within the basin of attraction such that @xmath38 . continuity of the local inverse means that @xmath39 is also a neighbourhood of @xmath35 . however all points in @xmath37 are in the basin of @xmath0 and so @xmath39 is an open set in the basin of @xmath0 that is within @xmath1 . hence @xmath40 for any @xmath1 that is a neighbourhood of @xmath0 . @xmath41 in particular , this proposition ensures that continuous invertible flows with continuous inverse ( homeomorphisms ) do not exhibit unstable attractors with zero local basin . we have so far been unable to determine whether the assumptions of proposition [ prop : nounsts ] exclude the possibility that @xmath0 is an unstable attractor with positive measure local basin , though clearly under additional assumptions ( such as linear stability of @xmath0 ) we can exclude this case . as we will see below , unstable attractors occur robustly for certain classes of noninvertible dynamical systems . suppose @xmath42 is a space of dynamical systems under some suitable topology , such that one can obtain an unstable attractor for some @xmath43 . if there are unstable attractors for all @xmath44 in a neighbourhood of @xmath45 then we say the unstable attractors for @xmath45 are _ robust _ ; this means that they are stable to perturbations of the system within this @xmath42 . in section [ secuasforpulsecos ] we show robustness of unstable attractors with respect to perturbations of a pulse response function defining a class of pulse coupled oscillator systems . there is no simple relationship between the lingering set for a given neighbourhood of an attractor and the basin of attraction ; in particular there may be points that are in the lingering set and not in the basin , and vice - versa . for example , near a centre in a planar system the lingering set can be an open set but the basin just one point . similarly the flow of @xmath46 on the unit circle @xmath47 has a fixed point @xmath48 . the basin of the attractor @xmath49 is the whole of phase space , but the lingering sets of small open neighbourhoods of @xmath48 will not contain nearby points to the right of @xmath48 . unstable attractors are a related but different phenomena to riddled basin attractors @xmath0 @xcite . recall that an attractor @xmath0 has _ riddled basin _ if @xmath30 is such that @xmath50 for all open sets @xmath1 that intersect @xmath30 . in particular , any neighbourhood @xmath1 that contains @xmath0 will intersect @xmath30 in a set of positive measure ; this means that a positive measure set in the basin leaves @xmath1 . for unstable attractors almost all points in a small enough @xmath1 will leave @xmath1 ; by the comment above , this does not have any direct implication for the basin of attraction of @xmath0 , though it does imply that any ` local basin of attraction ' relative to @xmath1 will have zero measure . examples of riddled basin attractors so far all possess irregular or chaotic dynamics to provide attraction local to some points and repulsion local to others . by contrast the dynamics observed on unstable attractors can be very simple , even equilibrium . to illustrate the possible types of unstable attractors we consider the dynamical systems defined by the piecewise continuous mapping @xmath51\mapsto[0,1]$ ] according to @xmath52 and @xmath53\rightarrow[0,1]$ ] where @xmath54 except on @xmath55 in which case @xmath56 . these two maps are illustrated in figure [ figeg1 ] . one can verify that the only invariant sets for @xmath45 and @xmath44 are the two fixed points @xmath48 and @xmath57 . these are both minimal milnor attractors ; for @xmath45 their basins of attraction are respectively @xmath58}.\label{eq : basinf(0)}\ ] ] whereas for @xmath44 they are @xmath59.\label{eq : basing(0)}\ ] ] in both cases the disjoint union of the two basins is the interval @xmath60 $ ] . on the other hand , any open set @xmath61 that contains @xmath62 has lingering set @xmath23 that is just the point @xmath63 . hence in both cases this point is an unstable attractor . considering the definitions ( [ eq : ualocalbasin ] ) and ( [ eq : uanolocalbasin ] ) above , we see that the attractor @xmath64 is an unstable attractor with zero local basin for @xmath45 and an unstable attractor with positive local basin for @xmath44 . and @xmath44 exhibit an unstable attractor @xmath65 with ( a ) zero and ( b ) positive local basin . [ figeg1],width=302 ] as noted in @xcite , unstable attractors can form networks such that a positive measure of sufficiently small perturbations from the unstable attractor @xmath66 are in the basin of attraction of some other unstable attractor @xmath67 . we show that this is analogous to robust heteroclinic networks . a set of unstable attractors @xmath66 , @xmath68 forms a _ network _ given by the directed graph with vertices @xmath66 . there are directed edges from @xmath66 to @xmath67 if and only if for any neighbourhood @xmath69 of @xmath66 we have @xmath70 if in addition @xmath71 we say the network is _ closed_. all trajectories near a closed network can not leave a neighbourhood of the network . in such a case the dynamics of the system is well described by the network even in the presence of small perturbations to the system . we illustrate such a network schematically in figure [ fignetworkofuas ] . if a network is not closed , we only have @xmath72 and such a network of unstable attractors @xmath73 may still manifest itself in transients . if there is a proper inequality in ( [ eq : notclosed ] ) then a positive measure of initial conditions starting near the network are not asymptotic to states that are in the network . a network is called _ transitive _ if there is a directed path between any two attractors . an arbitrary closed network will contain subnetworks that are transitive networks of unstable attractors . may form a network where a link from @xmath73 to @xmath74 implies that the basin @xmath75 intersects every neighbourhood of @xmath73 in a set of positive measure . in the example sketched above , there is a transitive subnetwork @xmath76 . [ fignetworkofuas ] , width=151 ] we briefly demonstrate that one can obtain unstable attractors from smooth semiflows with saddles by including small perturbations that induce collapse onto their stable manifolds . consider a hyperbolic saddle equilibrium @xmath77 for a ( non - invertible ) semiflow @xmath2 on a compact @xmath3 with stable manifold @xmath78 . we say @xmath17 exhibits _ collapse onto the stable manifold _ of @xmath79 if there is an open set @xmath80 and a @xmath81 such that @xmath82 . since @xmath83 is a saddle , and has thus an unstable manifold , such an @xmath1 can not contain @xmath83 . similarly , one can define _ collapse onto the unstable manifold _ of a saddle @xmath83 . this can occur on a neighbourhood that includes @xmath83 and in this case @xmath1 may contain @xmath83 . an example of an unstable attractor constructed by using a pair of saddle equilibria is shown in figure [ fig : twosaddles ] ; this is essentially the mechanism that causes the appearance of unstable attractors in the system we discuss in section [ secpulsecos ] . if there is a connecting orbit from @xmath84 to @xmath83 then collapse onto the unstable manifold of @xmath84 is simultaneously a collapse onto the stable manifold of @xmath83 ; the points in the open set @xmath34 intersecting the unstable manifold of @xmath85 tend to @xmath79 . unstable attractors in this sort of scheme can be robust to perturbations of the system if the connecting orbits are in invariant subspaces . is an unstable attractor . the set @xmath86 is a heteroclinic connection from @xmath84 to @xmath83 . we assume that the open set @xmath34 in the neighbourhood @xmath87 of another equilibrium @xmath84 collapses in finite time onto the stable manifold @xmath88 of @xmath79 . hence @xmath89 ; moreover , almost all perturbations to @xmath83 within @xmath90 will leave @xmath90 and so @xmath83 is an unstable attractor . [ fig : twosaddles ] , width=302 ] smooth dynamical systems with symmetries can exhibit heteroclinic cycles or homoclinic cycles as robust milnor attractors , if the connecting orbits lie within symmetry - forced invariant subspaces ; see for example @xcite . we indicate briefly how one can use these to construct a large class of systems that have networks of unstable attractors . consider a smooth semiflow @xmath2 on @xmath3 with an absorbing open region @xmath91 such that the only milnor attractor in @xmath92 consists of the set @xmath93 for hyperbolic saddle equilibria @xmath94 and the unstable manifolds @xmath95 . we assume that the unstable manifolds are one dimensional and write the connecting orbits from @xmath94 to @xmath96 by @xmath97 given such an attractor @xmath98 , we can construct piecewise smooth semiflows @xmath99 such that the set of @xmath94 forms a network of unstable attractors , and such that @xmath17 and @xmath100 are identical except on a set of arbitrarily small measure . to do this , pick a section @xmath101 to each connection @xmath102 and a neighbourhood @xmath103 of @xmath101 of arbitrarily small measure . define @xmath100 to be a semiflow that is equal to @xmath17 when the trajectories are outside of the @xmath103 . on entering the @xmath69 we require that the semiflow collapses onto the connecting orbit in finite time , as illustrated in figure [ figuasfromhets ] . the illustration depicts this for one - dimensional connections but there can clearly be higher - dimensional connections with qualitatively the same features . can be locally perturbed in sets of arbitrarily small measure to obtain ( b ) a network of unstable attractors with the same connection topology . here , trajectories entering the balls hit one of the connecting orbits after a finite time and continue out after a uniformly bounded time . [ figuasfromhets ] , width=302 ] note that robust networks of unstable attractors can appear under the same conditions as for robust heteroclinic attractors @xcite by adding collapse onto _ unstable _ manifolds because in this case unstable manifolds are contained within stable manifolds of other saddles . this can arise in cases where there is a discontinuity in the semiflow , but we note that it is not necessary ; one could construct an unstable attractor similarly to figure [ fig : twosaddles ] for a smooth semiflow , as long as this maps some open set onto a zero measure subset of the stable manifold of @xmath85 . in this section we consider the system of oscillators with delayed pulse interactions investigated in @xcite . first we will show that this system , although formally infinite dimensional , can be reduced under general assumptions to a finite dimensional system . the number of dimensions remaining depends on the system parameters . the system of oscillators we consider is technically a hybrid dynamical system . a hybrid dynamical system is defined by a smooth flow in some regions of phase space with boundaries ; when the flow hits one of the boundaries it is then mapped instantaneously by a deterministic mapping to some other point on the boundary . a trajectory from a hybrid dynamical system hence consists of intervals of continuous evolution interrupted by events that may be discontinuous . an evolution of the system is a _ zeno state _ @xcite ( also referred to as a _ chattering state _ ) if there can be an infinite number of applications of the map in some finite time interval ; in such cases one needs to have extra rules to permit evolution beyond the first accumulation point in time of an infinite number of applications of the map . we will show that there are no zeno states in the system considered below . [ def : phasehistory ] we define @xmath104 to be the set of piecewise smooth ` phase history ' functions @xmath105\rightarrow[0,1)$ ] , @xmath106 such that : * _ well defined limit from below . _ for some discrete ( possibly finite ) , monotone increasing set of @xmath107 , @xmath108 we have @xmath109 is smooth for @xmath110 . * _ no zeno states . _ if there are infinitely many discontinuities @xmath111 then @xmath112 . note that ( i ) means that the limit from below @xmath113 is defined for all @xmath114 . condition ( ii ) means that we allow no zeno states in the past . the system @xcite of @xmath115 pulse - coupled oscillators with delay @xmath116 , where the phase histories are @xmath117 , is defined in terms of a semiflow on @xmath118 for all @xmath119 @xmath120 such that @xmath121 for all @xmath122 . on the interval @xmath123 $ ] the function @xmath124 is defined below . the boundaries of the intervals on which the phase history function @xmath125_i$ ] of oscillator @xmath126 is smooth are now denoted by @xmath127 , @xmath128 . if @xmath129 we say oscillator @xmath126 _ fires _ at time @xmath130 . given that the system has evolved a time @xmath131 from some initial condition @xmath132 at time @xmath133 , we list the past firings @xmath134 of oscillator @xmath126 as @xmath135 such that the @xmath136 are the times since the @xmath137th last firing of the @xmath126th oscillator before or at time @xmath131 . note that the @xmath138 , @xmath139 count the firing events whereas the @xmath140 , @xmath139 count all discontinuities that may occur . we write the phase of the oscillator @xmath126 as @xmath141 , namely as the most recent point in the phase history of that oscillator , and the previous firing times as @xmath138 . the future dynamics will be determined only by the present phases and the times of the past firings ; formally there is a map @xmath142 such that @xmath143 thus the map @xmath144 directly gives the current phases @xmath145 and the past firing times @xmath136 , @xmath108 , of the oscillators @xmath146 . similarly , it will be useful to define a map onto the phases and the @xmath147 past firing times : @xmath148 the semiflow @xmath149 in ( [ eq : semiflowforoscs ] ) is given as a hybrid system with continuous time evolution @xmath150 for all @xmath126 and @xmath147 , except when one of a number of possible discrete events occur . these events are determined by a constant delay @xmath151 between one oscillator being reset and ` sending ' a pulse ( ` firing ' ) and its ` reception ' by another oscillator . these events ` firing ' and ` reception ' are the only types of discrete events interrupting the continuous time evolution . we say that at a time @xmath32 * _ oscillator @xmath126 fires _ , + denoted @xmath152 , if @xmath153 . * _ oscillator @xmath126 receives a pulse from oscillator @xmath137 _ , + denoted @xmath154 , if @xmath155 for some @xmath147 and @xmath156 . * _ oscillator @xmath126 fires induced by reception of a pulse from oscillator @xmath137 _ , + denoted @xmath157 , if @xmath154 and this results in an immediate firing of oscillator @xmath126 ( see below ) . observe that for any @xmath126 it is not possible that @xmath158 and @xmath154 ( for any @xmath137 ) occur simultaneously , whereas if @xmath154 occurs it is possible that @xmath159 , @xmath160 also occurs at the same time . the phases @xmath161 and firing times @xmath162 , and thus the evolution @xmath163 occurs as ( [ eq : phisdotone ] ) until one of @xmath158 , @xmath164 or @xmath165 occur at time @xmath166 . by definition [ def : phasehistory ] , we take @xmath167 and @xmath168 for all @xmath169 and all @xmath131 . at such an event we define the new phases using the list below . one can verify that all variables stay in their respective ranges if they are started in that range and evolving in this way . * _ for each @xmath126 such that @xmath158 occurs _ + we set @xmath170 * _ for each @xmath126 such that @xmath154 occurs _ + we set @xmath171 where @xmath172 is defined below , @xmath173 , @xmath174 , and @xmath175 is the number of ` input pulses ' that arrive at time @xmath176 . * once all the @xmath154 have been checked , we say there is _ firing @xmath165 induced by pulse reception _ if @xmath154 occurs for some @xmath177 and results in @xmath178 . in this case we set @xmath179 , @xmath180 and renumber @xmath181 for all @xmath182 , as if @xmath152 occurred . one can verify that the order of ` processing ' simultaneous events with different @xmath126 does not affect the final outcome ; hence the map is well - defined . the condition @xmath183 in ( [ eq : kinputsnow ] ) excludes self - interaction as in @xcite which we will consider exclusively in the following . the response function @xmath184 expresses the response of an oscillator at phase @xmath185 to an input @xmath186 . the model is specified once the delay @xmath187 , the coupling strength @xmath188 and the function @xmath34 are known . observe that given an initial @xmath189 and @xmath190 for @xmath191 we can define @xmath192 the phase history at time @xmath81 . this way we complete the definition of @xmath193 ( [ eq : semiflowforoscs ] ) for times @xmath123 $ ] . for continuous functions @xmath194\rightarrow { \mathbb{r}}$ ] we define their _ difference increments _ @xmath195 we say @xmath34 is _ uniformly strictly monotonic increasing _ on @xmath60 $ ] if @xmath196 for all @xmath197 ( note that this implies that @xmath172 is strictly monotonic increasing in @xmath185 ) . [ response functions with increasing response ] [ def : cv ] we define @xmath198 be the set of response functions @xmath199\times [ 0,c ] \rightarrow [ 0,\infty)$ ] for some @xmath200 with @xmath201 satisfying * _ continuity : _ @xmath172 is jointly continuous in @xmath185 and @xmath202 . * _ zero response to zero input : _ @xmath203 for all @xmath185 . * _ uniform strict monotonicity : _ for any fixed @xmath174 the function @xmath204 is uniformly strictly monotonic increasing in @xmath185 on @xmath60 $ ] . * _ upper bound linear in @xmath202 : _ for any fixed @xmath174 , monotonicity of @xmath34 means there is a unique @xmath205 that solves @xmath206 which gives the maximum response ; we assume this is bounded above by a linear function of @xmath202 . more precisely we assume there is a @xmath207 such that @xmath208 for all @xmath209 . condition ( iv ) allows one to obtain a bound on the total change in phase over one period . note that @xmath210 for all @xmath211 and that for @xmath212 , the response @xmath172 is such that the updated phase ( [ eq : epshat ] ) is supra - threshold , resulting in an induced firing ; hence @xmath213 is the largest response to an input of magnitude @xmath202 . we say that response functions @xmath214 give _ increasing response _ because the response increases with @xmath185 . in particular , concave ( down ) potential functions give rise to this form of response ( see lemma [ lemma : uv ] below ) . we give this space the structure of a metric space with norm @xmath215\times[0,c ] } @xmath216 . closeness of @xmath34 and @xmath217 in this norm imply that both the functions and their differences increments are close , by the following lemma . [ lemma : deltabound ] suppose that @xmath218 ; then @xmath219 . * proof : * suppose that @xmath218 , fix @xmath202 and drop the second argument of @xmath172 for convenience . then @xmath220 for all @xmath185 . the triangle inequality implies that @xmath221 for all @xmath222 and @xmath223 . for a given value of @xmath185 there is a sequence @xmath224 such that @xmath225 as @xmath226 . by considering @xmath227 in ( [ eq : dv2delta ] ) we can conclude that @xmath228 . similarly by considering @xmath229 such that @xmath230 as @xmath226 we have using ( [ eq : dv2delta ] ) that @xmath231 . this implies @xmath232 for all @xmath185 and @xmath202 . @xmath41 the resetting rule @xmath233 with a twice continuously differentiable potential function @xmath1 was considered previously @xcite to describe a phase jump induced by a reception ( @xmath154 ) of a pulse of strength @xmath202 . this ` mirollo - strogatz ' case is a special case of the more general reception rule we use here , if we substitute @xmath234 we now show that , if @xmath1 is increasing concave downwards , this fits into the more general reception scheme of definition [ def : cv ] . in figure [ fig : vanduplots ] we illustrate a typical choice of @xmath1 and its consequence for @xmath34 for given @xmath185 and @xmath202 . [ lemma : uv ] consider a response defined by @xmath235 in ( [ eq : phi_uphi],[eq : vfromu ] ) where @xmath236 and @xmath1 is an unbounded twice continuously differentiable function @xmath237 that satisfies @xmath238 , @xmath239 , @xmath240 and @xmath241 for @xmath242 $ ] where @xmath243 ( @xmath244 ) are positive constants . then @xmath245 for small enough @xmath246 . note that the details of the definition of @xmath1 are only critical in the range @xmath60 $ ] . this is because any time that @xmath247 we will get resetting ; the definition on @xmath248 is simply for expositional convenience . * proof : * one can directly verify that the conditions ( i)(ii ) in the definition of @xmath198 are satisfied . to show ( iii ) , note that @xmath1 monotonic increasing and surjective on its range means it has a unique inverse @xmath249 that is also monotonic increasing . to see that ( iii ) also holds , first observe that @xmath250 the condition @xmath251 means that @xmath1 is strictly monotonic increasing . the condition @xmath252 means that @xmath253 for all @xmath15 satisfying @xmath254 . together with the monotonicity of @xmath249 we obtain , substituting @xmath255 , @xmath256 similarly , @xmath257 means that @xmath258 is monotonic decreasing ; @xmath259 implies @xmath260 for all @xmath185 and thus , using ( [ eq : ujumpbounded ] ) , @xmath261 substituting into ( [ eq : vprime ] ) this leads to @xmath262 hence for fixed @xmath174 we have @xmath263 for all @xmath264 , and so ( iii ) holds . finally , to show ( iv ) , note that for @xmath265 eq . ( [ eq : phim ] ) , which determines @xmath205 , becomes @xmath266 so that by definition ( [ eq : phim ] ) , the maximum change at a reset is @xmath267 clearly , @xmath268 for all @xmath174 . moreover , @xmath269 by definition ( eq . ( [ eq : vonems ] ) ) , @xmath270 and @xmath271 ^ 3}<0\ ] ] because @xmath251 and @xmath257 in the relevant range . thus @xmath272 is increasing and concave and so we have the linear upper bound @xmath273 where @xmath274 . @xmath41 satisfying the conditions for lemma [ lemma : uv ] that hence giving a resetting function @xmath172 with increasing response . @xmath275 in this case is defined in ( [ eq : ub ] ) with @xmath276 . [ fig : vanduplots ] , width=226 ] we proceed to show that the system of pulse coupled oscillators with increasing response will reduce to a finite dimensional system for an open set of parameter values . theorem [ thmfinite ] gives an estimate @xmath147 for the largest number of ` past firings ' one needs to consider in the limit of large time . we can effectively reduce to a system of dimension at most @xmath277 after a finite time . the proof works by considering the maximum number of resets that can occur to an oscillator in any time interval @xmath278 $ ] . this depends on the number of firings that occurred in the previous time interval of the same length , giving the recurrence inequality in lemma [ lemestimate ] . we reduce the dynamics from a hybrid delay differential equation with infinite dimensional phase space to a finite dimensional system by restricting to phase histories of the form : @xmath279 namely , those that have at most @xmath147 firings within the delay time @xmath116 . [ lem : finite ] suppose that @xmath189 and @xmath280 are in @xmath281 and @xmath282 . then @xmath283 for all @xmath7 . * proof : * this result states that any two initial states in @xmath281 whose past @xmath147 firings and phases are the same will have precisely the same phases ( and hence firings ) at all times in the future . this is because by definition all firings beyond the @xmath147th in the past will not affect the future phases . @xmath41 the previous lemma does not necessarily imply that @xmath284 . [ thmfinite ] suppose that @xmath285 . then there is a @xmath286 ( depending on @xmath202 , @xmath116 and @xmath34 ) such that for any @xmath287 there is a finite @xmath81 ( depending on @xmath189 , @xmath202 and @xmath34 ) such that @xmath288 for all @xmath289 . moreover , after time @xmath32 the number of firings of an oscillator that can appear in any time interval of length @xmath116 is at most @xmath290 where @xmath291 is the smallest integer less than or equal to @xmath15 . for any @xmath147 such that the conclusion of theorem [ thmfinite ] holds we say the dynamics _ reduces _ to @xmath292 . we prove theorem [ thmfinite ] using the following lemma . for any initial condition @xmath132 let @xmath293 ( @xmath294 ) and suppose that the number of firings of @xmath295 in the interval @xmath296 is @xmath297 . [ lemestimate ] for all @xmath298 we have @xmath299 * proof : * suppose that there have been at most @xmath300 firing events of each of the @xmath115 oscillators during the time interval @xmath296 . consider the events that affect the phase @xmath301 of an arbitrary oscillator @xmath302 in the future time interval @xmath303 ; there will be @xmath304 events in @xmath303 where at each event @xmath305 , @xmath306 pulses of the oscillators @xmath307 , @xmath177 are simultaneously received by @xmath126 . the total number of received pulses is @xmath308 . the total change @xmath309 in phase of oscillator @xmath126 up to time @xmath116 is hence at most @xmath310 hence oscillator @xmath126 will have fired at most @xmath311 times in the interval @xmath303 . since @xmath126 is arbitrary , we have in general at most @xmath312 firing events of each oscillator in the interval @xmath313 $ ] and the lemma follows . @xmath41 * proof : * * ( of theorem [ thmfinite ] ) * since @xmath314 applying lemma [ lemestimate ] inductively we have @xmath315 and so because @xmath316 , after some finite time @xmath317 ( which can be estimated from ( [ eq : timeestimate ] ) given @xmath318 ) we have @xmath319 the remainder of the theorem is a consequence of applying lemma [ lem : finite ] . @xmath41 the estimate of @xmath147 will typically be a very poor upper bound . if the number of oscillators @xmath115 is small one can improve on these bounds by noting that the pulses are ` quantized ' into multiples of @xmath320 . in particular , one can replace the upper bound ( [ eq : vone ] ) by requiring only that there is a @xmath207 such that @xmath321 for all @xmath322 . for larger @xmath115 this approaches ( [ eq : vone ] ) . as a consequence of lemma [ lemestimate ] we can now justify defining the dynamics as a semiflow on @xmath323 . a system with @xmath214 that satisfies @xmath324 will remain free of zeno states for any @xmath174 and @xmath151 . this is because , although the number of firings in an interval @xmath278 $ ] can increase if @xmath325 is large , the maximum rate is given in lemma [ lemestimate ] and there can not be infinitely many firings in any time interval of finite length . another consequence of the proof of theorem [ thmfinite ] is that the semiflow on @xmath326 has a restriction to @xmath327 that is a well - defined semiflow . in the case that @xmath330 we can apply theorem [ thmfinite ] to see that @xmath331 , and so @xmath329 . in fact , this is not a very tight estimate , and in many cases where ( [ eq : k=1 ] ) does not hold the dynamics still reduces to @xmath329 . for a particular initial condition @xmath189 one can practically check the dimension it reduces to by finding the @xmath332 of the number of firings of one oscillator in any time interval of length @xmath116 for a neighbourhood of that initial condition . if at most one previous firing is needed to determine the phases uniquely in the future , the dynamics reduces to a semiflow on a @xmath328 dimensional space @xmath333 where we write @xmath334 in shorthand . for the oscillators we consider , the semiflow @xmath335 for @xmath119 is well defined for @xmath202 and @xmath116 small enough ; appendix [ app : k=1 ] gives an explicit form on this reduced space . the main result in this section is a proof that unstable attractors exist and are robust in the system of delay pulse coupled oscillators discussed in section [ secpulsecos ] . for this section we consider the potential function @xmath336 used in @xcite . computing the associated response function @xmath34 defined by ( [ eq : vfromu ] ) we find an affine relationship between @xmath185 and @xmath34 : @xmath337 and one can check that @xmath214 ; this is a particular case of lemma [ lemma : uv ] ; the relationship between @xmath1 and @xmath34 is illustrated in figure [ fig : vanduplots ] . [ thm : uasexist ] consider the system of oscillators defined in section [ sect : defsemiflow ] with @xmath1 defined by ( [ eq : ub ] ) with @xmath338 , @xmath339 and @xmath340 and @xmath276 . this system has a periodic orbit that is an unstable attractor with zero measure local basin . * proof : * by applying theorem [ thmfinite ] one can directly verify that any initial condition @xmath341 will reduce to the @xmath342 dimensional space @xmath343 after a finite time , i.e. such that only the phase and last four firings of each oscillator contribute to the dynamics at any point in the future . now consider the open set of initial conditions defined by @xmath344 where @xmath345^{8}\subset \pi_{1 } \left({\mathcal{p}}^4\right).\ ] ] as a slight abuse of notation , we will also write @xmath346 as meaning @xmath347 if the context is clear . note that all the @xmath348 , meaning that @xmath349 . theorem [ thmfinite ] means that at points in the future we may in principle arrive at states that are in @xmath350 for @xmath351 but not for @xmath352 . however , we will verify below that @xmath353 for all @xmath12 . note that @xmath354 is an open set , as is its image under @xmath355 is open for any @xmath356 ; we will show that it is in the basin of attraction of a saddle periodic orbit . in what follows we write @xmath357 to denote the @xmath147th intersection of the trajectories in @xmath358 with one or more of the ` events ' @xmath152 or @xmath154 . since the information about the original initial condition of the @xmath359 has been lost we list only the first four components . we compute that @xmath360\times[0.749,0.751]\times[0.349,0.351]\times\{0\ } \label{eq : c1}\ ] ] due to the first event being @xmath361 . the next event is @xmath362 which will give @xmath363~~\}\label{eq : c2}\ ] ] and so we have reduced to a one dimensional subspace of phase space at this point . a very tedious direct numerical verification listed in table [ tab : c0traj ] shows that _ all _ of the points starting in @xmath364 after a further 32 events reach a single point @xmath365 where @xmath366 and @xmath367 can be computed exactly ( see appendix b ) . we present in figure [ fig_return ] the values of the coordinates depending on choice of @xmath368 over a larger range than in @xmath364 ; observe that a one dimensional set of possible initial values converges to exactly the same point . the point @xmath369 ( [ eq : c22p ] ) is a cross section to a periodic orbit where two pairs of oscillators remain synchronized . since the periodic orbit through @xmath369 contains the open set @xmath346 in its basin it is a milnor attractor for any reduction of the system to finite dimensions , for example to @xmath370 . @xmath371 dependence of dynamics on the initial phases @xmath372 . panel ( a ) shows the phases ( @xmath373(@xmath374 ) , @xmath375(@xmath376 ) , @xmath377 ( @xmath41 ) , @xmath378(@xmath379 ) ) after @xmath380 further intersections with discontinuities of the semiflow ( ` events ' ) in dependence of @xmath368 . observe that there is a large range of @xmath368 for which the dynamics arrives at precisely the same point . panel ( b ) shows the actual attractor reached from the above initial conditions ( shown after the @xmath381 reset of oscillator @xmath382 . surprisingly , the same attractor is reached by several initial conditions whose trajectory after 34 events differs substantially . , width=340 ] now we show that the orbit through @xmath369 is unstable . in order to do this we show that almost all perturbations of the first two coordinates grow to finite size under the action of the local return map . a complete stability analysis is presented in the appendix [ app : unstable ] . in particular , as is shown below , this proves instability in general , not only linear instability . in this analysis we consider the return map from the plane @xmath383 to itself . the pulses @xmath359 are kept locked to the phases @xmath301 such that the nonlinear return map ( [ eq : f(delta ) ] ) @xmath384 is three - dimensional and maps a perturbation @xmath385 to the periodic orbit started at ( [ eq : c22p ] ) . @xmath386 to the perturbation @xmath387 when the plane @xmath388 is reached the next time . if we take a general perturbation ( satisfying @xmath389 ) just before arriving at @xmath390 , observe that the @xmath391 are all larger than @xmath116 and hence will not affect the future of the phases . starting with such a perturbation , when it hits the next event we can arrive at an arbitrary perturbation satisfying @xmath392 such that immediately after the perturbation , events occur in the order @xmath393 . note that the order of the events @xmath394 and @xmath361 does not play a role as they become synchronized at their first firing . as demonstrated explicitly in appendix b , there is a small @xmath395 such that the return map ( [ eq : fdelta ] ) defined in the three - dimensional region @xmath396 is given by a smooth map with @xmath397 ( and such that @xmath398 is a line that intersects @xmath399 ) . in particular , given the map @xmath400 in terms of an arbitrary function @xmath214 ( see appendix b ) we analyse of the fully nonlinear local dynamics of a sufficiently small perturbation @xmath401 , @xmath402 , to the periodic orbit . for the following estimate , we write @xmath403 for @xmath404 in shorthand notation . first note that requirement ( iii ) in def . [ def : cv ] implies that there is a @xmath405 such that for all @xmath406 , @xmath407 for all @xmath174 , in particular , @xmath408 is monotonic increasing in @xmath185 . this implies that @xmath409 hence we find that @xmath410 implies that any perturbation @xmath411 satisfying @xmath412 increases ( e.g. in maximum norm ) . because the function @xmath413 is non - decreasing in @xmath414 , we see that there is a @xmath395 for which any sufficiently small perturbations ( satisfying @xmath415 ) leave the region @xmath399 in finite time @xmath416 , bounded by @xmath417 by the permutation symmetry of the system , the analysis in the case @xmath418 is analogous . this means that almost all sufficiently small perturbations @xmath419 leave a neighbourhood of the attractor in finite time , proving that the attracting periodic orbit considered is also unstable . the case @xmath420 can be treated analogously by the symmetry of the globally coupled network . taken together there is a neighbourhood @xmath421 of @xmath422 such that all points starting in @xmath421 with @xmath423 will iterate away until they leave this neighbourhood . this proves instability of the orbit through @xmath369 ( [ eq : c22p ] ) . @xmath41 figure [ fig : xc ] shows the dynamics starting from one initial condition ( [ eq : c0 ] ) used in the proof of the theorem . the convergence in finite time to the attractor a small random perturbation to the phases of all oscillators is expanded by the instability , yielding convergence towards another attractor . , @xmath340 , @xmath339 , @xmath424 , @xmath276 ) exhibiting an unstable attractor . at time zero , initial phases in @xmath346 are shown . the time axis is discrete and shows the phase when oscillator @xmath425 fires . just after the firing event @xmath426 of oscillator @xmath425 at times @xmath427 , all phases @xmath428 are plotted ( @xmath373(@xmath374 ) , @xmath375(@xmath376 ) , @xmath377 ( @xmath41 ) , @xmath378(@xmath379 ) ) . there is convergence of the trajectory in state space in finite time towards the periodic orbit passing through ( [ eq : c22p ] ) . an instantaneous small random perturbation ( @xmath429 ) to the phases at time @xmath430 leads to a switching to another attractor , demonstrating instability of the original periodic orbit . [ fig : xc],width=264 ] [ thm : uasrobust ] the unstable attractors of theorem [ thm : uasexist ] are robust in the sense that there is an open neighbourhood in @xmath198 of @xmath431 given by ( [ eq : vfromu],[eq : ub ] ) and an open neighbourhood of @xmath432 such that a system defined by any @xmath217 , @xmath433 , and @xmath434 in these neighbourhoods , has an unstable attracting periodic orbit that remains close to that of theorem [ thm : uasexist ] . * proof : * we refer to the return map @xmath435 ( [ eq : f(delta ) ] ) derived in the appendix b and already discussed in the proof of theorem [ thm : uasexist ] . note that in the derivation of the return map @xmath435 , the form of the transfer function @xmath436 and the precise values of the parameters @xmath116 and @xmath202 are not important , because @xmath435 continuously depends on @xmath436 , @xmath116 , and @xmath202 and changes its form only if the sequence of events occurring along the trajectory on or near the periodic orbit is altered . thus , an equivalent periodic orbit exists if the dynamical system is slightly perturbed , i.e. if @xmath436 , @xmath116 , and @xmath202 , are slightly changed from their original values . in particular , one can choose @xmath437 , where @xmath34 is sufficiently close to the original . since the mechanism of dimensional reduction and attractivity depends only on the supra - threshold synchronization mechanism given by the induced firing events @xmath165 , the slightly perturbed periodic orbit will also be an attractor . moreover , for a sufficiently small perturbation to @xmath34 , @xmath116 , and @xmath202 we still obtain @xmath438 which implies instability . the proof of the previous theorem ) . again , since the return map @xmath435 does not change form if the parameters @xmath436 , @xmath116 , and @xmath202 are slightly perturbed , also the proof of instability is independent of these parameters of the system . this completes the proof . @xmath41 this result implies that qualitatively the same dynamics is obtained by any flow defined by a @xmath217 nearby the original @xmath34 ( [ eq : vfromu ] ) ; in particular it does not have to be defined in terms of a any potential function @xmath1 by ( [ eq : vfromu ] ) . since the proof of theorem [ thm : uasexist ] is rather involved , we discuss some intuition in terms of what we term a dimension jump of the semiflow . given a semiflow @xmath439 that is not invertible , one can nevertheless define , for all @xmath440 , a set - valued inverse for any @xmath12 by @xmath441 and the semiflow can be extended uniquely to an invertible flow precisely when @xmath442 is a single point for all @xmath15 and @xmath12 . given sufficient regularity on @xmath443 , for example sufficient to ensure that @xmath442 is a union of manifolds , we can compute its dimension . if all @xmath442 are non - empty then @xmath444 will be a non - decreasing function of @xmath131 . the _ dimension jump _ @xmath445 at @xmath15 is of particular interest and we refer to @xmath446 as the _ dimension jump set _ , shown schematically in figure [ fig_jumpset ] . such behaviour can be found in systems exhibiting sliding modes @xcite , and we suggest that such systems may display unstable attractors . that has a unique inverse @xmath447 for @xmath448 . if we define @xmath449 then @xmath31 is in @xmath450 the jump set ( in this case @xmath451 ) and for any @xmath452 the dimension of the set @xmath453 is greater than zero . [ fig_jumpset],width=377 ] observe that trajectories converging to an invariant set that pass through @xmath450 may cause locally unstable invariant sets to be unstable attractors . for the system of delay coupled oscillators with @xmath329 there are two events on which dimension jump can occur : 1 . for one of the firing events @xmath158 or @xmath165 we lose dependence on the delayed phase @xmath454 . 2 . at a resetting @xmath455 that induces firing in two oscillators @xmath456 and @xmath457 we synchronize the oscillators @xmath458 and @xmath459 beyond that point ; this also causes a dimension jump . in out system , these events have ` jump sets ' that are codimension one and typically they cause a loss of one dimension ; however when the oscillators become synchronized , several such events can occur at once resulting in a loss of more dimensions . the proof of theorem [ thm : uasexist ] works by choosing an open set of initial conditions that reduces dimension down to one dimension after two further discrete events . the final loss of dimension that reduces the open set down to one point in the section occurs only a while later , and this is because there is a global connection in phase space that we need to follow . in this paper we give a rigorous definition for a novel type of milnor attractor , the _ unstable attractor _ , introduced in @xcite . we classify these attractors in terms of presence or absence of local basin . these attractors are surprising in many ways ; notably they have ` local repelling ' properties in that they have neighbourhoods such that almost all points leave the neighbourhood after a finite time . we show that unstable attractors not only exist , but even appear robustly in certain classes of hybrid dynamical systems of interest as physical and biological models ; namely in networks of pulse - coupled oscillators . the robustness that we have demonstrated is limited in the sense that we only consider perturbations that keep the time delay identical for all connections and the coupling function identical for all oscillators . we do not consider perturbations that break the identical nature of the oscillators or the symmetry of the system . on introduction of such general perturbations we expect no robustness for such unstable attractors , although one may obtain attractors that remain ` nearby ' to where a network of unstable attractors used to exist , i.e. close to a collection of unstable periodic orbits and their heteroclinic connections . for perturbations that break only part of the symmetry , unstable attractors may still persist . for instance there is numerical evidence that unstable attractors can still exist even the network is not coupled all - to - all , cf . the exact consequences of non - identical oscillators and more general broken symmetry remains to be studied . if we perturb the system in such a way that it becomes smooth and invertible , the unstable attractors will disappear ; any saddle periodic orbit will have a zero measure basin simply by existence of a global stable manifold of dimension less than that of phase space ; however as indicated in section [ secunstableatts ] one should be able to make the system arbitrarily differentiable and still obtain unstable attractors , as long as non - invertibility is retained . in proposition [ prop : nounsts ] we have shown that certain conditions on a general dynamical system exclude the possibility of unstable attractors with zero local basin . to generalize this finding to also exclude unstable attractors with positive local basin , we believe it will be necessary to strengthen these conditions . generalization of our result is not straightforward because a ( local ) basin may be highly irregularly , as for riddled basin attractors . our proofs of the main results are not very elegant ; this is because they rely on global features of the hybrid dynamical system , and we have not found them easily reducible to simpler arguments . we believe however that it should be possible for exact numerical methods @xcite to identify when unstable attractors are present in such systems . the numerical results in @xcite suggest that networks of unstable attractors exist in a wide range of systems of pulse - coupled oscillators . preliminary studies have shown that the networks of unstable attractor may often be transitive for a wide range of parameters and system sizes . revealing the mechanisms underlying the existence , structure and prevalence of such networks of unstable attractors is a challenging task for future research . * acknowledgements : * pa thanks the max - planck - institut fr dynamik und selbstorganisation ( formerly : mpi fr strmungsforschung ) for their hospitality and financial support , as well as a grant from the leverhulme foundation for partial support . mt thanks leonid bunimovich , stefan grokinsky and raoul - martin memmesheimer for useful and interesting conversations relating to this work as well as the department of mathematical science , university of exeter for their hospitality and financial support . we give the semiflow of globally coupled oscillator networks explicitly for the case @xmath329 by the following algorithm . consider @xmath460 where the @xmath126th component represent the @xmath126th phases and the @xmath461th component represents the time since the @xmath126th oscillator fired . suppose that @xmath462 for all @xmath7 . we note that we can represent the flow on @xmath463 by @xmath464 for all @xmath465 such that @xmath466 and @xmath467 . after time @xmath468 we hit the first event and apply a mapping according to the following algorithm : * set @xmath469 . * for each @xmath126 such that @xmath470 we say @xmath455 occurs for all @xmath471 and set @xmath472 for @xmath186 defined as in ( [ eq : epshat ] ) . * if there is an @xmath126 such that @xmath178 then we say @xmath158 occurs and set @xmath179 and @xmath473 . after applying this algorithm we are assured that @xmath474 , as @xmath475 and @xmath476 for @xmath477 . thus we can take this state as a new initial condition and rerun the algorithm sequentially from event to event . for the main results of this paper ( theorems 2 and 3 ) we use a nonlinear return map @xmath400 of perturbations @xmath419 to the unstable attracting periodic orbits in a network of @xmath338 oscillators using the semiflow detailed in appendix [ app : k=1 ] . we detail the derivation of this return map and analyze the periodic orbit that is the unstable attractor . we choose @xmath478 where @xmath479 , and consider cases where the initial @xmath480 . first we analyse the dynamics on the periodic orbit . to establish instability we then consider , without loss of generality , perturbations to the periodic orbit phases at a point on the orbit where the time since last pulse is greater than @xmath116 for all oscillators . this means that the future depends only on the current phases and we can effectively consider @xmath481 for all @xmath482 and all @xmath131 satisfying @xmath483 where @xmath484 defines the time of the last event such that @xmath485 and @xmath486 the time of the next event analogous to ( [ eq : algort ] ) . within this convention , we consider the return map from the three - dimensional plane defined by @xmath388 to itself and establish instability for this return map to prove instability of the periodic orbit . the following analysis applies for the parameters @xmath340 , @xmath339 , and the function @xmath424 ( [ eq : ub ] ) where @xmath276 . it will turn out that it also applies for a neighbourhood thereof , showing robustness of the instability of the periodic orbit . for compact notation , we introduce the _ transfer function _ @xmath487 this transfer function mediates the action of an incoming pulse of strength @xmath202 on the phase @xmath185 . we obtained the sequence of discrete events and the numerical values in the following analysis for the specific choice ( [ eq : vfromu],[eq : ub ] ) . first we determine the dynamics of the unperturbed orbit ( table [ tab : unperturbed_po ] ) . throughout the table , the phases after each event are labelled @xmath488 where the first index labels the oscillator @xmath482 and the second index counts the events @xmath147 occurring for a _ perturbed _ trajectory ( see table [ tab : perturbed_po ] below ) starting with @xmath489 for the initial condition ( first row of table [ tab : unperturbed_po ] ) . since the second oscillators @xmath425 and @xmath490 in the two synchronous groups have the same phases as the oscillators @xmath382 and @xmath491 , respectively , we define @xmath492 for all @xmath147 . we show that the unperturbed orbit indeed is periodic with period @xmath493 given by the time @xmath131 in the last row of table [ tab : unperturbed_po ] where@xmath494 is also read off the table . here we use the abbreviation @xmath495 for the strength of an individual pulse . . events occurring during one period of one of the unstable attracting periodic orbits in the system of four coupled oscillators ( see text for explanation ) . [ tab : unperturbed_po ] [ cols="<,<,^,^",options="header " , ] table [ tab : perturbed_po ] displays the time evolution of a general perturbation event by event . as for the unperturbed orbit described above , the left column gives the sequence of events and the second left column gives the time of these events . the right four columns give the phases of oscillators @xmath496 right after an event has occurred . the first row gives the initial condition at @xmath176 , just after the signal has been sent , @xmath426 , by oscillator @xmath425 . before , at time @xmath497 , the signal of oscillator @xmath382 has been sent , denoted @xmath498 ; this signal will travel until time @xmath499 . the initial phases ( [ eq : phi(0 ) ] ) are displayed in the first column . here @xmath500 by definition . throughout the table , the phases after each event are labelled @xmath501 where the first index labels the oscillator @xmath126 and the second index counts the events @xmath147 starting with @xmath489 for the initial condition as was also used in table [ tab : unperturbed_po ] above . for instance , the first event ( after the initial condition ) , @xmath329 , at time @xmath499 is the reception @xmath502 of the signal sent by oscillator @xmath382 . since there are no self - interactions , the phase of oscillator @xmath382 is only shifted in time , @xmath503 the phases of all other oscillators @xmath504 additionally jump because of the incoming subthreshold signal , @xmath505 again , as for the unperturbed dynamics , going through the table row by row , the phases are calculated event by event . the phases @xmath501 for @xmath496 right after the events are approximated by the unperturbed phases @xmath488 defined in table [ tab : unperturbed_po ] , @xmath506 for all @xmath147 for which the @xmath488 are defined , @xmath507 . here @xmath508 denotes a function of all @xmath509 , @xmath510 , that approaches zero at least linearly if all @xmath511 , i.e.@xmath512 . we thus obtain the return map@xmath513 from the last row of table [ tab : perturbed_po ] . here the @xmath514 again depend on the original perturbation vector @xmath419 . from this identification we obtain the components @xmath515 note that the component functions @xmath516 in ( [ eq : a : f_i ] ) are of order @xmath517 ; moreover the dynamics clearly depends only on the initial perturbation @xmath518 to this orbit . it is important to note that although @xmath519 and @xmath520 in general , this group is resynchronized at the next return , @xmath521 . 10 p. antsaklis , w. kohn , a. nerode and s. sastry ( eds ) _ hybrid systems . papers from the @xmath522 workshop held at cornell university , ithaca , new york ( 1994 ) . lecture notes in computer science , 999 . springer , berlin , ( 1995 ) . p. ashwin , riddled basins and coupled dynamical systems . to appear in `` dynamics of coupled map lattices and of related spatially extended systems '' , chazottes and b. fernandez , springer lecture notes in physics ( 2005 ) .

classical be ( or b - emission ) stars are rapidly rotating , hot stars with optical spectra that show hydrogen emission lines , and , frequently , emission lines from singly ionized metals . it has been recognized since the days of @xcite that recombination in a flattened disk of circumstellar gas can reproduce the basic features of the spectroscopically observed h@xmath3 line profiles of be stars . however , it is now generally accepted that this model is too simplistic . the detailed mechanism(s ) that creates and maintains be star disks remains unclear @xcite , although rapid rotation of the central b star certainly plays a role @xcite . in addition to this lack of a successful dynamical model , the evolutionary status of be stars and related stars ( such as b[e ] stars ) is not well understood , and possible evolutionary connections between these groups are ambiguous . these problems are compounded by the additional uncertainty in their rotation rates @xcite and lack of understanding of how rotation and evolution interplay . the suggestion by @xcite that the rotation rates of be stars may be systematically underestimated , combined with new interferometric data , has rekindled interest in finding suitable dynamical models for these stars . until recently , the bulk of the observational evidence for the presence of disks surrounding be stars has been spectroscopic and polarimetric in nature . the spectroscopic evidence includes the profiles of the emission lines observed in the optical and ir spectra , as well as continuum excesses at infrared wavelengths . the net linear polarization observed in be stars arises via electron scattering in the non - spherical distribution of gas , as first suggested by @xcite . however , interferometric observations that spatially resolve the disks in several wavelength regimes are becoming available , ( see for example * ? ? ? * ; * ? ? ? * ; * ? ? ? interferometric observations , in concert with other observables , allow the disk physical properties to be determined with greater accuracy and this may contribute to the development of more successful dynamical models . @xcite collected long - baseline , optical , interferometric observations of the classical be stars @xmath4cas and @xmath5per at the navy prototype optical interferometer ( npoi ) and interpreted these observations using simple models of the intensity distribution on the sky , such as uniform disks , rings , or disks with a gaussian intensity distribution . in this paper , we extend the work of @xcite by computing detailed theoretical models for the intensity distribution on the sky produced by the central b star and disk of a be system . we then compare these models to observations collected at the npoi . the theoretical models are computed with the bedisk code developed by @xcite . assuming a density distribution of the disk gas , the code can compute the temperatures in the disk given the energy input from the central stars photoionizing radiation field , by enforcing radiative equilibrium . as the disk density model contains several adjustable parameters , values for these parameters can then be extracted from the match to observations . we present new observations and detailed models for the be stars @xmath6dra , @xmath7psc , and @xmath8cyg . we also use near - contemporaneous observations of the h@xmath3 spectral lines for these stars as additional constraints on the density model . as we shall demonstrate , the additional spectroscopic constraint of detailed line profiles is often an important ingredient in selecting among models consistent with the interferometric observations . the overall goal of this study is to place tight constraints on physical conditions in the circumstellar regions of these be stars . the theoretical disk models presented in this investigation were constructed using the bedisk code developed by @xcite . a brief overview of the code and assumptions relevant to this work are presented below . the reader is referred to @xcite for more details . the circumstellar disk is assumed to be axi - symmetric about the star s rotation axis and symmetric about the mid - plane of the disk . in this case , cylindrical geometry is appropriate and we use @xmath9 for the radial distance from the star s rotation axis and @xmath10 for the height above the equatorial plane . the radial density distribution in the equatorial plane is assumed to be given by an @xmath11 power - law following the works of @xcite . at each radial distance from the rotation axis , the vertical structure of the disk is determined by the requirement of hydrostatic equilibrium in the @xmath10 direction which balances the @xmath10 component of the star s gravitational acceleration with the gradient of the gas pressure . typical values for the radial power - law index @xmath12 for theoretical models of be stars usually fall in the range of 2 to 3.5 based on models that fit the ir continuum @xcite . models with the highest values of @xmath12 have a faster decrease in the density distribution with increasing @xmath9 . the density distribution within the disk is determined by an assumed value of the density at the stellar surface in the equatorial plane , @xmath13 , by the power - law index , @xmath12 , and by the assumption of pressure support perpendicular to the equatorial plane . since @xmath13 and @xmath12 significantly affect the theoretical predictions , we conduct a 2-dimensional parameter search by varying these input parameters over all reasonable values for comparison with interferometric and spectroscopic observations . given this density model , the bedisk code is used to find the temperature structure of the disk by enforcing radiative equilibrium . the microscopic heating and cooling rates implied by a gas with a solar chemical composition are computed and the temperature that balances heating and cooling is found . to compute the atomic level populations required by this procedure , the statistical equilibrium equations are solved for each chemical element included , accounting for the bound - bound and bound - free collisional and radiative processes that set the rates in and out of each atomic level . to compute the hydrogen line profiles , we solved the transfer equation along a series of rays through the star plus disk system as viewed at an angle @xmath14 relative to the line - of - sight ( where @xmath15 is pole - on , and @xmath16 is equator - on ) . the disk is assumed to be in pure keplerian rotation , and the equatorial rotational velocity of the star was chosen so that the measured @xmath17 of the star was recovered for the adopted value of @xmath14 . for rays terminating on the stellar surface , we adopted a photospheric h@xmath3 line profile computed using the synthe code and the stark broadening routines of @xcite ; these profiles are based on the hydrogen populations computed for the appropriate lte , line - blanketed model atmosphere adopted from @xcite . for rays passing through the disk , the equation of radiative transfer was solved along the ray using the short - characteristics method of @xcite . to compute the h@xmath3 opacity and emissivity , the hydrogen level populations computed for the thermal solution were used . here , hydrogen was represented by a 15 level atom and all implied radiative and collisional bound - bound and bound - free processes from and between the 15 levels were included . the radiative bound - bound rates were treated by the escape probability approximation in which the net radiative bracket for the line @xcite is replaced by a single - flight escape probability . while the escape probability approximation can provide a reasonable description of the line thermalization , the use of escape probabilities is an important approximation . the bedisk code assumes static escape probabilities based on the complete redistribution over a doppler profile for all lines ( for a justification of this procedure see * ? ? ? if the functional form of the escape probability function is changed , for example , to a simple @xmath18 dependence on optical depth for @xmath19 , the predicted equivalent width of h@xmath3 can change by @xmath20 @xmath21 ( due to the change in the thermal structure of the disk and in the hydrogen level populations ) . thus in matching to the observed h@xmath3 profiles , the largest uncertainty is likely from the uncertainty in the theoretical profile and not from the observational uncertainties ( see section 3 for an estimate of the observational uncertainties ) . our approach allows the full dependence of the h@xmath3 line emissivity and opacity on the physical conditions throughout the disk to be retained . the stark broadening routines of @xcite were also used to compute the local h@xmath3 line profile throughout the disk . this is important as these routines can handle high disk densities for which the assumption of a simple gaussian or voigt profile for the h@xmath3 line would be invalid . the final h@xmath3 line profile from the unresolved system was obtained by summing over the rays weighted by their projected area on the sky . the h@xmath3 line profile was computed over a wavelength bin of @xmath22 from line centre at each area element on the sky . to obtain the final h@xmath3 line profile to compare with spectroscopic observations , the profile was convolved with a gaussian of fwhm of @xmath23 to bring the resolving power of the computed profile down to @xmath24 to match the observations . the predicted interferometric visibilities also followed from the same numerical model . the monochromatic h@xmath3 image of the system projected on the plane of the sky was computed by integrating the stellar continuum and h@xmath3 line flux over a 150 @xmath21 spectral window centered at h@xmath3 to represent the interferometric spectral channels ( see section 3 ) . thus the calculation of the h@xmath3 profiles and the h@xmath3 interferometric visibilities used the same model calculations . we obtained long - baseline interferometric and spectroscopic observations of three be stars , @xmath0 dra , @xmath1 psc , and @xmath2 cyg . the interferometric observations were acquired using the navy prototype optical interferometer ( npoi ) utilizing baselines ranging from 18.9 to 64.4 m in length . the npoi is described in detail in @xcite and the technique used to extract interferometric observables from the spectral channel containing the h@xmath3 emission line has been discussed by @xcite . table [ obs_table ] shows all the interferometric observations obtained for the three stars and the number of data points obtained on each night in the spectral channel containing the h@xmath3 line . the interferometric observations of all stars were obtained over a period of less than a month , with @xmath1 psc and @xmath2 cyg observed during 2005 , and @xmath0 dra observed during 2006 . to complement our interferometric observations we have also obtained near contemporaneous spectroscopic observations of all three targets . the spectroscopic observations were obtained using an echelle spectrograph at the lowell observatory s john . s. hall telescope and covered the spectral region around the h@xmath3 line . the properties of the spectrograph and the description of the reduction methods have been described elsewhere ( see 3.2 in * ? ? ? * and references therein ) . the continuum signal - to - noise ratio ( snr ) near h@xmath3 is more than 200 in the spectra of all three stars . however , the equivalent width and the shape of the line is determined with respect to the normalized continuum level . because the line profile must be normalized by a smoothly varying function that is fit to the continuum ( or where one thinks the continuum is located ) , the choice of this normalizing function affects the shape of the h@xmath3 line . in addition , the presence of telluric lines , as well as their temporal variability from night to night ( and season to season ) can affect how well the continuum level is determined ; we expect this to be the dominant source of uncertainty associated with the shape of the observed h@xmath3 line . we have estimated the magnitude of this effect by analyzing a set of reduced spectra all obtained from the same raw spectrum , but all obtained with slightly different quadratic functions fitted to slightly different continuum regions . the variations in the resulting normalized continuum were at the 2 to 3% level , and to be conservative we adopt a 3% uncertainty for our continuum level determination and , in turn , the shape and equivalent width measure of the observed h@xmath3 emission line . for @xmath1 psc and @xmath2 cyg the spectra were obtained on 2005 sep 16 , and for @xmath0 dra the spectrum was obtained on 2006 mar 17 , and therefore in all three cases the spectroscopic observations overlapped the interferometric observing runs ( table [ obs_table ] ) . the equivalent widths of the observed h@xmath3 spectra lines are -15.3 , -24.8 , and -22.2 @xmath21 for @xmath1 psc , @xmath2 cyg , and @xmath0 dra , respectively . ( we adopt the usual convention that a negative in the quoted value of the equivalent width indicates a net line flux above the continuum . ) .observing log for interferometric observations [ obs_table ] [ cols="<,^,^,^ " , ] there is considerable uncertainty in the assigned spectral types for be stars due to rapid rotation and spectral variability . the effects of gravity darkening , and the potential obscuration of portions of the stellar surface by the disk , further compound the uncertainty in classification and the corresponding stellar parameters . since the stellar parameters fix the photoionizing radiation field that is assumed to be the sole source of energy input into the circumstellar disk , realistic stellar parameters are essential to the construction of reasonable disk models . for each of the stars in this study , we searched the literature in order to find the most reasonable stellar parameters available . see table [ stellar_param ] for the adopted stellar parameters for @xmath0 dra , @xmath1 psc , and @xmath2 cyg . the rationale for adopting the particular stellar parameters is explained and compared with other published values in each subsection for a given star . lccccccccr star & hr & @xmath25&spectral & stellar radius & stellar mass & @xmath26 & @xmath17 & reference + name&number & & type & @xmath27 & @xmath28 & k & km / s + @xmath0 dra & hr4787 & 3.5 & b6iiipe & 6.4 & 4.8 & 14000 & 170 & 1 + @xmath1 psc & hr8773 & 4.0 & b6ve & 3.6 & 4.7 & 15500 & 90@xmath2915 & 2 + @xmath2 cyg&hr8146 & 4.0 & b2vne & 4.7 & 6.8 & 19800 & 173@xmath2910&3 + @xmath0 dra is a known spectroscopic binary and giant star of intermediate spectral type . it is variable on a variety of timescales ; see @xcite for a detailed discussion of these variations . @xcite modeled the central star of @xmath0 dra by comparing a variety of spectroscopic and photometric observations to a grid of non - lte models . we have adopted their derived best - fit stellar parameters for our study . we used their parameters of @xmath26 , log@xmath30 , and mass to calculate the stellar radius . @xcite also used these parameters for the interpretation of @xmath31-band observations of @xmath6dra collected at the chara array interferometer . the parameters of @xcite are also in reasonable agreement with stellar parameters listed for the spectral class of b5iii in @xcite . figure [ kd_grid ] shows the density parameter grid for the disk of @xmath6dra . this grid was selected after running a series of models over a much wider range in @xmath32 and @xmath12 on a coarser grid . a comparison of the observed and predicted h@xmath3 equivalent widths allowed us to determine a finer grid to sample for detailed comparisons with interferometric observations . the final density parameter grid consisted of 278 models . we have compared the output from each model to the interferometric observations using the technique described in detail by @xcite . the reduced @xmath33 values corresponding to these models are presented in figure [ kd_chi2 ] as a function of @xmath32 ( g/@xmath34 ) and @xmath12 . the darkest patches in this figure represent the models with reduced @xmath35 . the white plus signs correspond to models with the lowest values ( ranging from 1.18 to 1.29 ) of the interferometric reduced @xmath36 . in total , there are 25 models with reduced @xmath35 for this star , therefore , it is necessary to further constrain our models in order to determine the best - fit model . figure [ halpha_2d ] shows the predicted h@xmath3 equivalent widths as a function of @xmath37 , and the power - law index , @xmath12 , for the entire grid of models . the predicted h@xmath3 profiles were constructed assuming an inclination of @xmath38 . we note that this value of @xmath14 is consistent with other values determined from spectroscopy and polarization analyzes presented in the literature . for example , @xcite , estimates @xmath39 based on the observed polarization angle , and @xcite find @xmath40 based on spectroscopy . we also experimented by varying @xmath14 and found that an inclination angle for this system of @xmath41 seemed to produce profiles that matched the observed h@xmath3 line . we also note that although changing the value of @xmath14 by @xmath29 @xmath42 changes the shape of the h@xmath3 profile , the effect on the equivalent width is @xmath43 . the lightest coloured symbols are closest to the value of the observed equivalent width of -22.2 . there are 14 models that predict an h@xmath3 equivalent width within @xmath44 @xmath21 of the observation and these are represented by white circles . models which correspond to predicted h@xmath3 equivalent widths greater than the observed emission have a plus sign within the corresponding symbol in figure [ halpha_2d ] . we note that this subset of 14 models does not represent the 14 best models using interferometry alone . although this set matches the observed h@xmath3 equivalent width within @xmath29 2 , using interferometric observations as a constraint this same set corresponds to reduced @xmath33 from 1.18 ( the minimum value ) up to 16.0 . in fact , 6 of the 14 models in this set had a reduced @xmath33 greater than 2.0 . it is clear from a comparison of figure [ kd_chi2 ] and figure [ halpha_2d ] that the results from h@xmath3 interferometry and h@xmath3 line profile are complimentary but that each provides extra information to help reduce the degeneracy in finding the best - fit model . multiple models , of pairs of @xmath12 and @xmath32 , that produce h@xmath3 profiles with equivalent widths that match the observations within uncertainties can be found . however , not all of these are the best - fits in terms of interferometry . that is , it is possible to match the equivalent width for a range of @xmath12 by an appropriate adjustment in @xmath32 , but that these combinations do not necessary have the correct density distribution as a function of radial distance from the star . however , by combining results from spectroscopy and interferometry , we can find the best pairs of @xmath12 and @xmath32 for @xmath0 dra . next , we compared the subset of 8 models which represent the best - fits from both interferometry and h@xmath3 equivalent width . a subset ( 4 models with the lowest reduced @xmath33 from interferometry ) of the corresponding h@xmath3 profiles is shown in figure [ halpha_dra ] . notice the variation in the shape of the profile as @xmath12 is increased from 2.7 in the left - most panel to 4.2 in the right - most panel . the predicted equivalent widths are -21.9 , -23.5 , -22.4 , and -20.3 , with the corresponding @xmath33 values from interferometry of 1.18 , 1.29 , 1.18 , and 1.18 . the model with @xmath45 does not have the the correct line profile shape compared to the observed profile . there is too much absorption at line centre , and the profile is clearly too narrow in the wings . all of the predicted profiles in figure [ halpha_dra ] have an absorption feature at line centre . this feature is the most pronounced for low values of @xmath12 and gradually decreases for models with higher values of @xmath12 . if we increase the inclination , this absorption feature becomes more and more pronounced as the system is viewed closer and closer to edge on . if we decrease @xmath14 , so the system is viewed close to pole - on , the absorption feature remains for the models with the lowest values of @xmath12 . in addition , with @xmath45 , the wings of the predicted profile are not broad enough for any reasonable values of @xmath14 . the panel with the model corresponding to @xmath46 and @xmath47 in figure [ halpha_dra ] shows that the predicted line is too strong and this results in an h@xmath3 equivalent width which is too large . models corresponding to @xmath48 have profiles with wings wider than the observation . the model that corresponds to the profile with @xmath49 matches the shape of the observed profile best and has a minimum reduced @xmath33 of 1.18 from interferometry . we prefer this model with @xmath49 and @xmath50 g/@xmath34 in terms of both spectroscopy and interferometry . figure [ fig_dra_fits ] shows the interferometric observations compared with this model for @xmath0 dra . we note that as previously discussed the uncertainty due to the approximations in the computational code ( see section 2 ) are larger than those in the observed h@xmath3 profiles ( see section 3 ) . we searched the literature to find the best possible stellar parameters for the disk systems studied in this investigation ( see section 4.1 , 4.2 , 4.3 ) . however , generally for be stars , not only are the stellar parameters not well established but for some stars , even the spectral type may not be accurate . we now wish to test the dependence of model parameters on our results . we experimented by varying the stellar parameters in turn , to see how these changes affect the predicted h@xmath3 equivalent widths and the interferometric fits for the star @xmath0 dra . similar types of uncertainties would be expected for the other two systems presented in this paper . one might expect that the @xmath26 of the central star , which supplies the disk with energy , may be the source of the greatest uncertainty . we changed the @xmath26 by @xmath29 1000 @xmath51 and reproduced all of the profiles . this resulted in a change in the equivalent width of the h@xmath3 line by a maximum of @xmath52 2 . to assess the effect of uncertainties in the fundamental stellar parameters combined , we performed a monte carlo simulation in which new stellar parameters were randomly realized . these parameters were used to compute a new circumstellar disk model , corresponding to our preferred model of @xmath53 and @xmath54 . this procedure was repeated to generate new h@xmath3 profiles and equivalent widths . errors assumed for the stellar parameters were @xmath55% in mass , @xmath55% in radius , and @xmath56% in @xmath26 ( corresponding to @xmath57k ) ; each stellar parameter was assumed equally probable within its errors . using a gaussian distribution for the errors , we found our results did not change significantly . the mean of h@xmath3 equivalent width was @xmath58 with a standard deviation of @xmath59 . we have one star in common with the work of @xcite : @xmath60dra . they use a disk density model which is very similar to ours and derive two sets of parameters @xmath61 based on fits to k-band interferometric visibilities : @xmath62 , @xmath63 for a single star model , and @xmath64 , @xmath65 for a binary model . @xcite prefer the binary fit to the single star model because of the peculiar solution they obtain in the single star case ; namely a very small exponent in the radial density power law ( @xmath66 ) and a best solution with @xmath67 which is at odds with the resolved h@xmath3 line profile . nevertheless , the particular properties of the introduced binary are not constrained by any independent observations . both of these solutions seem at odds with our results . it is well documented that the h@xmath3 equivalent width and the infrared excess are correlated ( see for example * ? ? ? @xcite present a figure which shows the change in h@xmath3 equivalent width from the early 1970 s to 2004 for @xmath0 dra ( see @xcite , figure 4(d ) ) . although there is certainly scatter of data values in their figure , one can easily see from this figure that the h@xmath3 equivalent width varies by @xmath52 15 @xmath21 ( from about - 5 to about -20 ) in a period of @xmath52 6 years . this is a substantial change in the h@xmath3 emission line and we expect there would be a corresponding increase in the infrared excess . the interferometric observations for @xmath0 dra presented in @xcite were obtained in april ( 3 dates ) and december ( 2 dates ) of 2005 . their spectroscopic observations were obtained in april 2005 . our contemporaneous observations were obtained in 2006 . the difference in our results and @xcite is partially accounted for by the fact that we model this variable star at different times . this further supports the requirement of contemporaneous observations . other possible causes for the difference in results could be due the fact that the h@xmath3 observations are sensitive to different regions within the disk as well as the different spatial resolution of the two interferometers . there are also other differences between our calculations and @xcite that should be highlighted : as noted in section 2 , the calculation of the h@xmath3 visibilities and the h@xmath3 line profiles in our work are obtained from essentially the same calculation that determined the radiative equilibrium temperature structure for the disk given the adopted density model . the h@xmath3 opacity and emissivity at each point in the disk were computed from the local temperature , pressure , and radiative field by solving the set of statistical equilibrium equations for a 15-level hydrogen atom . while our fits to the observed h@xmath3 line profiles are not perfect , they are predicted naturally from the models and the assumption of pure keplerian rotation by the disk . to fit the h@xmath3 , h@xmath68 , and br@xmath3 lines , @xcite found it necessary to convolve their profiles with a unit normalized lorentzian to broaden the line wings ; they discuss numerous possible sources for this extra broadening . in the current work , we convolved our line profiles only with a unit normalized gaussian of fwhm of @xmath69 to bring the resolving power of the computed profile to @xmath24 to match the observations . among the reasons cited by @xcite as to possible broad h@xmath3 wings , they discuss the possibility of stark broadening for the hydrogen lines , but then note that the densities they find are too low for stark broadening to be important . this is consistent with their best fit densities which are about 1 - 2 orders of magnitudes smaller than those derived here . previous studies have shown that self - consistent disk thermal structures are crucial in order to interpret observations correctly ( see for example , * ? ? ? * ; * ? ? ? we computed the thermal structure for @xmath0 dra with our bedisk code using the model parameters , @xmath64 , @xmath65 , r@xmath70 , listed in @xcite . we find substantial variation in disk temperature near the star and equatorial plane , and an overall average disk temperature of @xmath52 13000 k. @xcite adopt an isothermal disk temperature of @xmath52 9300 k for @xmath0 dra based on previous studies @xcite , but we note that this study as well as @xcite do not investigate the disk densities as low as the @xcite best - fit model . finally , we note that the value , @xmath49 , we obtain for @xmath0 dra is in agreement with typical values previously predicted in the literature for be star disks ( see for example * ? ? ? * ; * ? ? ? @xmath1 psc is a b6ve star that has long been known to exhibit emission in the hydrogen lines ( see for example * ? ? ? this star is similar in spectral type to @xmath0 dra , but it is a main sequence dwarf . we have included @xmath1 psc in this investigation because comparing the disk models with the interferometric observations lead to unique density parameters , and it is useful to compare the results for this star with @xmath0 dra . we adopt the stellar parameters from @xcite and note that these parameters are also consistent with interpolated values in @xcite . the interferometric observations are from 12 nights between 28 aug . 2005 and 26 sept . 2005 ( see table [ obs_table ] ) , and the h@xmath3 spectroscopy for @xmath1 psc ( and @xmath2 cyg ) were obtained on 16 sept . 2005 and therefore are definitely representative of the h@xmath3 line during the period of the interferometric observations . we note that a subsequent observation of the h@xmath3 line for @xmath1 psc on 3 oct . 2006 had a 20% weaker equivalent width . the procedure for selecting the grid of models to be compared in detail with interferometric observations was obtained by the same selection process as done for @xmath0 dra . the final parameters to be searched covered the range @xmath71 to @xmath72 and @xmath73 g/@xmath34 to @xmath74 g/@xmath34 . in this range , 193 models were constructed for comparison with interferometric observations . the corresponding @xmath33 values had a minimum value of 0.83 corresponding to the model with @xmath75 and @xmath76 g/@xmath34 and a maximum value of 438 corresponding the model with @xmath77 and @xmath78 g/@xmath34 . the @xmath33 values revealed only one other model , @xmath79 and @xmath80 , which was statistically the same as the best - fit . for this model @xmath81 . in order to check the predicted h@xmath3 profiles , we had to adopt a value for @xmath14 . the observed sharp emission line profile ( shown in figure [ halpha_psc ] ) suggests a small value for @xmath14 . however , this star has a @xmath82 of 90@xmath83 km / s ( see table [ stellar_param ] ) , so for very small values of @xmath14 , the star will exceed its critical velocity . the estimated critical velocity of a star of this spectral type is @xmath84 km / s @xcite . therefore , we argue that the inclination of this system must be at least 15@xmath83 , and we have adopted a value of @xmath85 . we calculated the h@xmath3 equivalent widths for a subset ( @xmath86 models ) chosen by the best - fit @xmath33 values , and we considered @xmath33 values up to @xmath87 . statistically , the two models corresponding to @xmath33 values of 0.83 , and 0.84 are equally as good while the other models in the subset with larger values of @xmath33 are poorer fits to the interferometric data . nevertheless , to be complete we computed profiles for every model in the subset . the equivalent widths for the two models corresponding to the minimum @xmath33 values had reasonable line shapes compared to the observed profile but the model corresponding to @xmath88 had an equivalent width of -25.0 @xmath89 while the model corresponding to @xmath90 had an equivalent width of -21.6 @xmath89 which is closer to the value for the observed spectra of -15.3 . figure [ halpha_psc ] shows the h@xmath3 profile for the best - fit model ( @xmath91 , @xmath92 g/@xmath34 ) in the left - most panel in terms of interferometric observations and spectroscopy . models corresponding to other values of @xmath12 did not produce h@xmath3 profiles that agreed with observations . figure [ halpha_psc ] shows the behavior of the h@xmath3 profile with increasing values of @xmath12 . the reader can clearly see that the wings of the line in the middle and right - most panels are too wide , resulting in predicted emission which is too large . these profiles correspond to models with @xmath93 , @xmath94 g/@xmath34 , equivalent width of -30.4 , and @xmath95 , @xmath96 g/@xmath34 , equivalent width of -23.4 , and have interferometric reduced @xmath33 values of 1.13 , and 1.21 , respectively . models with values of @xmath97 had large @xmath33 values based on the interferometric results , and were eliminated . figure [ fits_psc ] shows the interferometric observations for our best fit model with @xmath91 , and @xmath98 g/@xmath34 for @xmath1 psc . @xmath2 cyg has long been recognized as a be star with strong h@xmath3 emission @xcite . although several authors have noted that this emission stays relatively constant over periods as long as decades ( see , for example , * ? ? ? * ) others have noted periodic outbursts in h@xmath3 emission , typical of many be stars @xcite . the interferometric observations for @xmath2 cygnus are from 12 nights between 28 aug . 2005 and 26 sept . 2005 , and the h@xmath3 line profile was acquired on 16 sept . 2005 ( see table [ obs_table ] ) . we note that we also have observations from 17 oct . 2005 that confirm the spectroscopic stability of the star . @xcite suggested that the outbursts , which occur on periods of @xmath52 years , are the result of multi - periodic , non - radial pulsations . @xcite derived the stellar parameters for this star by detailed modeling constrained by photometric and spectroscopic data . three of their models include veiling effects , and gravitational darkening for a range of stellar rotational velocities , 0.80 , 0.90 , and 0.95 of the critical velocity . ( see @xcite for more detail . ) we have chosen to adopt the stellar parameters from their model d which has an intermediate value of the rotational velocity ( that is , 0.90 of critical velocity ) . also , @xcite finds a value for the inclination angle of @xmath99 . the stellar radius listed in table [ stellar_param ] corresponds to the equatorial radius given by @xcite . this is a reasonable choice since the star is nearly pole - on . the stellar parameters adopted for @xmath2 cyg are reasonable compared to other values presented in the literature of this spectral type ( see for example * ? ? ? the procedure for selecting the grid of models to be compared with interferometric observations was obtained by the same selection process as done for the two other stars in this investigation . the final selection of parameters to be searched ranged from @xmath100 to @xmath101 , and @xmath102 g/@xmath34 to @xmath103g/@xmath34 , giving a total of 193 models constructed for comparison with interferometric observations . the corresponding @xmath33 values had a minimum value of 1.14 corresponding to the model with @xmath47 and @xmath104 g/@xmath34 and a maximum value of 471 corresponding the model with @xmath105 and @xmath106 g/@xmath34 . the @xmath33 values revealed several models that statistically fit the observations equally well . again , we turn to spectroscopy to remove the degeneracy . in order to predict h@xmath3 profiles we adopt @xmath107 which is consistent with @xcite . we calculate the h@xmath3 equivalent widths for a subset ( @xmath108 models ) chosen based on the interferometric reduced @xmath109 . statistically , the models with larger values of @xmath33 are poorer fits to the interferometric data . the equivalent widths for the subset ranged from a minimum of -19.0 @xmath89 to a maximum of -43.1 . the observed h@xmath3 line has an equivalent width of -24.8 , and there were three models which had reasonable line shapes and equivalent widths that agreed within @xmath110 @xmath21 and had statistically similar interferometric @xmath33 values . figure [ halpha_cyg ] shows the h@xmath3 profiles corresponding to these models . table [ cyg_best ] lists the values of @xmath12 , @xmath32 , h@xmath3 equivalent width , and the reduced @xmath33 values based on comparisons from interferometry for these 3 models . the models with @xmath111 and @xmath112 in figure [ halpha_cyg ] have predicted profiles that are too narrow and also do not fit the peak of the observed profile . since the @xmath33 values are statistically similar , we adopt the model corresponding to @xmath113 and @xmath104 g/@xmath34 in figure [ halpha_cyg ] as the best - fit . as a further illustration , figure [ halpha_cyg_bad ] shows h@xmath3 profiles corresponding to models with smaller and larger values of @xmath12 than those given in table [ cyg_best ] . specifically , these profiles correspond to models with @xmath114 , @xmath115 g/@xmath34 and @xmath116 , @xmath117 g/@xmath34 with reduced @xmath33 values of 1.37 , and 1.29 , respectively . the shapes of these profiles do not match the observed line . the predicted profile corresponding to @xmath114 has too much emission ( with an equivalent width of -33.7 ) and is too wide in the line wings . the predicted profile corresponding to @xmath116 , has too little emission ( with an equivalent width of -20.3 ) and much of the profile is narrower than the observation . figure [ cyg_fits_interfer ] shows a comparison of the interferometric observations and the best - fit model for @xmath2 cyg . lccr @xmath12 & @xmath32 & h@xmath3 equivalent & @xmath33 + & g/@xmath34&width [ ] & + 4.0 & @xmath118&-26.0&1.14 + 4.1&@xmath119&-24.4&1.16 + 4.2&@xmath120&-23.7&1.30 + in this work , disk density models were obtained for the classical be stars @xmath6dra , @xmath7psc , and @xmath8cyg by matching the observed interferometric h@xmath3 visibilities with fourier transforms of synthetic images produced by theoretical models . it was demonstrated that the additional constraint obtained by the ( co - temporal ) observed h@xmath3 line profile was critical to selecting a density model . this was particularly evident in the case of @xmath0 dra . we have demonstrated that in the case of these 3 stars , only by additional observational constraints can the degeneracy in the interferometric best - fitting models be removed . the best fit models to both the interferometric observations and the observed h@xmath3 line profiles for the three stars are summarized in table [ summary ] . the range of @xmath12 is consistent with previous determinations based on other diagnostics ( such as ir excess see we note that the range of base " densities , @xmath32 , varies by over two orders of magnitude between the three stars . the navy prototype optical interferometer is a joint project of the naval research laboratory and the us naval observatory , in cooperation with lowell observatory , and is funded by the office of naval research and the oceanographer of the navy . we thank the lowell observatory for the telescope time used to obtain the h@xmath3 line spectra presented in this paper . this research was supported in part by nserc , the natural sciences and engineering research council of canada . we thank l. thomson for contributions to this work as an nserc usra in 2006 . _ facilities : _ , armstrong , j. t. , mozurkewich , d. , rickard , l. j. , hutter , d. j. , benson , j. a. , bowers , p. f. , elias , n. m. ii , hummel , c. a. , johnston , k. j. , buscher , d. f. , and 5 coauthors 1998 , , 496 , 550 ballereau , d. , chauville , j. , & zorec , j. 1995 , , 111 , 423 barklem , p. s. , & piskunov , n. e. 2000 in _ model ling of stellar atmospheres _ , iau symposium 210 , n. e. piskunov , w. w. weiss , & d. f. gray eds . , p.e28 campbell , w. w. 1895 , , 2 , 177 carciofi , a. c. , & bjorkman , j. e. 2006 , , 639 , 1081 chesneau , o. , meilland , a. , rivinius , t. , stee , p. , jankov , s. , domiciano de souza , a. , graser , u. , herbst , t. , janot - pacheco , e. , koehler , r. , & 5 co - authors . 2005 , , 435 , 275 clarke , d. 1990 , , 227 , 151 cot , j. , & waters , l. b. f. m. 1987 , , 176 , 93 cox , a. n.(editor ) 2000 , allen s astrophysical quantities , fourth edition , springer - verlag , new york coyne , g. v. , & kruszewski , a. 1969 , , 74 , 528 cranmer , s. r. 2005 , , 634 , 585 dougherty , s. m. , waters , l. b. f. m. , burki , g. , cote , j. , cramer , n. , van kerkwijk , m. h. , & taylor , a. r. 1994 , , 290 , 609 fleming , m. 1891 , astronomische nachrichten , 128 , 403 gies , d. r. , bagnuolo , w. g. , jr . , baines , e. k. , ten brummelaar , t. a. , farrington , c. d. , goldfinger , p. j. , grundstrom , e. d. , huang , w. , mcalister , h. a. , mrand , a. , & 15 coauthors 2007 , , 654 , 527 hirata , r. 1995 , , 47 , 195 juza , k. , harmanec , p. , hill , g. m. , tarasov , a. e. , & matthews , j. m. 1991 , bull . czechoslovakia , 42 , 39 kurucz , r. f. 1993 , kurucz cd - rom no.13 . cambridge , mass : smithsonian astrophysical observatory levenhagen , r. s. , & leister , n. v. 2004 , 127 , 1176 mcalister , h. a. , brummelaar , t. a. , gies , d. r. , huang , w. , bagnuolo , w. g. , jr . , shure , m. a. , sturmann , j. , sturmann , l. , turner , n. h. , taylor , s. f. , & 6 co - authors . 2005 , , 628 , 439 merrill , p. w. , humason , m. l. , & burwell , c. g. 1925 , , 61 , 389 mihalas , d. 1978 , stellar atmospheres ( san francisco : w. h. freeman ) neiner , c. , floquet , m. , hubert , a. m. , frmat , y. , hirata , r. , masuda , s. , gies , d. , buil , c , ; martayan , c. 2005 , , 437 , 257 nuclei , ( mill valley : univ . books ) owocki , s. p. 2003 , _ rotation and mass ejection : the launching of be - star disks _ in maeder a. , eenens , p. , eds , iau symposium 215 , stellar rotation , 1 olson , g. l. , & kunasz , p. b. 1987 , j. quant . transfer , 38 , 325 peters , g. j. 1979 , , 39 , 175 porter , j. m. 1996 , mnras , 280 , 31 porter , j. m. , & rivinius , t. 2003 , , 115 , 1153 saad , s. m. , kubt , j. , koubsk , p. , harmanec , p. , koda , p. , korkov , d. , krtika , j. , lechta , m. , boi , h. , ak , h. , hadrava , p. , votruba , v. 2004 , , 419 , 607 sigut , t. a. a. , & jones , c. e. 2007 , 668,481 struve , o. 1931 , , 73 , 94 townsend , r. h. d. , owocki , s. p. , & howarth , i. d. 2004 , , 350 , 189 tycner , c. , hajian , a. r. , mozurkewich , d. , armstrong , j. t. , benson , j. a. , gilbreath , g. c. , hutter , d. j. , pauls , t. a. , & lester , j. b. 2003 , , 125 , 3378 tycner , christopher , gilbreath , g. c. , zavala , r. t. , armstrong , j. t. , benson , j. a. , hajian , arsen r. , hutter , d. j. , jones , c. e. , pauls , t. a. , & white , n. m. 2006 , , 131 , 2710 tycner , christopher , jones , c. e. , sigut , t. a. a. s. , schmitt , h. r. , benson , j. a. , hutter , d. j. , & zavala , r. t. 2008 , , submitted van kerkwijk , m. h. , waters , l. b. f. m. , & marlborough , j. m. 1995 , , 300 , 259 waters , l. b. f. m. 1986 , 162 , 121 waters , l. b. f. m. , cot , j. , & lamers , h. j. g. l. m. 1987 , , 185 , 206

based on their broad - band spectral and variability properties , six of the persistently bright neutron star low - mass x - ray binaries ( lmxbs ) were classified as z sources @xcite , after the z - like tracks they trace out in x - ray color - color ( cds ) and hardness - intensity diagrams ( hids ) . these sources are gx 17 + 2 , cyg x-2 , gx 5 - 1 , gx 340 + 0 , sco x-1 , and gx 349 + 2 . the z - like tracks consist of three branches , which , from top to bottom , are referred to as the horizontal branch ( hb ) , the normal branch ( nb ) , and the flaring branch ( fb ) . it is generally believed that the parameter that determines the position along the z track is the mass accretion rate , increasing from the hb to the fb . note that the definition of the mass accretion rate in this respect is rather vague it is often defined as the mass accretion rate through the inner disk and onto the neutron star surface . in addition to spectral changes along the z - track , some of the z sources show long term changes in the shape and position of the z - track in the cd and hid . these secular changes , as they are referred to , are clearly observed in cyg x-2 @xcite , gx 5 - 1 @xcite , and gx 340 + 0 @xcite ( the cyg - like sources ; the other z sources are referred to as the sco - like sources ) , and more recently also in gx 17 + 2 @xcite . it has been suggested that they are related to the relatively high inclination at which these sources are seen @xcite , or to different magnetic field strengths of the neutron stars @xcite . the power spectra of the z sources show several types of quasi - periodic oscillations ( qpos ) and noise components ( see * ? ? ? * for a review ) . it was found that their presence and properties are very well correlated with the position of the source along the z track @xcite , even when the z - tracks show secular changes ( e.g. * ? ? ? three types of low frequency ( @xmath1100 hz ) qpos are seen in the z sources : the horizontal branch ( hbos ) , normal branch ( nbos ) and flaring branch qpos ( fbos ) . their names derive from the branches on which they were originally found . the hbo is found on the hb and nb with a frequency ( 1560 hz ) that gradually increases along the hb towards the nb . when the sources move from the hb onto the nb the frequency increase flattens off . in gx 17 + 2 and cyg x-2 it was found that when the source passes a certain point on the nb , the hbo frequency starts to decrease @xcite . the nbo and fbo are most likely the same phenomenon . they are found on the nb and fb ( near the nb / fb vertex ) but not on the hb . on the nb the qpo has a frequency of @xmath057 hz , which rapidly increases to @xmath020 hz when the source moves across the nb / fb vertex @xcite . in recent years two high frequency ( or khz ) qpo were found in the z sources ( @xcite , see @xcite , for a review ) . they are often observed simultaneously , with a frequency difference of @xmath0300 hz , and have frequencies between 215 hz and 1130 hz , which increase from the hb to the nb . in sco x-1 the frequency difference was found to decrease with increasing qpo frequency @xcite . the frequency difference in the other z sources is both consistent with the behavior seen in sco x-1 and with being constant @xcite . three types of noise are seen in the z sources . they are the very low frequency noise ( vlfn ) , the low frequency noise ( lfn ) , and the high frequency noise ( hfn ) . the vlfn and hfn are found on all branches , whereas the lfn is only observed on the hb and nb . the vlfn , which is found at frequencies below 1 hz , can be described by a power law . the hfn and lfn are both band limited components , with cutoff frequencies of , respectively , 10100 hz , and 210 hz . many competing models have been proposed for the origin of the qpos and noise components . it is beyond the scope of this introduction to mention these models in detail - most of them will be discussed in section [ gx17 + 2_discuss_sec ] . in this paper we present a study , based on data acquired with the _ rossi x - ray timing explorer _ ( _ rxte _ ) , of the correlated spectral and variability properties of the z source gx 17 + 2 . it is a continuation of the work by @xcite . the current paper constitutes the first report on the very large observing campaign of 1999 , which more than doubled the total coverage of the source . this campaign was undertaken with the express purpose of investigating if a non - monotonic relation exists between the frequency of the khz qpos and the hbo in gx 17 + 2 . section [ 17 + 2_obs_sec ] deals with the observations and analysis . the spectral results are presented in section [ spectral_sec ] and the results for each power spectral component in sections [ timing_sec ] and [ sec : res - khz ] . a number of qualitatively new results is found in our greatly expanded data set . in particular , we find that when the upper khz qpo frequency is below 1030 hz , it correlates with the hbo frequency , whereas above 1030 hz they anti - correlate . we also find that khz qpo frequency difference is not constant and that the q values of the hbo and khz qpos are probably not explained by life time broadening . these and other results are compared to observations of other z sources and discussed in terms of current models in section [ gx17 + 2_discuss_sec ] . finally , we argue that changes in the spectral and variability properties of the z sources may not be not driven by changes in the mass accretion rate . the data used for the analysis in this paper were all obtained with the proportional counter array ( pca , @xcite ) on board _ rxte _ @xcite . the pca consists of five xenon - filled proportional counter units ( pcus ) , each with an effective area of @xmath01250 @xmath2 ( at 10 kev ) . although the five pcus are in principle identical they all have a slightly different energy response . these responses change continuously due to slow processes such as gas leakage and the aging of the electrodes . in addition , the high voltage settings of the instruments are occasionally altered ( gain changes ) , resulting in rather more drastic changes in the detector response . these changes have , at the time of writing , been applied three times during the life time of _ rxte _ thereby defining four gain epochs . occasionally one or more pcus are not operational . they can be switched off by an internal safety mechanism , or by the ground control crew , for reasons of detector preservation . therefore , the number of active detectors varies between the observations . all our _ rxte_/pca observations of gx17 + 2 were done between 1997 february 2 and 2000 march 31 . a log of the observations is given in table [ obs_tab ] . we do not include the observations done in february 1996 , which were used by @xcite . the reasons for this are the limited time resolution in the energy range of interest and difficulties with scaling the @xmath3 ( see below ) due to an incomplete z track . data taken during satellite slews and earth occultations were removed . the nine type i x - ray bursts that were observed are the subject of a separate article by @xcite ; two of those were also studied in this paper . the total amount of good data that remained was @xmath0600 ks . data were collected in several modes with different time and spectral resolutions . two of these modes , ` standard 1 ' and ` standard 2 ' , were always operational . in addition to these two modes , other modes were active that varied between the observations . the properties of all modes are given in table [ gx17 + 2_modes_tab ] . the standard 2 data were used to perform a broad - band spectral analysis . the data were background subtracted , but no dead time corrections were applied ; these were in the order of 25% . for each 16 s data segment ( i.e. the intrinsic resolution of the standard 2 mode ) we defined two colors , which are the ratios of count rates in two different energy bands , and an intensity , which is simply the count rate in one energy band . the energy bands used for the colors ( soft and hard color ) and intensity , are given in table [ colors_tab ] . the lower energy boundaries for the soft color and intensity were chosen relatively high in order to avoid the lowest and least reliable energy channels . by plotting the two colors against each other a color - color diagram ( cd ) was produced . a hardness - intensity diagram ( hid ) was produced by plotting the hard color against the intensity . to produce the cds and hids we only used data obtained with those pcus that ( for each gain epoch ) were always on . for gain epoch 3 these were pcus 0 , 1 and 2 , and for gain epoch 4 pcus 0 and 2 . due to the different numbers of detectors and the differences in the detector settings we decided to produce the cds and hids separately for the two gain epochs . however , in choosing the energy channels we tried to take channels whose energy boundaries were as close as possible . due to the aging processes mentioned earlier , observations with the same source spectrum that are made more than a few weeks apart end up at a different location in the cd . to correct for this effect we analyzed a number of _ rxte_/pca observations of the crab pulsar / nebula ( which is assumed to have a constant spectrum , see also @xcite ) that were taken around the time of our gx 17 + 2 observations . for all crab observations we produced the colors in the same energy bands as we used for gx 17 + 2 . we found that the observed colors of the crab indeed changed . for each observation we calculated multiplicative scaling factors , for both colors , with respect to those of the first crab observation . we used these factors to scale the colors of the gx 17 + 2 observations back to those of the first gx 17 + 2 observation . this procedure could only be applied to the epoch 3 observations , where , as expected , we found it to result in a narrower track in the cd . no corrections were applied for the intensity . since no crab observations were available for the march 2000 observations , no corrections could be applied to epoch 4 data . the shifts in that data set appeared to be small anyway . our power spectral analysis was based on selecting observations as a function of the position along the z track in the cd . we used a method that is based on the rank number and @xmath3 parameterization methods introduced by @xcite and @xcite , that has been gradually refined in similar studies @xcite . in this method in its current form , all points in the cd are projected onto a bicubic spline ( see * ? ? ? * ) whose normal points are placed by hand in the middle of the z track ( see figure [ cd - hid_fig ] ) . the hb / nb and nb / fb vertices are given the values @xmath3=1 and @xmath3=2 , respectively . the rest of the z track is scaled according to the length of the nb . problems arise when sharp curves are present in the track , in our study most notably around the nb / fb vertex . due to scatter , points which have fb properties end up on the nb and vice versa . this is not a problem that is only intrinsic to the @xmath3 parameterization ; it is a limitation that applies to all selection methods based on colors . no observable parameter , apart from the power spectra , has been identified that could be used to distinguish nb and fb observations around the vertex better than the x - ray colors . in order to improve on the results of @xcite we wished to combine the epoch 3 and epoch 4 data sets . unfortunately , the tracks traced out in the cds of epoch 3 and 4 are not the same and two different splines had to be used for the @xmath3 parameterization . since the normal points for these splines were drawn by hand , the @xmath3 scales for the epoch 3 and epoch 4 cds were unlikely to be exactly the same . therefore , we first transformed the @xmath3 scale of epoch 4 to that of epoch 3 . to accomplish this , we measured the frequency of either the hbo or nbo at several places along the z track of epoch 3 , and determined the @xmath3 interval corresponding to the same frequency in the epoch 4 data . we found that @xmath4 , showing that , although the scales are the same ( within the errors ) , a small shift is present . we subsequently scaled the @xmath3 values of epoch 4 using the above expression . the above scaling method assumes that the hbo and nbo frequencies are strongly related to @xmath3 . previous studies of z sources ( see e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , as well as the fact that only little scatter was present around the linear relation seem to confirm this . power spectra were created from the data in modes with high time resolution ( @xmath5 ; see table [ gx17 + 2_modes_tab ] ) using standard fast fourier transform techniques ( see * ? ? ? * and references therein ) . the data were not background subtracted and no dead time corrections were applied prior to the fourier transformations . we made power spectra in several energy bands , with several frequency resolutions and nyquist frequencies . we finally settled on 0.06254096 hz power spectra in the 5.160 kev ( epoch 3 ) and 5.860 kev ( epoch 4 ) bands , since the qpos were most significantly detected in those bands ( note that the source does actually not contribute significantly to the power spectra above @xmath040 kev , however , our data modes did not allow us to discard data above this energy ) . this choice of frequency range means that the properties of the vlfn , which dominates the power spectrum below 0.1 hz , could not always be measured satisfactorily , but it does allow us to follow the power spectral evolution on time scales down to 16 s. note that the power spectra of epoch 4 were produced in a slightly higher energy band than those of epoch 3 . due to the gain changes and the limited choice of energy channels this was the best possible match . the power spectra were selected on the basis of @xmath3 , as determined from the cd . the @xmath3 selections usually had a width of 0.1 and did not overlap - no power spectrum was represented in more than one @xmath3-selection . different widths were used in cases where the power spectrum changed rapidly as a function of @xmath3 ( narrower selections ) or when the powers were weak ( wider selections ) . all the power spectra in a selection were averaged , and the resulting power spectrum was rms normalized according to a procedure described in @xcite . the properties of the power spectra were quantified by fitting functional forms to them . the low frequency ( 0.0625256 hz ) and high frequency ( 1004096 hz ) parts of the power spectrum were fitted separately . the high frequency part was fitted with one or two lorentzians ( for the khz qpo(s ) ) and with the function @xmath6 for the dead time modified poisson level @xcite ; no separate term was used for the contribution of the very large events ( vle ) count rate since it was absorbed by this fit function . at low frequencies the noise that was present could not be fitted consistently , and varying combinations of functionals had to be used : * @xmath3@xmath10.0 : a lorentzian ( lfn+vlfn ) * @xmath3=0.00.1 : a cut - off power law ( lfn+vlfn ) * @xmath3=0.11.4 : a power law ( vlfn ) and a cut - off power law ( lfn ) * @xmath3=1.41.6 : a power law ( vlfn ) and a lorentzian ( lfn+nbo ) * @xmath3=1.65.0 : a power law ( vlfn ) and a lorentzian ( lfn ) * @xmath3@xmath75.0 : a power law ( vlfn ) the expression for a power law is @xmath8 , that for a cut - off power law is @xmath9 ( where @xmath10 is the cut - off frequency ) , and for a lorentzian @xmath11 $ ] ( where @xmath12 is the centroid frequency and @xmath13 is the full - width - at - half - maximum ) . the hbo , its second harmonic and the nbo / fbo were each fitted with a lorentzian . below the hbo , at about half its frequency , a broad bump was present that was also fitted with a lorentzian , whose frequency was sometimes fixed to zero . at low frequencies the dead time modified poisson level was fitted with a constant . an example of a fit to a power spectrum ( @xmath3=0.40.5 ) , displaying the contributing of the several components , is shown in figure [ pds_comp_fig ] . note that changes in the fit function , such as using a cut - off power law instead of a lorentzian or adding an extra component , may lead to minor changes in the values of other parameters . errors on the fit parameters were determined using @xmath14 . upper limits were determined by fixing some or all of the parameters of a component , except the rms amplitude , to values similar to those obtained in the closest @xmath3 selection where it was found to be significant , leaving all other fit parameters free , and using @xmath15 ( 95% confidence ) . a study of the energy dependence and time lag properties of the qpos and noise components will be presented elsewhere . the cds and hids of both epochs are shown in figure [ cd - hid_fig ] . in both cds the hb / nb vertex is not well defined , since the hb is almost a continuation of the nb . this is mainly due to the relatively high energies that we chose for our soft color ( see section [ 17 + 2_obs_sec ] ) . at lower energies the turn - over is much clearer . in figure [ sz - count_fig ] we show the count rate in several energy bands as a function of @xmath3 , for epoch 4 . it shows that the hb / nb vertex in the hid is entirely due to the count rates at low energies . at high energies the hb is a perfect continuation of the nb , and no vertex is present . in the hid of epoch 3 the z track appears to be segmented , which was already noted by @xcite . although no corrections for the slow detector aging processes were applied to the intensity in this hid , we note that a quick look showed that they would only have made the segmentation more apparent . the shifts in the hid might be due to the secular motion that has been observed in other z sources . the shifts , of up to @xmath05% in intensity , do not show up in the cd . this is because colors are ratios of intensities and are therefore not very sensitive to overall intensity changes ; exactly for this reason cds are preferable over hids for our purpose . moreover , the width of the tracks in the cd is about 5% , so changes smaller than this are hard to observe . the bottom panel of figure [ hist - vz_fig ] shows the distribution of the time spent by the source in each part of the z track . the source spent 28% of the time on the hb ( @xmath3@xmath161 ) , 44.2% on the nb ( 1@xmath1@xmath3@xmath162 ) , and 27.8% on the fb ( @xmath3@xmath72 ) . the average speed along the z track as a function of @xmath3(@xmath17 ) is shown in the top panel of figure [ hist - vz_fig ] . the speed at a given @xmath18 is defined as @xmath19 ( see also * ? ? ? * ) , where @xmath20 is used to number the points in order of time . as expected the @xmath17 increases considerably when the source enters the fb , but also at the top of the hb ( @xmath21 ) . combined with the small amount of time spent in the upper hb we can conclude that the source reaches to @xmath3 values this low only in brief , quick dashes . for a more detailed analysis of the spectral changes as a function of the position along the z track in gx 17 + 2 we refer to et al . ( in prep . ) . figure [ pds_fig ] shows the low frequency part ( 0.0625256 hz ) of the power spectrum for nine different @xmath3 selections . the power spectra shown in figure [ pds_fig ] are selected from epoch 3 and epoch 4 and are therefore a combination of 5.160 kev and 5.860 kev data . the percentage of epoch 3 and epoch 4 data varies between the @xmath3-selections . the contribution of epoch 3 data is highest at the extremes of the @xmath3 range , reaching 100% , and gradually decreases from both ends to @xmath010% around @xmath3=1 . several components can be seen , which change in strength and shape as a function of @xmath3 . the different components are identified in figure [ pds_fig ] ( see also figure [ pds_comp_fig ] ) and the @xmath3ranges in which they were detected are given in table [ sz - comp_tab ] . the component identified as the sub - harmonic of the hbo is rather broad and does not have the appearance of a qpo . however , based on the frequency ratios ( see below ) it is referred to as the hbo sub - harmonic . some difficulties were experienced while fitting the low frequency part of the power spectrum between @xmath3=1.4 and @xmath3=1.6 , where two or more components with similar frequencies were simultaneously present . above @xmath3=1.4 the nbo appeared , as a broad feature on top of the lfn . we were not able to distinguish the two components and decided to fit them together with a single lorentzian . the fit values are not used in figures and tables , since they do not represent any of the individual components . above @xmath3=1.6 , the nbo and lfn could be distinguished more easily . the fit function used for the lfn , which was underlying the nbo , was changed to a lorentzian to be more consistent with fits at higher @xmath3 ( fits with a cut - off power law gave equally good @xmath22 at this @xmath3 ) . in the following sections the results for each of the components will be presented . the hbo was detected between @xmath3=@xmath230.6 and @xmath3=2.1 and its second harmonic between @xmath3=@xmath230.6 and @xmath3=1.0 . the second and third columns of figure [ hbo_fig ] show their rms amplitudes , fwhm , and frequencies as a function of @xmath3(see also table [ low_tab ] ) . the frequency of the hbo increased from 21.3 hz at @xmath3=@xmath230.43 to 60.3 hz at @xmath3=1.45 and then decreased to 48.2 hz at @xmath3=2.04 . the second harmonic of the hbo had a frequency that was on average 1.941@xmath240.007 times that of the first harmonic , which is significantly different from the expected value of 2 . an explanation for this discrepancy is proposed in section [ disc - hbo_sec ] . between @xmath3=@xmath230.6 and @xmath3=1.0 both the hbo and its second harmonic decreased in strength , with the decrease of the second harmonic occurring somewhat faster than that of the hbo . in that same @xmath3interval the fwhm of the hbo and its second harmonic were fairly constant ( showing a slight increase ) , although considerable scatter was present around the average values ( which were , respectively , 7.9@xmath240.3 hz and 14.4@xmath240.6 hz ) . the q values of the hbo and its harmonic are shown in figure [ q - values_fig ] . in the @xmath3 range where both were detected their q values were consistent with each other . when the second harmonic was not significantly detected anymore ( @xmath3@xmath71.0 ) the rms amplitude and the fwhm of the hbo both decreased from , respectively , 2.45@xmath240.08% and 9.7@xmath247 hz ( @xmath3=0.51.0 ) to 1.48@xmath240.06% and 5.1@xmath240.4 hz ( @xmath3=1.01.5 ) . between @xmath3=1.0 and @xmath3=1.5 the relation between the @xmath3 and the hbo frequency started to flatten . above @xmath3=1.5 the frequency of the hbo clearly dropped , initially quite abruptly and later on more smoothly . this frequency drop coincided with an increase in the fwhm to 15.7@xmath241.5 ( @xmath3=1.52.1 ) ; the rms amplitude showed a small increase to 2.1@xmath240.1% ( @xmath3=1.76 ) , followed by a decrease to 1.2@xmath240.2% ( @xmath3=2.04 ) . underlying the hbo and its second harmonic , but with a central frequency lower than that of the hbo , we found a broad feature that was also fitted with a lorentzian . it was significantly detected between @xmath3=@xmath230.6 and @xmath3=1.3 . the properties of the broad feature are shown in the first column of figure [ hbo_fig ] ( see also table [ low_tab ] ) . below @xmath3=0 the frequency of the lorentzian was fixed to zero . between @xmath3=0.5 and @xmath3=1.5 the frequency of the broad feature is on average 0.539@xmath240.015 times that of the hbo . this could mean that the broad feature is the sub - harmonic of the hbo , certainly when one takes into account that the frequency of a broad feature is rather sensitive to the shape of the continuum . whereas the rms amplitude and the frequency of the sub - harmonic both change strongly with @xmath3 the fwhm remains more or less constant , with an average value of 29.5@xmath241.5 hz . the nbo and fbo were detected between @xmath3=1.6 and @xmath3=2.7 . the fit results for the nbo / fbo are shown in figure [ nbo_fig ] ( see also table [ low_tab ] ) . below @xmath3=2.1 the nbo ( represented by the filled circles in figure [ nbo_fig ] ) had a fairly constant frequency , with values between 6.3 hz and 7.0 hz . its rms amplitude increased from 1.7% to 3.2% , and the fwhm was approximately @xmath02.1 or @xmath03.8 . in the @xmath3=2.02.1 selection both the nbo and fbo were present . this is likely an artifact of the @xmath3selection method , since a careful inspection of dynamical power spectra showed no evidence for simultaneous presence of both qpos . it is interesting to note though , that in the @xmath3=2.02.1 selection the frequency of the fbo was 2.0@xmath240.1 times that of the nbo , which could mean that the two nbo and fbo are harmonically related ( however , see below ) . the fbo increased in frequency from 13.9 hz ( @xmath3=2.04 ) to 23.1 hz ( @xmath3=2.65 ) , while its fwhm increased from @xmath05 hz to @xmath013 hz . the rms amplitude of the fbo initially continued the trend of the nbo rms amplitude ; it increased from 2.8% ( @xmath3=2.04 ) to 5.6% ( @xmath3=2.23 ) , but then decreased to 1.8% ( @xmath3=2.65 ) . to study the transitions between the nbo and fbo more carefully , we inspected all dynamical power spectra of observations with @xmath3values around 2 . although no clear transitions were found , mainly due to the limited quality of the dynamical power spectra , we did in some cases see qpos with intermediate frequencies ( @xmath0 10 hz ) , suggesting that the frequency does not jump directly from @xmath07 hz to @xmath014 hz . fitting the power spectra in consecutive time intervals ( longer than 16 s ) , rather than inspecting dynamical power spectra with the eye , will probably lead to more conclusive results , but is beyond the scope of the current paper . the time scales on which the nbo / fbo frequency changed from @xmath07 hz to @xmath014 hz and back were as short as a few tens of seconds . we also studied the behavior of the nbo / fbo during two long type i x - ray bursts . the first one started on 1998 november 18 at 08:51:26 utc , the second one on 1999 oct 10 at 09:10:47 utc . their exponential decay times were , respectively , 189@xmath242 s and 144@xmath242 s @xcite . both bursts occurred near the nb / fb vertex and in the power spectra of their respective observations the nbo / fbo is clearly detected . no other qpos were detected in the power spectra of these bursts . figure [ burst - nbo_fig ] shows the dynamical power spectra of both bursts , together with their 260 kev light curves . during the brightest part of the bursts the nbo / fbo seemed to disappear apparently , the burst flux was not modulated at the nbo / fbo frequencies with the same amplitude as the persistent flux . to quantify this , we determined upper limits on the nbo / fbo strength and compared those with the values outside the bursts . the results are shown in table [ burst - nbo_tab ] . clearly , during the brightest part of the bursts the rms amplitude of the nbo / fbo was significantly weaker , not only as a fraction of the total flux , but also in absolute terms ( and hence as a fraction of the persistent flux , if the persistent emission is assumed to continue during the bursts ) . these measurements constitute the first determination of the effect of x - ray bursts on the nbo / fbo . the fact that the bursts suppress the qpo can have important consequences for our understanding of its formation mechanism ( see section [ disc - nbo_sec ] ) . lfn was detected between @xmath3=@xmath230.6 and @xmath3=5.0 . as mentioned before , the appearance of the nbo around @xmath3=1.4 , kept us from putting firm constraints on the lfn parameters between @xmath3=1.4 and @xmath3=1.6 . the fit results for the lfn are shown in figure [ lfn_fig ] ( see also table [ tab : noise ] ) , in two separate columns : one for the fits with a cut - off power law fit ( @xmath3=0.01.4 ) and one for the fits with a lorentzian ( @xmath3@xmath10.0 and @xmath3=1.65.0 ) . the strength of the lfn was in both cases defined as the integrated power spectral density between 1 hz and 100 hz . we note that below @xmath3=0.1 the vlfn was not fitted separately from the lfn . an inspection of 1/2564096 hz power spectra below @xmath3=0.1 showed that a weak power law component was present at frequencies below 0.1 hz . this component was probably vlfn ; its power in the 1100 hz range was much smaller than that of the lfn , so , although some vlfn power was absorbed by the lfn , this did not affect the lfn rms amplitudes significantly . the strength of the lfn changed considerably as a function of @xmath3 . it showed a narrow peak between @xmath3=@xmath230.6 and @xmath3=0.2 of @xmath05% rms . between @xmath3=0.2 and @xmath3=1.7 it gradually decreased from @xmath04.5% rms to @xmath02.5% rms . another decrease was observed between @xmath3=2.7 and @xmath3=5.0 , from @xmath03.8% rms to @xmath00.9% rms . the behavior between @xmath3=1.7 and @xmath3=2.7 was quite erratic , probably due to interactions with the fit functions of nbo and fbo . the centroid frequency of a lorentzian and the cut - off frequency of a cut - off power law can not be directly compared . since we wanted to see how the typical frequency of the lfn evolved with @xmath3 , we chose to plot @xmath25 , which is the maximum in a @xmath26 plot and the frequency at which most of the power is concentrated per logarithmic frequency interval ( belloni , psaltis , & van der klis 2001 , in prep . ) . for a lorentzian @xmath25 is @xmath27 and for a cut - off power law @xmath25 is @xmath28 ( see section [ 17 + 2_obs_sec ] for analytical expressions for a lorentzian and cut - off power law ) . as can be seen from figure [ lfn_fig ] , between @xmath3=@xmath230.6 and @xmath3=1.7 @xmath25 smoothly increased from @xmath01.2 hz to @xmath014.7 hz . above @xmath3=1.7 the errors on @xmath25 were larger and the behavior was less clear . between @xmath3=1.7 and @xmath3=2.3 @xmath25 decreased to @xmath06.5 hz , and above @xmath3=2.3 it increased again to @xmath015 hz . we tested whether the change of the fit functions at @xmath3=0.0 affected the values for @xmath25 and the other power spectral parameters , by swapping the fit functions used below and above @xmath3=0.0 ; no significant changes were found . vlfn was detected over almost the whole @xmath3 range . although we only started fitting the vlfn separately from the lfn above @xmath3=0.1 ( in the 16 s power spectra ) , vlfn was present at frequencies below 0.1 hz in the 256 s power spectra below @xmath3=0.1 . in the 16 s power spectra it was only significantly detected above @xmath3=0.3 . the fit results for the vlfn are shown in figure [ vlfn_fig ] ( see also table [ tab : noise ] ) . between @xmath3=0.3 and @xmath3=1.4 the vlfn strength decreased from @xmath00.9% rms to @xmath00.4% rms . above @xmath3=1.4 its strength increased , to a peak value of @xmath01.3% rms at the nb / fb vertex . on the lower fb the strength decreased again , to a value of @xmath00.5% rms , and above @xmath3=3.0 it increased to @xmath01.6% rms at the top of the fb . the index of the vlfn slowly increased from @xmath00.6 to @xmath01.0 between @xmath3=0.1 and @xmath3=2.4 , with a small peak between @xmath3=1.6 and @xmath3=1.9 , where the index had a value of @xmath01.5 . between @xmath3=2.4 and @xmath3=3.2 the vlfn was much steeper . the indices were not well constrained and had values between 1.7 and 4.0 . above @xmath3=3.2 , where the error bars are much smaller , the index slowly decreased from 2.1 to 1.8 . our fits did not require an additional component to fit possible hfn ; no such component was found . possible explanations for this are discussed in section [ disc : noise ] . both the lower and upper khz qpo were clearly detected ; the lower khz qpo between @xmath3=0.5 and @xmath3=1.5 , the upper khz qpo between @xmath3=0.3 and @xmath3=1.7 . the results can be found in table [ khz_tab ] and are shown in figure [ khz_fig ] . the qpo frequencies showed a clear increase with @xmath3 , although both relations flattened at their low frequency ends . the fwhm of the lower khz qpo was consistent with being constant at @xmath0100 hz , whereas the rms amplitude showed a peak near @xmath3=1.05 with a value of 3.6@xmath240.3 . the rms amplitude and fwhm of the upper khz qpo both decreased with @xmath3 . figure [ khz - diff_fig]a shows the frequency difference of the two khz qpos as a function of the frequency of the upper khz qpo . some fits were made to the data ; they are also shown in figure [ khz - diff_fig]a . the best fit to the frequency difference with a constant gave a value of 282@xmath244 hz . the @xmath29 for this fit was 18.6/9 , which means that at a 97% confidence level the frequency difference was not constant . this is the first time that this is observed for gx 17 + 2 . fits with first and second order polynomials ( taking into account errors in both coordinates ) resulted in , respectively , @xmath29=15.0/8 and @xmath29=3.7/7 . although the latter two fits show that with 99.76 % confidence ( 3.2@xmath30 ) a decrease towards higher frequency is not monotonic , it is not clear either whether the decrease towards lower frequencies is significant or not . this would be the first time that such a decrease towards lower frequencies is observed in any khz qpo source with a non - constant frequency difference . to test this , two fits with a broken line ( not shown ) were performed , where in one case the slope of the low frequency part was fixed to zero . they resulted in @xmath29=6.28/7 ( slope fixed ) and @xmath29=2.52/6 ( slope free ) , which suggests that the frequency decrease towards lower frequency is only significant at a 97.6% confidence level ( 2.4@xmath30 ) . in figure [ khz - diff_fig]b we show additional fits to the data with theoretical curves for the radial epicyclic frequency vs.azimuthal frequency , which in the relativistic precession models are associated with the frequency difference and the upper khz qpo frequency . the solid line is the best fit for circular orbtis @xcite , which yields a @xmath22 of 7.3 ( d.o.f = 9 ) for a mass of 2.07 m@xmath31 ; the dashed line is the best fit for eccentric orbits @xcite , yielding a @xmath22 of 1.14 ( d.o.f = 8) for a mass of 1.86 m@xmath31 . both curves were calculated for non - rotating stars ; rotation of the neutron star gives only insignificant improvements . for comparison we also plotted the data for sco x-1 from @xcite . clearly , both fits ( especially the solid line ) do not fit the decrease of the frequency difference towards lower frequencies , which occurs in a frequency range for the upper khz qpo that was not observed in sco x-1 , very well . the q values ( frequency / fwhm ) of the two khz qpos were consistent with each other and with that of the hbo and its harmonic ( see figure [ q - values_fig ] ) ; between @xmath3=1.0 and @xmath3=1.5 the q values of the hbo and khz qpos increased from @xmath05 to @xmath010 . this unexpected finding may provide a key to understanding the formation mechanism of the three qpos this will be further discussed in section [ q - values_sec ] . when comparing figures [ khz_fig ] and [ hbo_fig ] one can see that above @xmath3=1.5 , where the hbo frequency starts to decrease , the upper khz qpo frequency still increases . in figure [ khz - hbo_fig ] we plot the frequencies of both qpos against each other . for values of the upper khz frequency lower than 1030 hz the frequencies are well correlated , but above there is a clear anti - correlation . the solid line is the best power law fit to the points below 1000 hz . the power law index is 2.08@xmath240.07 . this is the first time in any atoll or z source showing khz qpos that an anti - correlation is observed between the frequencies of the low and high frequency qpos ( see section [ disc - hbo_sec ] for a discussion ) . we performed a detailed study of the low and high frequency power spectral features of the neutron star low - mass x - ray binary and z source gx 17 + 2 . as was found in previous studies of gx 17 + 2 and other z sources , the properties of most power spectral features correlated well with the position of the source along the z track in the x - ray color - color diagram , suggesting that the spectral and variability changes are caused by one and the same parameter . some new results were found , the most interesting being the fact that the frequency separation of the khz qpos was not constant and probably not monotonic ( fig . [ khz - diff_fig ] ) , that their q values were consistent with those of the hbo and its harmonic ( fig.[q - values_fig ] ) , and that the frequency of the upper khz qpo was anti - correlated with that of the hbo when the latter started to decline on the nb ( fig . [ khz - hbo_fig ] ) . we also found a sub - harmonic of the hbo and showed that the absolute amplitude of the nbo / fbo is suppressed during type i x - ray bursts ( fig . [ burst - nbo_fig ] ) . in the remainder of this section these and other results will be discussed in more detail . for a more detailed discussion , including a comparison of our results with those obtained from with other satellites ( like _ exosat _ and _ ginga _ ) and of other z sources , we refer to @xcite . in our discussion of the hbo and khz qpo results , we will only consider models that try to explain the behavior of both types of qpos . currently these models are the relativistic precession model , the sonic point beat frequency model and the two - oscillator model . in the latter model ( see * ? ? ? * ) an angle between the magnetospheric equator and the disc plane is defined , which depends on the frequencies of the hbo and khz qpos and is supposed to be constant . however , @xcite show , by using our gx 17 + 2 results and theirs of gx 5 - 1 , that this is definitely not the case . for this reason , we will only consider the first two models . in the relativistic precession model ( rpm , @xcite ) three frequencies are predicted ; they correspond to the fundamental frequencies of motion of a test particle in the vicinity of a compact object and are identified with the hbo and the two khz qpos . the lowest predicted frequency , which should explain the hbo , is a factor 2 lower than the low frequency qpos in the atoll sources and the sub - harmonic of the hbo in the z sources . hence , a way to produce second harmonics is required . note that no actual mechanism is proposed for the production of qpos themselves . however , in a recent model by @xcite the accretion disk acts as low band - pass filter with resonances close to the three frequencies predicted in the rpm . it is unclear to what extent the details of the rpm ( i.e. the turnover in the relation between hbo and upper khz frequency ) carry over to the @xcite description . we also note that recent investigations of the rpm by @xcite have cast some doubt on the frequencies predicted by the rpm and relations between them . in the sonic point beat frequency model ( spbfm , * ? ? ? * ; * ? ? ? * ) the hbo is explained by the original beat frequency model of @xcite and @xcite . the observed khz frequencies are close to the orbital frequency at the sonic point of this disk and the beat of this frequency with neutron star spin frequency . the model requires the presence of a magnetosphere and a solid surface , and can therefore not explain the low and high frequency qpo in the black hole candidates . the decrease of the hbo frequency on the nb in gx 17 + 2 was first observed in _ rxte _ data that were not used in our analysis @xcite . there were interesting features in the hbo properties which were reproduced in that data and ours . in both cases the @xmath3 dependence of the hbo frequency was not symmetric around its peak value ; below , the frequency slowly flattened off to its maximum value , but above it showed a fast decrease by a few hz that was followed by a more or less linear decrease with @xmath3 . also , simultaneously with the frequency drop , the fwhm increased by a factor of more than two . possible explanations for the hbo frequency decrease are discussed in section [ hbo - khz_disc ] . in all z sources where an harmonic of the hbo has been observed , the frequency ratio of the harmonic and the fundamental is generally less than two . this can be explained , at least in gx 17 + 2 , by looking at the @xmath3 dependence of the rms amplitudes of both components . in a certain @xmath3 selection both frequencies tend to be dominated by those with the higher rms amplitudes , and hence by lower @xmath3 and lower frequencies ; this effect will be stronger for the harmonic than for the fundamental , since the @xmath3 dependence of its rms amplitude is steeper ( see figure [ hbo_fig ] ) . as a result the frequency ratio will end up to be less than two . to test this hypothesis we calculated the weighted averages of two harmonically related linear functions , representing the hbo frequency-@xmath3relation between @xmath3=0.0 and @xmath3=1.0 . as weight we used linear functions that were fit to the hbo rms-@xmath3 relation between @xmath3=0.0 and @xmath3=1.0 . the ratio of the weighted frequency averages was 1.93 , which is very close to the observed value of 1.94 . peaked features with frequencies intermediate to those of the lfn and the hbo have been reported for gx 340 + 0 @xcite , sco x-1 @xcite , gx 5 - 1 @xcite , and now also for gx 17 + 2 . based on the average frequency ratio of this sub - hbo component and the hbo in gx 17 + 2 , 0.539@xmath240.015 , we suggest that , similar to what was suggested for gx 340 + 0 , it may be the sub - harmonic of the hbo . a comparative study by @xcite of low frequency ( hbo - like ) qpos and band limited noise components in the neutron star z and atoll sources and black holes revealed strong correlations between the typical frequencies of those components . when plotting the frequency of the low frequency qpo versus the break frequency of the band limited noise they found that the branch traced out by the z sources , although lying parallel to that of the atoll source and black holes , was slightly offset . it is interesting to see that this discrepancy between the z sources on the one hand and the atoll sources and black holes on the other disappears @xcite , at least for gx 17 + 2 , sco x-1 and gx 340 + 0 , when one uses the frequencies of the sub - hbo component , instead of the hbo frequency ( see also * ? ? ? if the low frequency qpo in the atoll sources and the sub - hbo component of the three z sources would indeed be the same qpo , this would mean that the(se ) z sources do not anymore follow the relation between the low frequency ( hbo - like ) qpo and the lower khz qpo as defined by atoll and black hole systems @xcite . it is interesting to see that the three z sources would then end up near or slightly above a second relation that is traced out by atoll sources like 4u 1728 - 34 and 4u 160852 . at the moment , however , it seems that the relation found by @xcite and the main relation of @xcite are mutually exclusive for the two types of sources . note that @xcite found that if in their plot for sco x-1 the frequencies of the sub - hbo component and the hbo were used , instead of the lfn and hbo frequencies , the discrepancy between sco x-1 and the non - z sources also disappeared . this reinterpretation has the advantage of preserving the consistency with the @xcite relation as well . the dominance of the even harmonics over the odd harmonics in the low frequency ( hbo - like ) qpos , as suggested by the presence of a sub - harmonic of the hbo , is not only observed in the z sources , but also a common feature in bhcs such as gs 112468 @xcite and xte j1550564 @xcite . it suggests that a two - fold symmetry is present in the production mechanism of these low frequency qpos . for the spbfm one could think of an asymmetry in the magnetic field , which results in different areas for the polar caps ( see e.g. * ? ? ? * ) . in the rpm a two - fold symmetry is introduced by the two points at which the slightly inclined orbits of test particles cross the plane of the accretion disk . the khz qpos in gx 17 + 2 were discovered with _ rxte _ @xcite . in our data the lower khz qpo was found between 517 hz and 794 hz ( @xmath3=0.51.5 ) and the upper khz qpo between 618 hz and 1087 hz ( @xmath3=-0.31.7 ) , similar to what was reported by @xcite . our findings of the khz qpos in gx 17 + 2 are compatible with the general properties of khz in other z sources @xcite . an important new result is that the frequency difference is no longer consistent ( at a 97% confidence level ) with being constant ( see figure [ khz - diff_fig ] ) . the frequency separation reached a maximum value of 308 hz , when the upper peak frequency was about 915 hz , and fell off in both directions to values of 263 hz and 239 hz , when the frequencies of the upper peak were , respectively , 780 hz and 1033 hz . note that the decrease in both directions , when fitted with a parabola , resulted in significantly ( 3.1 @xmath30 ) better fits than that with a constant . the decrease in the frequency separation at the low frequency side of the upper khz qpo , although itself only significant at a 2.3@xmath30 level , has not been seen before in any khz qpo source . in the other lmxbs that show an non - constant frequency separation , sco x-1 @xcite , 4u 160852 @xcite , 4u 173544 @xcite , and 4u 172834 @xcite , the frequency separation was not measured at upper peak frequencies as low as in gx 17 + 2 . although both the spbfm and the rpm are capable of explaining the observed decrease in frequency difference towards higher upper peak frequencies , they have severe problems with a decrease towards lower frequencies . such a decrease is not possible in the spbfm , and , although it is in principle predicted by the rpm , that model only yields reasonale fits to our data when one allows for highly eccentric orbits ( see figure [ khz - diff_fig]b ) . the principal aim of our observing campaign in 1999 was to check on the relation between hbo frequency ( whose increase was known to turn into a decrease towards higher @xmath3 ) and upper khz qpo frequency . clearly , a decrease , or even flattening , of the upper khz qpo frequency , is not observed to accompany the decrease in hbo frequency ( see figures [ hbo_fig ] , [ khz_fig ] , and [ khz - hbo_fig ] ) . in a pure rpm , this can not be explained . however , if classical precession due to oblateness of the neutron star also plays a role , such an explanation is possible . for certain combinations of neutron star parameters , the prediction is that above a certain value of the upper khz qpo frequency , the frequency of the hbo in fact starts to decrease ( see * ? ? ? * ) , due to the increasing importance of the classical precession . in gx 17 + 2 this turn - over of the hbo frequency occurs at a value of @xmath01 khz for the upper khz qpo with a peak frequency of @xmath060 hz for the hbo . judging from figure 2 in @xcite , the relation between the hbo and upper khz qpo frequencies in gx 17 + 2 would require a high neutron star spin frequency ( @xmath0500700 hz ) , a neutron star mass of more than 2 @xmath32 , and a hard equation of state ( assuming the observed hbo frequency is four times the predicted frequency ) . in the rpm , well below the turn - over , the hbo frequency is expected to scale quadratically with the upper khz qpo frequency . the best power law fit to the data below 1000 hz yielded an index of 2.08@xmath240.07 ( see figure [ khz - hbo_fig ] ) , which is consistent with a quadratic relation . in the spbfm a turnover in the hbo - upper khz qpo relation is not predicted , but as the two frequencies do not directly depend on each other , the model is flexible enough to account for it ; in case the turnover in hbo frequency can be explained by the two flow model put forward by @xcite , the steady increase of the upper khz qpo frequency could in principle be explained by assuming that most of the radial flow falls back to the disk and rejoins the disk flow before the sonic point is reached , so that @xmath33 at that location , and hence the upper khz qpo frequency keep increasing . however , since the two radii at which the different qpos are generated are very close , this seems a rather unlikely option . the q values of the khz qpos are remarkably similar to each other , and to those of the hbo and its harmonic ( see figure [ q - values_fig ] ) . they also show an almost simultaneous increase above @xmath3=1 , when the hbo harmonic is no longer detected . if the qpos were artificially broadened by our selection method , averaging together peaks with different centroid frequencies , one would expect correlations of the fwhm ( and q value ) with the width of the @xmath3 selection and with @xmath34 . such correlations , if present at all , are far too weak to explain the observed similarities . the fwhm of a qpo can be caused by several mechanisms ; intrinsic frequency variations , phase jumps , the simultaneous presence of several frequencies , and a finite lifetime of the signal will all cause a signal to appear as a broadened peak in a power spectrum . the data quality did not allow us to test the phase jump option . lifetime broadening can probably not explain our results . in the spbfm there is no reason why the lifetimes of the low and high frequency oscillations , which are determined by different processes , would lead to similar q values . in the rpm , where all frequencies are generated by a single blob of matter , the lifetime for all variations should to first order be the same . however , in that case the q values of the khz qpos should be much higher ( by a factor 1020 ) than those of the hbo , since more khz cycles fit in a blob s lifetime . of course , it is possible that the lifetime of the oscillations is not determined by the lifetime of the blob , but rather by a damping factor . assuming that these factors are not the same for the different qpos , the similarity of the q values could in principle be explained . however , there is no obvious reason why the ( independent ) damping factors would be fine tuned in a such a way that they lead to similar q values . explaining the q values by broadening due to the presence of several frequencies does not work either . below the turnover in the hbo - upper khz qpo frequency relation , the frequency of the hbo scales quadratically with the frequency of the upper khz qpo . a given range of upper khz frequencies would therefore lead to a q value for the hbo that is half the q value for the upper khz qpo . this is not consistent with what we find . frequency modulation could in principle explain the similar q values , but only if the modulation is caused by a different parameter than the one that determines the frequency changes along the z track . frequency modulation only produces similar q values if the qpo frequencies vary with the modulating parameter by the same factor , i.e. proportionally to each other . the observed quadratic relation between the hbo frequency and that of the upper khz qpo as well as the subsequent turnover , shows that this is not the case for the parameter that causes the motion along the z track . however , assuming that the modulating parameter is not responsible for motion along the z ( see section [ disc - spec ] ) , we discuss the possibility of frequency modulation in a bit more detail . for broadening by frequency modulation to be observable in our power spectra , the time scale of these variations should be longer than that of the hbo . additionally , the strength of those variations should anti - correlate with the q values of the qpo . assuming that the parameter that causes the frequency modulations also causes some changes in the count rate we could try to identify it in the power spectrum . the only reasonably strong components that we see in our power spectra and whose typical frequencies fulfill the above frequency requirement ( @xmath35 hz ) are the vlfn , the lfn and the nbo / fbo . the vlfn might be associated with the motion along the z ( see section [ disc : noise ] ) ; disentangling that effect from a possible additional one on qpo frequencies is beyond the scope of this work . the strength of the lfn , which is of interest below @xmath3=1.5 , anti - correlates with the q values below @xmath3=1.5 . above @xmath3=1.5 the q values decrease again to a value between 2 and 5 . in that case the broadening might be due to the appearance of the nbo / fbo and the increase in the lfn strength . it is interesting to note that in sco x-1 it has already been found that the khz qpo properties are clearly modulated by the nbo @xcite . for the first time we were able to study the properties of the nbo / fbo during type i x - ray bursts . we found that the absolute rms amplitude significantly decreased when the radiation from the neutron star surface increased . this is clearly different from the behavior of the @xmath01 hz qpos in the dipping lmxbs 4u 132362 @xcite , exo 0748676 @xcite , and 4u 174637 @xcite , whose absolute rms amplitudes increased during type i x - ray bursts . it is thought that these qpos are not related to the nbo / fbo in the z sources . the behavior of the nbo / fbo in gx 17 + 2 shows that nbo / fbo mechanism is very sensitive to the radiation field , thereby lending strong support for radiation feedback mechanisms ( see below ) . in the framework of such mechanisms it is easy to understand that the nbo / fbo disappears during bursts : the nbo / fbo is only observed in a narrow @xmath3 range , suggesting that it requires a delicate balance of certain parameters , among which the radiation field . the observations during the burst tails indicate that the nbo / fbo mechanism does not switch on instantly but rather becomes stronger in a more gradual way . spectral fits during the bursts @xcite suggest that the spectral properties of the accretion flow itself are not affected much by the increase in radiation . the frequency of the nbo / fbo changed from being more or less constant below @xmath3=2 to being strongly correlated with @xmath3 , above @xmath3=2 . in our data set we found the nbo between @xmath3=1.6 and @xmath3=2.1 , although it may already have been present as a very broad feature as early as @xmath3=1.4 . between @xmath3=2.0 and @xmath3=2.1 , the nbo ( 6.86 hz ) was detected simultaneously with the fbo ( 13.9 hz ) . above @xmath3=2.1 the fbo frequency increased , to @xmath023 hz at @xmath3=2.65 . note that the simultaneous presence of the nbo and fbo between @xmath3=2.0 and @xmath3=2.1 is likely an artifact of the @xmath3 selection method . inspection of dynamical power spectra showed that they were never present at the same time . it is interesting though , that in that @xmath3 selection their frequencies differed by a factor of 2 , which is suggestive of mode switching of the nbo / fbo . however , there is evidence for qpos with intermediate frequencies , which undermines that idea . the only way to see whether the frequencies really gradually transform into each other is to study the power spectral changes as a function of time . however , the quality of the gx 17 + 2 data is not sufficient to perform such a study . the presence of the nbo / fbo in the z sources has often been related to the fact that these sources accrete at near eddington mass accretion rates . it is thought that at these high mass accretion rates a significant fraction of the accretion flow is in the form of a thick and perhaps spherical flow . also , the effects of radiation pressure are thought to play an important role in this regime . the model for the nbo by @xcite is based on a radiation feedback mechanism in a spherical accretion flow . another model for the nbo was proposed by @xcite , in which the nbo frequency is basically that of sound waves in a thickened accretion disk . this model does not explain how the nbo changes into the fbo . a major problem for both models could be the recently discovered nbo / fbo - like qpos in the two atoll sources 4u 182030 @xcite and xte j1806246 @xcite . certainly in the first source the luminosity is believed to be only @xmath02040 % of the eddington luminosity , while near - eddington luminosities are required in both models . the lack of hfn in our data , compared to , e.g. , in the _ exosat_@xcite data can be explained by instrumental effects and different approaches to fitting the data . a study of _ exosat _ data of several sources by @xcite showed that a component with an rms amplitude of @xmath03% is always present in the frequency range where the hfn was often found in the _ exosat _ data . when hfn was found at lower frequencies , it was , based on strength and frequency , most likely the component that we identified as the sub - harmonic of the hbo . past measurements of the energy spectrum of the vlfn suggest that it is caused by motion of the source along the z @xcite . the strength of the vlfn ( fig . [ vlfn_fig ] ) and the speed along the z track ( @xmath36 , see figure [ hist - vz_fig ] ) should therefore have a similar dependence on @xmath3 . our results are inconclusive ; although some correlations are observed ( the increase above @xmath3=3 , and the steep increase around @xmath3=2 ) there are also some differences . most notable is the dip in the vlfn strength around @xmath3=3 , which is not observed as a decrease in @xmath17 . however , it is possible that due to the steepness of the vlfn around @xmath3=2.5 most of its power was outside the range in which we determined the strength . the lfn followed the behavior of the hbo , both in strength and frequency , until the latter disappeared ( @xmath37 ) . between @xmath3=2.1 and @xmath3=2.8 the behavior is rather complex and probably considerably influenced by the presence of the nbo / fbo . above @xmath3=3.0 the behavior was rather clean , with a decrease in strength and an increase in frequency . it is not clear , however , whether the lfn on the fb is the same as that on the hb and nb . unfortunately no hbo was observed above @xmath3=2.1 to compare its behavior with . van der klis et al . ( 1997 ) suggested that in sco x-1 the nbo / fbo emerged from the lfn . this seems not to be the case in gx 17 + 2 , since the rms amplitude of the lfn already started to decrease well before the nbo appeared . moreover , @xmath25 shows a gradual increase between @xmath3=-0.6 and @xmath3=1.4 to values well above the nbo frequency . a comparative study by @xcite strongly suggests that lfn has a similar origin as the band - limited noise component in atoll sources , bhcs , and the millisecond x - ray pulsar sax 1808.4 - 3658 . they suggest that this noise component is produced in the accretion disk , and does not require the presence of magnetosphere or solid surface . in our observations gx 17 + 2 traced out the well known z - like tracks in the cd and hid . in addition to motion along this track , we also observed secular motion of the z track itself in the hid of epoch 3 ( see figure [ cd - hid_fig ] ) , as was already reported by @xcite ( see also * ? ? ? * for a possible shift in the exosat hid of gx 17 + 2 ) . below @xmath3=1 the fluxes in the hard and soft energy bands behaved quite differently ( fig.[sz - count_fig ] ) ; unlike the soft flux , the hard flux continued the trend that is observed above @xmath3=1 . hence , the hb / nb vertex in the cd is entirely due to the turn over of the soft spectral flux . it is not clear what causes the soft spectral flux to decrease below @xmath3=1 . although it is partly an instrumental effect the peak of the soft component slowly shifts out of the pca effective energy range as it becomes softer most likely intrinsic source changes occur at @xmath3=1 . most variability components increase in strength below @xmath3=1 , suggesting that they are related to the hard spectral component . it is commonly believed that the motion of the source along and its position on the z track are determined by the mass accretion rate ( @xmath33 ) through the inner accretion disk and onto the neutron star , increasing from the hb to the fb . even though the count rates actually decrease on the nb and in some sources also on the fb ( implying an anti - correlation between @xmath33 and flux ) there are several observational results that support this view : ( 1 ) the frequencies of the hbo and khz qpos , that in many models are directly related to @xmath33 , gradually increase from hb to nb ( and in sco x-1 even to the fb , see * ? ? ? ( 2 ) the optical and uv flux increase from hb to fb @xcite ; in x - ray binaries they are thought to be due to reprocessing of x - rays from the central source in the outer accretion disk . there are several reasons to believe that the optical and uv flux are betters tracers of the mass accretion rate than the x - ray flux @xcite . this would then solve the apparent anti - correlation between @xmath33 and the ( x - ray ) flux @xcite . ( 3 ) motion of a source is always along the z track ; jumps between branches are never observed , in accordance with the assumption that @xmath33 varies continuously . it has been found that secular motion of the z tracks in the cds of the cyg - like sources does not affect the relation between the rapid variability properties and source position along the track ( see , e.g. @xcite , however see @xcite ) . in view of the @xmath33-driven picture mentioned above , these changes can therefore not be due to changes in @xmath33 , but should have a different origin . it has been suggested that the cyg - like sources are viewed at a higher inclination than the sco - like sources @xcite . the differences in shapes of the z tracks and some of the variability properties are also explained in this model @xcite . @xcite suggest that the difference among the z sources might be explained by a difference in the magnetic field strength , with the sco - like sources having a lower magnetic field . in the last few years results have been obtained that may challenge the above described @xmath33-driven picture . these results are mainly due to the arrival of _ rxte _ and _ bepposax _ , whose high quality data revealed new phenomena and allowed for a better comparison between different classes of x - ray binaries . in the following we discuss how some of our results of gx 17 + 2 conflict with the @xmath33-driven picture . the frequency of the hbo in gx17 + 2 starts to decrease when the source is half way through the nb . although there are several explanations for this ( see section [ disc - hbo_sec ] ) , it means that the first assumption given above ( i.e. qpo frequencies increase with @xmath33 ) is not as solid as it seemed a few years ago . the theory of type i x - ray bursts predicts that their properties change and correlate with @xmath33 @xcite . a study of the type i x - ray bursts in gx 17 + 2 by @xcite showed that their properties did not correlate well with the position along the z track at which they occurred , and therefore not well with the inferred @xmath33 . although other explanations are possible ( see * ? ? ? * ) , this could mean that @xmath33 does not determine the position of the source along the z track . however , note that there does seem to be a correlation between burst occurrence and position along the z track , which is usually explained by a changes in @xmath33 onto the neutron star ( along the z track ) . recently , @xcite reported the discovery with bepposax of a hard tail in the energy spectrum of gx 17 + 2 . it increased in strength from the hb / nb vertex ( where only upper limits on its strength could be determined ) onto the hb . they also showed that the total 0.1200 kev flux increased monotonically from the nb / fb vertex ( @xmath38 ) to the top of the hb ( @xmath39 ) . this is different from the behavior of the 2.920.1 kev count rate ( see figure [ sz - count_fig]a ) , which increased from nb / fb vertex to the hb / nb vertex , but decreases afterwards . as can be seen from figure [ sz - count_fig]b the behavior of the hard count rate confirms the findings of @xcite . also , there is no evidence for a change in its behavior near the hb / nb vertex . unfortunately no bepposax data were taken on the fb to compare to our rxte data . hard power law tails have also been found in the z sources gx 5 - 1 @xcite , cyg x-2 @xcite , sco x-1 @xcite and gx 349 + 2 @xcite . except for sco x-1 , where no clear correlation with spectral state was found @xcite , in all these sources the strength of the hard power law tail seems to decrease in the same direction as was observed in gx 17 + 2 , i.e. from hb to fb . the behavior of the 0.1200 kev flux on the nb and hb suggest that @xmath33 actually increases in the opposite direction from what is commonly assumed . however , since the flux changes are rather moderate ( @xmath020% ) , other options are that @xmath33 does not change at all along the z track , or at least that it shows no good correlation with the position along the z. in that case motion along the z track ( i.e. spectral change ) is caused by another parameter , for example the inner disk radius @xcite . a similar possibility was suggested by @xcite to explain the observed behavior of the black hole transient xte j1550564 . they found that spectral transitions occurred at different levels of the inferred @xmath33 . the spectral changes were ascribed to an unknown parameter , which they suggested to be the inner disk radius ( based on variability ) , the size of the comptonizing region ( based on spectral hardness ) , and/or the accretion flow geometry ( based on radio brightness ) . note that options like the latter one may lead to changes in the @xmath33 onto the neutron star , while the @xmath33 through the inner disk remains constant ; in that way the occurrence of type x - rays bursts might become dependent on the position along the z track . if the above also applies to the z sources , the secular motion that is observed could perhaps be explained by changes in the @xmath33 . the apparent lack of a correlation between the type i burst properties and the position along the z track could then also be explained , since @xmath33 would determine the properties of the bursts but no longer the spectral properties . it is interesting to see that gx 17 + 2 and the z sources , show in fact several similarities with xte j1550564 and other black hole sources . the first was already briefly mentioned above , and concerns the independent behavior of the hard and soft flux on the hb . this independence of the hard and soft spectral components is commonly observed in black hole candidates ( bhcs , see * ? ? ? moreover , there are several properties that correlate well with the spectral hardness in bhcs and also with the spectral hardness in z sources , which increases from the nb to the hb ( for the moment we will exclude the fb from the comparison ; it will be discussed later ) . these properties are the radio flux , the optical flux , and to a certain degree also qpo frequencies . ( 1 ) the radio flux in bhcs is correlated with the spectral hardness ( * ? ? ? * ; * ? ? ? * and references therein ) . this is also true for gx 17 + 2 , where the radio flux increases from the nb to the hb @xcite . sco x-1 @xcite and cyg x-2 @xcite show similar behavior , whereas gx 5 - 1 only showed one radio flare on the nb @xcite . the behavior of the radio flux suggests that as the spectral hardness increases , part of the flow onto the neutron star evolves into an outflow . ( 2 ) @xcite found that in xte j1550564 the optical flux decreased during several small hard x - ray flares . although the optical counterpart of gx 17 + 2 @xcite has never been studied in much detail , this behavior is similar to the anti - correlation between the hard color and optical / uv flux on the nb and hb in the z sources cyg x-2 @xcite and sco x-1 @xcite . ( 3 ) in xte j1550564 the frequencies of the low and high frequency qpo anti - correlate with the hard flux @xcite . this is also found in gx 17 + 2 for the khz qpos and hbo ( compare figure [ sz - count_fig]b with figures [ hbo_fig ] and [ khz_fig ] ) and other z sources . the only exception to this , together with the odd behavior of the hard tail in sco x-1 ( see above ) is the correlation between the hbo frequency and spectral hardness that we observed on the lower nb . in that respect it is interesting to mention that @xcite report a switch from a correlation to an anti - correlation between the qpo frequency and spectral hardness in several bhcs , when the latter increases beyond a certain value . in addition to the qpo frequencies also the band - limited noise behaves similarly . it becomes stronger with spectral hardness , both in bhcs and z sources @xcite it is tempting to compare the different black hole states @xcite with those of the z sources . based on the above similarities , we suggest that the hb corresponds to the low state , the nb to the intermediate and/or very high state state , and the fb to the high state . however , there are still considerable differences : the low state spectra are much harder than those of the hb , and also the noise in the low state is much stronger . moreover , flaring is not observed in the black hole high state . however , in both the high state and fb the variability at high frequencies ( @xmath750 hz ) is extremely weak . it is not clear if and how the presence of a solid surface can account for these differences . our most important findings are the variable frequency difference of the khz qpo , the non - monotonic relation between the hbo and the upper khz qpo , the similar q values of the hbo , its second harmonic , and the khz qpos and the suppression of the nbo amplitude during type i x - ray bursts . in the following we summarize what the implications of each of our new findings are for some of the proposed models . the models we considered for the hbo and khz qpos were the relativistic precession model ( rpm ) , the disk - filter model , the sonic point beat frequency mode ( spbfm ) , and the two - oscillator model . the latter model can be discarded on the basis of our results ( see * ? ? ? * ) . for the disk - filter model we assumed that it produces the same frequencies as the rpm and it will therefore not be discussed it separately . \(1 ) hbo - khz qpo frequencies : the rpm can only explain a turnover of the hbo - khz qpo frequency relation when one assumes that the observed hbo is four times the predicted frequency and then only for rather extreme neutron star parameters : a high neutron star spin frequency ( @xmath0500700 hz ) , a neutron star mass of more than 2 @xmath32 , and a hard equation of state . the spbfm needs some extension ; for example , decoupling of part of the accretion flow from the disk flow outside the magnetosphere ( to explain the hbo frequency decrease ) and a subsequent recoupling of this flow with the disk flow before the sonic point radius is reached ( to explain the increase in the khz qpo frequencies ) . how this should exactly work is not clear . \(2 ) variable khz qpo separation : although not significant at a 3 sigma level , the khz qpo separation seems to decrease both towards low and high values for the upper khz qpo frequency . the rpm basically predicts such behavior , but in its current form it is inconsistent with our results , as is apparent from figure [ khz - diff_fig ] . the spbfm can not explain the possible decrease of the frequency separation at the low frequency side . \(3 ) similar q values : in the spbfm the low and high frequency qpos are produced independently from each other and there is no a priori reason why their q values should be similar . in the rpm , all the oscillations are produced by a single test particle . lifetime broadening and the presence of multiple test particles both lead to different q values for the low and high frequency qpo . frequency modulation could work for both models , but only if the modulating parameter is not the same as the one that causes the motion along the z track . the observed behavior of the nbo during two type i x - ray bursts showed that this qpo is rather sensitive to the strength of the radiation field from the neutron star . this seems to confirm the basic idea behind the feedback mechanism model that was proposed by @xcite , i.e. it requires a delicate balance between the inwards directed forces and the radiation field . finally , we compared the behavior of gx 17 + 2 ( and the other z sources ) with that of ( some ) black hole lmxbs . many similarities exist , particularly with regard to the relation between the hard spectral component on the one hand , and variability properties ( qpo frequency , broad band noise strength ) and radio flux on the other . based on recent findings in black hole lmxbs we conclude that the mass accretion might not be the parameter that determines the position along the z track after all . this work was supported in part by the netherlands organization for scientific research ( nwo ) grant 614 - 51 - 002 and by nwo spinoza grant 08 - 0 to e.p.j . van den heuvel . rw was supported by nasa through the chandra postdoctoral fellowship grant number pf9 - 10010 awarded by the chandra x - ray center , which is operated by the smithsonian astrophysical observatory for nasa under contract nas8 - 39073 . mm is a fellow of the consejo nacional de investigaciones cientficas y tcnicas de la repblica argentina . whgl is grateful for support from nasa . the authors thank the referee for his / her thorough reading and usefull remarks . finally , we would like to thank draza markovi for helpfull comments and for providing theoretical fits to our data . , m. & barret , d. 2000 , in proceedings of the fifth compton symposium , held in portsmouth , nh , usa , september 1999 . melville , ny : american institute of physics ( aip ) , 2000 . edited by mark l. mcconnell and james m. ryan aip conference proceedings , vol . 510 . , p. 157 , y. & lewin , w. h. g. 1995 , in x - ray binaries ( cambridge astrophysics series , cambridge , ma : cambridge university press , |c1995 , edited by lewin , walter h.g . ; van paradijs , jan ; van den heuvel , edward p.j . ) , p. 126 , m. 1989 , in proceedings of the nato advanced study institute on timing neutron stars , held in eme , . izmir , turkey , april 415 , 1988 . editors , h. gelman and e.p.j . van den heuvel ; publisher , kluwer academic , dordrecht , the netherlands , boston , massachusetts , p. 27 , m. 1995 , in proceedings of the nato advanced study institute on the lives of the neutron stars , held in kemer , turkey , august 29-september 12 , 1993 . editor(s ) , m.a . alpar , u. kiziloglu , j. van paradijs ; publisher , kluwer academic , dordrecht , the netherlands , boston , massachusetts , 1995 . , p. 301 ( epoch 4 data only ) . the hb / nb vertex in the 2.920.1 kev band ( a ) is caused by the contribution from the low energies ( b ) . this vertex does not show up at high energies ( c ) . [ sz - count_fig],width=377 ] = 0.40.5 power spectrum , showing the relative contribution of the individual components . the combined fit is represented by the thick solid line , the three lorentzians / qpos by the thin solid lines , the cut - off power law / lfn by the dashed line , and the power law / vlfn by the dotted line . the poisson level was subtracted . the reduced @xmath40 of this fit was 1.03 for 204 degrees of freedom . [ pds_comp_fig],width=377 ] . the left column shows the results for the fits with a lorentzian , the right column those for fits with a cut - off power law . in both cases the rms amplitude is the integrated power in the 1100 hz range . @xmath25 is the frequency at which most of the power is concentrated ( see text for the expressions for @xmath25 . ) for reasons of comparison we also plotted the values of the other fit function ( gray ) in the panels for the rms amplitude and @xmath25.[lfn_fig],width=377 ] 23 . for comparison we also plotted the data for sco x-1 ( gray ) from @xcite . for additional details see text[khz - diff_fig],title="fig:",width=245 ] 23 . for comparison we also plotted the data for sco x-1 ( gray ) from @xcite . for additional details see text[khz - diff_fig],title="fig:",width=245 ] . a log of all _ rxte_/pca observations used in this paper . mode 7 was not always active during the november 1998 observations . note that none of the observations represents an uninterrupted interval . each is a collection of observations that were done around the same time . these observations were separated in time from each other for various reasons , such as earth occultations , passages of the south atlantic anomaly , or observations of other sources.[obs_tab ] [ cols="<,<,^,^,^",options="header " , ]

[ sec : s1 ] the compactness of its core is an important structural characteristic of a presupernova star that affects whether it will explode as a supernova . a shallow density gradient around the iron core , as typically exists in more massive stars , implies a higher accretion rate and ram pressure @xcite surrounding the iron core during its collapse that must be overcome by neutrino energy deposition or other energy deposition to turn what is initially an implosion into an explosion . the connection between this structure and the likelihood of explosion has been noted many times @xcite , but little attention has been given to explaining just why the compactness has the values that it does , or why different groups obtain different core structures for models with similar main sequence mass . recently , @xcite quantified this compactness in terms of a parameter , @xmath1 where m=2.5 @xmath0 was chosen as the relevant mass for quantifying the density gradient outside of the iron core , which itself typically has a mass in the range 1.4 to 2.0 @xmath0 . in oconnor et al s definition , the time , t@xmath2 , chosen for evaluating @xmath3 is when the core has collapsed to its maximum ( super - nuclear ) density . as we shall show ( [ sec : s2.3 ] ) , for m = 2.5 @xmath0 , no substantial accuracy is lost if this parameter is evaluated for the `` presupernova model '' defined by when the collapse speed first reaches 1000 km s@xmath4 . the choice of m = 2.5 @xmath0 is justified as being larger than the iron core itself , yet deep enough in the star to sample matter that might accrete , especially in a failed explosion . since @xmath5 is inversely proportional to the radius of a fiducial mass outside the iron core , it is small when the density around the iron core falls off rapidly with radius and greater when the density gradient is shallow . that is , when @xmath5 is large , one need go only a shorter distance to include more mass . the systematics of neutrino - powered supernova explosions were explored as a function of @xmath5 by @xcite and @xcite , both of whom found that a small value of @xmath5 favored explosion . @xcite plotted @xmath5 as a function of main sequence mass for a variety of surveys due to @xcite , @xcite and @xcite , and a subset of that data is shown in fig . [ fig : f1 ] . note especially the distinctly non - monotonic behavior in the vicinity of 25 @xmath0 and 40 @xmath0 . for a polytrope of constant index and composition supported by a constant fraction ( @xmath6 ) of ideal gas pressure , the compactness defined by equation [ eq : compact ] will actually _ decrease _ monotonically with increasing mass when evaluated at a constant central temperature and small radius . this is because , for polytropes , the central temperature and density obey a relation ( e.g. * ? ? ? * ) @xmath7 for a given @xmath8 the central density thus declines as @xmath9 and the radius required to enclose a given mass increases with m. as will be discussed in [ sec : s3.1 ] , equation [ eq : poly ] works reasonably well for massive stars that are burning hydrogen and helium in their centers . the inner regions of such stars are non - degenerate and can be well represented by polytropes of constant index . as massive stars evolve , however , they develop nested cores with different compositions and entropies . lighter stars develop more degenerate cores in their late stages , especially during oxygen burning . the large gravitational potential at the edge of these degenerate cores implies a small pressure scale height that results in a steep decline in the local density @xcite . as a result , in heavier stars , the higher entropy actually leads to more extended configurations and larger values of @xmath5 . one does not have to to go out as far in the `` envelope '' of the compact core to encompass 2.5 @xmath0 , while in lighter stars that mass is reached farther out . this accounts for a general tendency of the compactness parameter for presupernova stars to increase with mass , especially for low metallicity stars where mass loss is less important . @xmath10 ) models from @xcite ( thick ) and for the solar metallicity models from @xcite ( dot - dashed ) . see also fig . [ fig : f9 ] of @xcite . note the non - monotonic behavior in both sets of models around 25 @xmath0and 40 @xmath0 . since the mass loss was ignored in the low metallicity models , the mass of the co core and the compactness continues to increase for those models at very high masses . [ fig : f1],scaledwidth=35.0% ] fig . [ fig : f1 ] also shows that the core compactness has significant non - monotonic behavior above about 20 @xmath0 , and possibly some fine structure below 20 @xmath0 . if this behavior is real and robust , it would have interesting implications for the explosion mechanism(s ) for massive stars , the masses of their compact remnants , and stellar nucleosynthesis . to address the `` real and robust '' issue , it is necessary to understand why this non - monotonic structure exists and its sensitivity to uncertainties and variations in the stellar models . the present study has several parts , some dealing with the systematics of @xmath5 found in new surveys of massive stars , others with the uncertainties one should assign to those results . in [ sec : s2 ] , three new surveys of massive stellar evolution are presented for stars of solar metallicity , very low metallicity ( 10@xmath11 solar ) , and solar metallicity with suppressed mass loss . these new surveys , though using the same physics as in previous works @xcite , are needed to provide a finer mass grid for examining rapid fluctuations in @xmath5 and to give additional data not archived in previous work . in [ sec : s3 ] , we address the heart of the matter : why is the compactness a non - monotonic function and , in some places , almost chaotic function of main sequence mass . the timing and location of several carbon and oxygen convective shells are found to play a major role . because there are multiple episodes of shell burning , the final compactness can be quite complex . in [ sec : s4 ] the sensitivity of these results to uncertain assumptions in the physics used in two stellar evolution codes , kepler and mesa , is explored . these uncertainties , especially the treatment of semiconvection and convective overshoot mixing , lead to large variations in the final co core mass that emerges for a given main sequence mass and account for much of the diversity of published results for presupernova evolution . since these uncertainties cloud the interpretation of the compactness plot , a further study is carried out in [ sec : s5 ] for bare co stars using both codes . provided the carbon mass fractions at carbon ignition are the same as in the full star models , the same trends seen previously for the compactness parameter are also found in these simpler cores , albeit with an offset due to surface boundary pressure in the full star . thus the non - monotonicity of @xmath5 can be considered robust and the co core mass is a somewhat better indicator of pre - supernova structure ( and to some extent explosion dynamics ) than the main sequence mass . finally , [ sec : s6 ] offers some conclusions . [ sec : s2 ] the models published in @xcite were aimed at sampling supernova nucleosynthesis and light curves for a broad range of non - rotating stars of solar metallicity . the resolution in mass necessary for that purpose was not particularly fine , so that study has been repeated using the same code and physics , but with a finer mass grid . the prescriptions for nuclear reaction rates , mass loss , opacity , convective overshoot mixing , and semiconvection are the same as in the previous survey . for the assumed mass loss rates @xcite , the entire hydrogen envelope is lost for stars with initial masses above 33 @xmath0 . heavier stars thus die as wolf - rayet ( wr ) stars with smaller helium cores than would have existed had mass loss been neglected . for the mass loss rate adopted during the wr - phase @xcite , the core shrinks to the extent that the final compactness becomes small again , thus the compactness of very heavy massive stars when they die depends upon an uncertain prescription for mass loss . is evaluated when the collapse speed anywhere in the iron core first reaches 1000 km s@xmath4 ( see fig . [ fig : f7 ] for other choices ) . notice especially the low values of @xmath5 for stars lighter than 22 @xmath0 , the irregular variations from 18 to 22 @xmath0 , and the subsequent rise to a maximum at 23 - 24 @xmath0 followed by a decline . for heavier stars additional structure , including a second rise , is found at higher masses ( fig . [ fig : f3]).[fig : f2],scaledwidth=35.0% ] the new survey ( fig . [ fig : f2 ] ) includes non - rotating stars with solar metallicity and masses in the range 15 to 30 @xmath0 calculated in increments of 0.1 @xmath0using the 1d implicit hydrodynamic code kepler ( table 1 ) . altogether , 151 models were simulated using the standard physics ( for variations see [ sec : s4 ] ) . nuclear energy generation was calculated , up to oxygen depletion ( central oxygen mass fraction less than 0.04 ) , using a 19 isotope network @xcite . a quasi - statistical equilibrium network and a nuclear statistical equilibrium network were used thereafter . however , a much larger `` adaptive '' network of up to 1500 isotopes was carried in parallel with the main structure calculation @xcite and was used to calculate changes in the electron mole number , @xmath12 . this was particularly important during oxygen burning when the quasi - equilibrium network could not be employed , but substantial weak interactions were already changing the central structure . this is the same approach that was used by @xcite , but a substantial improvement over what was done in @xcite . these models are collectively referred to as the `` s series '' stars - for `` solar '' . in addition to confirming the distinct bump previously seen around 23 @xmath0 the new study ( fig . [ fig : f2 ] ) shows other interesting features . fine scale variation of the compactness parameter persists throughout most of the mass range , but is particularly apparent between 19 and 21 @xmath0 . these features are further explored in [ sec : s3 ] . for this series of models , @xmath5 is small for main sequence stars lighter than 22 @xmath0 . it then rises rapidly to a peak , but declines again to lower values for main sequence masses between 26 @xmath0 and 30 @xmath0 . above about 30 @xmath0 , results for these solar metallicity stars are clouded by the effects of mass loss reducing the helium core mass , but if mass loss is suppressed , the compactness stays large above 40 @xmath0 with a slight dip around 50 @xmath0 ( fig . [ fig : f1 ] and @xmath13 ) . these features pose a challenge to the conventional notion that the difficulty of exploding a star is a monotonically increasing function of its initial mass and instead imply that models below 22 @xmath0 and between 26 @xmath0 and 30 @xmath0 may be easier to blow - up than other masses . this could alter how we think about the galactic chemical evolution @xcite . in order to facilitate the comparison with low metallicity stars where the mass loss rate may be negligible , another series of solar metallicity models was calculated for stars above 30 @xmath0 in which mass loss was artificially suppressed . these are the sh models covering the mass range of 30 - 60 @xmath0 with varying increments - 31,32,33,35,40, .. ,60 @xmath0(table 1 ) . this survey was truncated at 60 @xmath0 because the effect of the pulsational pair instability became noticeable around 70 - 80 @xmath0 . the 60 @xmath0 model of the sh - series has a helium core mass of 27.5 @xmath0 ( fig . [ fig : f4 ] ) , far greater than that of any of the solar metallicity stars calculated with mass loss . for all runs , the convective time history was recorded and checkpoints registered at representative points along the evolution including : 1 ) helium depletion - when helium reached 1% by mass in the stellar center ; 2 ) carbon ignition - when the central temperature was @xmath14 k ; 3 ) carbon depletion - when the central carbon mass fraction fell below 1% ; 4 ) oxygen ignition - when the central temperature first reached @xmath15 k ; 5 ) oxygen depletion - when the central oxygen abundance declined to 4% ; 6 ) silicon depletion - when silicon mass fraction reached 1% in the center ; and 7 ) presupernova - when any point in the core collapsed faster than 1000 km s@xmath4 . subsequent discussion will refer to these points as representative of the stellar models at these different times as he.dep . , presn . , etc . [ sec : s2.2 ] stars with equal masses on the main sequence , but differing initial metallicity can have different presupernova structure for a variety of reasons . most importantly , metallicity affects the mass loss . if the amount of mass lost is very low or zero , the presupernova star , including its helium core , is larger , and that has a dramatic effect on the compactness ( fig . [ fig : f1 ] ) . there are also other less dramatic , but important effects . low metallicity implies a smaller initial helium mass fraction ( and more hydrogen ) . the final helium core mass is sensitive to this and is reduced . nitrogen has an abundance just prior to helium ignition that is directly proportional to the initial metallicity and `` nitrogen burning '' by @xmath16n(@xmath17f(@xmath18o is an important exoergic , convective phase in the star s life that precedes helium burning . low metallicity affects the energy generation during hydrogen shell burning by the cno cycle , and this affects the boundary conditions for the helium core . low metallicity also affects the opacity , and the combined effects of opacity , energy generation , and mass loss determine if the star is a red supergiant or a blue one when it dies . this is especially true for the non - rotating stars studied here . the more compact radiative structure of a blue supergiant envelope places greater surface boundary pressure on the helium core therein and can affect its evolution . to illustrate the systematics of compactness in stars of low metallicity , we include here a previously unpublished set of non - rotating presupernova stars by heger and woosley with metallicity of 10@xmath11 @xmath10 . these used the same physics as the s series , but had an initial composition of 76% hydrogen and 24% helium with only a trace , 10@xmath11 solar , of heavier elements . the masses of these stars were 10 to 95 @xmath0 , though only a subset , 15 - 65 @xmath0 , is considered here . this restricted set , called the u series , contains 86 models with varying mass increments ( 0.2@xmath0between 15 - 25@xmath0 , 0.5@xmath0between 25 - 35 @xmath0 , 1@xmath0between 35 - 45@xmath0 and 5@xmath0between 45 - 65 @xmath0 ; ( table 1 ) . other details of these models not related to their compactness will be published elsewhere . are shown for the s - series ( solar metallicity ; thick line ) and u - series ( 10@xmath11 solar metallicity ; dashed ) stars of mass less than 30 @xmath0 . notice a slight shift of @xmath5 near the first `` bump '' . bottom : same plot for the u - series ( dashed ) and sh - series ( continuous ) models for masses over 30 @xmath0 . [ fig : f3],scaledwidth=48.0% ] . below 35 @xmath0 , the core masses are not very sensitive to metallicity . above 35 @xmath0 , the helium core is somewhat larger for the higher metallicity stars reflecting a more active hydrogen - burning shell during helium burning . [ fig : f4],scaledwidth=48.0% ] the compactness of these low metallicity models is shown , along with those from s and sh series , in fig . [ fig : f3 ] . without mass loss , a much larger range of helium and co - core masses is accessible ( figure 4 ) . one sees a continuation of the overall increase in compactness parameter with increasing mass all the way up to 65 @xmath0 , but with new features . the compactness rises rapidly above 30 @xmath0 to a broad peak around 40 @xmath0 and then , following a dip at 50 @xmath0 , resumes its rise . all of the stars above 30 @xmath0 , with no mass loss , will probably be very difficult to explode by any solely neutrino - powered mechanism . mass fraction at the time of core carbon ignition is shown for s ( thick upper curve ) , sh ( dashed ) and u ( thinner lower curve ) stars . note that the carbon mass fraction is smaller in stars that ignore mass loss and that the sh series tracks the u series very well . [ fig : f5],scaledwidth=35.0% ] below 30 @xmath0 , the compactness curve for the u series shows non - monotonic structure with peaks qualitatively similar to those observed for the s series in figure 2 , but with an offset of about a solar mass . the offset is most pronounced in the range 20 - 25 @xmath0 where the compactness starts to rise in the u series for a lower value of main sequence mass . the main cause of this shift is the lower mass fraction of carbon produced by helium burning in the u stars ( fig . [ fig : f5 ] ) . that value , in turn , affects the critical mass where carbon ceases to burn convectively in the star s center ( [ sec : s3.1 ] ) , shifting it to lower values , which causes the compactness curve to rise earlier ( [ sec : s3 ] ) . thus end their lives as blue supergiants . [ fig : f6 ] , scaledwidth=35.0% ] this variable nucleosynthesis of carbon for the s and u stars results from differing amounts of helium being mixed into the convective helium core as it grows during the final stages of helium burning . in fact , the carbon mass fractions for s and u stars in this mass range are very similar half way through helium burning , but as the helium mass fraction declines below 10% , the u stars are more effective at bringing additional helium in from their outer regions . each helium convected down turns a carbon into an oxygen . the u stars have somewhat higher helium core luminosities at this point and the structure at the outer edge of the helium convection is also influenced by surface boundary pressure from the hydrogen envelope which is different for red and blue supergiants . below 30 @xmath0 , the u - series stars are blue supergiants while the s - series stars are red ones ( fig . [ fig : f6 ] ) . the pressure at the edge of the helium core does not decline as steeply in a blue supergiant and the entropy barriers inhibiting convection in the outer helium core are slightly reduced . this makes the growth of the helium convective core easier . convective dredge up in a red supergiant could also possibly reduce the helium core mass , though this effect seems to be small in the present models . with rotation , many of these lighter u - stars would also be red supergiants in the end because of primary nitrogen production at the hydrogen - helium interface , and the effect would be diminished . above 30 @xmath0 the structure of the sh and u stars are very similar because both are red supergiants . in the past , it has sometimes been assumed that the helium core mass uniquely determines the presupernova structure of a star and hence that structure would be roughly independent of the metallicity for two stars that made the same helium core mass . while this is qualitatively true , the figures here show that this assumption is not very accurate for supernova progenitors below 30 @xmath0 . [ sec : s2.3 ] with the new surveys , it is possible to address a point of possible concern - the choice of mass ( 2.5 @xmath0 ) and time ( `` presn model '' ) for evaluating the compactness in equation [ eq : compact ] . the full evolution of a massive star to the point that its iron core collapses and possibly powers an explosion is being characterized by a single number here . what motivates the choice of this particular point in space and time ? @xcite explored the effect of evaluating @xmath19 at different fiducial masses and concluded that 1.75 @xmath0 might be a better discriminant of explosion characteristics rather than the 2.5 @xmath0 chosen here and in @xcite . the time of core bounce rather than initial collapse ( `` presn '' ) also seemed a more relevant time for its evaluation . recently , @xcite also examined the choice of 1.75@xmath0 , since the early neutrino signal is more sensitive to the structure around the neutron star . obviously smaller choices than 1.75 @xmath0 would not be sensible since they often lie _ within _ the collapsing iron core itself , and depend on different physics that occurs after that core has already reached high density , but what about values in between ? to address these questions , the collapse of the s series was continued until the central density reached @xmath20 g @xmath21 . this is about 100 times greater than the central density of presupernova s - series stars . beyond this point , neutrino trapping and , ultimately , the nuclear equation of state , would influence the dynamics ( fig . [ fig : f7 ] ) . as expected , @xmath19 rises with increasing central density since a smaller radius encloses the same mass . it also rises as the mass chosen for its evaluation decreases , since the average density enclosed by r increases faster than r@xmath22 . running kepler well beyond @xmath23 g @xmath21 to nuclear density gives values of questionable accuracy , but strongly suggests that very little further evolution will occur in the compactness so long as the sampling mass remains greater than 1.75 @xmath0 . as the figure shows , the compactness evaluated at 2.5 @xmath0 changes very little during the collapse from the presupernova star to high central density . at that larger radius , the hydrodynamical response time is longer than the time for the denser part of the core to collapse . ( red ) , and later when the central density reaches @xmath24 g @xmath21 ( blue ) . each pair of curves represent @xmath19 evaluated at the same time in evolution , but at different points in fiducial mass : 1.75 @xmath0(top ) , 2.0 @xmath0 ( middle ) and the default 2.5 @xmath0 ( bottom ) . little evolution occurs during the collapse from about @xmath25 g @xmath21 to @xmath26 g @xmath21 if the compactness is evaluated at 2.5 @xmath0 , but for smaller fiducial masses the contrasts seen at 2.5 @xmath0 are significantly amplified with time . [ fig : f7],scaledwidth=35.0% ] substantial variation does occur , however , for smaller sampling masses . smaller fiducial masses show a non - linear amplification of the structure in the compactness curve with time . large values get much larger than adjacent smaller ones , suggesting the development of islands of stars that may be hard to blow up . this is due to the tendency of high density regions to collapse faster under the influence of their own gravity and is particularly apparent for stars in the 18 to 22 @xmath0 mass range . this may account for some of the variability in outcome seen by @xcite for supernovae in this mass range . the robustness of @xmath5 evaluated at 2.5 @xmath0 , however , and its strong correlation with the compactness evaluated at smaller masses suggests that we can continue to use our standard choices of time and mass for its evaluation . it should be kept in mind , however , that structures that seem small in some of the plots for presupernova stars may become amplified by the further collapse . [ sec : s3 ] surveys of stellar evolution find a complex , non - monotonic behavior for the compactness parameter as a function of main sequence mass and metallicity . why is this so ? why does nt the compactness vary smoothly with mass as it would in a polytrope with a single index ? four factors drive the development of the compactness profile for massive presupernova stars . one obvious effect is the contraction of the core to higher densities in order to burn heavier fuels . this contraction increases the average density inside 2.5 @xmath0 and causes @xmath5 to grow with time . another is the tendency of lighter stars to have lower entropy cores and be more degenerate . degeneracy is responsible for the transition between stars that make planetary nebulae and those that make supernovae around 8 @xmath0 , but the effects of degeneracy on the post - carbon burning evolution continue to be important throughout the entire mass range studied here . third , as has been noted previously @xcite , is the disappearance , around , 20 @xmath0 , of central convective carbon core burning . above 20 @xmath0 , carbon fuses away without contributing a large excess of energy generation over what neutrinos are carrying away in the center of the star . model , a typical supernova mass , ( top : ) from carbon burning until silicon shell ignition , and ( bottom : ) from the end of first oxygen shell burning until 5 minutes before collapse . only the inner 4 @xmath0 is plotted and the time axis is plotted as the logarithm of time until iron core collapse ( @xmath27 ) measured in years . cross hatched regions indicate convection and integer numbers annotate convective carbon burning shell episodes . the convective carbon core burning starts about 1000 years before death and is followed by three episode of carbon shell burning . oxygen burning starts about two years before death and is preceded by a brief stage of neon burning . silicon burning is the last convective episode in the center . at silicon depletion , no oxygen shell is active , only a carbon shell at about 1.9 @xmath0 . shortly afterwards though , the third oxygen shell ignites in the incompletely burned ashes of the second shell . this final vigorous shell merges with the carbon and some neon burning shells shortly before the star dies . the operation of these shells , especially the ones between 1.5 and 2.5 @xmath0 , affects the compactness parameter.[fig : f8 ] , scaledwidth=48.0% ] fourth and frequently overlooked ( though see barkat 1994 ) are the effects of convective carbon and oxygen shell burning . lighter stars can have three or more carbon burning shells in addition to core carbon burning ( fig . [ fig : f8 ] ) . as these stars evolve , their central regions become increasingly degenerate , especially after carbon depletion in the star s center , and , by oxygen ignition , the concept of a chandrasekhar mass has some approximate meaning , especially for stars lighter than 30 @xmath0 . shells that burn outside the effective chandrasekhar mass , roughly 1.7 @xmath0 depending on thermal corrections , but inside the point where the compactness is measured at 2.5 @xmath0 , will considerably modulate the compactness parameter . [ sec : s3.1 ] fig . [ fig : f9 ] shows that @xmath5 remains nearly independent of the main sequence mass until after carbon ignites in the core . prior to this time , the fiducial point at 2.5 @xmath0 lies well within a much larger star or helium core that can be characterized by a single polytropic index . the tendency of @xmath28 to decrease as @xmath29 , as given by equation [ eq : poly ] , is offset by the slight increase of the burning temperature with mass , resulting in a nearly flat curve . as a function of main sequence mass for the s series models up to the point when the carbon mass fraction decreases below 1% in the center of the star ( `` c.dep '' ) . at each successive stage , the star has contracted to higher density and the average curve for @xmath5 rises . below about 20 @xmath0 carbon burns convectively in the core and this reduces @xmath5.[fig : f9 ] , scaledwidth=35.0% ] by the time carbon has disappeared from the center of the star though , things have started to change . a pronounced dip in @xmath5 develops for the lower mass stars , and the curve shows an abrupt , small rise at 20 @xmath0 . for current code physics and solar metallicity , 20 @xmath0 is the mass below which carbon burns convectively ( exoergically ) at the stellar center as opposed to radiatively ( endoergically ) . convection brings additional fuel into the burning region increasing the effective supply by approximately the ratio of the convective mass to the mass of the burning region . because of the high temperature sensitivity of the carbon fusion reaction , the energy generating region is small , so the enhancement is significant . during this longer time , neutrino losses carry away both energy and entropy , not only from the carbon convective core , but from the hot , overlying helium - rich layers supported by it . this loss of entropy exacerbates the natural tendency of lighter cores to have greater degeneracy and accelerates the development of a compact , white dwarf - like core structure - a dense degenerate core surrounded by a much less dense extended envelope . , averaged over the inner 1 @xmath0 of the star is shown as a function of initial mass at the times of ( top : ) carbon ignition and core carbon depletion , ( bottom : ) oxygen ignition , oxygen depletion and silicon depletion . very negative values of @xmath30 correspond to a non - degenerate gas and large positive values imply degeneracy . the cores of the lower mass models grow much more degenerate due to neutrino losses between carbon depletion and oxygen ignition . the structure around 20 @xmath0 in the curve for oxygen ignition reflects the brief appearance of strong , deeply - sited carbon convective shells in this mass range ( [ sec : s3.2 ] and fig . [ fig : f11 ] ) . these degeneracy curves at oxygen ignition and depletion are nearly mirror images of the @xmath5 curves at these times ( fig . [ fig : f13 ] ) showing the key role played by degeneracy in determining the compactness.[fig : f10],scaledwidth=48.0% ] as fig . [ fig : f10 ] makes clear , _ central _ carbon burning , by itself , is not the whole story though . most of the increase in central degeneracy occurs after carbon has been exhausted in the star s center . it is during the period between central carbon depletion and oxygen depletion that neutrinos cool the core appreciably and , for masses below about 30 @xmath0 , give it a white dwarf - like structure . during this long cooling - off time , contraction of the inner core is frequently held up by two , or even three vigorous carbon burning shells ( fig . [ fig : f11 ] ) , the first igniting shortly after carbon core depletion typically about 0.5 @xmath0 from the center . as we shall see , the presence and location of these shells is strongly correlated with how carbon burns in the center , so 20 @xmath0 remains a critical mass . it is ultimately the timing and location of these carbon convective shells , however , not just central carbon burning , that account for the structure developed during and after oxygen burning . greatly affects the location and extent of all three convective shells . for slightly larger masses , the first convective shell ( `` 1 '' ) disappears . after that mass carbon burns from the center radiatively out to the base of the former second shell . the location of the third shell is drawn as a dashed line when it that shell exists for less than a year at the end of the life of the star . ( bottom : ) difference between starting times and core collapse for various convective episodes ( oxygen burning - green and carbon burning - blue ) as a function of initial mass . [ fig : f11],scaledwidth=48.0% ] [ sec : s3.2 ] fig . [ fig : f12 ] shows the evolution of the compactness parameter for the s series stars from the time carbon is depleted at the center of the star until oxygen is similarly depleted . this is a critical period when the major features of the compactness emerge ( fig . [ fig : f7 ] ) . , as a function of initial mass at the time of central carbon depletion ( lower curve ; see also fig . [ fig : f9 ] ; oxygen ignition ( central temperature equals 1.6 gk ) ; and oxygen depletion ( central oxygen mass fraction equals 0.04 ; top curve ) . non - monotonic behavior caused by carbon burning shells is clearly evident above 18 @xmath0 by the time oxygen ignites , and a broad peak in @xmath5 is emerging around 18 - 24 @xmath0 with a `` notch '' between 19 and 21 @xmath0 . note the strong anti - correlation of @xmath5 with degeneracy below 22 @xmath0 ( fig . [ fig : f10 ] ) especially around 20 @xmath0 . [ fig : f12],scaledwidth=35.0% ] the typically low values for @xmath5 below 18 @xmath0 result from the operation of the first two convective carbon shells ( fig . [ fig : f11 ] ) . while the star is supported by these shells , its inner regions radiate neutrinos and become cool and degenerate . the rapid variation of @xmath5 in the range 18 to 21 @xmath0 reflects the response of these three shells to the disappearance of the convective core . when that happens , _ the first carbon burning shell disappears _ , and the second and third convective shells move inwards dramatically in response to its loss . these events do not happen simultaneously . the center of the star ceases to be convective at 20.3 @xmath0 , but the first shell has already gone out at 19.9 @xmath0 . the second and third shells inwards until this mass is reached , and then move outwards for bigger stars . the offsets in these masses and the rapid motion , first in , then out , of shells two and three underlie some of the complex structure in degeneracy and compactness seen around 20 @xmath0 . is plotted in each case and the format is similar to fig . [ fig : f8 ] , but here all stars , except the 20@xmath0 model ( which has a very tiny convective core ) , burn carbon radiatively in their centers . prominent in all figures is a powerful carbon burning shell igniting roughly 10 years before the star dies . note the migration outwards of the base of this shell as the mass increases . the shell that ignites farther out supports the star while the inner core emits neutrinos and becomes increasingly degenerate . the shorter kelvin - helmholtz time for the larger carbon - depleted core in the heavier star also leads to oxygen ignition at a substantially earlier time . the 20@xmath0 and 23 @xmath0 models end up with a large value of @xmath5 , while 20.5@xmath0 and 28 @xmath0 models give a smaller value . see also fig . [ fig : f13 ] . [ fig : f13],scaledwidth=48.0% ] from 21 to 30 @xmath0 , the compactness is most sensitive to whether , when oxygen core burning ignites , there is , or very recently has been an active carbon convective shell inside of about 2.0 @xmath0 ( fig . [ fig : f13 ] ) . an active shell supports the overlying star and relieves pressure on the oxygen core , allowing it to evolve as if it had a smaller effective mass . smaller oxygen cores burn longer , emit more neutrinos , cool more effectively , and become more degenerate . the relevant carbon shell can either be the third ( for 18 - 20 @xmath0 ) or the second ( for 21 - 30 @xmath0 ) shell . if that shell is inside 1.4 @xmath0 , the oxygen core lacks sufficient mass to ignite @xcite , so the most important mass range for the convective shell is 1.4 to 2.0 @xmath0 . the location of the boundaries of these convective carbon shells are given in fig . [ fig : f11 ] . the third convective shell is given there as a solid curve only for the mass range where it has a lifetime longer than one year . shorter lived shells do not greatly affect the evolution prior to oxygen depletion . based upon this criterion , the third shell only has a major effect in the range 19 to 21 @xmath0 . for heavier stars , the second shell dominates ( note that , for the sake of continuity , we persist in calling it the `` second '' shell even though the `` first '' convective shell has been replaced by radiative burning ) . this behavior is illustrated by the convective history during carbon and oxygen burning for stars of four masses : 20.0 , 20.5 , 23.0 and 28.0 ( fig . [ fig : f13 ] ) . none of these four stars experience central convective carbon core burning or `` first shell '' burning ( in 20 @xmath0 there is a very tiny convective carbon core ) . the relevant carbon shells are thus `` 2 '' and `` 3 '' , and they are labeled as such in the figure . at oxygen depletion , the 20.5 @xmath0 and 28.0@xmath0 models have small values of @xmath5 ( 0.13 and 0.15 , respectively ) , while the 20.0 and 23.0 have larger ones ( 0.19 and 0.21 ) . in the 20 @xmath0 model , shells 2 and 3 complete their major burning about a year before oxygen ignites in the stellar center . shell 3 is located at 1.32 @xmath0 , which is too small to allow oxygen to ignite . later , following a brief , inconsequential `` blip '' of convective neon burning , oxygen burning ignites in a core devoid of any nuclear energy sources out to more than 3 @xmath0 . the entropy is higher and the degeneracy is less at both oxygen ignition and depletion ( fig . [ fig : f10 ] ) . oxygen burning ignites late , with less than a year remaining in the star s life . ultimately the star dies with a relatively large compactness parameter , 0.26 . the 20.5 @xmath0 model , though only slightly different in mass , has a very different evolution . the third carbon shell now has its base at 1.62 @xmath0 which is sufficiently large for oxygen burning to ignite before carbon shell burning is over . when oxygen does ignite at the center , the resulting expansion extinguishes the carbon shell , leaving unburned carbon in the region outside 1.62 @xmath0 . some neutrino losses have already occurred in the oxygen before it ignited , but over the course of its burning , which commences much earlier than in the 20.0 @xmath0 model , more cooling occurs and the core becomes very degenerate . by oxygen depletion , the compactness parameter has a value of 0.13 , which declines still further to 0.10 when the star dies . for 23.0 @xmath0 , the evolution resembles that of the 20.0 @xmath0 model more than it does the 20.5 @xmath0 model . by now , the third carbon shell is igniting so late as to be unimportant . shell 2 , burning at 1.33 @xmath0 , however , again delays oxygen ignition to a late time . the oxygen burns in a core that , at the time has essentially no active burning shell except the helium shell far above it ( off - scale ) . it is easier for the layers above the core to contract since there is a time with no active burning , neither core nor shell . once it ignites , the oxygen core burns rapidly and at low density , producing a less degenerate silicon core . by oxygen depletion , the compactness parameter is 0.21 which rises to 0.42 in the presupernova star , the largest compactness of any model under 30 @xmath0 . for 28.0 @xmath0 and nearby masses , the evolution resembles that of the 20.5 @xmath0model , but with no third shell . the second shell has now moved out to 1.91 @xmath0 which is enough to allow core oxygen burning to ignite . there is always support by an active burning front , reducing the contraction of the outer core . oxygen burning lasts longer and occurs in a core of smaller effective mass . the compactness at the end of oxygen burning is back down to 0.15 , though this rises to 0.28 in the presupernova star . beyond 28 @xmath0 , the second carbon shell continues to move out and the effective core mass when it ignites oxygen grows . when the shell crosses 2.5 @xmath0 at 30 @xmath0 , the compactness parameter rapidly increases once more . this marks the end of the epoch of carbon shells . all heavier stars will have large compactness parameters ( fig . [ fig : f3 ] ) with relatively small variations introduced by the _ oxygen _ burning shells . the location and strength of the carbon shells prior to oxygen depletion thus sculpt a highly variable result out of what was a comparatively smooth curve at carbon depletion . in some sense , 20.5 @xmath0 and 28 @xmath0 stars resemble each other more at death than do 20.5 @xmath0 and 20 @xmath0 stars . above 30 @xmath0 , the carbon shells lose their influence . after oxygen finishes burning in the center , powerful convective _ oxygen_-burning shells develop that can also modify the structure either by themselves , or by interacting with existing carbon shells . the continued evolution serves chiefly to amplify the features already present at oxygen ignition ( fig . [ fig : f14 ] ) . having developed a degenerate core , further shell burning pushes matter out at 2.5 @xmath0 and reduces @xmath5 . the location of the first oxygen shell at silicon depletion correlates tightly with @xmath5 . its base is located between 0.8 and 1.2 @xmath0 and is set by the extent of the oxygen convective core . its extent correlates tightly with the location of its base . similarly , the second oxygen shell , when there is one , sets atop the extent of the first . the farther out these shells and the later their ignition , the more extended the core , which results into a shallower density gradient around the iron core later on ( higher @xmath5 ) . the peak in @xmath5 in fig . [ fig : f14 ] , around 23 to 25 @xmath0 , is enhanced by the migration outwards in mass of this oxygen burning shell . unlike the carbon burning shells , however , the outwards movement of the oxygen burning shell is not monotonic in mass , since the extent of the oxygen core varies due to the effect of previous carbon shells . as it recedes inwards from 24 to 28 @xmath0 , the @xmath5curve declines , leading to a pronounced dip at 28 to 30 @xmath0 . above 30 @xmath0 , the oxygen burning shells move out rapidly . ) and silicon depletion . the tendency of existing features to be amplified by the contraction , as seen previously in fig . [ fig : f13 ] , continues . bottom : the location of the outer extent of the oxygen convective core , which is also the base of the first oxygen shell is given as the lower curve labeled `` o.core '' , while the extent of the first oxygen shell is the curve labeled `` 1st o.shell '' . note the correlation between shell locations and the compactness . [ fig : f14],scaledwidth=45.0% ] [ sec : s3.4 ] fig . [ fig : f15 ] shows that , while the peak in @xmath5 at 23 @xmath0 continues to grow due to core contraction and the operation of the oxygen shells , the principal qualitative change in @xmath5 during this final stage of the star s life is a `` chopping up '' of the peak that existed at silicon depletion around 20 @xmath0 into finer structures . given that these structures will be amplified in the collapse ( fig . [ fig : f7 ] ) , they may yet affect the outcome of the explosion . , as a function of initial mass at silicon depletion and for the presupernova star . the peak at 24 @xmath0 grows due to core contraction and fine structure is imprinted between 19 and 22 @xmath0 by a strong oxygen burning shells in the last few days of it s life . [ fig : f15],scaledwidth=35.0% ] this rapid variation for a narrow range of masses results from an almost random decision by the star whether to burn oxygen or silicon in a shell first , the decision depending upon fine details of all that happened before . fig . [ fig : f16 ] shows the history of the convection following central silicon depletion in two stars of the s series with masses 20.1 and 20.2 @xmath0 . note especially the timing of the ignition of silicon shell burning compared with the growing extent of the strong oxygen burning shell with a base at 1.71 @xmath0 in the 20.1 @xmath0 star and 1.59 @xmath0 in the 20.2 @xmath0 star . in both stars this convective oxygen shell is very powerful , and eventually grows to include both the carbon and neon burning convective shells in one large aggregate shell shortly before the star dies . the oxygen shell in the 20.1 @xmath0 star drives a similar linking of convective layers and creates a large positive energy generation rate , but only _ after _ the silicon shell has already burned out and the star is close to collapsing . in the 20.2 @xmath0 star , on the other hand , the strong oxygen shell ignites _ before _ silicon shell burning and operates for a much longer time . ( top ) and 20.2 @xmath0 ( bottom ) stars of s series . each plot shows the evolution inside inner 4.5 @xmath0 during the -2 to -6 in the log of time until collapse . this last few days of evolution roughly corresponds to the time - evolution from the start of silicon ignition in the core until the presupernova stage . shaded regions describe convection . notice in 20.2@xmath0 , where the @xmath5(measured at presupernova ) is low , the silicon ignites int the core earlier and the shell c - o burning is very different . [ fig : f16 ] , scaledwidth=48.0% ] in the 20.1 @xmath0 star , oxygen shell burning initially generates , only a limited convection zone that slightly alters the structure of the star before silicon shell burning ignites and slows the contraction . in the 20.2 @xmath0 star , the oxygen shell becomes strong first and has already engulfed a large fraction of the co core before the silicon shell ignites . the 20.1 @xmath0 star ends up with a high @xmath5 , the 20.2 @xmath0 star with a low one . the strong oxygen shell in the latter pushes matter out to larger radii and leaves a less extended structure inside . .[fig : f17],scaledwidth=35.0% ] this sensitivity of the presupernova compactness to the timing of oxygen and silicon shell burning suggests that a correlation might exist between the time from silicon core depletion to death and @xmath5 . this is indeed observed ( fig . [ fig : f17 ] ) . the longer the strong oxygen shell supports the star after the initial iron core has formed from central silicon burning , the smaller @xmath5 for the presupernova star . these energetic , merged oxygen , neon and carbon burning shells during the last hours of a massive star s life are a robust feature seen in many models of stars above about 15 - 20 @xmath0 , both here and in previous studies @xcite . their study would be an interesting , though perhaps challenging topic for 3d simulation . if such small differences in the central structure of the star at silicon core depletion can lead to major changes in presupernova structure , even for stars differing in mass by only 0.1 @xmath0 , how sensitive might the results be to other `` hidden variables '' , such as the time step criteria and zoning . [ fig : f18 ] shows that , for some masses , a substantial variation in outcome can also be caused by varying these variables . the location of the oxygen shell is a sensitive function of the many convective episodes that went on in the same star during its earlier evolution , and to the physics used to treat that convection . and b : 20.1 - 20.5 @xmath0 . 1a , b - different zoning : default ( thick ) and @xmath31 of default ( dot - dash ) . 2a , b - time step : default ( thick ) , half ( dot - dash ) and double ( thick - dashed ) . [ fig : f18 ] , scaledwidth=48.0% ] as previously noted by @xcite , this sensitivity to small changes is especially strong for the stars in the 19 to 21 @xmath0 range . to quantify this behavior better , a subset of the s series was calculated using the same code , compiler , and physics , but slightly altered zoning and time step criteria . in each case , the standard calculations were recomputed using 50% finer resolution ( i.e. , @xmath31 the default zone size ) , and , in separate calculations , the default time step was multiplied by @xmath32 and @xmath33 . even these relatively small changes in operational parameters were frequently sufficient to provoke the star into two different final states . for comparison , we studied two distinct regions of main sequence masses , 17.1 - 17.5 @xmath0- region `` a '' , and 20.1 - 20.5 @xmath0- region `` b '' . as shown in figure 2 , the observed variations of @xmath5 in these two regions are quite different . panels 1a and 2a of fig . [ fig : f18 ] show that the solutions in region a are comparatively robust . decreasing the time step and or changing the zoning does little to alter the final values of @xmath5 . increasing the time step by a factor of @xmath33 does cause some mild variation , @xmath34 . panels 1b and 2b of fig . [ fig : f18 ] , however , show a much more pronounced sensitivity near 20 @xmath0 . the resulting values of @xmath5seem to be bimodal . small variations in mass , zoning , or time step can send the star down one path or the other . typical models in the surveys used 20,000 time steps to reach the presupernova stage and employed approximately 1000 mass shells of variable thickness . it is certainly feasible to double or even quadruple both the resolution and time steps , but the values used here are already conservative . to test the effect of finer time steps , the 21 s - series models from 19 to 21 @xmath0 were rerun with twice as many time steps . while @xmath5 for some individual stars did change significantly , the overall appearance of the pattern , including its rapid , but bounded variation between @xmath5 = 0.1 and 0.25 , did not . above 30 @xmath0 , if mass loss is neglected , the structure of the core becomes much simpler . the carbon shells have moved far out and no longer affect the solution . the oxygen core is larger and the oxygen shells farther out . because of the larger mass , the degeneracy is reduced and the compactness is always large . the most notable feature above 30 @xmath0 is a pronounced dip in @xmath5 around 50 @xmath0 . the dip is not large , but is seen in both the u - series and sh - series stars , albeit at slightly different masses ( fig . [ fig : f3 ] ) . it is also present in the compactness plot for the bare co cores studied later in [ sec : s5 ] . this behavior can be traced to the presence of a strong , extended convective oxygen burning shell during the post - silicon burning evolution of stars over about 50 @xmath0 . the lighter stars lack this shell ; the heavier ones have it . starting at 50 @xmath0 for the u - series , this shell is present at silicon depletion with a base at 1.8 @xmath0 . moving to heavier masses , the shell grows larger and its base moves outwards , reaching 2.5 @xmath0 at 65 @xmath0 . there is a sharp density decline at the base of the shell and because of this migration outside the fiducial point for measuring @xmath5 , the compactness parameter rises again as the star mass passes about 60 @xmath0 . whether this shell is present or not depends upon the timing of silicon core ignition and oxygen shell burning . recall the key role played by the carbon shell and oxygen ignition for stars in the range 21 - 30 @xmath0 ( [ sec : s3.2 ] ) . when the carbon shell was situated far enough out , oxygen burning would ignite before carbon shell burning was done with major consequences for the compactness . here , the oxygen shell plays the role of the former carbon shell . if it burns far enough out , the silicon core can ignite earlier . in this case , however , igniting silicon does not blow out the oxygen shell . it persists until the end . and has little effect on @xmath5 in this mass range . the bases of the 1st shells are not plotted for clarity , but they almost always perfectly match with the extents of the cores . notice , how the silicon core size responds as the oxygen core overgrows the silicon shell near 45 - 50 @xmath0 . [ fig : f19],scaledwidth=35.0% ] fig . [ fig : f19 ] shows the locations of various silicon and oxygen burning episodes as a function of mass for the sh - series models . though it lacks the time dimension of a full convective history plot , the figure shows that the size of the oxygen convective core increases monotonically with mass for these heavy stars . where the oxygen convective shell ignites is pegged to the extent of that convective core . at about 50 @xmath0 , the extent of the oxygen convective core passes the value needed to ignite silicon core burning . an oxygen shell ignites at the edge of that depleted core at about 1.8 @xmath0 . this is sufficiently large that silicon core burning also ignites , and after that , silicon shell burning , but both ignite without putting out the oxygen shell . indeed , the oxygen shell persists until the iron core collapses and forces the compactness parameter down . above 50 @xmath0 , the compactness parameter continues to increase until about 70 @xmath0 , where the onset of the pulsational pair instability becomes important @xcite . [ sec : s4 ] studies of massive stellar evolution by different groups yield substantially different presupernova structures and nucleosynthesis . a key quantity is the co core mass . the yield of elements heavier than helium , for an energetic explosion , is the difference between the mass of the collapsed remnant ( neutron star or black hole ) and the co core mass . as we shall see , the compactness is also quite sensitive to this quantity . table 2 shows the disparate values for the co core mass obtained by various groups studying the problem . for example , a 20 @xmath0 star may end up with a co core of anywhere from 2.71 to 6.59 @xmath0 for various choices of convective physics and assumptions about rotation . even for non - rotating stars the range is 2.71 to 4.99 @xmath0 . why are the values so different ? might using the co core mass as a discriminant for the presupernova compactness alleviate some of the sensitivity to uncertain code physics ? the effect of rotation is not surprising . rotation adds new mixing processes ( e.g. * ? ? ? * ) that increase the masses of both the helium and co core . shear mixing in differentially rotating stars can also erase some of the sensitivity to the treatment of semiconvection and convective shell boundaries . the large residual range of co core masses in non - rotating stars reflects these uncertainties in the treatment of convection and warrants discussion . uncertainties in the treatment of convection for non - rotating massive stars can be lumped into three categories : a ) the use of time - dependent mixing length theory ( mlt ) to describe the overall transport ; b ) uncertainties in semiconvection , i.e. , what to do about stellar regions that are unstable to convection by the schwarzschild criterion , but not the ledoux criterion ; and c ) the treatment of convective boundaries ( i.e. , overshoot and undershoot ) . a detailed review is beyond the scope of this paper . most modern studies use time - dependent mlt , if only for the lack of a viable , implemented alternative . semiconvection and convective overshoot can sometimes have similar effects since both allow convective regions on their tops and bottoms to grow smoothly when circumstances warrant it , and prevent the unrealistic fracturing of convective regions into multiple concentric shells . semiconvection additionally slowly mixes regions that , owing to composition barriers , would not have mixed if the ledoux criterion were strictly applied . a relevant case in point is the convective helium core . it has been known for some time that the co core mass is quite sensitive to the efficiency of semiconvection during helium burning @xcite . in situations where semiconvection is very small ( or zero ) , a mathematical instability can develop that leads to the bifurcation of the helium convective core and the production of an unusually small co core . in an algorithm that looks at the composition to determine convective instability ( i.e. , ledoux criterion ) , a single zone that is marginally unstable might temporarily become flagged as stable due to a trivially small increase in the mean atomic weight ( helium plus carbon and oxygen ) in the zone beneath . if this zone is flagged as stable against convection , burning during the next time step will increase the atomic weight farther in the region that is still convectively coupled to the burning at the center of the star without raising it in the outer core . this makes convection between the two shells even more difficult . the helium convective core splits permanently into two pieces . in the outer one , helium never burns to completion and the resulting co core for the star is small . if enough semiconvective mixing ( or rotation ) is included though , the splitting of the the convective core does not happen . in fact , considering the real situation in three dimensions , this numerical instability seems unphysical and should be avoided . overshoot mixing can also help bridge this artificial segregation . it also leads to an increased growth of the helium convective core at the end of helium burning that affects not only the co core mass , but the carbon mass fraction when carbon burning ignites . simulations by meakin and collaborators ( e.g. * ? ? ? * ; * ? ? ? * ) confirm that some type of mixing will always take place at convective boundaries defined by either the ledoux or schwarzschild criterion , although the exact formulation of this mixing remains uncertain . the treatment of convective overshoot and semiconvection in kepler has been discussed previously @xcite . semiconvection is parametrized by a diffusion coefficient that is manufactured from the local convective diffusion coefficient and the radiative one : @xmath35 where @xmath36 is the convective diffusion coefficient the zone would have had based upon the schwarzschild criterion , and @xmath37 is a free parameter that multiplies the radiative diffusion coefficient , @xmath38 . usually @xmath39 so that @xmath38 dominates the transport . the actual value of @xmath37 is not known , or even that such a simple formula captures the essential features of semiconvection . in past surveys using kepler , @xmath37 has usually been taken to be 0.1 , and , unless otherwise specified , that is the value used in this paper . in practice , this generally assigns the semiconvective diffusion coefficient a value of near 10% of the radiative one . overshoot mixing is presently treated very crudely in kepler . single zones at the top and bottom of regions flagged as unstable to ledoux convection are slowly mixed with a diffusion coefficient given by equation [ eq : semid ] . the convective diffusion coefficient is calculated based upon the schwarzschild criterion : @xmath40 with @xmath41 assumed to be a parametrized fraction , @xmath42 and with @xmath43 having a default of 0.01 . typically the convective diffusion coefficient so calculated is much greater than @xmath44 so that the mixing occurs on a time scale given by this quantity . this prescription lacks a physical basis , but at least allows convective zones the liberty of slowly growing into regions where mild entropy gradients would otherwise disallow mixing . to explore the sensitivity of the compactness and the co core mass to semiconvection and overshoot mixing , a subset of 16 solar metallicity models in the mass range of 15 to 30 @xmath0 was calculated multiple times using various settings for the both semiconvection and convective overshoot mixing multipliers . the masses studied were the integers from 15 to 30 and a total of 176 new models was calculated ( tables 1 and 2 ) . in these , the semiconvective multiplier , @xmath37 , was varied from 0.001 to 0.1 ( 16 models each with @xmath37 = 0.001 , 0.0025 , 0.005 , 0.0075 , 0.01 , 0.0125 , 0.015 , 0.0175 , 0.02 , 0.05 , and 0.075 in addition to the standard models with @xmath37 = 0.1 ) and are collectively called ss series . additionally , a set of 16 models was run with standard semiconvection ( @xmath37 = 0.1 ) , but zero overshoot mixing ( @xmath45 ) = 0 ( so series ) . all models were generated on the zero age main sequence and run until core collapse . the co core mass does not change significantly after carbon depletion , however , and presupernova values can safely be compared with other studies that stop at an earlier time . fig . [ fig : f20 ] shows that the compactness exhibits great deal of diversity depending on the efficiency of semiconvection . the mass resolution in these figures is significantly degraded by considering only 16 masses points between 15 and 30 @xmath0 , but the location of the peak in @xmath5 , when one exists , is clear . reduced values of @xmath37 give smaller co cores , more than spanning the space of published values . the carbon mass fraction at carbon ignition is also affected by the choice of convection physics , both in response to the altered co core mass and the mixing of helium at the outer boundary of the helium core that is altered when semiconvection is turned off . consequently the compactness , which is quite sensitive to both the co core mass and the carbon mass fraction , vary significantly . ) are shown in comparison with the s - series models ( top panel ) with @xmath46 . the s - series is plotted with the same @xmath47 increments as was used in ss series calculations . [ fig : f20],scaledwidth=48.0% ] for small reductions in @xmath37 , the peaks found for @xmath5 in the standard runs is robust . both the locations and amplitudes do not vary much . though not explored , we expect that small changes upwards of @xmath37 would have given similar results . when @xmath37 declines below 0.02 however , significant changes occur . with smaller @xmath37 , it takes a higher mass main sequence star to give a specific co core mass , so the peak that was at 23 @xmath0 shifts rapidly to higher values , ultimately leaving the grid . by @xmath37 = 0.01 , the peak is no longer clearly discernible in the mass range examined . continued reduction in @xmath37 ( bottom panel , fig . [ fig : f20 ] ) results in a compactness curve that loses most of its non - monotonic structure and increases steadily with mass . however , such small values imply a co core that is small compared with all previous studies in the literature ( table 2 ) . for @xmath37 = 0.001 , the 15 , 16 and 17 @xmath0 models closely resembled previous models calculated with kepler and other codes for stars with mass near 10 @xmath0 @xcite . oxygen ignition occurs off center in these models and propagates as a convectively bounded flame to the stellar center ( e.g. * ? ? ? a similar result was obtained for @xmath37 = 0.0025 for the 15 and 16 @xmath0 models . such stars are known to have a very steep density gradient surrounding the iron core , which might be a boon to those trying to blow up stars in a wider mass range , but contribute almost nothing to the nucleosynthesis of intermediate mass elements . were the observational limit on the maximum mass supernova to be pegged not far above 18 @xmath0 @xcite , models of this sort would be inconsistent with the theory of stellar nucleosynthesis . a model with 12 @xmath0 and @xmath37 = 0.001 failed even to ignite oxygen burning and would likely become an agb star with a degenerate neon - oxygen core and not a supernova . since this would be in contradiction with a large volume of observational data showing that stars down to about 8 @xmath0 do explode , we must consider values of @xmath37 much below 0.01 as unrealistic . turning off convective overshoot mixing in kepler also has a significant effect on the co core mass and carbon mass fraction , though not so extreme as that caused by large variations in semiconvection ( table 2 ) . [ fig : f21 ] shows the co core sizes and final presupernova @xmath5 of the so - series models compared with those from s series . the elimination of overshoot reduces the co core sizes ( top - panel ) , and as a consequence , the @xmath5 curve is shifted to higher masses . for example , central carbon burning shifts to being radiative at a higher mass compared to the s - series models with overshooting . . [ fig : f21],scaledwidth=48.0% ] these results , which imply that the success or failure of exploding a presupernova model of given mass will vary appreciably depending upon who calculated the model and the description they used for convection physics , are troubling . however , semiconvection and overshoot mixing are real physical processes that will eventually be better understood . some of the very small values for @xmath37 used here may not be physically realistic ( e.g. * ? ? ? * ) and the high threshold mass that they imply for making a supernova is inconsistent with observations . zeroing overshoot mixing is also unrealistic . bifurcation of the helium convective core probably does not occur . real stars rotate , and rotationally induced mixing reduces the sensitivity to semiconvection and overshoot mixing . we believe that our standard choices for @xmath37 and @xmath45 ( 0.1 and 0.01 ) are a good compromise for non - rotating stars since they avoid the unphysical splitting of the convective core , agree reasonably well with the results for rotating models ( table 2 ) , and are not presently at odds with any existing physical prediction . clearly this is an area where more work is needed . see @xcite and references therein for some recent insights . as a function of initial main sequence mass ( top ) and final co core mass ( bottom ) for the s - series models ( continuous ) , a subset of ss - series models with @xmath48 ( triangles ) and so - series models ( diamonds ) . though the definition of boundary of the co core inside a massive star is ambiguous , its mass can be a better discriminant of @xmath5 than the initial mass in most cases . [ fig : f22],scaledwidth=48.0% ] our results also suggest that @xmath5 , and presupernova structure in general , will be more similar for two models having the same helium and co core masses than for two that simply started with the same main sequence mass . carrying this to the extreme , the outcome of the explosion of two stars with the same co core mass will be more similar than for those having the same main sequence mass , but different co core masses . fig . [ fig : f22 ] shows this to be approximately true . the offset in peaks seen in fig . [ fig : f21 ] for the stars calculated with and without overshoot mixing is greatly reduced if @xmath5is plotted against co core mass rather than main sequence mass . similarly , the large discrepancy with the s series found for low values of semiconvection ( e.g. , @xmath49 ; fig . [ fig : f20 ] ) is significantly improved . note that the allowed range of co masses in the figure is much more limited for 15 to 30 @xmath0 stars if semiconvection is reduced significantly . just what constitutes the `` co core mass '' inside a massive star is ambiguous however , because there usually is no sharp decline of pressure at its edge ( [ sec : s5 ] ) . here we arbitrarily took the value inside of which the carbon mass fraction rose above 0.20 . differing values of the carbon mass fraction at carbon ignition due to changes in the initial composition or code physics must also be considered . even similar physics often has different implementations in various stellar evolution codes , so to further explore the sensitivity of the compactness curve to the way it is calculated , we repeated part of our study using a completely different code . mesa @xcite is a popular code for the study of stellar evolution which has been successfully applied to a large variety problems in astrophysics and is freely available . mesa has recently been applied to the calculation of presupernova models @xcite . it uses similar nuclear physics to kepler for energy generation during hydrogen through oxygen burning and a very similar equation of state and opacity tables . it can be constrained to use similar mass loss rates . here the red giant mass loss rates of @xcite was used in both codes : @xmath50 however , the two codes have quite different prescriptions for semiconvection and overshooting . given the previous discussion , those differences can be expected to cause changes that might be educational . mesa also uses different nuclear physics during silicon burning and iron - core formation that affect the post - oxygen evolution . in particular , mesa continues to use a 19 isotope network to describe the energy generation and changes in composition after oxygen depletion , while kepler uses a 125 isotope `` quasi - statistical equilibrium '' ( qse ) network @xcite . the latter is more stable and less prone to spurious temperature transients and is also able to follow accurately the appreciable changes in electron mole number that happen in these late phases . since the major features of the pre - supernova @xmath5(m ) curve , including the existence and location of peaks , have already been imprinted by oxygen depletion , we elected to compare the results of the two codes at that point , not for the pre - supernova star . this way , the post - oxygen burning differences are avoided and also the run time greatly reduced . mesa is a continually evolving code while kepler is more `` mature '' , and for this study we used mesa version 4930 . the stars studied are a subset ( table 1 ) of our s series , that is , non - rotating stars with solar metallicity . this version of mesa employed ledoux convection . for the mixing length parameter , @xmath51 , the free parameter @xmath52 was taken to be 2.0 , which is within the range found in the literature . in those zones flagged as semiconvective , the @xcite model was used . this calculates the diffusion coefficient as : @xmath53 where @xmath54 is the specific heat at constant pressure and the efficiency is tuned by the free parameter @xmath55 . we carried out a brief survey varying this parameter between @xmath56-@xmath57 , and found its effect to be rather complicated . for the masses tested , this range of @xmath55 gave a spread of about @xmath58 in the co core mass , @xmath47 difference in final mass , and a one order of magnitude change in radius . in general , for larger values ( @xmath59 ) the mesa results approached that of the pure schwarzschild case . the very low values , on the other hand , gave results that were very similar to one another , but quite different results were obtained using the larger values . the greatest sensitivity for the co core size was found to be between @xmath60 and @xmath61 . in our survey @xmath55 was taken to be @xmath62 as a midpoint between these two extremes , which is also well within the range used in literature ( e.g. * ? ? ? we also note that the work to eliminate or greatly constrain this parameter is underway @xcite . overshoot mixing in mesa is accomplished either by fully mixing the zones flagged as overshooting ( step overshooting ) or by applying an exponential cutoff to the diffusion coefficient calculated for convection in zones that are unstable by the ledoux or schwarzschild criterion . in our survey we have employed the exponential decay formalism , that is : @xmath63 where @xmath64 is the diffusion coefficient from mlt , @xmath65 is the height , and @xmath66 , the pressure scale height . the efficiency is then controlled by the free parameter @xmath67 ( see @xcite,@xcite for further details ) . , as a function of initial mass curves at oxygen depletion in the core for each series . [ fig : f23],scaledwidth=48.0% ] in general , the value of @xmath67 is quite uncertain as well @xcite . a value of 0.016 was used by @xcite for his studies of agb stars . since the stellar context here is quite different , we have performed some additional test runs . the effects of five different choices for @xmath67 ranging from 0 to 0.025 were explored ( table 2 ) . in the end , a value of 0.025 was selected for detailed study since it produced very similar co core sizes to the s - series models calculated using kepler(fig . [ fig : f23 ] ) . a value as small as herwig s had little effect and values much bigger than 0.025 resulted in the bump in compactness around 23 @xmath0 becoming very broadened and shifted to lower masses . though our understanding of overshoot mixing is not adequate to rule this out , there was no compelling region to use larger values that would have given larger co core masses for the heavier stars than previously published by others ( table 2 ) . , evaluated at carbon ignition , carbon depletion , oxygen ignition and oxygen depletion . compare with fig . [ fig : f9 ] and fig . [ fig : f13 ] for the equivalent runs using kepler . [ fig : f24],scaledwidth=35.0% ] two sets of mesa models were ultimately calculated - one with overshooting ( mov series ) and another without ( m series ) . each set had 16 models in the mass range of @xmath68 with @xmath69 increments , and both used the ledoux criterion ( table 1 ) . as expected , the results from kepler and mesa were nearly identical during hydrogen and helium burning , but started to diverge following helium depletion when the structure became more sensitive to the treatment of semiconvection and overshoot mixing . as shown in fig . [ fig : f24 ] , by carbon depletion , the discrepancy had become significant , especially around the critical mass for the extinction of central carbon convection , 20 @xmath0 , and continued to grow during oxygen burning . [ fig : f23 ] and table 2 show the comparison at oxygen depletion between the kepler s series and mesa mov series . the quantity @xmath5 for the mov series shows a non - monotonic structure similar to that in the s series , but shifted in mass by about 2@xmath0 . since the co cores are nearly identical for mov- and s - series models , the difference must result from either a different composition at carbon ignition or a different treatment of convection after carbon ignites , or both . the carbon mass fractions at carbon ignition are indeed different in the two studies . mass fractions in the mov series range from 0.34 to 0.24 , declining roughly linearly as the mass rises from 15 to 30 @xmath0 , while the corresponding limits for the s series are 0.21 to 0.18 . the larger carbon abundance probably accounts for most of the shift in the peak of the mov plot . as we shall see ( [ sec : s5 ] ) , however , the treatment of post - carbon burning convection also matters . even bare co cores with identical initial compositions often have significantly different compactness when evolved to oxygen depletion in the two codes . the results from the m series are not shown in fig . [ fig : f23 ] , but table 2 shows that the resulting co cores are much smaller , and the compactness thus differs appreciably , especially for the lighter stars . despite the shift , it is encouraging that the non - monotonic structure of @xmath5 as a function of mass in fig . [ fig : f23 ] and , qualitatively , the location of its peak , can be achieved using standard settings for convection physics in two very different codes . the study also highlights the dependence of the outcome on the co core mass . because of the sensitivity of the results to the carbon mass fraction at carbon ignition , outcomes will depend on the rates used for its creation and destruction . three major reaction rates are involved , each of which has some associated uncertainty : 3@xmath72 , @xmath70c(@xmath71o , and @xmath70c+@xmath70c . it is the competition of 3@xmath72 and @xmath70c(@xmath71o that sets the ratio of carbon to oxygen produced by helium burning . increasing the former raises the carbon yield while increasing the latter makes it smaller . as we have shown , small reductions in carbon mass fraction tends to shift the compactness curve downwards , especially the critical mass , 20 @xmath0 , above which carbon burns radiatively . the effects of varying the three uncertain reaction rates on massive stellar evolution , and nucleosynthesis in particular , have been previously explored by @xcite , @xcite , and @xcite for @xmath73 , and @xmath70c(@xmath71o and by @xcite for @xmath70c+@xmath70c , but none of these works explicitly focused on how these rates affect the _ structure _ of presupernova stars . we will not attempt a survey of all possibilities at the present time , but focus here on just the one rate for @xmath70c(@xmath71o . reaction rate . curves are labeled by a multiplier times the rate of @xcite . a value of 1.2 was used in the rest of the paper . the current 1@xmath74 range is approximately 0.9 to 1.3 . [ fig : f25],scaledwidth=35.0% ] historically , the large possible range for this reaction rate has been a source , perhaps even _ the _ major source of nuclear physics uncertainty in massive stellar evolution . recent developments ( e.g. * ? ? ? * ) have narrowed the error bar for this rate to less than 20% , which is not so much greater than the uncertainty in 3@xmath72 @xcite . for small errors in the rates , it is the ratio of these rates that matters most , so variation of @xmath70c(@xmath71o can act , qualitatively , as a surrogate for exploring the uncertainty in 3@xmath72 . the standard value used in this paper for @xmath70c(@xmath71o is 1.2 times @xcite who used s(300 kev ) = 146 kev b ; that is , our effective s - factor at 300 kev , near the gamow energy for helium burning , is 175 kev b. @xcite now report laboratory measurements and analysis that gives s(300 kev ) = 161 @xmath75 19 stat - 2 sys + 8 kev b , or a range 0.9 to 1.3 times the buchmann value . a subset of our s - series models was thus recalculated with varying multipliers on the buchmann rate . thirteen different masses of presupernova star were calculated in the mass range of 19 to 25@xmath0 ( 0.5@xmath0 intervals ) for each multipliers of 0.7 , 0.9 , 1.3 , 1.5 and 1.9 times the buchmann rate ( see table 1 ) . these are the sb series ( table 1 ) . [ fig : f25 ] and tables 2 and 3 show the results . decreasing the rate for @xmath76 significantly increases the carbon mass fraction at helium ignition , but affects the mass of the co core very little . the greater ( or smaller ) carbon mass fraction shifts the peak in @xmath5 , formerly seen at 23 @xmath0 for a multiplier of 1.2 to higher ( or lower ) values . for a multiplier of 1.9 , the peak would disappear altogether , and is not plotted for clarity . however , such large values are outside the current experimental range . the current ( 1 @xmath74 ) error bar for the rate translates into a multiplier between 0.9 and 1.3 , but even this smaller range can shift the explodabilty of massive stars enough to have a dramatic potentially observable effect on presupernova masses and nucleosynthesis . a total uncertainty of less than 10% in the combined error of 3@xmath72 and @xmath76 may be needed to pin down the mass range of supernovae that explode by the neutrino transport model to an accuracy of less than 2 @xmath0 . [ sec : s5 ] the importance of the co core mass motivates a separate study of the compactness of presupernova stars evolved from _ bare _ co stars of constant mass . it is to be emphasized that there is no direct correspondence between the results of evolving an isolated co core and those obtained for a complete massive star with the same co core mass embedded inside a helium core . the lack of a precipitous drop in density and pressure at the edge of a co core inside a massive star results in an evolution that is qualitatively different . the definition of a co core there , even by composition , is ambiguous @xcite . how does one count a partly burned helium shell ( and these are a common case ) ? nevertheless , co cores are simple to evolve and bypass the uncertainties associated with convection during the helium burning phase . if they display the same sort of compactness systematics seen for the full stars , one can have greater confidence in the result . as a function of mass at central oxygen depletion for bare co stars studied with kepler and mesa . top : ks ( solid ) and ku ( dotted ) ; bottom : ms(solid ) and mu(dotted ) . [ fig : f26],scaledwidth=48.0% ] a total of 380 co cores were evolved using both kepler and mesa ( table 1 ) . mesa used @xmath77 and @xmath78 and kepler used @xmath79 and @xmath80 . the calculations followed the evolution from central carbon ignition until the presupernova stage for the kepler cores and until the core oxygen depletion stage for mesa models . the core masses ranged from 3 to 25@xmath0 with varying increments : 0.1 @xmath0 increments were used for 3 to 10 @xmath0 , 0.5 @xmath0 was used for 10 to 20 @xmath0 , and 1 @xmath0was used for 20 to 25 @xmath0 . co core masses of 3 to 9 @xmath0 correspond , approximately , to the main sequence masses of 15 to 30@xmath0 studied in the s series . for low metallicity models ( u series ) or other models with no mass loss ( sh series ) , the maximum co core mass ( at 65 @xmath0 ) was 25 @xmath0 , and that sets the upper limit . the physics used to calculate the co core models was identical to that in the full star calculations . the different mass fractions of carbon for the s and u stars ( figure 5 ) , however , necessitated separate surveys . because of convection , the composition of a co core of given metallicity and mass can be taken as roughly constant . a good fit to the abundances in fig . [ fig : f5 ] is @xmath81 the mass fraction of oxygen was @xmath82 in each case . these 4 sets of calculations are denoted as : ks , ku , ms and mu . , i.e. - ks stands for kepler cores with initial composition from solar metallicity full stars ( s series ) , and mu stands for mesa cores with initial composition from low metallicity full stars ( u series ) . [ fig : f26 ] shows @xmath5 at core oxygen depletion point for the 4 series ( see also table 1 ) . the same sort of non - monotonic variation in @xmath5 seen previously in the full star models is evident in all cases but , as expected , cores with initial composition from u series ( ku and mu ) , have their first peak at a lower mass than the cores with initial composition from solar metallicity stars ( ks and ms ) . this same sort of offset was also seen in the @xmath5 curve for full stars of different metallicities ( fig . [ fig : f3 ] ) . for presupernova stars derived from bare co cores of a given mass and composition . ( top : ) for the mass range 3 to 12 @xmath0 , corresponding to main sequence masses from about 15 @xmath0 to 30 @xmath0 . note the large bump near 6 @xmath0 corresponding to the similar peak for main sequence masses 22 to 24 @xmath0shown in fig . [ fig : f3 ] . the offset between solar ( ks ; green ) and low metallicity ( ku ; red ) is also similar . ( bottom : ) a similar plot for the heavier co cores . the agreement with fig . [ fig : f3 ] is striking.[fig : f27],scaledwidth=48.0% ] above co core masses of 11 @xmath0 , the differences between runs using kepler and mesa are minimal . below that , one sees similar structures for ks and ms , for example , but there are differences . the major peak for the ks stars at 7 @xmath0 , is near 5.5@xmath0 for the ms stars and the peak of ku stars is also shifted down about 1 @xmath0 in the mu stars . even after removing the major dependence on co core mass by running bare co cores , there is still some residual difference resulting from the sensitivity of the location of carbon burning shells to convective and semiconvective physics in the two codes . the overall similarity of the patterns is encouraging , however . the compactness is non - monotonic with mass and there are regions of mass that will easier to explode than others . the kepler calculations of co cores were continued until the presupernova stage , and fig . [ fig : f27 ] shows the compactness parameter as a function of core mass for both the ks and ku series at this final stage . a comparison with fig . [ fig : f3 ] is interesting . the peaks at 22 and 23 @xmath0 in the full star models for the u and s series are reflected well in their equivalent mass co cores , 5.8 @xmath0 and 6.2 @xmath0 , respectively . the good agreement continues to heavier masses where the dip at 50 @xmath0 for the u - series appears as a dip at 18.5 @xmath0 in the ku - series . the mass of the co core in the u - model 50 @xmath0 star was indeed 18.5 @xmath0 . similarly , although not so precisely , the co core of the sh - series 55 @xmath0 star was 20.6 @xmath0 , close to the dip at 24 @xmath0 seen in the lower panel of fig . [ fig : f27 ] . the maximum mass co core in the sh series was 27 @xmath0 for the 65 @xmath0 model . for the u series , the corresponding co core mass was 26 @xmath0 . [ sec : s6 ] while characterizing the complex structure of a presupernova massive star with a single number is a gross over - simplification , considerable insight can be achieved by studying the systematics of the compactness parameter , @xmath5 . the compactness of a given presupernova core turns out to be sensitive to a variety of inputs , including not only the star s initial mass and composition , but the way convection is handled in the code , the nuclear physics employed , and the code used for the calculation . stellar modeling , at least of massive stars , is , in a sense , a statistical science . different groups will almost universally obtain different results for the structure and composition of a non - rotating presupernova star resulting from a given main sequence mass star , even without the complication of mass loss . however , for the same input physics , there should exist a robust pattern of results that can be sampled by running a large set of models . this is a paper about one such pattern , the compactness parameter . our study confirms the non - monotonic nature of @xmath5 noted in previous works , and explores its causes . [ fig : f7 ] shows that the choice of 2.5 @xmath0 as a fiducial point for evaluating the compactness , though arbitrary , is a good indicator of the behavior at other locations deep inside the star , but outside the iron core . whether @xmath5 is evaluated at the presupernova model ( @xmath83 km s@xmath4 ) , or at core bounce matters little at 2.5 @xmath0 , but for points deeper in , small variations seen in @xmath5 for the presupernova model are amplified during the collapse . small variations in main sequence mass might therefore result in significant differences in outcome for an explosion model @xcite . extensive new surveys of presupernova evolution are reported for both solar metallicity and low metallicity ( 10@xmath11 z@xmath84 ) stars ( [ sec : s2 ] ; table 1 ) . while the emphasis here is on presupernova structure and compactness , this is clearly a rich data set for exploring nucleosynthesis and supernova explosion physics , and this will be done elsewhere . major differences are found in the presupernova compactness for stars of the same initial mass , but different metallicity ( [ sec : s2.2 ] ) . the most obvious differences result from the differing degrees of mass loss expected for the two compositions , but there are subtle secondary effects involving the amount of helium convected into the helium core near central helium depletion . the growth of the helium convection zone results from a different structure at its outer edge which , in turn , is affected by the strength of the hydrogen shell , whether the star is a red or blue supergiant , and how much hydrogenic envelope remains following mass loss ( in the solar metallicity stars ) . even with mass loss suppressed , a substantial offset remains between presupernova stars of solar and low metallicity . generally speaking , the low metallicity stars become difficult to blow up at a lower mass ( fig . [ fig : f3 ] ) and may make more black holes . the underlying cause for the non - monotonic behavior of @xmath5 with initial mass derives from the interaction of carbon - burning and oxygen - burning shells with the carbon - depleted and , later , oxygen depleted core ( [ sec : s3 ] ) . as has been noted previously , the end of convective carbon core burning at around 20 @xmath0 for standard stellar physics , has a major effect , but not so much due to the entropy decrease during central carbon burning itself as to the effect this has on the ensuing shells . above 20 @xmath0 , both central carbon burning and what was the first convective shell switch to radiative transport . energy generation in excess of pair neutrino losses is unable to support core against further contraction , and the next two carbon convective shells are pulled down . their gradual migration outwards again causes the rise and decline , and rise again of @xmath5 above 20 @xmath0 . above 30 @xmath0 , @xmath5 rises to large values and , in the absence of mass loss , keeps on rising . there is an appreciable dip at about 50 @xmath0 , though , due to the outward migration of the oxygen shell which essentially plays the same role as the carbon shells did for lower mass stars . broadly speaking , then , the evolution of massive stars separates into four mass intervals characterized by similar values of @xmath5 . for a standard choice of physics and solar metallicity these are stars from 8 to 22 @xmath0 , 22 - 25 @xmath0 , 25 - 30 @xmath0 , and above 30 @xmath0(e.g . , fig . [ fig : f3 ] ) . stars below 22 @xmath0 and between 25 and 30 @xmath0 have low @xmath5 and might more easily explode . between 22 and 25 @xmath0and above 30 @xmath0 , it will be more difficult for the traditional neutrino - transport model to succeed . there can be large variations within those ranges and not all stars lighter than 22 @xmath0 will be easy to explode @xcite , but the distribution of stars that make black holes , for example , or important contributions to nucleosynthesis could be bi - modal . galactic chemical evolution studies using such non - monotonic yields have yet to be done , but may give quite different results , e.g. , for the s - process and oxygen synthesis @xcite . while the pattern seen in @xmath5 _ vs _ mass is robust , the locations of its peaks and valleys are sensitive to the treatment of semiconvection and convective overshoot in the code . these cause variations in the mass of the co core and the carbon mass fraction when carbon burning ignites . reducing semiconvection or convective overshoot makes the co core smaller and this is a leading cause of the large variations seen in the literature for many studies of presupernova ( table [ tab : cocore ] ) . smaller co cores can mimic the effect of reducing the main sequence mass in a calculation where the convection physics is held constant and thus make cores with steep surrounding density gradients up to a higher mass . however these cores , while easier to blow up will be essentially nucleosynthetically barren . extremely small values for semiconvective efficiency and overshoot mixing can lead to a discrepancy between the lightest stars expected to explode as supernovae based upon observations , about 8 @xmath0 , and that calculated by theory . caution must be exercised when using the co core mass as a discriminant , however , because the outcome is also very sensitive to the carbon mass fraction , and even to the treatment of convection after helium depletion . most of the models in this paper were calculated using the kepler code , but the mesa code was also used to calculate the compactness curve at oxygen deletion where its essential features have already been determined . reasonable agreement is achieved between the two codes provided that some amount of semiconvection and convective overshoot mixing is used in both . the non - monotonic nature of the compactness plot , in particular , is apparent in the results of both surveys . the effects of varying the uncertain reaction rate for @xmath70c(@xmath71o within its error range were also explored and found to have an appreciable effect . a total error of just 10% in the combined error for 3 @xmath72 and @xmath70c(@xmath71o can shift features in the compactness plot by a solar mass or more . to the extent the compactness is related to the difficulty of exploding a given model using neutrinos , the changes for nucleosynthesis and supernova `` cut - off mass '' are significant at that level . and binding energies for presupernova stars of the s series ( top : ) the binding energy above the iron core shown as a function of initial mass . despite the general increasing baseline trend , the relative variations closely follow that of @xmath5 . ( bottom : ) the final value of @xmath5 is plotted against the binding energy outside of the iron core . [ fig : f28],scaledwidth=48.0% ] because the co core mass plays such a major role in determining @xmath5 , we explored its systematics using a grid of `` bare '' co - cores . these differ from the co cores of massive stars which are defined by composition changes and do not have sharp declines in density and pressure at their edges . nevertheless , the same non - monotonic compactness curve results . this both confirms the robustness of the @xmath5 relation found in the other surveys and suggests that , for a given presupernova model , its final co core mass , not its main sequence mass might be a more accurate structural indicator . the same behavior was found for co cores studied using mesa . picking the co core mass circumvents some , though certainly not all of the uncertainty introduced by ambiguous treatments of semiconvection and convective overshoot mixing during core helium burning . for wr stars and other stars that have experienced severe mass loss , the presupernova co core mass will be the best indicator of presupernova structure . even for bare co cores , however , the mass fraction of carbon at carbon ignition remains an important `` hidden variable '' that will be sensitive to both the code physics and the metallicity of the star . while this paper has emphasized the relation between compactness of the core and its explodability in terms of the neutrino transport model , the results are actually more general . [ fig : f28 ] shows the binding energy of the matter outside of the iron core of the presupernova star as a function of its compactness . stars with larger values of @xmath5 will be more difficult to explode by _ any _ mechanism , both because of the higher ram pressure immediately after collapse and the larger energy needed to eject the matter outside the iron core . we gratefully acknowledge numerous valuable discussions with alexander heger and bill paxton regarding the kepler and mesa codes and their application to this problem . we also benefited from alex s expert insights into massive stellar evolution on many occasions , and his permission to include unpublished details of the `` u series '' low metallicity models . casey meakin and evan oconnor provided helpful comments on an earlier draft . we thank the anonymous referee for a careful reading of the submitted manuscript and many helpful suggestions . this research has been supported by the national science foundation ( ast 0909129 ) , the nasa theory program ( nnx09ak36 g ) , and the university of california lab fees research program ( 12-lr-237070 ) . kepler & s & zams-@xmath24 g @xmath21 & 15 - 30 & 151 & default + & sh & zams - presn & 30 - 60 & 9 & no mass - loss + & u & zams - presn & 15 - 65 & 86 & @xmath85 , no mass - loss + & ss & zams - presn & 15 - 30 & 176 & weak semiconvection + & so & zams - presn & 15 - 30 & 16 & no overshooting + & sb & zams - presn & 19 - 25 & 65 & varied @xmath76 + & ks & c.ign-presn & 3 - 25 & 95 & co core with s composition + & ku & c.ign-presn & 3 - 25 & 95 & co core with u composition + + mesa & m & zams-o.dep & 15 - 30 & 16 & default + & mov & zams-o.dep & 15 - 30 & 16 & with overshooting + & ms & c.ign-o.dep & 3 - 25 & 95 & co core with s composition + & mu & c.ign-o.dep & 3 - 25 & 95 & co core with u composition + ccccccccccc 4.84 & 2.97 & 7.17 & 4.94 & 9.65 & 7.20 & @xmath86 & @xmath87 ( ? ) & c.dep & @xmath88 & @xcite + 4.0 & 2.02 & 6.0 & 3.70 & 8.0 & 5.75 & & o.dep & @xmath89 & @xcite + 4.21 & 2.44 & 6.27 & 4.13 & 8.50 & 6.27 & @xmath90 & @xmath87 & si.dep & @xmath91 & @xcite + 5.68 & 3.76 & 8.65 & 6.59 & 10.0 & 8.63 & @xmath90 & @xmath87 & si.dep & and rotation & @xcite + 3.85 & 2.19 & 5.70&3.54 & 5.18 & 7.73 & @xmath92 & @xmath93 & o.dep & & @xcite + 4.97 & 2.56 & 7.31 & 3.83 & 9.80 & 5.48 & & presn & @xmath94 & @xcite + 5.37 & 3.59 & wr&5.53 & wr&6.63 & & presn & and rotation & @xcite + 4.35 & 3.14 & 6.21 & 4.93 & 8.25 & 6.95 & @xmath95 & @xmath96 & presn & @xmath97 & kepler , s + 4.44 & 3.14 & 7.18 & 5.61 & 9.07 & 7.35 & @xmath95 & @xmath96 & presn & and rotation & kepler + 4.29 & 3.08 & 6.27 & 4.99 & 8.50 & 7.18 & @xmath95 & @xmath96 & presn & @xmath98 & kepler , ss + 4.15 & 2.93 & 6.22 & 4.59 & 8.40 & 6.39 & @xmath95 & @xmath96 & presn & @xmath99 & kepler , ss + 4.09 & 1.96 & 6.00 & 2.95 & 8.30 & 4.40 & @xmath95 & @xmath96 & presn & @xmath100 & kepler , ss + 3.98 & 1.63 & 6.08 & 2.71 & 8.42 & 4.11 & @xmath95 & @xmath96 & presn & @xmath101 & kepler , ss + - & - & 6.22 & 4.96 & 8.28 & 7.00 & @xmath95 & @xmath96 & presn & 1.5@xmath102buchmann & kepler , sb + - & - & 6.21 & 4.90 & 8.26 & 6.98 & @xmath95 & @xmath96 & presn & 1.3@xmath102buchmann & kepler , sb + - & - & 6.17 & 4.96 & 8.23 & 6.99 & @xmath95 & @xmath96 & presn & 0.9@xmath102buchmann & kepler , sb + - & - & 6.16 & 4.94 & 8.21 & 6.99 & @xmath95 & @xmath96 & presn & 0.7@xmath102buchmann & kepler , sb + 4.19 & 2.70 & 6.11 & 4.38 & 8.14 & 6.27 & @xmath95 & @xmath96 & presn & @xmath103 & kepler , so + 3.77 & 1.96 & 5.81 & 3.47 & 7.31 & 3.91 & @xmath95 & @xmath96 & o.dep & @xmath104 & mesa , m + 4.01 & 2.21 & - & - & 7.80 & 5.16 & @xmath95 & @xmath96 & o.dep & @xmath105 & mesa + 4.20 & 2.39 & - & - & 8.06 & 5.47 & @xmath95 & @xmath96 & o.dep & @xmath106 & mesa + 4.41 & 2.61 & - & - & 8.34 & 5.81 & @xmath95 & @xmath96 & o.dep & @xmath107 & mesa + 4.85 & 3.05 & 6.92 & 4.77 & 9.28 & 6.84 & @xmath95 & @xmath96 & o.dep & @xmath108 & mesa , mov + [ tab : cocore ]

the ab initio band structure calculations of strongly - interacting low - dimensional ( quasi - one - dimensional ( q1d ) or two - dimensional ( 2d ) ) systems ( high-@xmath1 superconductor yba@xmath2cu@xmath3o@xmath4 as well as charge - density - wave ( cdw ) systems k@xmath5moo@xmath3 and bavs@xmath3 being examples @xcite ) usually reveal several bands in the vicinity of the fermi level that are , in the leading approximation , decoupled from the rest of the band structure . in the tight - binding approximation , the bare dispersions of such a multiband model are well understood in terms of the strong hybridization among several properly chosen orbitals per unit cell , while the hybridization with the other electronic levels is neglected . @xcite the bare single - particle tight - binding hamiltonian is thus taken to include commensurate / incommensurate cdw or sdw ( spin - density wave ) average fields and several first - neighbour bond energies . in the two - particle hamiltonian one usually retains the long - range coulomb interactions , the largest local ( hubbard ) interactions together with the most important short - range interactions . associated with different terms in the single - particle hamiltonian are fluctuations in the charge and spin densities on different orbitals in the unit cell and on the corresponding bonds . all these ingredients make strongly - interacting multiband electronic problems extremely complicated . to explain most of experimental results on typical single - band systems , it usually suffices to describe properly the coupling between @xmath6 monopole charge fluctuations and external electromagnetic ( em ) fields and combine this coupling hamiltonian with the ordinary transport equations . @xcite however , in multiband electronic systems , even in the cases when only one band intersects the fermi level , the analysis is considerably more complicated . not only the interband physics but also the intraband physics depends drastically on both single - particle and two - particle parameters of the hamiltonian and are affected differently by various types of fluctuations . therefore , to understand the results of transport measurements @xcite and reflectivity measurements @xcite of multiband electronic systems in detail , we describe the @xmath6 multipolar fluctuations in the dynamical conductivity tensor in a way consistent with the results of experimental methods which probe charge and spin fluctuations which presumably occur at finite wave vectors @xcite . the @xmath6 charge - charge correlation function is thus just one of several equally important correlation functions defined in terms of symmetrized intraband and interband charge and spin fluctuations . common to all these correlation functions is a complicated structure of the intraband and interband electron - hole propagators . therefore , to understand the contribution of different types of fluctuations to these correlation functions , in particular to the dynamical conductivity , we must first find the solution to quite general bethe salpeter equations for the electron - hole propagators and then combine the result with the very definition of the response function under consideration . this question , which is of great importance in studies of strongly - interacting low - dimensional systems , is in the focus of the present analysis . in this article , we thus present a general response theory which uses both an essentially exact semiclassical description of the coupling between valence electrons and external em fields @xcite and a quite general description of the intraband and interband charge fluctuations @xcite . the result is the current - dipole kubo formula for the dynamical conductivity , @xcite with the structure of the intraband electron - hole propagators determined by means of the quantum transport equation ( i.e. , the intraband bethe salpeter equation ) . to emphasize the connection between the present theory and the ordinary transport theory , @xcite we determine the explicit form of the intraband conductivity in the memory - function approximation . @xcite we also find the structure of the interband conductivity in both cases , with real and with purely imaginary interband dipole vertices , to show in which way the lorentz local field corrections @xcite affect the dynamical conductivity of interacting low - dimensional systems . the article is organized as follows . to make the reading of the article easier , we give in section 2 an overview of the results . then in section 3 the notation is introduced to describe electrons and boson modes in a general multiband electronic model with boson - mediated electron - electron interactions . a complementary multiband electronic model with nonretarded electron - electron interactions will be studied in a separate article @xcite which is focused on the commensurability effects in the single - particle conductivity tensor . starting with the quantum electrodynamical description of the interaction between valence electrons and external em fields , we derive the exact semiclassical form of the electron - em field coupling hamiltonian in section 4 . in section 5 we define the dynamical conductivity tensor and briefly discuss the relaxation - time approximation . section 6 gives the structure of the self - consistent bethe salpeter equations for the dipole vertex functions and for the related auxiliary electron - hole propagators ( to be defined latter ) in a general multiband case . we show the equivalence of the bethe salpeter equation for the auxiliary intraband electron - hole propagators and the quantum transport equation . the generalized drude formula is derived from the quantum transport equation by using the memory - function approximation . the bare interband conductivity is considered in section 7 and the related local field effects in section 8 . section 9 contains the concluding remarks . before discussing the details of the dynamical conductivity model proposed in this article , we present an overview of the results . in this way principal physical messages of the present work are separated from the technical details . we are interested here in a general low - dimensional model with multiple electronic bands in the presence of boson - mediated electron - electron interactions . the single - particle part of the total hamiltonian is assumed to be exactly solvable . the bloch energies @xmath7 in the band labeled by the band index @xmath8 and all relevant bare vertex functions ( the dipole vertices @xmath9 , for example ) are thus taken as known functions . to study the influence of the boson - mediated electron - electron interactions on single - electron propagators @xmath10 and , in addition , on different electron - hole propagators encountered in the response functions of interest , we use high - order diagrammatic perturbation theory of the metallic state , as usual in investigations of low - dimensional electronic systems . @xcite the electron - hole propagators are associated with the dispersive , undamped bosons which can either correspond to the external degrees of freedom coupled to electrons / holes or the internal degrees of freedom built of the electron - hole excitations at finite values of * q*. in sections 68 , the emphasis is on the acoustic and raman - active and infrared - active optical phonons . the scattering by phase and amplitude phonon modes of the common peierls cdw model will be studied in detail within the microscopic lee rice anderson model in the accompanying article @xcite ( to be referred to as article ii ) . ) for electron green s functions . @xmath8 is the electron band index and @xmath11 is the boson branch index . the black circle and the big gray circle represent , respectively , the bare and the renormalized electron - boson vertex . here and herafter , the tiny and bold solid ( dashed ) lines are the bare and renormalized electron ( boson ) propagators . ] in high - order perturbation theory , the electron green s function @xmath10 satisfies the dyson equation @xcite @xmath12 { \cal g}_l ( { \bf k } , { \rm i } \omega_n ) = \hbar , \label{eq1}\end{aligned}\ ] ] where @xmath13 is the self - energy function of the electron ( see figure 1 ) . in this case @xmath13 is a function of @xmath10 , resulting in a system of two coupled equations . a strong reduction of the quasi - particle pole weight in @xmath10 at the fermi level is predicted by different strictly 1d theories @xcite and observed in angle - resolved photoemission spectroscopy ( arpes ) experiments on q1d systems @xcite . in this article we present a systematic method of calculating the dynamical conductivity tensor in interacting low - dimensional systems which is consistent with the described treatment of the single - electron green s functions @xmath10 . in semiclassical electrodynamics applied to a general exactly solvable single - particle problem with multiple electronic bands , the electric component of the em field parallel to the cartesian axis @xmath14 , @xmath15 , couples to the currents of free charges and to the dipole moments of bound charges in a way described by the coupling hamiltonian @xmath16 here , @xmath17 is the total dipole density operator . it is the sum of the intraband and interband contributions @xmath18 with @xmath19 representing the corresponding intraband ( @xmath20 ) and interband ( @xmath21 ) dipole vertices , and @xmath22 . it is not hard to verify that these vertices are directly related to the vertices in the current and charge density operators , @xmath23 and @xmath24 . the relation is given by the expression @xmath25 ( @xmath26 in the longitudinal case ) . the imaginary currents and the imaginary dipole moments do not have a straightforward physical meaning , but they make the notation used throughout this article as simple as possible . for example , they can be used to express the gauge invariance of the expression ( [ eq2 ] ) simply by rewritting @xmath27 in terms of @xmath24 and @xmath23 and using the standard relations between @xmath28 and the scalar potential @xmath29 and the vector potential @xmath30 . in order to examine how electrons respond to applied em fields , we introduce the dynamical conductivity tensor @xmath31 , @xcite which is the response function relating the induced total current @xmath32 to @xmath33 ; @xmath34 . we use here the finite temperature formalism . the dynamical conductivity @xmath31 is in this case given by analytical continuation ( @xmath35 ) of @xmath36 , which is the matsubara fourier transform of @xmath37 . the compact form of the semiclassical coupling hamiltonian ( [ eq2 ] ) allows us to express @xmath37 in the form of the current - dipole kubo formula @xcite @xmath38 \rangle_{\rm irred}. \label{eq5}\end{aligned}\ ] ] here , @xmath39 is the total current density operator defined in a way analogous to ( [ eq3 ] ) ( see ( [ eq23 ] ) ) . in order to extract the most important physical consequences that follow from this expression , one usually uses the relations ( [ eq4 ] ) to rewrite @xmath37 either in terms of the charge - charge or the current - current kubo formulae . @xcite in contrast to these approaches , we use here ( [ eq5 ] ) , because this expression will be shown below to provide the direct link of the present analysis to the common transport theory @xcite and to the textbook approaches to the dynamical long - range coulomb screening @xcite . in a general multiband case . the dynamical conductivity @xmath36 , equation ( [ eq5 ] ) , is the rpa - irreducible part of @xmath40 . @xcite ] to complete the response theory , we must take into account the relaxation processes describing the way in which electrons are scattered by static disorder , by other electons or by different types of boson modes . since our primarily interest is in the low - dimensional systems where the quasi - particle pole weight in the single - electron green s function is strongly reduced by boson - mediated electron - electron interactions , we must calculate @xmath37 beyond the common weak coupling theory ( corresponding to second - order perturbation theory for the single - electron self - energy function ) . this means that in order to work out @xmath37 we must use the general bethe salpeter expression for the screened conductivity tensor illustrated in figure 2 and the definition @xmath41 , where @xmath42 is the external electric field @xcite . the bold solid lines in these two diagrams are the single - electron green s functions which satisfy the dyson equation ( [ eq1 ] ) . the intraband and interband contributions to @xmath43 can be represented by the same expression , @xmath44 equally well for @xmath20 as for @xmath21 . here the @xmath45 are the renormalized dipole vertex functions which satisfy the self - consistent bethe salpeter equations for the renormalized dipole vertex , and @xmath46 . in order to emphasize the connection between the present approach and the common transport theory , as well as to simplify the notation , we introduce the auxiliary electron - hole propagator @xmath47 by comparing the relation @xmath48 with ( [ eq6 ] ) . therefore , our task is to solve the self - consistent equations for @xmath47 rather than that for the renormalized dipole vertices . we limit our discussion to the low - dimensional systems where the scattering by static disorder and the scattering processes in which electrons change the band can both be neglected . these processes have negligible influence on the properties studied here . however , the self - consistent equations still have a very complex structure in particular as a consequence of the following features : the nesting properties of the fermi surface , the resonant nature of the scattering by the ( quasi)static cdw or sdw potentials and the local - field effects in the interband conductivity channel . the results of the present analysis are the following : _ ( i ) _ we show the formal solution to the bethe salpeter equations for @xmath49 and to the ( intraband ) quantum transport equation , _ ( ii ) _ we compare the intraband conductivity to that obtained by using the memory - function approximation , _ ( iii ) _ we calculate the bare interband conductivity and _ ( iv ) _ we rederive the well - known lorentz lorenz form of the optical conductivity in the presence of scattering by boson modes . _ ( i ) _ although it is tempting to completely neglect the electron - em field vertex corrections in @xmath36 , this is not justified in the intraband channel , at least in the cases with low - dimensional conductivity studied here . the method by which the intraband conductivity tensor is calculated here is very general and is based on a systematic analysis of the self - consistent equation for the auxiliary electron - hole propagator @xmath50 in which the single - electron self - energy contributions and the related vertex corrections are treated on an equal footing . the equation is of the form @xmath51 \phi_\alpha ( { \bf k},{\bf k}_+ , { \rm i } \omega_n , { \rm i } \omega_{n+ } ) \nonumber \\ & & \hspace{5 mm } = \frac{1}{\hbar^2 } \big [ { \cal g } ( { \bf k } , { \rm i } \omega_n ) - { \cal g } ( { \bf k}_+ , { \rm i } \omega_{n+ } ) \big ] p_\alpha ( -{\bf q } ) , \label{eq8}\end{aligned}\ ] ] with @xmath52 and @xmath53 . for simplicity we assume that only one band intersects the fermi level and omit explicit reference to the conduction band index . @xmath54 is the electron - hole self - energy , the structure of which is derived here for the case in which conduction electrons are scattered only by various boson modes . these boson modes are assumed to dissipate momentum by their own means ( impurities , umklapps , etc . ) . pinned phasons of the commensurate cdw problem , for example , enter in the present theory in the same way as the infrared - active optical phonons from section 8.2 . this question is discussed in more details in article ii . the electron - hole self - energy is found to be the difference between two auxiliary single - electron self - energies @xmath55 , and the contributions of the @xmath6 forward scattering processes are found to drop out of both @xmath55 and @xmath54 , in full agreement with the charge continuity equation . equation ( [ eq8 ] ) is valid under quite general conditions , and is therefore applicable to a rich variety of problems , including purely electronic models with nonretarded local and short - range interactions . for example , it is not hard to verify that the normal scattering processes do not lead to the resistivity in this case , as required by the general principle of microscopic reversibility . @xcite this question will be discussed in more details in ref . . _ ( ii ) _ in the ordinary 3d metallic regime , equation ( [ eq8 ] ) leads to the ordinary transport equation , with @xcite @xmath56 being the resulting intraband electrical conductivity , and @xmath26 again . here , @xmath57 is the unity of mass , commonly taken as the free electron mass , and @xmath58 is the intraband relaxation rate . in the drude @xmath59 limit , where @xmath60 , the @xmath61 dependent term in the denominator of ( [ eq9 ] ) can be ignored , and we obtain the intraband conductivity ( [ eq10 ] ) . for @xmath62 and @xmath63 , on the other hand , the result is the well - known thomas fermi expression @xcite for @xmath64 and for the dielectric susceptibility @xmath65 . _ ( iii ) _ the general expression for the interband single - particle conductivity is derived and compared to the approximate expression which is frequently encountered in the literature . @xcite _ ( iv ) _ if the scattering of valence electrons is sufficiently weak , the electron - boson vertex renormalizations are of no importance and the electron - hole propagators of figure 2 satisfy the bethe salpeter equations shown in figure 3 with two types of contributions , the ladder contributions and the rpa ( random phase approximation ) contributions . @xcite these equations can be completely rewritten in terms of the renormalized single - electron green s functions and the dynamically screened electron - electron interactions . this approximation is used here to illustrate how the conductivity tensor is affected by the short - range dipole - dipole interactions . the result is the well - known lorentz - lorenz form of the dynamical conductivity for 3d metallic @xmath0 models @xcite which is believed to be valid in the q1d conductors , as well @xcite . from the result @xmath66 we can conclude that the contributions to the conductivity tensor of the free charges and of the bound charges are decoupled and that the lorentz dipole - dipole interactions affect only the bound ( interband ) charge contribution to the conductivity tensor . let us first introduce the quantities characterizing a general multiband electronic model with boson - mediated electron - electron interactions . the total hamiltonian is the sum of two contributions , @xmath67 . here , @xmath68 \label{eq12 } \end{aligned}\ ] ] is the bare electron - boson hamiltonian , and @xmath69 is the scattering hamiltonian which describes the scattering of valence electrons by each of the boson modes . as already mentioned , the bosonic modes can be external to electronic system ( e.g. phonons ) or built from electron - hole fluctuations at finite wave vectors . in ( [ eq12 ] ) , @xmath7 is the bare electron dispersion , @xmath70 is the bare boson frequency and @xmath71 is the corresponding `` mass '' parameter . finally , @xmath72 is the generalized scattering potential . for convenience , it is assumed to be a product of the coupling constant @xmath73 , the dimensionless electron - boson vertex @xmath74 and the boson field @xmath75 , @xmath76 in the intraband scattering approximation , in which the electron does not change the band when it is scattered by bosons , the scattering potential @xmath72 is diagonal in the band index . ) for boson green s functions . ] in the finite temperature formalism , the effects of @xmath77 on the single - electron and boson properties are described in terms of two exact matsubara green s functions @xcite @xmath78 \rangle , \nonumber \\ & & \hspace{-11 mm } { \cal d}_\nu ( { \bf q } ' , \tau ) = - \langle t_\tau \big[q_{\nu { \bf q}'}(\tau ) q^\dagger_{\nu { \bf q}'}(0 ) \big ] \rangle . \label{eq15 } \end{aligned}\ ] ] their fourier transforms satisfy two dyson equations , equation ( [ eq1 ] ) and @xmath79 - \hbar \sigma_\nu ( { \bf q } ' , { \rm i } \nu_n ) \big\ } { \cal d}_\nu ( { \bf q } ' , { \rm i } \nu_{n } ) = \hbar \label{eq16}\ ] ] ( see figure 4 ) . it is convenient to adopt an abbreviated notation by writting @xmath80 this expression can also be shown in the form @xmath81 , where @xmath82 \rangle , \label{eq18}\end{aligned}\ ] ] is by definition the force - force correlation function @xcite ( ffcf ) associated with the potential @xmath72 . ( the big gray circle ) and the completely irreducible four - point vertex function @xmath83 ( the white rectangle ) . ] in high - order perturbation theory the self - energy function of the electron , @xmath84 , is usually represented by the sum of skeleton diagrams shown in figure 5 , leading to a simple analytical expression @xmath85 here , @xmath86 is the renormalized ffcf obtained by replacing one bare electron - boson vertex in @xmath87 , @xmath88 , by the renormalized vertex @xmath89 . equation ( [ eq19 ] ) is nothing but the three - point vertex representation of @xmath84 . in the same representation the boson self - energy takes the form @xmath90 we have just reemphasized that the three - point vertex representation is very useful in describing the closed system of four equations characterizing strongly - interacting electron - boson systems in the absence of external em fields ( equations ( [ eq1 ] ) , ( [ eq16 ] ) , ( [ eq19 ] ) and ( [ eq20 ] ) ) . however , the conductivity tensor of such an electron - boson model has a compact form if the structure of the bethe salpeter equations of section 6 is also simple , at least from the formal point of view . for this purpose , it is much more convenient to use the four - point vertex representation . this representation is based upon the knowledge of the irreducible four - point interaction @xmath91 . the equivalence of the two representations in the skeleton approximation is illustrated in figure 5 , where @xmath91 is represented by the white rectangle . the expression for @xmath84 is now of the form @xmath92 a full discussion of the bethe salpeter equations is postponed to section 6 . in quantum electrodynamics of tightly bound valence electrons in solids , whether metallic or insulating , the coupling between electrons and external em fields is given by the minimal gauge invariant substitution . the expansion to the second order in the vector potential @xmath93 gives the coupling hamiltonian @xmath94 , with @xcite @xmath95 here , @xmath96 are , respectively , the total current density operator and the bare diamagnetic density operator . to obtain the semiclassical expression for the coupling hamiltonian , equation ( [ eq2 ] ) , we have to take into account the fact that both the interband and intraband current - current contributions to the diamagnetic current must vanish in the normal metallic or insulating state . the cancellation of the interband contribution is a consequence of the effective mass theorem , @xcite @xmath97 leading to @xmath98 , with @xcite @xmath99 , \nonumber \\ & & \hspace{-5 mm } h^{\rm ext}_2 = \frac{e^2}{2mc^2 } \sum_{l { \bf q } { \bf q}'\alpha \beta } a_{\alpha } ( { \bf q}-{\bf q}')a_{\beta } ( { \bf q } ' ) \hat{\gamma}^{ll}_{\alpha \beta } ( -{\bf q } ) . \label{eq25 } \end{aligned}\ ] ] here , @xmath100 , @xmath21 , is the interband part in dipole density operator ( [ eq3 ] ) , and @xmath101 is the diamagnetic density operator . the reciprocal effective mass tensor @xmath102 is the vertex function in @xmath101 . the intraband current - current contribution to the diamagnetic current cancels @xmath103 , confirming the assertion made in section 2 that @xmath104 we emphasize the simple structure of this coupling hamiltonian . the most direct way to obtain this expression for @xmath27 is to consider the coupling of the scalar potential @xmath105 to the charge density operator . @xcite let us derive now the relations ( [ eq4 ] ) among three types of intraband and interband vertex functions underlying the gauge invariance of the present response theory . to do this , we use very general microscopic operator equations @xmath106 , \nonumber \\ & & \hspace{-10 mm } { \rm i } \hbar \hat j_\alpha({\bf q } ) = { \rm i } \hbar \frac{\partial}{\partial t } \hat p_\alpha({\bf q } ) = \big[\hat p_\alpha({\bf q } ) , h_0^{\rm el } \big ] , \label{eq26}\end{aligned}\ ] ] where @xmath107 is the electronic part in the bare hamiltonian ( [ eq12 ] ) . furthermore , @xmath108 is the total charge density operator , with @xmath109 being the charge vertex function , while @xmath39 and @xmath110 are given by ( [ eq23 ] ) and ( [ eq3 ] ) . the two commutators in ( [ eq26 ] ) are easily worked out , leading to the required relations @xmath111 the explicit form of each of these vertices is often obtained by calculating the coefficients in the linear term in @xmath27 in ( [ eq22 ] ) . the long - wavelength intraband vertices are of particular interest . they are given by model independent expressions @xmath112 here , @xmath113 is the bare electron group velocity and @xmath52 is the bare dipole vertex . notice that the thermodynamic averages of the density operators in ( [ eq26 ] ) are the usual induced macroscopic densities . for example , @xmath114 , @xmath115 and @xmath116 represent the common notation for the induced macroscopic current , charge and dipole densities . they are related by two equations @xcite @xmath117 the first one is the usual charge continuity equation . we now define the total dynamical conductivity tensor @xmath31 as the response function to the macroscopic electric field @xcite @xmath118 that is @xmath119 here , @xmath114 is the induced total current density from ( [ eq29 ] ) , @xmath42 is the external field and @xmath120 is the depolarization field ( equal to @xmath121 for the longitudinal polarization of the field in the thin slab ) . the conductivity tensor is given by analytical continuation to the real axis of @xmath36 ( see figure 6 ) , where @xmath36 is the matsubara fourier transform of @xmath38 \rangle_{\rm irred}. \nonumber \end{aligned}\ ] ] therefore , to find the dynamical conductivity tensor of an interacting electron - boson system , we must determine the structure of the current - dipole kubo formula ( [ eq5 ] ) which agrees with the definition of the macroscopic electric field ( [ eq30 ] ) . from figure 2 , shown in the three - point vertex representation ( left diagram ) and in the four - point vertex representation ( right diagrams ) . the collection of diagrams in the renormalized four - point interaction must be consistent with the definition of @xmath122 in ( [ eq30 ] ) ; i.e. , the renormalized four - point interaction is irreducible in the long - range rpa contribution . ] it is instructive first to show the structure of the bare dynamical conductivity ( [ eq5 ] ) in the simplest multiband model , namely , a two - band model with the lower band partially occupied by electrons and the upper band empty , in the limiting case where the scattering is sufficiently weak . as explained in more details in article ii , this model is of importance in accounting for the properties of the ordered state of some simple cdw systems . @xcite to do this , we first calculate the ideal bare conductivity tensor @xmath123 which is given by the first term in figure 6 , with @xmath124 replaced by the bare green s function @xmath125 . summation over @xmath126 can be performed easily to give @xmath127 . after analytical continuation , we obtain the ideal bare intraband ( @xmath20 ) and interband ( @xmath128 ) contributions described by @xmath129 } { \hbar \omega + { \rm i } \hbar \eta + \varepsilon_{l}({\bf k})- \varepsilon_{l'}({\bf k}_+ ) } . \label{eq32}\end{aligned}\ ] ] weak scattering processes can be inserted into this expression by replacing the adiabatic term @xmath130 by the relaxation rate @xmath131 independent of frequency and wave vector . this is the generalization of the common relaxation - time approximation . as required , the associated transport relaxation time @xmath132 differs from that found in the single - electron propagators . the bare intraband contribution is of the drude form @xmath133 with @xmath134 ^ 2 \bigg(-\frac{\partial f_l({\bf k})}{\partial \varepsilon_l ( { \bf k})}\bigg ) \label{eq34}\end{aligned}\ ] ] representing the effective number of conduction electrons , @xcite and @xmath135 by assumption . @xmath136 includes the conduction holes from the lower band and the thermally activated electrons from the upper band . the bare interband contribution is the sum of two off - diagonal terms in ( [ eq32 ] ) , with @xmath130 replaced by @xmath131 : @xmath137 . the main disadvantage of this simple procedure of calculating @xmath138 is that it does not apply to low - dimensional systems with sizeable interactions . the reason for that is the following . first , comparison with experimental data shows that the relaxation rates @xmath131 are frequency dependent . second , the causality principle requires that @xmath139 is just the imaginary part of a complex quantity which will be called here the electron - hole self - energy function . finally , it turns out that in the strong coupling limit the electron - hole self - energy is not a simple function of frequency @xmath140 . instead , it depends quite drastically on two internal electron frequencies too . we shall therefore present now a systematic treatment of scattering processes in interacting electron - boson systems with multiple electronic bands which treats all elements in the electron - hole self - energy consistently with the charge continuity equation ( [ eq29 ] ) , with the causality principle and with the definition of @xmath122 . the resulting self - consistent equations turn out to be the generalization of the equations used by vollhardt and wlfle @xcite to study the scattering by static disorder in a single - band case . the term bethe salpeter equations @xcite refers to the self - consistent equations for the exact electron - hole propagators @xmath141 shown in figure 7 . a widely - used simplification in these equations is illustrated in figure 3 . in this case , we have both the ladder contributions and the rpa - like contributions to the irreducible four - point interaction , but the three - point vertex renormalizations are neglected both in these equations and in the related dyson equations . the interaction lines and the electron propagators are renormalized , so that these equations must be treated consistently with the dyson equations , as already pointed out in section 2 . however , in strictly 1d metallic systems as well as in strongly - correlated low - dimensional systems , it is necessary to use the bethe salpeter equations beyond this approximation by taking into account the three - point vertex renormalizations as well . to simplify the analysis , it is convenient to multiply the left - hand side of the diagrams in figure 7 by the corresponding bare vertex function and then sum over @xmath142 and @xmath143 . for the electron - hole propagators in the dynamical conductivity ( [ eq5 ] ) , the bare vertex in question is the bare dipole vertex function , and the bethe salpeter equations from figure 7 turn into the bethe salpeter equations for the auxiliary electron - hole propagators @xmath144 the equations are illustrated in figure 8 . with little loss of generality , we restrict the analysis to the case where the irreducible four - point interaction @xmath145 can be approximated by @xmath146 . in this case , the explicit form of the intraband ( @xmath20 ) and interband ( @xmath21 ) equations is the following @xmath147 \nonumber \\ & & \hspace{0 mm } \times \bigg\ { -p_{\alpha}^{l'l}({\bf k}_+,{\bf k } ) - \frac{1}{\beta } \sum_{l_1l_1 ' } \sum_{{\bf q } ' { \rm i } \nu_m } u^{l'l_1',l_1l } ( { \bf q } ' , { \rm i } \nu_{m } ) \nonumber \\ & & \hspace{10 mm } \times \phi^{l_1l_1'}_\alpha ( { \bf k}+{\bf q}',{\bf k}_++{\bf q } ' , { \rm i } \omega_m , { \rm i } \omega_{m+ } ) \bigg\ } , \label{eq36}\end{aligned}\ ] ] with @xmath148 and @xmath149 . the @xmath150 satisfy the dyson equations for electrons ( [ eq1 ] ) and @xmath146 is a simple generalization of the completely irreducible four - point interaction @xmath151 from ( [ eq21 ] ) . ) for auxiliary electron - hole propagators . ] in the electronic multiband model we are interested in here , there are @xmath152 bands in the vicinity of the fermi level . the bands are built of @xmath152 orbitals per unit cell . associated with each of these orbitals are the intracell charge and spin fluctuations , and these intracell fluctuations are possible sources of scattering . the monopole charge fluctuations underlying the present analysis of the dynamical conductivity tensor might be also strongly affected by the presence of the intracell fluctuations . as a result , @xmath31 will have considerably more structure than that characterizing the ordinary single - band model , even in the case where only one band out of @xmath152 valence bands intersects the fermi level . in the remainder of this section we consider the intraband term in ( [ eq36 ] ) from this point of view . we assume that there is only one conduction band and again drop reference to the band index . we use the intraband scattering approximation where the scattering processes in which the electron changes the band are neglected . as discussed in article ii , this approximation is easily seen not to be adequate , for example , to study fluctuations in the order parameter in the pseudogap state of the ordinary cdw systems . however , it is convenient to illustrate the main ideas regarding the quantum transport equations . finally , the three - point vertex renormalizations in the renormalized interactions are also neglected , as usual in the common non - crossing approximation for the electron - boson systems . @xcite the resulting intraband bethe salpeter equation is of the form @xmath153 \bigg\ { p_{\alpha}(-{\bf q } ) \nonumber \\ & & \hspace{-5 mm } - \frac{1}{\beta } \sum_{{\bf q } ' { \rm i } \nu_m } { \cal f } ( { \bf q } ' , { \rm i } \nu_{m } ) \phi_\alpha ( { \bf k}+{\bf q}',{\bf k}_++{\bf q } ' , { \rm i } \omega_m , { \rm i } \omega_{m+ } ) \bigg\}. \nonumber \\ \label{eq38}\end{aligned}\ ] ] the structure of ( [ eq38 ] ) is simplified by defining the auxiliary single - electron self - energy function @xmath154 and the electron - hole self - energy @xmath155 equation ( [ eq38 ] ) now reduces to the quantum transport equation @xmath156 \phi_\alpha ( { \bf k},{\bf k}_+ , { \rm i } \omega_n , { \rm i } \omega_{n+ } ) \nonumber \\ & & \hspace{5 mm } - \frac{1}{\hbar^2 } \big [ { \cal g } ( { \bf k } , { \rm i } \omega_n ) - { \cal g } ( { \bf k}_+ , { \rm i } \omega_{n+ } ) \big ] p_\alpha ( -{\bf q } ) \nonumber \\ & & \hspace{-5 mm } = - { \rm i } \pi^i ( { \bf k},{\bf k}_+ , { \rm i } \omega_n , { \rm i } \omega_{n+ } ) \phi_\alpha ( { \bf k},{\bf k}_+ , { \rm i } \omega_n , { \rm i } \omega_{n+ } ) . \label{eq41}\end{aligned}\ ] ] equation ( [ eq41 ] ) is the self - consistent equation for the auxiliary electron - hole propagator , in contrast to the ordinary transport equation , equation ( [ eq47 ] ) , which is formulated in terms of the induced density @xmath157 defined by @xmath158 the conditions under which ( [ eq41 ] ) reduces to the ordinary transport equation are discussed below . to illustrate the role of the electron - em field vertex renomalizations in ( [ eq41 ] ) , it is useful to define the auxiliary single - electron propagator @xmath159 as the single - particle propagator which satisfies the following dyson equation @xmath160 \widetilde { \cal g } ( { \bf k } , { \rm i } \omega_n ) = \hbar . \label{eq43}\end{aligned}\ ] ] the forward scattering processes cancel identically out in @xmath161 as a result of the charge continuity equation . the fact that the auxiliary propagator @xmath162 has the quasi - particle pole even for relatively strong scattering processes is consistent with the fact that the residue associated with the quasi - particle pole in the exact propagator @xmath163 is proportional to @xmath164 and vanishes in the strictly 1d limit @xmath165 . @xcite this interesting observation is in the background of a variety of approximate treatments of various pseudogapped states . in these approaches , the forward scattering contributions to the ffcf @xmath166 are neglected from the outset . the expression for the intraband conductivity which correctly describes the @xmath59 drude limit , but which is not necessarily correct in the @xmath167 thomas fermi limit , is the following @xmath168 this is nothing but ( [ eq32 ] ) for the ideal intraband conductivity in which @xmath169 is replaced by @xmath54 . in the weak coupling limit , the self - energy @xmath54 is a non - singular function of two fermion frequencies and in the simplest approximation can be written in the form @xmath170 . @xmath171 is called the memory function . @xcite the momentum distribution function @xmath172 appearing in ( [ eq44 ] ) in the memory - function approximation reduces in the weak coupling limit to the fermi dirac distribution function @xmath173 . we now define the electron - boson coupling function @xmath174 , the mass enhancement factor @xmath175 , the intraband relaxation rate @xmath176 and the collision integral @xmath177 by @xmath178 summation of ( [ eq41 ] ) over @xmath126 , together with analytical continuation @xmath35 and multiplication by @xmath179 , gives the ordinary transport equation @xcite @xmath180 \delta n ( { \bf k } , { \bf q } , \omega ) + \frac{m}{m_{\alpha \alpha}^*({\bf k } , \omega ) } e \hbar v_\alpha^0 ( { \bf k } ) \nonumber \\ & & \hspace{20 mm } \times \frac{\partial f({\bf k})}{\partial \varepsilon ( { \bf k } ) } e_\alpha ( { \bf q } , \omega ) = \hbar i({\bf k } , \omega ) , \label{eq47}\end{aligned}\ ] ] where @xmath181 . in solving ( [ eq47 ] ) it is convenient to follow the usual textbook procedure . @xcite this equation can be solved either for @xmath182 or for @xmath183 , where @xmath184 are , respectively , the induced charge and current densities of conduction electrons , and the @xmath185 , @xmath186 , are the contributions to @xmath187 which are proportional to @xmath188^i$ ] . for further considerations of the random - phase approximation in section 8.1 , the second route is more appropriate choice . in this case , we consider the thin slab with the longitudinal polarization of the electric fields , introduce the scalar fields @xmath29 and @xmath189 , by the relations @xmath190 and @xmath191 , and use the definition of the dielectric susceptibility @xmath192 : @xmath193 , or @xmath194 \rho^{\rm c } ( { \bf q},\omega ) = v^{\rm ext } ( { \bf q},\omega ) . \label{eq49}\ ] ] it is readily seen that the intraband conductivity is of the form @xmath195 ^ 2 \big ( -\partial f({\bf k})/\partial \varepsilon ( { \bf k})\big ) } { \omega + { \rm i } \gamma_{1\alpha}({\bf k},\omega ) -q_\alpha^2 v^2_\alpha ( { \bf k})/\omega}. \label{eq50}\end{aligned}\ ] ] in the drude @xmath59 limit , we set @xmath61 in the denominator equal to zero , with @xmath196 and @xmath197 , to obtain @xmath198 an expression known as the generalized drude formula . here , @xmath199 means the average over the fermi surface . the expression ( [ eq51 ] ) remains valid when more than one band intersects the fermi level , provided that @xmath200 . the generalization to the case with the relaxation rates @xmath201 dependent on the band index @xmath8 is straightforward . although we have arrived at ( [ eq41 ] ) and ( [ eq47 ] ) by considering in the auxiliary self - energy @xmath55 only the scattering by boson modes , these expressions are of a more general validity , namely all relevant scattering mechanisms have to be included in @xmath202 . therefore , we can use ( [ eq41 ] ) to study the effects of local and short - range electron - electron interactions on electrodynamic properties of strongly - correlated low - dimensional systems or to study similar effects associated with the scattering by different periodic cdw / sdw potentials . in analogy to ( [ eq44 ] ) , we can write the bare interband conductivity in the form @xmath203 with the electron - em field vertex corrections neglected , resulting in @xmath204 . the electron - hole self - energy @xmath205 is replaced by @xmath206 , in the memory - function approximation , and by @xmath207 , in the relaxation - time approximation . in those multiband systems in which the interband dipole vertex @xmath208 is purely imaginary , the local dipole moment vanishes and there are no local field effects . @xcite the interband conductivity is given by @xmath209 . this result can be equally well understood to represent the solution to the rpa equation @xmath210 \rho ( { \bf q},\omega ) = v^{\rm ext } ( { \bf q},\omega ) \label{eq53}\ ] ] for the total dielectric susceptibility @xmath211 , where @xmath212 . in the opposite case , the interband dipole vertex is real and its dispersionless part is directly related to the intracell dipole transitions . particularly interesting are the examples in which @xmath213 is completely local . in this case we can use the procedure from the following section to determine the effects of the lorentz local fields . at this point caution is in order regarding the relaxation - time approximation . it is well known that in those systems in which the damping energy @xmath214 is comparable to the interband threshold energy ( the pseudogapped state of the ordinary cdw systems is an example ) the relaxation - time approximation of ( [ eq52 ] ) overestimates the in - gap part of the interband conductivity . in this regime , @xmath215 is found to be better described by the approximate textbook expression @xcite @xmath216 } { \hbar \omega + { \rm i } \hbar \gamma_{2\alpha}({\bf k } ) - \varepsilon_{l'}({\bf k}_+ ) + \varepsilon_{l}({\bf k})}. \label{eq54}\end{aligned}\ ] ] in order to briefly illustrate our results , the typical mean - field single - particle conductivity of the peierls cdw model at temperatures well below the mean - field transition temperature @xmath217 is shown in figure 9 . the dashed and the dot - dashed lines represent , respectively , the interband contributions calculated by using ( [ eq54 ] ) and the relaxation - time version of ( [ eq52 ] ) . notice the difference between the two curves at @xmath59 . the solid line is the total single - particle conductivity which is the sum of the contribution of thermally activated charge carriers in ( [ eq51 ] ) and the interband contribution ( [ eq54 ] ) . this result agrees reasonably well with the experimental spectra @xcite measured at temperatures well below the cdw transition temperature . the detailed discussion of the mean - field approximation as well as of the effects of fluctuations in the order parameter is given in article ii . , at temperature @xmath218 k. the bare electron dispersion is @xmath219 , with @xmath220 ev and @xmath221 mev . the mean - field transition temperature is assumed to be @xmath222 k and @xmath223 mev is the cdw order parameter at temperature @xmath224 k. the damping energies are @xmath225 mev at @xmath226 , and @xmath227 mev at @xmath228 . the dot - dot - dashed line shows the inverse square - root singularity in @xmath229 associated with the limit @xmath230 . ] in the next section , we shall examine the lorentz local field corrections in the relaxation - time approximation , starting with @xmath231 given by ( [ eq54 ] ) . the damping energy @xmath232 is assumed to be small in comparison with the interband threshold energy . most q1d systems are characterized by a large number of atoms in the unit cell ( for example , in the blue bronze k@xmath5moo@xmath3 we have 86 atoms per unit cell ) . the bands in the vicinity of the fermi level are thus built of different molecular orbitals of the well - defined symmetry with the intermolecular distance large when compared to the first - neighbor atom - atom distance . therefore , the bands are narrow and the typical interband threshold energy is in the infrared part of the spectrum . in most cases of interest , the low - energy interband electron - hole excitations are local and correspond to dipolar transitions between one molecular orbital of the @xmath233 symmetry and another of the @xmath234 symmetry . consequently , to determine the proper structure of the total conductivity tensor @xmath235 in this case , it is important to take into account the local field effects too . this is not the only q1d case in which the local field corrections are important . in the ordered state of the strong bond - cdw systems the interband dipole excitations are also local . finally , the ordinary @xmath0 models with the interband threshold energy in the visible part of the spectrum also have @xmath236 of the form ( [ eq54 ] ) , with the dipole vertex @xmath237 real . @xcite to demonstrate the usual procedure for dealing with the local field corrections , @xcite we consider the model with the intraband currents and with one type of bound charges . we assume that the contribution of the bound charges is related to the intra - atomic or intramolecular dipolar transitions in @xmath236 . in section 8.2 , we shall briefly discuss the case of infrared - active optical phonons , as well . in this section , we make a simple modification of the notation . for example , the label intra is replaced by c and the label inter by b ( for conduction and bound charges ) . the bare intraband dielectric susceptibility is assumed to be of the form @xmath238 with @xmath239 given by ( [ eq50 ] ) , and the bare interband dielectric susceptibility is @xmath240 with @xmath241 given by ( [ eq54 ] ) . ( a ) and @xmath242 ( b ) . ] the electrons are assumed to be subjected to an external longitudinal electric field @xmath243 represented by the scalar potential @xmath244 . @xmath245 and @xmath246 are the fourier transforms of two types of macroscopic charge densities . @xmath247 describes the monopole - charge density of conduction electrons and @xmath248 the dipole - charge density of bound charges . the two - component random - phase approximation describes the long - range screening in such a case . according to figure 10 , the rpa equations can be shown in the form @xmath249 \rho_1 ( { \bf q } , \omega ) - v_{12 } \ , \rho_2 ( { \bf q } , \omega ) = v^{\rm ext } ( { \bf q},\omega ) , \nonumber \\ & & \hspace{-7 mm } - v_{21 } \ , \rho_1 ( { \bf q } , \omega ) + \big[1/\chi_{22}^0 ( { \bf q } , \omega ) - v_{22}\big ] \rho_2 ( { \bf q } , \omega ) = v^{\rm ext } ( { \bf q},\omega ) . \nonumber \\ \label{eq57}\end{aligned}\ ] ] here , the @xmath250 stand for two bare dielectric susceptibilities in ( [ eq55 ] ) and ( [ eq56 ] ) , and @xmath251 are the fourier transforms of different long - range interactions . the numbers @xmath252 , @xmath253 and @xmath254 correspond to the monopole - monopole , monopole - dipole and dipole - dipole interactions , respectively . @xcite @xmath255 are easily shown to represent the common depolarization factors for the longitudinal @xmath6 fields . @xcite it is not hard to verify that , after neglecting the lorentz term @xmath256 in @xmath257 ( i.e. , when @xmath258 ) , equations ( [ eq57 ] ) reduce to ( [ eq53 ] ) . equations ( [ eq57 ] ) may be rewritten now in the alternative form , in terms of two dipole densities @xmath259 , to give @xmath260 with the implicit dependence on @xmath261 and @xmath140 . here @xmath262 is the common notation for two bare polarizabilities . it must be emphasized that the expressions on the right - hand side of ( [ eq58 ] ) represent the macroscopic electric field @xmath122 , equation ( [ eq30 ] ) , which is the sum of the external electric field and two depolarization fields originating from the conduction electrons and from the induced dipole moments of bound charges . @xcite this is a useful observation which helps us to rewrite ( [ eq58 ] ) in the final form @xmath263 j^{\rm c}_{\alpha } ( { \bf q } , \omega ) = e_{\alpha}({\bf q } , \omega ) , \label{eq60 } \\ \big [ 1/\alpha^{\rm b}_{\alpha \alpha } ( { \bf q } , \omega ) \big ] p^{\rm b}_{\alpha } ( { \bf q } , \omega ) = e_{\alpha}({\bf q } , \omega ) , \label{eq61 } \end{aligned}\ ] ] with @xmath264 . \label{eq62}\ ] ] for the thin slab with the longitudinal polarization of the fields , the field @xmath243 is identical to the dielectric displacement field @xmath265 and the relation ( [ eq62 ] ) leads to the usual form of the longitudinal macroscopic dielectric function . if other high - energy on - site dipolar excitations are also included , the total dielectric function becomes @xmath266 here , @xmath267 and @xmath268 are two renormalized polarizabilities . in the notation of the previous sections , this result can be expressed by ( [ eq10 ] ) and ( [ eq11 ] ) . we can therefore conclude that the contribution of the monopole charges to the conductivity tensor is decoupled from the contribution of the dipole charges . the latter contribution is affected by the lorentz field in the usual way , giving also the possibility for the collective contribution in @xmath269 when @xmath270 . it was implied by our discussion of the quantum transport equation that ( [ eq5 ] ) , together with the density operators ( [ eq3 ] ) and ( [ eq23 ] ) , is able to capture all relevant aspects of the electron - phonon coupling in the dynamical conductivity tensor of multiband electronic systems , at least for the scattering by acoustic and raman - active optical phonons . in order to treat carefully infrared - active optical phonons , we must also include the direct coupling between phonons and em fields in @xmath27 and take into account the corresponding local field corrections . the generalization of ( [ eq63 ] ) to such a case with multiple dipolar bound - charge contributions is very simple . for our present purposes it is simplest to replace @xmath271 by the sum of all bare dipolar bound - charge polarizabilities . in the two - band case with one infrared - active phonon branch , we obtain @xmath272 with @xmath273^{-1}$ ] being the bare phonon polarizability . in this article , we have derived the dynamical conductivity tensor in a general multiband electronic model with sizeable boson - mediated electron - electron interactions . starting with the gauge - invariant form of the coupling between valence electrons and external em fields , we have derived the general form of the self - consistent bethe salpeter equations for the auxiliary intraband and interband electron - hole propagators . it is demonstrated that the intraband bethe salpeter equation is identical to the usual quantum transport equation and that it reduces to the ordinary transport equation in the weak coupling limit . in this way , the present response theory of multiband electronic systems is connected to the generalized drude formula for the intraband conductivity widely used in experimental analyses as well as to different fermi liquid theories based on the landau silin transport equations . in order to illustrate the formalism presented here , we shall give in the accompanying article @xcite the detailed description of the cdw mean - field approximation and the effects of fluctuations in the order parameter on the dynamical conductivity of the peierls cdw systems . this research was supported by the croatian ministry of science and technology under project 119 - 1191458 - 0512 . massidda s , yu j and freeman a j 1987 _ phys . lett . _ a * 122 * 198 whangbo m - h and schneemeyer l f 1986 _ inor . chem . _ * 25 * 2424 mattheiss l f 1985 _ solid state commun . _ * 93 * 791 emery v j 1987 _ phys . lett . _ * 58 * 2794 pines d and nozires p 1989 _ the theory of quantum liquids i _ ( new york : addison - wesley ) ziman j m 1972 _ electrons and phonons _ ( london : oxford university press ) ando y , komiya s , segawa k and kurita y 2004 _ phys . lett . _ * 93 * 267001 forr l , gal r , berger h , fazekas p , penc k , kzsmrki i and mihly g 2000 _ phys . lett . _ * 85 * 1937 uchida s , ido t , takagi h , arima t , tokura y and tajima s 1991 _ phys . 7942 degiorgi l , alavi b , mihly g and grner g 1991 _ phys . _ b * 44 * 7808 kzsmrki i , mihly g , gal r , barii n , akrap a , forr l , homes c c and mihly l 2006 _ phys . lett . _ * 96 * 186402 sugai s , suzuki h , takayanagi y , hosokawa t and hayamizu n 2003 _ phys . rev . _ b * 68 * 184504 takigawa m , reyes a p , hammel p c , thompson j d , heffner r h , fisk z and ott k c 1991 _ phys . _ b * 43 * 247 landau l d , lifshitz e m and pitaevski l p 1995 _ electrodynamics of continous media _ ( oxford : butterworth heinemann ) kupi i 2009 _ j. raman spectrosc . _ * 40 * 442 barii s and zelenko j 1990 _ solid state commun . _ * 74 * 367 kupi i and barii s 2007 _ phys . _ b * 75 * 094508 kubo r , toda m and hashitsume n 1995 _ statistical physics ii _ ( berlin : springer - verlag ) eliashberg g m 1961 _ zh . eksp . fiz . _ * 41 * 410 ( 1962 _ sov . phys . jetp _ * 14 * 886 ) abrikosov a a 1988 _ fundamentals of the theory of metals _ ( amsterdam : nort - holland ) vollhardt d and wlfle p 1980 _ phys . _ b * 22 * 4666 gtze w and wlfle p 1972 _ phys . _ b * 6 * 1226 giamarchi t 1991 _ phys . _ b * 44 * 2905 kupi i 2004 _ physica _ b * 344 * 27 adler s l 1962 _ phys . rev . _ * 126 * 413 wieser n 1962 _ phys . rev . _ * 129 * 63 upanovi p , bjeli a and barii s 1996 _ z. phys . _ b * 101 * 387 kupi i et al unpublished abrikosov a a , gorkov l p and dzyaloshinski i e 1975 _ methods of quantum field theory in statistical physics _ ( new york : dover publications ) dzyaloshinskii i e and larkin a i 1973 _ zh . eksp . fiz . _ * 65 * 411 ( 1974 _ sov . . jetp _ * 38 * 202 ) kupi i , rukelj z and barii s to be published ( in the main text referred to as article ii ) fetter a l and walecka j d 1971 _ quantum theory of many - particle systems _ ( london : mcgraw hill ) mahan g d 1990 _ many - particle physics _ ( new york : plenum press ) solyom j 1979 _ adv . * 28 * 201 giamarchi t 2004 _ quantum physics in 1d _ ( oxford : clarendon press ) vescoli v , zwick f , henderson w , degiorgi l , grioni m , gruner g and montgomery l k 2000 _ eur . phys . j. _ b * 13 * 503 mitrovic s , fazekas p , sndergaard c , ariosa d , barii n , berger h , clotta d , forr l , hchst h , kupi i , pavuna d and margaritondo g 2007 _ phys . _ b * 75 * 153103 ziman j m 1979 _ principles of the theory of solids _ ( london : cambridge university press ) wooten f 1972 _ optical properties of solids _ ( new york : academic press ) negele j w and orland h 1988 _ quantum many - particle systems _ ( new york : addison wesley ) kupi i 2009 _ phys . _ b * 79 * 235104 kupi i and barii s 2005 _ fizika _ a * 14 * 47 abrikosov a a and genkin v m 1973 _ zh . eksp . * 65 * 842 ( 1974 _ sov . phys . jetp _ * 38 * 417 ) kupi i 2003 _ physica _ c * 391 * 251 lee p a , rice t m and anderson p w 1974 _ solid state commun . _ * 14 * 703 kim k , mckenzie r and wilkins j w 1993 _ phys . lett . _ * 71 * 4015 ziman j m 1988 _ elements of advanced quantum theory _ ( london : cambridge university press ) millis a j and lee p a 1987 _ phys . _ b * 35 * 3394 niki h , tuti e and barii s 1995 _ physica _ c * 241 * 247 rozenberg m j , kotliar g and kajueter h 1996 _ phys . rev . _ b * 64 * 8452 kupi i 2002 _ physica _ b * 322 * 154

for atomization energies and geometries , density functional theory ( dft ) constitutes a cost - effective alternative to wavefunction - based _ ab initio _ methods , being capable of accuracies on the order of a few kcal / mol and a few picometer , respectively , if basis sets of polarized triple zeta quality are employed@xcite . as a result of this astonishing success , especially considering the modest computational cost of dft methods , a variety of new applications are now being explored@xcite , including , very recently , the calculation of molecular anharmonic potential energy surfaces @xcite . anharmonic potential energy surfaces ( see refs.@xcite for initial pioneering studies at the scf and cisd level ) allow for the prediction of molecular vibration - rotation spectra that can be compared directly to experiment , obviating the need to make estimations or approximations for the effect of anharmonicity . ccsd(t)/@xmath0 anharmonic force fields can achieve accuracies on the order of 10 @xmath1 or better for fundamentals ( e.g. @xcite and references therein ) , and even greater accuracy is achievable if still larger basis sets and corrections for inner - shell correlation are considered@xcite . + dft anharmonic force fields for small molecules have recently been the subject of two validation studies@xcite . our own validation study@xcite suggested that , for fundamental frequencies , an rms accuracy of about 18 @xmath1 can be attained when using the b97 - 1 functional with sufficiently large basis sets . the question arises as to how capable dft is to treat medium - sized organic molecules in this fashion . particularly relevant here is the study the study of miani et al.@xcite on the fundamental frequencies of benzene , which employed the b3lyp@xcite functional with the tz2p ( triple - zeta plus double polarization ) basis set . this paper , as well as a more recent study of furan , pyrrole and thiophene@xcite , in fact yielded even more accurate results on these medium - sized organic systems than suggested by the small - molecule validation studies cited above . + the azabenzenes are obtained by systematically replacing ch moieties in benzene by nitrogen atoms . numerous experimental spectra of these compounds are available in the literature , ranging from mono - azabenzene ( pyridine ) to 1,2,4,5-tetraazabenzene ( s - tetrazine ) . pentazine is quite elusive@xcite ; the equally elusive n@xmath2 ( hexazine or hexaazabenzene ) molecule has been the subject of extensive theoretical studies@xcite : at the ccsd(t)/cc - pvtz level , the hexazine ( hexaazabenzene ) structure undergoes distortion from the idealized @xmath3 ring to a @xmath4 geometry , which is a local minimum situated some 23 kcal / mol above the @xmath5 diazide global minimum@xcite . straka recently made the interesting suggestion that cyclic n@xmath2 may form very stable m(@xmath6-n@xmath2 ) complexes with m = ti , zr , hf , th@xcite . in practice , nine molecules are relevant to the present study : the parent benzene molecule , pyridine , pyridazine , pyrimidine , pyrazine , 1,2,3-triazine , 1,2,4-triazine , 1,2,5-triazine ( _ sym_-triazine ) and 1,2,4,5-tetrazine ( s - tetrazine ) . all molecules are displayed in figure 1 . + azabenzene skeletons serve as building blocks in nature , e.g. , pyridine in pyridixol ( vitamin b@xmath2 ) , pyridazine and pyrimidine in pteridine , itself found in folic acid ( vitamin vitamin b@xmath7 ) and riboflavin ( vitamin b@xmath8 ) . moreover the four dna bases are derivates of pyrimidine ( c , t ) and of purine ( a , g ) itself a pyrimidine - imidazole fused ring system . for these reasons , azabenzene - like compounds figure in drug design , e.g. the reverse transcriptase inhibitors azt ( contains pyrimidine ) or nevirapine ( contains pyridine ) . derivatives of pyridazine have been found to have potential therapeutic or plant growth inhibitory effects . + quite different applications include melamine - based plastics ( melamine is a derivative of s - triazine ) and the potential use of s - tetrazine in molecular data storage applications ( one of the relaxation mechanisms of the @xmath9 state leads to 2 hcn+n@xmath8.@xcite ) . + innes , ross and moomaw ( irm throughout this paper ) published a compilation and critical review of experimental vibration spectroscopic data current to 1988@xcite . one of us ( with c. van alsenoy ) carried out a harmonic - only b3lyp / cc - pvtz study of the vibrational frequencies and geometries.@xcite the limitations of this approach are obvious . + in the present contribution , we will consider dft _ anharmonic _ force fields and anharmonic vibrational spectra which are directly comparable to experiment for the azabenzenes , and demonstrate their power as a spectroscopic assignment tool for medium - sized organic molecules . in our validation study@xcite , we considered a wide variety of exchange - correlation functionals , as well as convergence in terms of the basis set and the integration grids ( both kohn - sham , ks , and coupled perturbed kohn - sham ) . especially for organic molecules , we found the best performance to be delivered by the b97 - 1 functional , which is handy s reparametrization@xcite of becke s 1997 hybrid functional@xcite . satisfactory basis set convergence was generally found to be achieved with the tz2p basis set@xcite . this is therefore the functional / basis set combination employed for the present study . we found@xcite results for anharmonic force field calculations to be exceedingly dependent on the ks integration grid , much less so on the cpks grid . for the present study , we ended up using a ( 200,974 ) grid , that is , the direct product of a 200-point euler - maclaurin radial grid with a 974-point lebedev angular grid . for the cpks steps , we employed a ( 75,194 ) grid , which considerably reduced the overall computational cost . neither grid was pruned . the quartic force fields were calculated by numerical central differences ( in rectilinear normal coordinates ) of analytical second derivatives , using a stand - alone driver program adapted from the cadpac electronic structure program system@xcite . the actual dft second derivative calculations were done using the gaussian 98 rev . a11 program package@xcite . the step size in a numerical derivatives calculation always represents a compromise between discretization error and roundoff error . based on experience , we determined the step size as a functions of both the absolute value of the harmonic frequency associated with the normal coordinate involved , and the associated reduced mass ( @xmath10 ) : @xmath11 in order to reduce roundoff error as much as possible , the ks and cpks equations were basically converged to machine precision . to calculate this factor , both the reduced mass of the mode @xmath12 and the eigenvalues of the normalisation matrix @xmath13 are needed . the last part of equation 1 , depending on the harmonic frequencies @xmath14 , additionally ensures that each displaced geometry has approximately the same energy difference to the minimum geometry . + in this manner , we obtain a complete cubic force field , as well as all the diagonal and semidiagonal quartic force constants . these are sufficient for second - order rovibrational perturbation theory analyses@xcite , which were carried out using the spectro@xcite and polyad@xcite programs developed in the cambridge group and at weizmann , respectively . the parent molecule of our series , benzene ( figure 1a ) , has been the subject of a few previous anharmonic force field studies , two at the scf / dzp level@xcite , and a very recent one at the b3lyp / tz2p level@xcite . an extensive experimental literature exists on the subject ( see refs.@xcite for reviews ) : high resolution data are available for many of the fundamentals , and all the assignments can be regarded as conclusive . ( an analysis of the b3lyp / cc - pvtz normal modes of benzene and the azabenzenes in terms of pulay s redundant internal coordinates can be found in table 17 of ref.@xcite . ) experimentally obtaining harmonic frequencies for a molecule this large , even with such high symmetry , is a nearly impossible task . in table [ tab1 ] we compare computed harmonic frequencies with selected prior calculations ( in particular ccsd(t)/@xmath0 data@xcite ) , as well as several sets of experimentally derived data . the first such set are the ` @xmath15 ' estimates of handy , murray and amos ( hma)@xcite , themselves obtained as averages of an empirical estimate by goodman , ozkabak , and thakur ( got)@xcite and their own combination of experimental fundamentals with the scf / dzp computed anharmonicities of maslen et al.@xcite . the second set are derived from the experimental fundamentals and a new scf / dzp anharmonic analysis by handy and willetts ( hw)@xcite . the third set is derived from the experimental fundamentals and the b3lyp / tz2p anharmonicities@xcite , and is expected to be the most reliable . + both sets of dft numbers compare about equally well with the ccsd(t ) data . the c - h stretching frequencies are underestimated by about 20 @xmath1 , while all other frequencies are being reproduced very accurately . ( as an aside , we note that the c h underestimation problem also occurs@xcite at the mller - plesset perturbation theory ( mp2 ) level . ) although the c h stretches are reproduced marginally worse by the b97 - 1 functional ( 3 @xmath1 ) compared to the reference ccsd(t)/@xmath0 values , the overall performance is improved : the mean absolute error decreases from 9.3 ( b3lyp ) to 6.5 ( b97 - 1 ) @xmath1 , and the rms deviation from 10.6 ( b3lyp ) to 9.6 ( b97 - 1 ) . the smaller decrease for the rms error ( compared to the mean absolute error ) can be attributed to the greater proportional weight given to larger errors . when excluding the c - h stretching frequencies , the rms errors decrease to 8.4 ( b3lyp ) and 4.3 @xmath1 ( b97 - 1 ) . considering the computational cost of density functional theory , the b3lyp results are already of spectacular compared to the reference _ ab initio _ data . the b97 - 1 functional , however , reduces the rms error of the non c - h stretching frequencies by an additional 50% . + both functionals yield excellent agreement with experiment for the fundamental frequencies ( table [ tab2 ] ) . the experimental data were taken from the earlier compilation of got@xcite and from more recent gas - phase measurements@xcite . the two sets of experimental data agree very well with each other , except for @xmath16 and @xmath17 which are involved in strong fermi resonances . to obtain the b97 - 1/tz2p fundamentals , we had to take the following fermi resonances into consideration : @xmath18 and @xmath19 ( with the deperturbed frequency @xmath19 at 3018 @xmath1 and the combination at 2947 @xmath20 ) , @xmath21 and @xmath17 ( unperturbed @xmath22 at 3040 and the combination band at 3082 @xmath20 ) , @xmath23 and @xmath24 ( with the unperturbed frequencies at 3064 and 3020 @xmath20 ) , and @xmath25 and @xmath16 ( with @xmath16 ( unperturbed ) at 1600 , @xmath25 at 1603 @xmath20 ) . + interestingly , because of the small differences in the harmonic frequencies between the b3lyp / tz2p and the b97 - 1/tz2p calculations and the different force field , now new fermi resonances are predicted . furthermore , the fundamentals affected by these resonances are the ones which differ the most when using the b3lyp or b97 - 1 functional . because of the fermi resonances , the difference between different functionals and methods becomes more important than on the harmonic level of approximation . the results of both functionals compared to experiment are , however , very similar , with mean absolute errors ( compared to the latest experimental results of miani et al.@xcite ) of 9.4 ( b3lyp ) and 8.3 ( b97 - 1 ) @xmath1 and rms errors of 17.0 and 13.7 @xmath1 , respectively . again , all frequencies are very well described with the exception of the c - h stretches , which are underestimated by 1630 @xmath1 ( discounting @xmath24 which is severely perturbed by resonances ) . without the c - h stretches , the mean absolute errors are reduced to 3.7 ( b3lyp ) and 3.9 ( b97 - 1 ) @xmath1 and the rms errors to 4.8 ( b3lyp ) and 4.5 @xmath1 ( b97 - 1 ) . hence , while clearly better for harmonic frequencies , the b97 - 1 functional is only slightly better for predicting the fundamental frequencies of benzene . nevertheless , the accuracy obtained by density functional theory , which yields rms errors of less than 5 @xmath1 for non c - h stretching frequencies , is very good . + in the third column , we added the b97 - 1/tz2p anharmonic corrections to the ccsd(t)/ano4321 harmonic frequencies of ref.@xcite , in order to see how such a ` hybrid ' ab initio - dft approach would perform . the principal improvement seen is for the c h stretches , which are thus all bought into the range of the experimental values , except for @xmath26 , which is in resonance with @xmath23 as discussed above . when discounting c h stretching frequencies , the improvement compared to a pure b97 - 1/tz2p calculation is quite modest , to 3.1 @xmath1 for the mean absolute error and to 4.1 @xmath1 for the rms error . the corresponding error statistics including c h stretching frequencies change rather more significantly , to a mean absolute error of 4.0 @xmath1 and an rms error of 7.4 @xmath1 . + summarizing , b97 - 1/tz2p is likely to be a useful tool for analysis of vibrational spectra ( and verification of their assignments ) of aromatic organic molecules in general and of the azabenzene series in particular . pyridine being the simplest azabenzene ( figure 1b ) and the closest to the parent molecule , we may anticipate similar accuracy as for benzene . as expected for this chemically important species , many experimental spectra are available , in both liquid and gas phases . most of the experimental results and the latest experimental assignments are summarized by klots@xcite . the computed and observed vibrational frequencies are presented in table [ tab3 ] . as expected , the c - h stretching frequencies are not well described . here , they are also heavily perturbed , and , considering performance for the corresponding bands in benzene , our method may not be sufficiently accurate to assist in the experimental assignment . thus , although we will attempt to elaborate on the c - h stretching frequencies for most azabenzenes , the results are more tentative . additionally , a small change in the original unperturbed fundamental frequencies will cause large changes in the perturbed frequencies . nevertheless , the two asymmetric stretches @xmath27 and @xmath19 do not follow the general trend of being underestimated by 2040 @xmath1 . while the perturbed @xmath27 is close to the experimental value , @xmath19 differs by 70 @xmath1 . especially the latter appears too large to be accounted for merely by deficiencies in our dft calculation . here , the severe fermi resonance seems to be responsible for this assignment , as its deperturbed value is at 3021 @xmath1 . this mode is resonating with ( deperturbed frequencies given ) @xmath27 at 3053 @xmath1 , @xmath28 at 3015 @xmath1 and @xmath29 at 3054 @xmath1 , with the perturbed @xmath28 now appearing at 3023 @xmath1 . it appears that the latter value is more in line with experiment , and that perhaps this assignment might be more reasonable . nevertheless , some caution is appropriate as the c h harmonic stretching frequencies are not as well described by dft as the remaining modes . large basis set ccsd(t ) harmonic frequencies would be helpful here , but would require inordinate amounts of cpu time . + another picture emerges for the non c - h frequency modes . here , we primarily compare to the medium resolution raman vapor data of klots ( klots2)@xcite : their assignments seem to be in line with most other experiments , although they differ from the most recent data obtained by partal et al.@xcite . comparing to these latter inelastic neutron scattering ( ins ) data , however , amounts to comparing apples and oranges , as the ins data are based on further refinement of a dft computed _ harmonic _ force field by maximizing agreement between simulated and observed ins spectral . the pseudo - harmonic frequencies from the refined force field correspond neither to true harmonic nor to true fundamental frequencies in effect , they are neither fish nor fowl . + we can confirm the assignment of klots@xcite of @xmath30 to the lower value of 1053 @xmath1 . difficult assignments seem to result from the fermi resonances for both @xmath31 and @xmath32 , since these are the only fundamentals which give errors of 10 @xmath1 or larger . the deperturbed mode of @xmath31 at 1578 @xmath1 resonates with @xmath33 at 1593 @xmath1 resulting in perturbed modes at 1575 and 1596 @xmath1 . klots may have misassigned the latter band to @xmath31 rather than @xmath33 . @xmath32 , on the other hand , seems to be very difficult to assign from experimental data because of its depolarized character . the problem is further exacerbated as @xmath32 is located at the shoulder of @xmath34 . although @xmath32 is involved in a fermi resonance , it is probably not the source of the disagreement with , since its unperturbed frequency at 1241 @xmath1 resonates with @xmath35 at 1282 @xmath1 , resulting in bands at 1237 and 1289 @xmath1 . with the exception of these two strongly perturbed bands ( @xmath31 and @xmath32 ) , the agreement between experiment and theory can only be described as stunning for bands other than c - h stretches . for the latter , probably only a full ccsd(t ) force field can resolve the assignment . even when including @xmath31 and @xmath32 ( but excluding all c - h stretching frequencies ) , the mean absolute error for the computed values compared to the experimental data of klots@xcite is 3.6 @xmath1 and the rms error is 5.3 @xmath1 ; without @xmath31 and @xmath32 they are reduced to 2.7 and 3.8 @xmath1 , respectively . + the double - harmonic infrared intensities are in good agreement with the b3lyp / cc - pvtz values@xcite and in reasonable agreement with the experimental numbers . even here , the c - h stretches differ , probably not only because of anharmonic contributions and fermi resonances , but also because of the poor description of these modes by dft . still , the excellent performance seen for benzene results is repeated for pyridine . pyridazine ( figure 1c ) has been of particular interest in the last couple of years @xcite . a large body of experimental data is available , generally measured in the gas phase but when certain modes were unavailable , the authors of ref.@xcite substituted their own liquid or solid phase measurements in order to get a full complement of frequencies . all of the spectra have been assigned with the aid of scaled hartree - fock , mp2 , blyp or b3lyp harmonic force fields using small basis sets . some of them have their force fields fitted to match the experimental data . despite the large amount of data , huge discrepancies exist , as shown in table [ tab4 ] . this seems to be especially true for the out - of - plane modes where experimental assignments and values seem almost arbitrary . for @xmath36 , for example , the experimental numbers range from 765 to 949 @xmath1 , and even the two most recent numbers appear at opposite ends of this range . these assignments have been done using different methods and scaling techniques , since no reliable theoretical data were available at the time . + the same problem with the c - h stretches that was encountered for benzene and pyridine occurs for pyridazine as well ; the situation is further complicated since the experimental data differs and thus we can not assign the frequencies of the experiment . despite the discrepancies between various experimental numbers , all the c - h stretches appear plausible within the error range of our method . + as the ins data set dramatically varies from the remainder of the experimental data , and because it did not appear to be reliable for pyridine , it was excluded from our analysis , although it is reported in table [ tab4 ] for completeness . + since such a large amount of experimental data is available , it is best to discuss each mode separately , starting with the in - plane modes . * for @xmath37 , all experiments except the first of ref.@xcite agree on an assignment to the 1570 @xmath1 band . our results , however , would seem to indicate that the assignment around 1555 @xmath1 might be more plausible . * the same occurs for @xmath31 . experimental values of refs.@xcite cluster around 1440 @xmath1 , with refs . @xcite around 1415 @xmath1 . our calculation clearly favors the former assignment . * for @xmath24 , significant differences between experiments of refs.@xcite and the computed value of 1145 @xmath1 can be found . the experimental results are thus untenable in light of this huge discrepancy of more than a hundred @xmath1 , and only ref.@xcite proposes an assignment of 1160 @xmath1 , which still appears to be on the high side . * for @xmath34 , most experiments have assignments around the computed 1157 @xmath1 ; however , ref.@xcite assigned this mode to 1120 @xmath1 because of their previous assignment . our calculated @xmath24 and @xmath34 nearly coincide , * for @xmath38 , all experimental datasets are in agreement with each other and our calculations . considering the generally small anharmonicities for this type of vibrations , the discrepancy of 40 @xmath1 between the computed @xmath39 and the observed @xmath40 seems to be a bit on the large side . the experimental assignment may have been complicated by the band s position in an ir band envelope going from 960 to 980 @xmath1@xcite . * as with @xmath24 , the experimental assignment for @xmath41 of vazquez et al.@xcite differs from the other obtained fundamentals , and is the only one that can be confirmed by our calculations . * for @xmath16 , only refs.@xcite differ from the computed values , while for @xmath42 , the experimental assignment of stidham and tucci@xcite both in the ir and the raman phase appear doubtful . @xmath43 agrees nicely with all experiments . * for @xmath32 and @xmath22 , the same assignment problems that we experienced for @xmath24 and @xmath34 seem to have occurred ; while most experiments assign values above 1100 @xmath1 to @xmath32 , only vazquez et al.@xcite seem to give the correct assignment . this results , however , in the value of @xmath22 thus ending up a bit on the low side , and only the value of ref.@xcite ends up close to our calculated number . the same seems to have happened to @xmath30 , where only ref.@xcite is in line with our calculations . hence , for the in - plane modes , the data set of vazquez et al.@xcite appears to be most reliable , excluding @xmath37 and @xmath34 ( which appear to be dubious assignments ) and perhaps @xmath40 and @xmath22 . * for the out - of - plane modes , the experimental assignment appears to be even more problematic . for @xmath44 , all experiments differ by 20 @xmath1 , and are either found at 998 or 842 @xmath1 . both assignments appear to be unlikely , since the harmonic b97 - 1 frequency is at 981 @xmath1 , which would put the fundamental too low for 998 and definitely way too high for assignment to the 842 @xmath1 band . * for @xmath45 , all experimental assignments are around the computed 751 @xmath1 , while for @xmath46 , all the experiments appear to yield assignments higher than our computed fundamental of 362 @xmath1 . * based on our data , all the @xmath47 symmetry modes ( @xmath48 through @xmath49 ) should probably be re - examined , since several experimental results differ drastically from each other and from our calculations . a discussion of these modes is almost impossible , and with the exception of refs . @xcite for @xmath49 , all experimental assignments lie beyond the likely error bars of the theoretical values . on the whole , the pyridazine vibrational spectra definitely merit further investigation , as the many experimental and theoretical datasets available are considerably at variance with each other . thus , it is impossible to give an error estimate for the method used based on this molecule . however if our computational results prove to be as accurate as for benzene and pyridine , the experimental spectra have to be reexamined . based on this study , one might wonder whether the addition of more nitrogens or the n - n bond is causing the deterioration of the calculated values . pyrimidine ( figure 1d ) , has been studied less in recent years . notable are the ir experiments compiled by irm@xcite , the ir experiments of billes et al.@xcite , and the inelastic neutron scattering work of navarro et al.@xcite . here , we compare mainly to the ir experiments , since the ins data for neither pyridine nor pyridazine agree well with our computed values and the other experiments , for reasons outlined above . nevertheless , for pyrimidine the ins data seem a lot closer to the ir data than for the previous two azabenzenes ( table [ tab5 ] ) . the values for the @xmath50 c - h stretches are in agreement with experiment , although this seems fortuitous . otherwise , all in - plane modes are within 8 @xmath1 of experiment , with the sole exception of @xmath42 . interestingly , both @xmath38 and @xmath41 undergo strong fermi type 1 and 2 resonances ; their resonance matrices are displayed in table[tab6 ] with their respective eigensolutions . for @xmath38 , very small changes in the unperturbed fundamentals would result in the value ending up not at 1071 @xmath1 , but at the experimental assignment of 1065 @xmath1 . the assignment of the two resonating modes appears to be purely arbitrary . as for @xmath41 , the resonating overtone of @xmath46 matches the experimental number exactly . since both modes are in resonance , the overtone might have `` borrowed intensity '' ( as described in ref.@xcite ) from @xmath41 , and thus the assignment . + when looking at the out - of - plane modes in table [ tab5 ] , only @xmath48 and @xmath22 yield different results compared to the first set of experimental results . both discrepancies have been mentioned before by the harmonic - only study@xcite , and @xmath22 has a very small intensity . moreover , by using scaled hf frequencies , wiberg@xcite has suggested that the assignment for @xmath48 might also be incorrect . the irm@xcite assignments for modes 5 , 6 , 13 and 21 appear more plausible than from ref.@xcite . neglecting the problematic modes 10 and 20 and reassigning mode 9 to 677 @xmath1 , the errors , including the c - h stretches , are 6.3 ( mean absolute ) and 11.4 @xmath1 ( rms ) , and without them are 4.1 and 5.3 @xmath1 , respectively . thus this method again appears to be very reliable for predicting fundamental frequencies . as with pyrimidine , only few experiments have been done for pyrazine ( figure 1e ) . the vibrational assignments are displayed in table [ tab7 ] . in ref.@xcite , assignments were done in the gas phase for the ir active modes and in a melt for the raman spectrum . overall , the agreement is again very good for modes other than c - h stretches . if the previously mentioned tendency to underestimate the c - h stretches is taken into account , all assignments appear plausible , with the possible exception of the assignment of @xmath36 from ref.@xcite . it should be noted that all of the c - h stretches are perturbed by fermi resonances . for the in - plane modes , especially @xmath30 appears to lie beyond the error bars of our method . it has been mentioned by the harmonic - only study in ref.@xcite that this ` kekul mode ' is problematic and it was concluded that this mode has significant multireference character , since it corresponds to the dissociation into 2 hcn + c@xmath8h@xmath8 . however , hybrid density functionals are usually capable to describe nondynamical correlation to a certain extent , since for example the atomization energy of ozone is reasonably well reproduced . for this mode , the assignment of ref.@xcite appears to be out of question . + for all the other modes , most the experiments scatter only by about 5 @xmath1 , with this agreement suggesting that their assignments and values are pretty accurate . the out - of - plane modes are generally quite well described by theory . note that the two @xmath51 modes since they are neither ir nor raman active ; for this reason they are not assigned in ref.@xcite . the ins values are somewhat disappointing for both frequencies , since the chief advantage of this method is the ability to predict such bands . another discrepancy is the first @xmath52 mode ( @xmath41 ) , although the second experiment of ref.@xcite is closer to our values . + this leads to mean absolute and rms errors compared to ref.@xcite ( ref.@xcite ) of 7.8 ( 7.9 ) and 10.4 ( 11.1 ) @xmath1 including the c - h stretches , and of 5.4 ( 5.1 ) and 6.9 ( 6.5 ) @xmath1 when excluding them . if the problematic @xmath53 and @xmath41 are excluded , the mean absolute and rms errors for the non c - h stretches reduces to 4.1 ( 4.1 ) and 5.4 ( 5.5 ) @xmath1 , respectively . thus , the general rms discrepancy of 5 to 7 @xmath1 is obtained for pyrazine as well . + 1,3,5-triazine ( figure 1f ) is the only azabenzene with a degenerate point group symmetry ( @xmath54 ) . because of this , the @xmath55 modes ( @xmath31 and @xmath24 ) are both ir and raman inactive . the ins @xcite value for @xmath56 coincides exactly with our calculation ( table [ tab8 ] ) ; the discrepancy of 124 @xmath1 for @xmath24 is outlandish even for the ins data and this band is probably misassigned . the inelastic neutron scattering data@xcite are clearly inferior to the other experimental datasets . this is hardly surprising as this method relies on fitting and scaling a harmonic force field which is obtained from dft in small split - valence basis sets . already for the harmonics , this introduces errors on the order of about 50 @xmath1 . the error for the c - h stretches is expected to be even larger . it would be interesting to reanalyze their data based on the present anharmonic force fields , although this still would not resolve the problems associated with the c - h stretches , for which large basis set coupled cluster frequencies would be desirable . the experimental ir / raman data of daunt et al.@xcite and lancaster et al.@xcite are very similar to those of irm@xcite and are not listed in table [ tab8 ] . with the exception of the first @xmath57 @xmath17 and the c - h stretches , which are again underestimated , all computed values are within 11 @xmath1 of experiment again , the errors are similar to the other azabenzenes ; including the c - h stretches , the mean absolute error is 7.6 @xmath1 and the rms error 11 @xmath1 . excluding the stretches ( and @xmath17 ) , these numbers reduce to 5.1 ( 3.4 ) and 7.5 ( 5.0 ) @xmath1 , respectively . for 1,2,3-triazine ( figure 1 g ) , only one ir and raman spectrum in kbr has been experimentally determined@xcite : the assignment was assisted by scaled mp2/6 - 31 g * harmonic frequencies . the spectrum looks very similar to that of pyrimidine ; even the same fermi resonances are present . + at first glance at the spectra in table [ tab9 ] , the agreement between theory and experiment is disappointing . for very few of the modes is there agreement between the spectra . however , the authors did not consider _ any _ fermi resonances in the lower spectrum ( although they noted that one might exist , involving @xmath42 ) , and simply assumed the ordering of the fundamentals to be identical to that of the mp2/6 - 31 g * harmonics . this however leads to obvious misassignments . for example , @xmath34 and @xmath43 lie almost on top of each other , while @xmath58 borrows intensity from @xmath34 because of the fermi resonance , as can be seen in table [ tab10 ] . if the resonance were not as strong as computed specifically , if the unperturbed fundamental were shifted only a few @xmath1 towards higher values , the perturbed @xmath58 , presently calculated at 1056 @xmath1 , might easily correspond to the band at 1064 @xmath1 , assigned to @xmath34 in ref.@xcite . the same actually applies to @xmath22 , which has a strong fermi resonance with @xmath59 + @xmath49 , that has , however , been calculated at a lower value of 644 @xmath1 . we have also reassigned the first @xmath50 c - h stretch ( mode 1 ) , since the discrepancy between experiment and theory appears too high even for a c - h stretch . needless to say , this mode also has a strong fermi resonance with the @xmath60 overtone , which has its largest component for the eigenvector with associated eigenvalue at at 3100 @xmath1 , again explaining the experimental value perfectly . the only other value which is still in question would be a small peak at 1597 @xmath1 , which might correspond to an impurity ; a band was found that was assigned to acetylene , a fragmentation product of 1,2,3-triazine . the value of 318 @xmath1 , which has been reported for @xmath49 is actually just a scaled hartree - fock harmonic frequency . + therefore , a dramatic change in the errors is obtained by reassigning the experimental spectrum . including the c - h stretches in the error analysis , the mean absolute error decreases from 17.7 to 4.7 @xmath1 and the rms error from 27.9 to 6.1 @xmath1 . by excluding the c - h stretches , the mean absolute and rms errors improve from 14.6 and 22.1 @xmath1 , respectively , to 4.1 and 5.2 @xmath1 . thus , accuracy consistent with the preceding azabenzenes can be obtained following some reassignments in the experimental spectrum . this also corroborates our conclusion concerning pyridazine that neither substitution of c - h by n nor n - n bonds seem to be responsible for the discrepancies between experimental and computed vibrational spectra . because 1,2,4-triazine ( figure 1h ) is highly unstable , only one `` full '' experimental spectrum has been measured as a liquid film@xcite , although more recently two high - resolution ir spectra have been recorded for four frequencies@xcite . for the c - h stretches ( table [ tab11 ] ) , possibly @xmath61 and @xmath53 may have been swapped , since they both take part in a large resonance matrix block . overall , ten frequencies and combination bands interact forming this large resonance block that include modes 1 to 3 . hence , it is not surprising that a correct assignment based merely on experimental data would be problematic . with this swap , all three c - h stretches are underestimated by the usual 10 - 30 @xmath1 . only @xmath36 appears questionable in view of our results . however , @xmath41 has a fermi resonance with the combination of @xmath42 and @xmath30 , and the resonance , eigenvector and eigenvalue matrices are displayed in table [ tab12 ] . the fermi resonances displayed in these tables are similar to pyrimidine , 1,2,3-triazine and 1,2,4-triazine . assuming that both of the fermi resonances are slightly stronger than predicted , since @xmath30 is underestimated by 3 @xmath1 compared to the high - resolution ir data , the new @xmath42 + @xmath30 could well end up lower than the predicted 1143 @xmath1 , explaining the value of 1136 @xmath1 that would then not be assigned to any mode . the assignment of @xmath36 in the experiment would then correspond to @xmath48 , in agreement with the calculated spectrum . the assignment of @xmath49 also appears to be incorrect . + of the four modes obtained by high - resolution ir , the newer data for @xmath17 validates our computational value of 1044 @xmath1 rather than the 1050 @xmath1 predicted by the older experiment . on the other hand , the low - resolution ir value of 769 @xmath1 for mode 19 is closer to the hi - res data compared to our predicted 779 @xmath1 . these new experiments show the usefulness of the theoretical value , since its resolution is apparently even comparable to that obtained by low - resolution ir data . + without making any reassignments to the experimental spectrum , the mean absolute error from experiment is 13.6 @xmath1 , and the rms error is 19.2 @xmath1 including the c - h stretches . this appears to be quite large in comparison to the other azabenzenes . with the abovementioned suggested reassignments , these errors change to 7.5 @xmath1 and 10.0 @xmath1 , respectively . excluding the c - h stretches , we once again end up with a mean absolute error of 5.1 @xmath1 and an rms error of 5.8 @xmath1 . for 1,2,4,5-tetrazine ( figure 1i ) , ccsd(t ) harmonic frequencies are available in the literature@xcite . unfortunately , they were calculated with all electrons correlated in a basis set ( 6 - 311 g * * ) that is only minimal in the inner - shell orbitals . the resulting errors in harmonic frequencies can easily exceed 20 @xmath1 for small molecules@xcite , several times larger than the actual effect of neglecting core correlation ( typically less than 10 @xmath1 in hcnof systems@xcite ) . we have recomputed the ccsd(t)/6 - 311 g * * frequencies correlating only valence electrons ( table [ tab13 ] ) . for the in - plane modes , differences with the all - electron calculations are quite minor ( 6 @xmath1 or less ) . the out - of - plane modes , however , are drastically affected , up to 60 @xmath1 ( table [ tab13 ] , especially the @xmath62 and the second b@xmath63 frequencies ) . this adds yet another stanza to the long litany of rejoinders in the literature ( e.g.@xcite ) against the practice of correlating inner - shell electrons in basis sets unsuitable for the purpose . + we have recalculated the harmonic frequencies at the ccsd(t ) level ( frozen core ) with basis sets of @xmath0 quality , specifically dunning s popular cc - pvtz @xcite and the almlf - taylor atomic natural orbital@xcite basis set used in ref.@xcite for benzene . the ccsd(t)/cc - pvtz and ccsd(t)/ano4321 frequencies ( table [ tab13 ] ) are in excellent agreement with each other , the largest difference being 6 @xmath1 for the lowest frequency ( @xmath64 ) . ( this is very unlike the case of benzene@xcite , where the two @xmath52 modes are hypersensitive to the basis set and differ significantly even between cc - pvtz and ano4321 . ) . the two corresponding modes in s - tetrazine display appreciable differences between 6 - 311 g * and cc - pvtz but barely between cc - pvtz and ano4321 . ) in contrast , the ccsd(t)/6 - 311 g * * frequencies differs significantly from both data sets , particularly for the two @xmath52 modes ( 39 and 33 @xmath1 w.r.t . ano4321 ) , followed by the lowest @xmath65 mode ( a.k.a . ` kekul mode ' , 27 @xmath1 ) and the @xmath51 mode ( 25 @xmath1 , but a relative error of 7% ) . so even for a valence - only ccsd(t ) calculation , the 6 - 311 g * * basis set is insufficient , and the rms deviation from the ccsd(t)/ano4321 results , 16.4 @xmath1 ( 17.3 when excluding the c - h stretching frequencies ) , is comparable to that of the much more cost - effective b97 - 1 calculations , 18.1 ( 15.1 ) @xmath1 . turning now to the b97 - 1/tz2p harmonic frequencies , the most notable differences with the ccsd(t)/ano4321 data are for the ch stretches ( underestimates of 34 @xmath66 for both ) and the kekul mode ( overestimated by 37 @xmath66 ) . possibly , self - interaction error would adversely affect@xcite the presently used functional s description of the potential surface along the kekul mode , and it would be interesting how a self - interaction corrected functional would perform . unfortunately no implementation of such a functional is available to the authors , let alone an implementation including analytical derivatives . exclusive of this mode and the ch stretches , the rms deviation between the ccsd(t)/ano4321 and b97 - 1/tz2p frequencies is about 11 @xmath1 , compared to just 4 @xmath1 for benzene . as expect , agreement between b97 - 1/tz2p fundamentals and experiment is likewise compromised ( table [ tab14 ] ) . we note that s - tetrazine is both the least stable and the most ` inorganic ' molecule of the series , and that it has a low - lying excited state at less than 2 ev@xcite . if we combine the ccsd(t)/ano4321 harmonic frequencies with the b97 - 1/tz2p anharmonic corrections , the picture brightens somewhat . compared to the assignment of irm@xcite , the rms deviation amounts to 9.8 @xmath1 if the symmetric c h stretch is excluded . the peculiarly large splitting of 76 @xmath1 between irm s symmetric and asymmetric ch stretches is impossible to reconcile with any of our calculations , all of which suggest a very small splitting on the order of just 3 @xmath1 , in agreement with the earlier film ir and raman study of sigworth and pace ( sp)@xcite . ( their band origin of 3089 @xmath1 for the symmetric ch stretch is quite compatible with our calculations : irm s value of 3010 @xmath1 was taken from single - vibronic - level fluorescence spectra@xcite . an older gas - phase study by franks , merer , and innes ( fmi)@xcite found both ch stretches at 3090 @xmath1 ; the symmetric one in the ir , the antisymmetric one in the raman spectrum . ) our calculations do not reveal a resonance interaction affecting @xmath61 that is severe enough to make a large ch symmetric / antisymmetric spllitting plausible . we hence conclude that @xmath61 of irm is erroneous . one other irm assignment which is difficult to reconcile with our best calculations concerns the highest @xmath67 mode ( calculated : 902 , irm : 929 @xmath1 ) for which the franks , merer , and innes@xcite value of 904 @xmath1 nearly perfectly matches our calculation . it would appear that even the somewhat degraded performance , exhibited by our model for this molecule , still ought to be quite useful for resolving assignment issues or guiding high - resolution spectroscopic measurements . we have assessed the performance of modern density functional theory ( particularly , b97 - 1/tz2p ) for the anharmonic vibrational spectra of a series of related medium - sized molecules . the c - h stretching frequencies are consistently underestimated by about 20 - 30 @xmath1 ; disregarding them , the level of agreement with experiment that can be achieved is quite astonishing ( on the order of 6 @xmath1 rms deviation for fundamentals ) . somewhat poorer accuracy is achieved for pyridazine and particularly for s - tetrazine . for several systems ( particularly 1,2,3-triazine ) , our calculations strongly suggest revised assignments of the observed frequencies . for pyridazine , unfortunately , the experimental results are mutually contradictory and our analysis inconclusive . for s - tetrazine , basically all predicted fundamentals agree less well with experiment ; the more ` inorganic ' character of the molecule may have adversely affected the performance of our calculations . nevertheless , even here a rms deviation of about 15 @xmath1 for dft and 10 @xmath1 for the combined ccsd(t)/dft results is obtained . + overall , b97 - 1/tz2p quartic force fields combined with second - order rovibrational perturbation theory shows great promise for the assignment of vibrational spectra of medium - sized organic molecules . a. d. boese is grateful for financial support from the feinberg graduate school . this research was supported by the lise meitner - minerva center for computational quantum chemistry , of which j. m. l. martin is a member , and the helen and martin a. kimmel center for molecular design . b97 - 1/tz2p quartic force fields in normal coordinates of the molecules studied are available for download ( in spectro format ) at the url http://theochem.weizmann.ac.il/web/papers/azabenzenes.html . 99 boese , a. d. ; handy , n. c. ; _ j. chem . phys . _ * 114 * 5497 ( 2001 ) . boese , a. d. ; martin , j. m. l. ; handy , n. c. ; _ j. chem . phys . _ * 2003 * , _ 119 _ , 3005 . colwell , s. m. ; handy , n. c. ; lee , a. m. ; _ phys . * 1996 * , _ 53 _ , 1316 . jamorski , c. ; casida , m. e. ; salahub , d. r. ; _ j. chem . phys . _ * 1996 * , _ 104 _ , 5134 . bauernschmitt , r. ; ahlrichs , r. ; _ chem . lett . _ * 1996 * , _ 256 _ , 454 . van caille , c. ; amos , r. d. ; _ chem . _ * 2000 * , _ 317 _ , 159 . amos , r. d. ; _ chem . * 2002 * , _ 364 _ , 612 . dressler , s. ; thiel , w.;_chem . lett . _ * 1997 * , _ 273 _ , 71 . clabo jr . , d. a. ; allen , w. d. ; remington , r. b. ; yamaguchi , y. ; schaefer iii , h. f. ; _ chem . _ * 1988 * , _ 123 _ , 117 . allen , w. d. ; yamaguchi , y. ; a. g. csszr , clabo , d. a. ; remington , r. b. ; schaefer iii , h. f. ; _ chem . phys . _ * 1990 * , _ 145 _ , 427 . martin , j. m. l. ; lee , t. j. ; taylor , p. r. ; franois , j. p. ; _ j. chem . phys . _ * 1995 * , _ 103 _ , 2589 . martin , j. m. l. ; lee , t. j. ; and taylor , p. r. ; _ j. chem . phys . _ * 1998 * , _ 108 _ , 676 . martin , j. m. l. ; _ chem . lett . _ * 1998 * , _ 292 _ , 411 . neugebauer , j. ; hess , b. a. ; _ j. chem . phys . _ * 2003 * , _ 118 _ , 7215 . boese , a. d. ; martin , j. m. l. ; submitted . miani , a. ; cane , e. ; palmieri , p. ; trombetti , a. ; handy , n. c. ; _ j. chem . _ * 2000 * , _ 112 _ , 248 . becke , a. d. ; _ j. chem . phys . _ * 1993 * , _ 98 _ , 5648 . lee , c. ; yang , w. ; parr , r. g. ; _ phys . b _ * 37 * , 785 ( 1988 ) burcl , r. ; handy , n. c. ; carter , s.;_spectrochim . acta a _ * 2003 * , _ 59 _ , 1881 . politzer , p. ; murray , j. s. ; seminario , j. m. ; miller , r. s. ; _ j. mol . ( theochem ) _ * 1992 * , _ 94 _ , 155 . saxe , p. ; schaefer iii , h. f. ; _ j. am . soc . _ * 1983 * , _ 105 _ , 1760 . tobita , m. ; bartlett , r. j. ; _ j. phys . * 2001 * , _ 105 _ , 4107 and references therein ; see also http://www.qtp.ufl.edu/~bartlett/downloads/polynitrogen.pdf . straka , m. ; _ chem . lett . _ * 2002 * , _ 358 _ , 531 . scheiner , a. c. ; schaefer iii , h. f. ; _ j. chem . phys . _ * 1987 * , _ 87 _ , 3539 m. schtz , hutter , j. ; lthi , h. p. ; _ j. chem . phys . _ * 1995 * , _ 103 _ , 7048 . stanton , j. f. ; and gauss , j. ; _ j. chem . phys . _ * 1996 * , _ 104 _ , 9859 . rubio , m. ; roos , b. o. ; _ mol . phys . _ * 1999 * , _ 96 _ , 603 . nooijen , m. ; _ j. phys . a _ * 2000 * , _ 104 _ , 4553 . innes , k. k. ; ross , i. g. ; moomaw , w. r. ; _ j. mol . spectrosc . _ * 1988 * , _ 132 _ , 492 . martin , j. m. l. ; van alsenoy , c. ; _ j. phys . chem . _ * 1996 * , _ 100 _ , 6973 . hamprecht , f. a. ; cohen , a. j. ; tozer , d. j. ; and handy , n. c. ; _ j. chem . phys . _ * 1998 * , _ 109 _ , 6264 . becke , a. d. ; _ j. chem . phys . _ * 1997 * , _ 107 _ , 8554 dunning , t. h. ; _ j. chem . phys . _ * 1971 * , _ 55 _ , 716 . the cambridge analytic derivatives package ( cadpac ) , issue 6.5 , cambridge , 1998 ; developed by amos , r. d. , with contributions from alberts , i. l. ; andrews , j. s. ; colwell , s. m. ; handy , n. c. ; jayatikala , d. ; knowles , p.j . ; kobayashi , r.;laidig , k. e. ; laming , g. ; lee , a. m. ; maslen , p. e. ; murray , c. w. ; palmieri , p. ; rice , j. e. ; simandiras , e. d. ; stone , a. j. ; su , m .- d . ; tozer , d. j. frisch , m.j . ; trucks , g.w . ; schlegel , h.b . ; scuseria , g.e . ; robb , m.a . ; cheeseman , j.r . ; zakrzewski , v.g . ; j.a . montgomery jr . , stratmann , r. e. ; burant , j. c. ; dapprich , s. ; millam , j. m. ; daniels , a.d . ; kudin , k.n . ; strain , m.c . ; farkas , o. ; tomasi , j. ; barone , v. ; cossi , m. ; cammi , r. ; mennucci , b. ; pomelli , c. ; adamo , c. ; clifford , s. ; ochterski , j. ; petersson , g.a . ; ayala , p.y . ; cui , q. ; morokuma , k. ; malick , d.k . ; rabuckm , a.d . ; raghavachari , k. ; foresman , j.b . ; cioslowski , j. ; ortiz , j.v . ; baboul , a.g . ; stefanov , b.b . ; liu , g. ; liashenko , a. ; piskorz , p. ; komaromi , i. ; gomperts , r. ; martin , r.l . ; fox , d.j . ; keith , t. ; m.a . al - laham , peng , c.y . ; nanayakkara , a. ; gonzalez , c. ; challacombe , m. ; gill , p.m.w . ; johnson , b. ; chen , w. ; wong , m.w . ; andres , j.l . ; gonzalez , c. ; m. head - gordon , replogle , e.s . ; and pople , j.a . ; _ gaussian 98 , revision a. 11 _ ( gaussian , inc . , pittsburgh , pa , 1998 ) . mills , i. m. ; in : molecular spectroscopy : modern research , eds . k. narahari rao , c. w. mathews , vol . 1 , academic press , new york 1972 , pp . 115 - 140 . d. a. clabo jr . , allen , w. d. ; remington , r. b. ; yamaguchi , y. ; schaefer iii , h. f. ; _ chem . phys . _ * 1988 * , _ 123 _ , 187 . allen , w. d. ; yamaguchi , y. ; csaszar , a. g. ; clabo , d. a. ; remington , r. b. ; schaefer iii , h. f. ; _ chem . phys . _ * 1990 * , _ 145 _ , 427 . schneider , w. ; and thiel , w. ; _ chem . lett . _ * 1989 * , _ 157 _ , 367 . hoy , a. r. ; mills , i. m. ; and strey , g. ; _ mol . _ * 1972 * , _ 24 _ , 1265 . gaw , j. f. ; willets , a. ; green , w. h. ; handy , n. c. ; in _ advances in molecular vibration and collision dynamics _ ; bowman , j. m. , ed . ; jai : grennwich , ct , 1990 . martin , j. m. l. ; `` polyad : a vibrational perturbation theory program including arbitrary resonance matrices '' ( weizmann institute of science , reovot , 1997 ) . maslen , p. e. ; handy , n. c. ; amos , r. d. ; jayatilaka , d. ; _ j. chem . phys . _ * 1992 * , _ 97 _ , 4233 . handy , n. c. ; willets , a. ; _ spectrochim . acta a _ * 1997 * , _ 53 _ , 1169 . goodman , l. ; ozkabak , a. g. ; thakur , s. n. ; _ j. phys . chem . _ * 1991 * , _ 95 _ , 9044 . martin , j. m. l. ; taylor , p. r. ; and lee , t. j. ; _ chem . _ * 1997 * , _ 275 _ , 414 . handy , n. c. ; murray , c. w. ; and amos , r. d. ; _ j. phys . chem . _ * 1993 * , _ 97 _ , 4392 . handy , n. c. ; maslen , p. e. ; amos , r. d. ; andrews , j. s. ; murray , c. w. ; and laming , g. j. ; _ chem . lett . _ * 1992 * , _ 197 _ , 506 . klots , t. d. ; _ spectrochim . acta a _ * 1998 * , _ 54 _ , 1481 . partal , f. ; fernandez gomez , m. ; lopez gonzalez , j. j. ; navarro , a. ; kearley , g. h. ; _ chem . _ * 2000 * , _ 261 _ , 239 . navarro , a. ; vazquez , j. ; montejo , m. ; lopez gonzalez , j. j. ; kearley , g. j. ; _ chem . lett . _ * 2002 * , _ 361 _ , 483 . billes , f. ; mikosch , h. ; and holly , s. ; _ j. mol . ( theochem ) _ * 1998 * , _ 423 _ , 225 . vazquez , j. ; lopez gozalez , j. j. ; marquez , f. ; boggs , j. e. ; _ j. raman spectrosc . _ * 1998 * , _ 29 _ , 547 . ozono , y. ; nibu , y. ; shimada , h. ; and shimada , r. ; _ bull . jpn . _ * 1986 * , _ 59 _ , 2997 . stidham , h. d. ; and tucci , j. v. ; _ spectrochim . acta a _ * 1967 * , _ 23 _ , 2233 . navarro , a. ; fernandez gomez , m. ; lopez gonzalez , j. j. ; fernandez liencres , m. p. ; martinez torres , e. m. ; tomkinson , j. ; kearley , g. j. ; _ j. phys . * 1999 * , _ 103 _ , 5833 . burcl , r. ; carter , s. ; and handy , n. c. ; _ chem . lett . _ * 2003 * , _ 380 _ , 237 . wiberg , k. b. ; _ j. mol . struct . _ * 1990 * , _ 244 _ , 61 . arenas , j. f. ; lopez navarrete , j. t. ; otero , j. c. ; marcos , j. i. ; _ j. chem . faraday trans . 2 _ * 1985 * , _ 81 _ , 405 . navarro , a. ; lopez gonzalez , j. j. ; kearley , g. j. ; tomkinson , j. ; parker , s. f. ; and sivia , d. s. ; _ chem . phys . _ * 1995 * , _ 200 _ , 395 . kearley , g. j. ; tomkinson , j. ; navarro , a. ; lopez gonzalez , j. j. ; fernandez gomez , m. ; _ chem . * 1997 * , _ 216 _ , 323 . daunt , s. j. ; shurvell , h. f. ; _ j. mol . spectrosc . _ * 1976 * , _ 62 _ , 373 . lancaster , j. e. ; stamm , r. f. ; colthup , n. b. ; _ spectrochim . acta _ * 1961 * , _ 17 _ , 155 . fischer , g. ; nwankwoala , a. u. ; olliff , m. p. ; scott , a. p. ; _ j. mol . spectrosc . _ * 1993 * , _ 161 _ , 388 . neunhoffer , h. ; hennig , h. ; _ chemische berichte _ * 1968 * , _ 101 _ , 3952 . palmer , m. h. ; maier , r. r. j. ; hegelund , f. ; newnham , d. a. ; _ j. mol . spectrosc . _ * 1998 * , _ 192 _ , 331 . hegelund , f. ; larsen , r. w. ; nelander , b. ; christen , d. ; palmer , m. h. ; _ j. mol . spectrosc . _ * 2003 * , _ 217 _ , 9 . martin , j. m. l. ; schwenke , d. w. ; lee , t. j. ; taylor , p. r. ; _ j. chem . phys . _ * 1996 * , _ 104 _ , 4657 . martin , j. m. l. ; _ chem . lett . _ * 1995 * , _ 242 _ , 343 . little can be added to this statement : `` using valence - optimized basis sets and including core - correlation is not only a waste of computer time , but a potential source of problems , as it can substantially increase bsse . this point is not well appreciated : the prevailing view appears to be that no harm can come of correlating the core when the basis set is inadequate . this is not so . '' ( taylor , p. r. ; in _ lecture notes in quantum chemistry _ ; roos , b. o. , ed . ; springer : berling , 1992 ; p. 406 ) . dunning jr . , t. h. ; _ j. chem . _ * 90 * , 1007 ( 1989 ) almlf , j. ; taylor , p. r. ; _ j. chem . phys . _ * 86 * , 4070 ( 1987 ) patchkovskii , s. ; ziegler , t. ; _ j. chem . phys . _ * 116 * , 7806 ( 2002 ) sigworth , w. d. ; pace , e. l. ; _ spectrochim . acta a _ * 1971 * , _ 27 _ , 747 brumbaugh , d. v. ; innes , k. k. ; _ chem . phys . _ * 1981 * , _ 59 _ , 413 . franks , l. a. ; merer , a. j. ; innes , k. k. ; _ j. mol . spectrosc . _ * 1968 * , _ 26 _ , 458 . dilella , d. p. ; and stidham , h. d. ; _ j. raman spectrosc . _ * 1980 * , _ 9 _ , 90 . wiberg , k. b. ; walters , v. a. ; colson , s. d. ; _ j. phys . chem . _ * 1984 * , _ 88 _ , 6067 ; wong , k. n. ; colson , s. d. ; _ j. mol . spectrosc . _ * 1984 * , _ 104 _ , 129 ; walters , v. a. ; snavely , d. a. ; colson , d. a. ; wiberg , k. b. ; wong , k. n. ; _ j. phys . chem . _ * 1986 * , _ 90 _ , 592 . kakiuti , y. ; akiyama , m. ; saito , n. ; saito , h. ; _ j. mol . spectrosc . _ * 1976 * , _ 61 _ , 164 . califano , s. ; crawford jr . , b. _ spectrochim . acta _ * 1960 * , _ 16 _ , 900 . sverdlov , l. m. ; kovner , m. a. ; krainov , e. p. , eds . ; j. wiley and sons : new york , toronto , 1974 ; p. 533 ff . & b97 - 1 & b3lyp@xcite & ccsd(t)@xcite & hma@xmath68@xcite & hw@xmath69@xcite & mcpth@xmath70@xcite + + a@xmath71 & 3189 & 3192 & 3210 & 3198 & 3191 & 3218 + & 1008 & 1012 & 1003 & 1001 & 1008 & 1003 + a@xmath72 & 1380 & 1392 & 1380 & 1378 & 1367 & 1392 + b@xmath72 & 1012 & 1017 & 1009 & 1000 & 990 & 1012 + & 718 & 723 & 709 & 712 & 718 & 717 + e@xmath72 & 3165 & 3168 & 3183 & 3182 & 3174 & 3210 + & 1628 & 1635 & 1637 & 1623 & 1607 & 1645 + & 1195 & 1201 & 1194 & 1185 & 1178 & 1197 + & 617 & 624 & 611 & 610 & 613 & 617 + e@xmath71 & 863 & 864 & 865 & 856 & 847 & 861 + a@xmath73 & 686 & 686 & 687 & 680 & 686 & 683 + b@xmath74 & 3155 & 3159 & 3173 & 3173 & 3174 & + & 1022 & 1031 & 1020 & 1016 & 1024 & 1030 + + b@xmath73 & 1330 & 1333 & 1326 & 1313 & 1318 & 1338 + & 1170 & 1178 & 1163 & 1158 & 1167 & 1163 + e@xmath73 & 987 & 992 & 985 & 978 & 967 & 987 + & 408 & 412 & 406 & 402 & 1058 & 1056 + e@xmath74 & 3180 & 3183 & 3200 & 3186 & 3181 & 3212 + & 1509 & 1519 & 1509 & 1503 & 1494 & 1522 + & 1055 & 1060 & 1056 & 1048 & 1058 & 1057 + + + + & mode & b97 - 1/tz2p & b3lyp / tz2p & ccsd(t)/b97 - 1 & got@xcite & mcp@xcite + + a@xmath71 & 1 & 3048 & 3051 & 3069 & 3074 & 3073 + & 2 & 992 & 995 & 987 & 993 & 993 + a@xmath72 & 3 & 1348 & 1351@xmath75 & 1348 & 1350 & 1350 + b@xmath72 & 7 & 987 & 997 & 984 & 990 & 993 + & 8 & 707 & 708 & 698 & 707 & 702 + e@xmath72 & 15 & 3032@xmath75 & 3028 & 3050 & 3057 & 3057 + & 16 & 1611@xmath75 & 1613@xmath75 & 1620 & 1601 & 1610 + & 17 & 1178 & 1181 & 1177 & 1178 & 1178 + & 18 & 613 & 615@xmath75 & 607 & 608 & 608 + e@xmath71 & 11 & 841 & 846 & 843 & 847 & 847 + a@xmath73 & 4 & 672 & 677 & 673 & 674 & 674 + b@xmath74 & 5 & 3004@xmath75 & 2988@xmath75 & 3022 & 3057 & 3057 + & 6 & 1006 & 1015 & 1004 & 1010 & 1014 + + b@xmath73 & 9 & 1309 & 1305 & 1305 & 1309 & 1309 + & 10 & 1156 & 1163 & 1149 & 1150 & 1148 + e@xmath73 & 19 & 964 & 972 & 962 & 967 & 968 + & 20 & 400 & 403 & 398 & 398 & 398 + e@xmath74 & 12 & 3031@xmath75 & 3023@xmath75 & 3051 & 3064 & 3048 + & 13 & 1486 & 1484 & 1486 & 1484 & 1484 + & 14 & 1045 & 1038 & 1046 & 1038 & 1038 + & & 15.2 ( 4.2 ) & 17.0 ( 4.8 ) + & & 12.2 ( 2.8 ) & 13.7 ( 4.5 ) + & & 7.8 ( 4.5 ) & 7.4 ( 4.1 ) + + & & & & & ins + & mode & harmonic&fundamental & ref.@xcite & ref.@xcite & ref.@xcite & ref.@xcite & ref.@xcite & ref.@xcite & ref.@xcite + + a@xmath76 & 1 & 3193 & 3061@xmath75 & 3068 & & 3082 & 3086 & 3064 & 3070 & 3086 + & 2 & 3166 & 3034@xmath75 & 3056 & & 3052 & 3071 & 3052 & 3053 & 3071 + & 3 & 1604 & 1558 & 1570 & 1555 & 1568 & 1570 & 1572 & 1570 & 1547 + & 4 & 1477 & 1446 & 1415 & 1440 & 1418 & 1444 & 1417 & 1441 & 1465 + & 5 & 1180 & 1145 & & 1340 & 1337 & 1160 & 1347 & 1352 & 1188 + & 6 & 1171 & 1157 & 1159 & 1153 & 1154 & 1119 & 1160 & 1150 & 1177 + & 7 & 1088 & 1064 & 1061 & 1055 & 1061 & 1061 & 1063 & 1063 & 1075 + & 8 & 1006 & 987 & 963 & 960 & 969 & 968 & 964 & 963 & 998 + & 9 & 676 & 669 & 629 & 622 & 622 & 668 & 630 & 632 & 670 + b@xmath8 & 14 & 3180 & 3056@xmath75 & 3085 & & 3082 & 3079 & 3083 & 3080 & 3079 + & 15 & 3162 & 3052@xmath75 & 3056 & & 3080 & 3057 & 3052 & 3041 & 3057 + & 16 & 1601 & 1556 & 1563 & 1540 & 1572 & 1563 & 1566 & 1564 & 1571 + & 17 & 1438 & 1406 & 1446 & 1408 & 1412 & 1413 & 1450 & 1401 & 1436 + & 18 & 1311 & 1284 & 1283 & 1283 & 1281 & 1281 & 1283 & 1287 & 1306 + & 19 & 1081 & 1058 & 1131 & 1112 & 1131 & 1049 & 1129 & 1113 & 1075 + & 20 & 1056 & 1039 & & 1058 & 1058 & 1027 & 1032 & 1052 & 1049 + & 21 & 629 & 622 & 664 & 663 & 673 & 622 & 667 & 660 & 628 + + b@xmath76 & 22 & 981 & 962 & & & 985 & 987 & 842 & 986 & 998 + & 23 & 762 & 751 & 760 & 760 & 745 & 745 & 759 & 755 & 782 + & 24 & 370 & 362 & 369 & 372 & 369 & 376 & 370 & 370 & 375 + a@xmath8 & 10 & 1019 & 1002 & & & 963 & 1025 & 938 & 970 & 1039 + & 11 & 943 & 926 & & & 765 & 945 & 861 & 786 & 949 + & 12 & 775 & 763 & & & 785 & 729 & 753 & 775 & 793 + & 13 & 372 & 365 & & & 377 & 367 & 410 & 363 & 372 + + & & & & ins + & mode & harmonic&fundamental & irm@xcite & ref.@xcite & ref.@xcite + + a@xmath76 & 1 & 3194 & 3074@xmath75 & 3074 & 3082 & 3074 + & 2 & 3161 & 3045@xmath75 & 3052 & 3053 & 3052 + & 3 & 3147 & 3025@xmath75 & 3038 & 3053 & 3038 + & 4 & 1607 & 1565 & 1570 & 1572 & 1580 + & 5 & 1433 & 1404 & 1398 & 1465 & 1390 + & 6 & 1160 & 1139 & 1147 & 1155 & 1139 + & 7 & 1077 & 1071@xmath75 & 1065 & 1065 & 1080 + & 8 & 1004 & 989 & 991 & 969 & 972 + & 9 & 691 & 685@xmath75 & 678 & 679 & 665 + b@xmath8 & 12 & 3151 & 3040@xmath75 & 3086 & 3047 & 3086 + & 13 & 1605 & 1562 & 1568 & 1569 & 1565 + & 14 & 1494 & 1464 & 1466 & 1411 & 1470 + & 15 & 1393 & 1365 & 1370 & 1356 & 1376 + & 16 & 1252 & 1225 & 1225 & 1224 & 1225 + & 17 & 1211 & 1173 & 1159 & 1158 & 1136 + & 18 & 1091 & 1072 & 1071 & 1071 & 1021 + & 19 & 630 & 623 & 623 & 621 & 628 + + b@xmath76 & 20 & 1026 & 1006 & 980 & 1033 & 1070 + & 21 & 981 & 962 & 955 & 980 & 980 + & 22 & 822 & 809 & 811 & 804 & 826 + & 23 & 736 & 723 & 721 & 719 & 722 + & 24 & 347 & 340 & 344 & 347 & 344 + a@xmath8 & 10 & 1001 & 982 & [ 927 ] & 960 & 1020 + & 11 & 404 & 396 & 399 & 398 & 344 + & & 5.3 & 21.2 ( 9.9@xmath68 ) & + + + deperturbed & @xmath77 & @xmath45 + @xmath46 + & 1060.0 & + & -10.9 & 1060.1 + + deperturbed & @xmath59 & 2 @xmath46 + & 682.7 & + & 3.7 & 680.0 + + + + + perturbed & @xmath77 & @xmath45 + @xmath46 + coefficient of & 1049.2 & 1070.9 + @xmath77 & -0.7085 & -0.7057 + @xmath45 + @xmath46 & -0.7057 & -0.7085 + + perturbed & @xmath59 & 2 @xmath46 + coefficient of & 677.4 & 685.3 + @xmath59 & -0.5760 & 0.8175 + 2 @xmath46 & 0.8175 & -0.5760 + & & & & ins + & mode & harmonic&fundamental & irm@xcite & ref.@xcite & ref.@xcite & ref.@xcite + + a@xmath78 & 1 & 3168 & 3043@xmath75 & 3055 & 3053 & 3054 & 3054 + & 2 & 1612 & 1567 & 1580 & 1579 & 1574 & 1571 + & 3 & 1251 & 1233 & 1233 & 1235 & 1230 & 1240 + & 4 & 1037 & 1020 & 1016 & 1015 & 1015 & 1015 + & 5 & 607 & 598 & 602 & 601 & 596 & 602 + b@xmath79 & 11 & 3147 & 3024@xmath75 & 3040 & 3062 & 3040 & 3040 + & 12 & 1576 & 1533 & 1525 & 1522 & 1529 & 1529 + & 13 & 1372 & 1342 & 1346 & 1353 & 1343 & 1359 + & 14 & 715 & 706 & 704 & 698 & 701 & 698 + b@xmath74 & 15 & 3148 & 2990@xmath75 & 3012 & 3017 & 3018 & 3018 + & 16 & 1513 & 1481 & 1483 & 1484 & 1482 & 1485 + & 17 & 1163 & 1136 & 1130 & 1135 & 1135 & 1062 + & 18 & 1030 & 1013 & 1018 & 1020 & 1020 & 1007 + b@xmath73 & 19 & 3162 & 3055@xmath75 & 3069 & 3069 & 3069 & 3069 + & 20 & 1440 & 1413 & 1411 & 1413 & 1416 & 1427 + & 21 & 1214 & 1174 & 1149 & 1337 & 1146 & 1146 + & 22 & 1084 & 1061 & 1063 & 1063 & 1062 & 1131 + + a@xmath80 & 6 & 1002 & 982 & [ 960 ] & & [ 997 ] & 1065 + & 7 & 347 & 338 & [ 350 ] & & [ 422 ] & 350 + b@xmath71 & 8 & 946 & 927 & 927 & 925 & 919 & 940 + b@xmath72 & 9 & 987 & 968 & 983 & 976 & 983 & 983 + & 10 & 775 & 762 & 756 & 755 & 754 & 754 + b@xmath63 & 23 & 804 & 791 & 785 & 785 & 786 & 818 + & 24 & 429 & 421 & 418 & 417 & 420 & 414 + & & 6.9 ( 5.3@xmath68 ) & 7.1 ( 6.7@xmath68 ) & 6.5 ( 5.5@xmath68 ) & + + + & & & & ins + & mode & harmonic&fundamental & ref.@xcite & irm@xcite & ref.@xcite + + a@xmath81 & 1 & 3164 & 3025@xmath75 & 3082 & 3042 & 3042 + & 2 & 1156 & 1137 & 1130 & 1137 & 1125 + & 3 & 1005 & 989 & 989 & 989 & 991 + a@xmath82 & 4 & 1400 & 1375 & [ 1556 ] & & 1375 + & 5 & 1168 & 1124 & [ 1381 ] & & 1000 + e & 8 & 3159 & 3034@xmath75 & 3081 & 3059 & 3056 + & 9 & 1594 & 1545 & 1560 & 1556 & 1555 + & 10 & 1439 & 1405 & 1404 & 1410 & 1414 + & 11 & 1194 & 1167 & 1168 & 1173 & 1165 + & 12 & 688 & 675 & 666 & 676 & 675 + + a@xmath83 & 6 & 947 & 932 & & 925 & 940 + & 7 & 755 & 740 & & 737 & 732 + e & 13 & 1043 & 1017 & & 1034 & 1030 + & 14 & 345 & 338 & & 339 & 333 + & & & 7.5 ( 5.0@xmath68 ) & + + + & & & ir & raman & ir reassigned + & mode & harmonic&fundamental & ref.@xcite & ref.@xcite & + + a@xmath76 & 1 & 3195 & 3032@xmath75 & 3107 & 3110 & 3046 + & 2 & 3165 & 3028 & & 3045 & + & 3 & 1589 & 1548 & 1597 & 1594 & 1546 + & 4 & 1384 & 1347 & 1336 & 1329 & 1336 + & 5 & 1136 & 1122 & 1080 & 1088 & 1124 + & 6 & 1092 & 1081@xmath75 & 1069 & 1064 & 1080 + & 7 & 1006 & 985 & 979 & 977 & 979 + & 8 & 675 & 667 & 660 & 660 & 660 + b@xmath8 & 14 & 3171 & 3042@xmath75 & 3046 & & 3046 + & 15 & 1584 & 1538 & 1545 & 1547 & 1545 + & 16 & 1438 & 1410 & 1410 & & 1410 + & 17 & 1219 & 1199 & 1195 & 1198 & 1195 + & 18 & 1101 & 1078 & 1124 & 1127 & 1080 + & 19 & 974 & 929 & 935 & & 935 + & 20 & 662 & 659@xmath75 & 653 & & 660 + + b@xmath76 & 10 & 1017 & 998 & & & + & 11 & 824 & 811 & 819 & & 819 + & 12 & 783 & 769 & 769 & & 769 + & 13 & 302 & 296 & [ 318 ] & & + a@xmath8 & 8 & 992 & 973 & & & + & 9 & 358 & 350 & & 365 & + & & 22.1 & & 5.2 + + deperturbed & @xmath84 & @xmath17 + @xmath49 + & 1074.5 & + & 10.6 & 1061.4 + + deperturbed & @xmath22 & @xmath41 + @xmath49 + & 655.1 & + & -6.3 & 647.6 + + + + + perturbed & @xmath84 & @xmath17 + @xmath49 + coefficient of & 1055.5 & 1080.7 + @xmath84 & -0.4804 & -0.8770 + @xmath17 + @xmath49 & 0.8770 & -0.4804 + + perturbes & @xmath22 & @xmath59 + @xmath49 + coefficient of & 644.0 & 658.7 + @xmath59 & 0.4939 & -0.8695 + @xmath59 + @xmath49 & 0.8175 & 0.4939 + & & & ir & hires - ir + & mode & harmonic&fundamental & ref.@xcite & refs.@xcite + + a & 1 & 3178 & 3051@xmath75 & 3090 & + & 2 & 3175 & 3075@xmath75 & 3060 & + & 3 & 3150 & 3007@xmath75 & 3035 & + & 4 & 1596 & 1556 & 1560 & + & 5 & 1563 & 1524 & 1529 & + & 6 & 1470 & 1438 & 1435 & + & 7 & 1401 & 1376 & 1380 & + & 8 & 1314 & 1286 & 1295 & + & 9 & 1179 & 1171@xmath75 & 1163 & + & 10 & 1139 & 1107 & 1136 & + & 11 & 1096 & 1067@xmath75 & 1113 & + & 12 & 1063 & 1044 & 1050 & 1043.5 + & 13 & 1011 & 989 & 955 & + & 14 & 726 & 717 & 713 & + & 15 & 632 & 629@xmath75 & & + + a & 16 & 1014 & 1001 & & + & 17 & 990 & 976 & & + & 18 & 862 & 853 & 851 & + & 19 & 791 & 779 & 768 & 768.7 + & 20 & 379 & 372 & & 367.9 + & 21 & 316 & 308 & & 311.3 + & & 16.1(5.8@xmath68 ) & + + + deperturbed & @xmath36 & @xmath42 + @xmath30 + & 1160.4 & + & -13.7 & 1153.4 + + deperturbed & @xmath19&2 @xmath30 + & 622.8 & + & 8.4 & 617.7 + + + + + perturbed & @xmath36 & @xmath42 + @xmath30 + coefficient of & 1142.7 & 1171.0 + @xmath36 & 0.6136 & -0.7896 + @xmath42 + @xmath30 & 0.7896 & 0.6136 + + perturbed & @xmath19&2 @xmath30 + coefficient of & 611.4 & 629.0 + @xmath59 & -0.5949 & 0.8038 + 2 @xmath30 & 0.8038 & -0.5949 + & & cc - pvtz & ano-4321 & cc - pvtz & tz2p + & @xcite & full@xcite & & @xcite & + + a@xmath85 & 3253 & 3223 & 3221 & 3223 & 3226 & 3196 & 3192 + & 1524 & 1456 & 1452 & 1455 & 1457 & 1472 & 1473 + & 1059 & 1018 & 1018 & 1025 & 1025 & 1050 & 1047 + & 758 & 741 & 740 & 747 & 745 & 756 & 751 + b@xmath74 & 3251 & 3222 & 3218 & 3222 & 3225 & 3199 & 3191 + & 1250 & 1212 & 1210 & 1218 & 1216 & 1232 & 1231 + & 1109 & 1088 & 1088 & 1096 & 1098 & 1099 & 1096 + b@xmath73 & 1505 & 1471 & 1465 & 1472 & 1470 & 1479 & 1478 + & 1167 & 1137 & 1134 & 1138 & 1137 & 1151 & 1147 + & 845 & 908 & 905 & 932 & 927 & 964 & 969 + b@xmath79 & 1611 & 1558 & 1552 & 1561 & 1561 & 1559 & 1560 + & 1345 & 1321 & 1319 & 1323 & 1323 & 1330 & 1325 + & 648 & 637 & 635 & 640 & 640 & 656 & 650 + + a@xmath86 & 344 & 257 & 319 & 344 & 346 & 354 & 344 + b@xmath72 & 985 & 947 & 957 & 998 & 996 & 1011 & 999 + & 803 & 772 & 783 & 816 & 816 & 837 & 829 + b@xmath63 & 930 & 896 & 904 & 919 & 921 & 940 & 930 + & 297 & 228 & 270 & 276 & 270 & 258 & 250 + & mode & b97 - 1 & ccsd(t)/b97 - 1 & irm@xcite & ref.@xcite & sp@xcite & fmi@xcite + + a@xmath85 & 1 & 3073 & 3107 & 3010 & 3040 & 3089 & 3090 + & 2 & 1424 & 1408 & 1415 & 1489 & 1417@xmath87 & 1418@xmath87 + & 3 & 1024 & 1000 & 1009 & 990 & 1015 & 1017 + & 4 & 741 & 735 & 736 & 737 & 734 & 736 + b@xmath74 & 11 & 3070 & 3104 & 3086 & 3070 & 3086 & 3090 + & 12 & 1204 & 1189 & 1204 & 1200 & 1204 & 1200 + & 13 & 1079 & 1081 & 1093 & 1106 & 1109@xmath70 & 1103@xmath70 + b@xmath73 & 14 & 1445 & 1437 & 1448 & 1434 & 1448 & 1440 + & 15 & 1112@xmath75 & 1102 & 1108 & 1187 & 1151 & + & 16 & 927 & 885 & 883 & 1085 & 893@xmath70 & 881@xmath70 + b@xmath79 & 8 & 1511 & 1511 & 1525 & 1543 & 1523@xmath87 & 1521@xmath87 + & 9 & 1294 & 1293 & 1290 & 1278 & 1302 & 1303 + & 10 & 645 & 638 & 640 & 679 & 649 & 651 + + a@xmath86 & 5 & 336 & 338 & 335 & 319 & 335 & 335 + b@xmath72 & 6 & 977 & 974 & 994 & 925 & 1015 & + & 7 & 814 & 801 & 801 & 775 & 799 & 800 + b@xmath63 & 17 & 911 & 902 & 929 & 890 & 929 & 904 + & 18 & 244 & 261 & 254 & 340 & 254 & 254 + & 15.2@xmath68(15.1,14.8@xmath69 ) & 54.8 ( 57.5 ) & 19.0 ( 19.4 ) & 16.0 ( 15.6 ) + & 12.2@xmath68(11.3,9.4@xmath69 ) & 63.5 ( 64.5 ) & 20.1 ( 20.3 ) & 11.0 ( 10.2 ) + + + + + +

it has been realized since quite some time that several extensions of the standard model ( sm ) share similar signatures at the lhc , as , for example , that of jets , leptons , and missing transverse energy @xcite . this is because such extensions typically have partners of the sm , with gauge quantum numbers equal to those of the sm particles @xcite . supersymmetric models , extra - dimensional ones , and little higgs models are indeed of this type . among them , supersymmetric models stand out because the partners have spins that differ by @xmath12 from those of the sm particles , whereas in the other extensions have the same spins . this has prompted several proposal of methods for determining the spin of heavy unstable particles @xcite . thus , supersymmetric models are also the only ones , at least so far , in which there exist irreducible renormalizable interactions of four matter partners , of type @xmath13 . such interactions may , therefore , offer a new handle to discriminate supersymmetry ( susy ) from other extensions of the sm . the lhc being a hadronic machine , quartic couplings of sleptons , @xmath14 , contribute only to processes whose cross sections are suppressed by many powers of the electroweak couplings . the signals of quartic couplings of squarks , @xmath15 , may not be easily disentangled from those due to a pair of squarks . the first , simplest scrutiny of quartic couplings , therefore , may be that of couplings of two sleptons and two squarks @xmath16 . combined with the strong interaction vertices of squarks and gluons , these give rise , at the quantum level , to the effective couplings @xmath17 , and therefore to the gluon - initiated production of a pair of slepton . since higgs bosons couple to both , a pair of fermions and a pair of scalars , reducible couplings of four sm partners , two quark partners and two lepton partners , can be built through the exchange of higgs bosons also in models in which the partner spins do not differ from those of the sm particles . thus , the gluon - fusion production of a pair of lepton partners is not a special feature of supersymmetric models . the possibility of discriminating susy through this type of studies strongly relies on the impact that the irreducible quartic couplings of supersymmetric models can have . to the best of our knowledge , the one - loop process @xmath17 via quartic couplings , was never studied before . surprisingly , very little exists in the literature also about the same process induced by the exchange of a higgs boson , @xmath18 . it was studied in ref . @xcite for @xmath19 , the lightest higgs bosons . since this has a mass clearly below the slepton - pair threshold , the cross section was found to be very small . the study of this process was put aside until the authors of ref . @xcite realized that the cross section could increase if @xmath20 is one of the two heavier neutral higgs bosons @xmath21 . neverheless , not much attention has been given to this proposal . although suppressed by a loop factor and by the coupling @xmath22 with respect to that of the standard drell - yan production @xmath23 , the gluon - fusion cross sections may be favoured by the fact that , at the lhc , gluon - initiated processes may dominate over the quark - initiated ones . given the importance of the quartic couplings that may be probed , it is worth studying them in detail . the paper is organized as follows . in sec . [ sec : qcxsec ] we study the slepton pair production processes from gluon fusion induced by quartic couplings . in sec . [ sec : completexsec ] we give the complete one - loop level results , including also the contribution from the higgs boson - exchange . we summarize our findings in sec . [ sec : conclu ] . the relevant quartic interactions in the scalar potential of the minimal supersymmetric standard model ( mssm ) consist of d - term interactions , proportional to gauge couplings squared , and f - terms interactions , proportional to yukawa couplings squared . these are respectively collected in : @xmath24 where @xmath25 is any of the fields @xmath26 , @xmath27 , @xmath28 , @xmath29 , of any generation , and @xmath30 is either the charged component of the su(2)@xmath31 doublet @xmath32 with hypercharge @xmath33 , or the su(2)@xmath31 singlet @xmath34 with hypercharge @xmath35 , also of any generation . the coefficients @xmath36 and @xmath37 are listed in table [ table : dandfcoeff ] , where @xmath38 and @xmath39 are the coupling constants of u(1)@xmath40 and su(2)@xmath31 , with the normalization @xmath41 . the yukawa coupling matrices , denoted by @xmath42 , @xmath43 , and @xmath44 , are assumed to be diagonal and real , with only the @xmath45 element large enough to be relevant for this discussion . these quartic couplings , together with the su(3 ) interactions for squarks : @xmath46 allow a realization of the effective couplings @xmath47 at the one - loop level , different from those due to the mediation of an s - channel higgs boson . in this expression , @xmath48 is the strong coupling and @xmath49 are the generators of su(3 ) . . [ cols="^,^,^,^ " , ] we note here that also in this case sum rules analogous to those in eq . ( [ eq : chsumrule1 ] ) exist for each generation of squarks : @xmath50 as a consequence , similar sum rules hold for the d - terms contributions to the couplings @xmath51 and @xmath52 . this mechanism of production of a pair of sleptons through gluon fusion may also be regarded as a test of susy , although triple couplings of partners of the sm particles are not unique of susy models . in particular , unlike the drell - yan production mechanism , the gluon - fusion mechanism mediated by the exchange of a higgs boson can probe the higgs sector , the @xmath6 parameter , @xmath53 , and the trilinear @xmath5 terms . the parton - level cross section for the gluon - fusion production of a pair of sleptons , can be expressed in terms of two form factors , which parametrize the effective lagrangian induced at the one - loop level by quartic - couplings and higgs - exchange diagrams : @xmath54 here @xmath55 is a fully antisymmetric tensor , with @xmath56 , @xmath57 and @xmath58 are the momenta of the two initial gluons , and @xmath59 is generated by the diagram with exchange of the pseudoscalar higgs boson @xmath5 , whereas all the other contributions are included in @xmath60 : @xmath61 in this expression , the tilde distinguishes the contributions from scalar loops ( @xmath62 is the truncated amplitude discussed in the previous section ) , and the subscript \{@xmath63 } refers to diagrams with exchange of a scalar higgs boson , _ i.e. _ @xmath64 . notice that , because of our assumption of imaginary @xmath65 couplings and real chirality - mixing terms in the sfermion sector , there is no contribution to @xmath66 from the exchange of the pseudoscalar higgs boson . in terms of @xmath60 and @xmath67 the parton - level cross section is @xmath68 the expression for @xmath69 and @xmath67 ( which strictly speaking should be labelled @xmath70 ) coming from fermion loops are @xmath71 a^{\,\widetilde{l}_l \widetilde{l}_r}_p \!&= & \!\ ! \quad \ , -\quad g_{{\cal a } \widetilde{l}_l \widetilde{l}_r } \left(\frac{v}{m_t}\right ) \left ( \frac{1}{1-\ ! m_a^2/\hat{s } + \!i m_a\gamma_a/\hat{s } } \right ) y_{att } \,p(\hat{s};m_t^2 ) , \label{eq : apfermion}\end{aligned}\ ] ] with @xmath72 and @xmath73 , that for @xmath74 is @xmath75 @xmath76 & & \left . \ \left ( s_q^2\ , g_{{\cal h}\widetilde{q}_l\widetilde{q}_l } + c_q^2\ , g_{{\cal h}\widetilde{q}_r\widetilde{q}_r } -2 ( s_q c_q ) g_{{\cal h}\widetilde{q}_r\widetilde{q}_l } \right ) s(\hat{s};\widetilde{m}_{q_2}^2 ) \right ] , \label{eq : assfermion}\end{aligned}\ ] ] with @xmath77 and @xmath78 , and @xmath79 spanning over all three generations . we remind that in the limit of vanishing chirality - mixing terms in the squark sector , @xmath80 , @xmath81 , and @xmath82 , @xmath83 . the function @xmath84 is given in eq . ( [ eq : functions ] ) in sec . [ sec : qcxsecanal ] , @xmath85 and @xmath86 are @xmath87 , \nonumber\\ p(\hat{s};m^2 ) & = & \tau \,f(\tau ) , \hspace*{3.5truecm } \tau \equiv 4 m^2/\hat{s } , \label{eq : functionf } \end{aligned}\ ] ] with @xmath88 already defined in eq . ( [ eq : functionf ] ) . the imaginary and real parts of the three functions @xmath89 , @xmath90 , and @xmath91 are plotted in fig . [ fig : sfpfuncs ] versus @xmath92 . we have taken @xmath93 for @xmath94 and @xmath95 , and the two values @xmath96 and @xmath97gev for @xmath98 . the two resulting functions of @xmath92 in this last case are called @xmath99 and @xmath100 in this figure . see the two olive lines , solid for @xmath99 and short - dashed for @xmath100 . the azure long - dashed line and the red dot - dashed one represent @xmath94 and @xmath95 , respectively . = 12.5 cm it should be noted here that , because of the sum rules in eqs . ( [ eq : chsumrule1 ] ) and ( [ eq : chsumrule2 ] ) , the constant term -1 in the definition of the function @xmath98 does not contribute to the evaluation of @xmath62 and @xmath101 , when the quartic and trilinear couplings considered here are determined by d - terms . in this case , the function @xmath98 contributes to these truncated amplitudes exactly like the function @xmath95 . the values of the relevant real part of @xmath98 is then obtained shifting upward by the constant value 1 the two olive lines in fig . [ fig : sfpfuncs ] . the functions @xmath94 and @xmath95 ( and @xmath98 , when the relevant couplings come from d - terms ) vanish in the limit @xmath102 , for fixed @xmath103 , and also in the limit @xmath104 , for fixed @xmath92 . for @xmath105 , both the real and imaginary part of @xmath94 and @xmath95 are at most of @xmath106 . nevertheless , the contribution from a bottom - quark loop , with truncated amplitudes analogous to those from the top quark given in eqs . ( [ eq : asfermion ] ) and ( [ eq : apfermion ] ) , is enhanced by the factor @xmath107 , and unsuppressed values of the couplings @xmath108 and @xmath109 in the case of large @xmath53 . depending on the value of @xmath53 , it may end up being the dominant contribution . the generalization of the formulae in eqs . ( [ eq : asfermion ] ) , ( [ eq : apfermion ] ) , and ( [ eq : assfermion ] ) to the case with nonvanishing chirality - mixing terms in the slepton sector can be easily obtained using the algorithm of eq . ( [ eq : genqcxsecgenslep ] ) applied to @xmath110 and @xmath111 . even before presenting numerical results , it is possible to draw some general observations on the truncated amplitudes induced by the higgs - boson exchange from an inspection of the couplings @xmath112 , @xmath113 , @xmath114 , and @xmath115 , as well as @xmath116 , and @xmath117 , for a given higgs spectrum . we shall examine below three different higgs spectra . spectrum * i * has all three higgs bosons , @xmath118 , @xmath119 , and @xmath5 , light and maximally mixed , _ i.e. _ @xmath120 . all the three states have masses below the threshold @xmath121 . thus , no resonance is encountered when integrating @xmath92 from @xmath122 to @xmath123 . * ii * has @xmath124 $ \sim$}}~}2 \widetilde{m}_{l}$ ] , _ i.e. _ @xmath124 $ \sim$}}~}200\,$]gev in our case study ; @xmath125 and @xmath126 are calculated as explained at the beginning of this section , and depend on @xmath53 and the specific squark sector chosen . in this case , it is in general @xmath127 , and @xmath128 is small , but nonnegligible . the resonant contributions of @xmath119 and @xmath5 are expected to lift up the values of our cross sections . spectrum * iii * has @xmath5 and @xmath119 rather heavy and completely decoupled from our problem . the mixing angle @xmath129 is such that @xmath130 and @xmath131 , to a very good accuracy . we consider first the case of vanishing chirality - mixing terms in the squark sector , as in the squark spectrum * a * specified in the previous section . in this context we start concentrating on the couplings @xmath132 and @xmath133 , the only ones giving a nonvanishing contribution to @xmath74 . because of the sum rule in eq . ( [ eq : chsumrule2 ] ) , also the contribution to @xmath74 from d - terms , as the contribution to @xmath134 , vanishes identically in the limit of a completely degenerate squark spectrum . the squark spectrum * a * , which we shall use in the following , was however conceived precisely with the aim of evading the effectiveness of this sum rule . unlike in the case of @xmath134 , f - terms give now contributions to @xmath74 , that are nonvanishing . by inspecting the couplings @xmath132 and @xmath133 in table [ table : mehiggsscalcoupls ] , it is easy to see that f - terms assign a privileged role to the @xmath135 squarks . all other f - term contributions to these couplings are small even for large values of @xmath53 . this holds in particular also for the @xmath136 squarks , whose couplings receive an f - term contribution always multiplied by @xmath137 . in addition , the f - term contributions to the couplings @xmath138 and @xmath139 are in general larger than the contributions from d - terms to any of the couplings @xmath132 and @xmath133 , for any @xmath79 . in the case of the @xmath118 exchange , when , however , the cross section is expected to be small , @xmath140 dominates over @xmath141 and @xmath142 by a considerable margin . similarly , @xmath143 dominates also over @xmath144 and @xmath145 , in the case of the heavier higgs boson exchange @xmath119 , except when @xmath146 is very close to its typical decoupling - limit value , @xmath147 . in this limit , however , @xmath119 is too heavy to be of any relevance for our cross sections . = 12.5 cm = 12.5 cm we notice also that , @xmath148 $ \sim$}}~}\vert y_{{\cal h } tt}\vert$ ] . therefore a comparison of the relative size of @xmath149 and @xmath150 , amounts practically to a comparison of @xmath151 versus @xmath152 , @xmath153 . the @xmath135 squark loop is clearly penalized by a factor @xmath154 with respect to the factor @xmath155 , constant in @xmath92 . moreover , as fig . [ fig : sfpfuncs ] shows , when integrated in @xmath92 , the function @xmath98 gives a smaller contribution to @xmath150 than the function @xmath94 gives to @xmath156 , at least as far as the real parts of these functions are concerned . in contrast , the imaginary part of @xmath98 can contribute more than the imaginary part of @xmath94 for very light squarks , _ i.e. _ @xmath157gev . thus , if the squark sector has vanishing chirality - mixing terms , we expect the dominant contribution to @xmath158 to be induced by the @xmath159-quark loop , followed by an in - general - smaller contribution coming from the @xmath135-squark loop , with exact size depending on @xmath122 and the two @xmath135-squark masses . finally there is a much smaller contribution from the @xmath136-squark loop , at least as far as @xmath160 for @xmath161 . since @xmath162 is also small , the production of same - chirality sleptons proceeds always through the d - term contributions to the couplings @xmath163 , for any of the values of @xmath164 considered here ( @xmath165 $ \sim$}}~}3 $ ] ) . therefore , the size of @xmath166 from @xmath135-squark exchange is expected to be in the same ballpark of the size of @xmath167 obtained before , except in the resonant region @xmath168 , where @xmath169 has the chance to exceed @xmath170 substantially . we have assumed here that also the slepton sector has vanishing chirality - mixing terms . = 12.5 cm these observations are confirmed by the results shown in figs . [ figplot : qcvshmed ] and [ figplot : qcvshmedleft ] , where we plot the cross sections for the production of right - handed and left - handed sleptons , respectively , versus @xmath171 , which is defined in sec . [ sec : qcxsecnumer ] . in both cases we give these results for the higgs spectrum * i * and * iii*. for spectrum * i * , we have used the values @xmath172gev , @xmath173gev , and @xmath174 . in spectrum * iii * , it is @xmath175gev . the higgs - mediated cross sections , shown by the red and blue solid lines , are almost horizontal , at @xmath176fb for the higgs spectrum * i * , and @xmath177fb for the higgs spectrum * iii*. that is , in spite of the fact that spectrum * iii * contributes with the exchange of @xmath118 only , whereas spectrum * i * contributes with @xmath118 and @xmath119 , the value of the higgs - exchange cross section is very similar in the two cases , provided the two scalar higgs bosons in spectrum * i * are practically degenerate and have mass very similar to that of the light higgs in spectrum * iii * , _ i.e. _ @xmath178 . this is true for both , @xmath179 and @xmath180 . it is easy to convince oneself of this by inspecting the relevant couplings in tables [ table : gehiggsscalcoupls ] and [ table : mehiggsscalcoupls ] . a splitting between @xmath181 and @xmath182 , however , is in general present in spectrum * i * , with size depending on the value of @xmath53 and the squark spectrum . this accounts for the roughly @xmath183 difference in the cross sections obtained for the higgs spectrum * i * and * iii*. being practically independent of @xmath184 , the higgs - mediated cross sections show explicitly that the contribution from @xmath180 is small compared to that from @xmath185 , as already anticipated . the quartic - coupling cross sections are shown by the black dot - dashed lines . they coincide with those in fig . [ figplot : qcxsca ] . the total gluon - initiated cross sections are given by the two red ( blue ) dashed lines , the upper one for spectrum * i * , the lower one for spectrum * iii*. although the interference effects can be significant , they are similar in size to the corresponding ones induced by quartic couplings only . we remind that the squark spectrum * a * used for these figures was chosen in such a way to maximize the size of @xmath170 , but not necessarily that of @xmath186 . indeed , the value of the lightest @xmath135-squark in the loop ranges between 200260@xmath187gev when @xmath171 spans the interval 100200@xmath187gev . enhancements of these cross sections are possible in the region in which the higgs boson @xmath119 and/or @xmath5 are resonant . the cross sections in this region are very sensitive to the widths @xmath188 , which we have evaluated with the program cpsuperh @xcite . we plot in fig . [ figplot : qchmres ] the cross sections obtained for the production of same - chirality sleptons at the lowest point in the squark spectrum * a * , as a function of @xmath126 . the smallest value of @xmath126 is that of the higgs spectrum * i * of figs . [ figplot : qcvshmed ] and [ figplot : qcvshmedleft ] , the largest value , is quite close to the @xmath126 of spectrum * iii * in these same figures . no exchange of the pseudoscalar higgs boson @xmath5 is possible in this case . the production of right - handed sleptons is shown by red lines , that of left - handed sleptons by blue lines . the solid lines show the cross sections induced by higgs exchange only , the dashed lines show the cross sections due to both production mechanisms . the enhancement over the values of the cross sections shown in figs . [ figplot : qcvshmed ] and [ figplot : qcvshmedleft ] is considerable for a rather large range @xmath126 . although the @xmath135-squark loop gives a subdominant contribution to the higgs - exchange cross section with respect to @xmath159-quark loop , a smeared peak is nevertheless visible at @xmath189 . far less visible is that at @xmath190 , because of the smoother behaviour at @xmath191 of the real and imaginary part of the function @xmath94 , with respect to that in the case of the function @xmath98 . ( see fig . [ fig : sfpfuncs ] . ) at the peak of these resonances , the production cross sections for a pair of right - handed sleptons are roughly twice as large as those for a pair of left - handed sleptons . this is mainly due to the fact that @xmath192 , for @xmath193 . as in the case of those shown in fig . [ figplot : qcxsca ] , also these cross sections apply to the production of slepton of first and second generation , as well as those of third generation for low @xmath53 . a value @xmath194 , for example , suppresses the @xmath195-lepton yukawa couplings sufficiently to reduce @xmath196 ( see discussion in sec . [ sec : qcxsecnumer ] ) to be about few percent , as requested in the squark spectrum * a*. a moderate cancellation between @xmath197 and @xmath198 takes place in the chirality mixing terms of the @xmath4 sector , for a value of @xmath199 gev and @xmath200gev , presumably not too large to jeopardize the metastability of the color- and charge - preserving vacuum . the suppression of @xmath201 , however , is not sufficient to reduce the couplings @xmath202 , @xmath203 and their conjugates , in which the parameters @xmath197 and @xmath6 appear in different combinations , to be negligible with respect , for example , to the couplings @xmath204 , @xmath205 coming mainly from d - terms . thus , the production of opposite chirality @xmath4-sleptons is possible even for chirality - mixing terms in both , the @xmath4-slepton mass matrix as well as in squark mass matrices that are nearly vanishing , and for a relatively low value of @xmath53 . this is in contrast to the quartic - coupling case , in which the the production of opposite - chirality @xmath4 sleptons requires a large value of @xmath53 . = 12.5 cm indeed , not only is the production of opposite - chirality @xmath4 s possible in this case , but it has a cross section larger than that shown for the same spectrum in fig . [ figplot : qchmres ] , at least in the resonant higgs region . this is because both the two heavier higgs bosons , @xmath119 and @xmath5 , can mediate the @xmath159-quark contribution to the cross section . notice that , because no chirality - mixing terms are present in the squark sector , the exchange of the pseudoscalar higgs boson is possible only with the @xmath159-quark loop . given the sharp behaviour of the function @xmath95 shown in fig . [ fig : sfpfuncs ] , the peak at the @xmath159-quark threshold is now considerably more pronounced than in the case of the cross section of fig . [ figplot : qchmres ] with exchange of only the higgs boson @xmath119 . the values of the cross section that this last figure shows on top of the higgs resonance are now quite interesting . we expect these values , as well as those for the production of same - chirality sleptons , to increase even further for nonvanishing chirality - mixing terms in the squark sector . depending on the values of the trilinear soft parameters , @xmath6 , and @xmath53 , also the @xmath136-squark contribution may become competitive with that due to the @xmath135-squark and possibly also the @xmath159-quark . the results shown here are rather promising . we believe these cross sections should be studied in full generality for more realistic sfermion spectra than the one considered here , and for all values of @xmath53 , in order to ascertain the possibility of probing , through them , important elements of the supersymmetric parameter space . we have discussed the possibility of producing slepton pairs through gluon fusion . our guiding motivation was that of assessing the impact that the quartic couplings @xmath206 can have at the loop level for slepton production . if unmistakably detected , coupling such as these could provide an important confirmation of supersymmetric models . we have found that the corresponding cross sections are small . that due to the only sizable quartic coupling from f - terms is small because it is proportional to the square of the product of the yukawa couplings of the @xmath207 quark and the @xmath195 lepton . the slepton produced are in this case @xmath4 sleptons . the values of these couplings can be increased for large @xmath7 . nevertheless , the cross section gets in general penalized by the mixing effects in the slepton and the squark sector , in spite of the fact that this last mixing is necessary to produce the effective coupling @xmath208 . the cross section for slepton production is small also when it is induced by quartic couplings from d - terms . the reason is due to a peculiar cancellation mechanism . like the glashow iliopoulos - maiani ( gim ) mechanism cancels the contributions from degenerate fermions of up or down - type quarks with different flavour in loop - induced flavour - changing - neutral - current processes , this mechanism cancels the contributions to slepton production from squarks that are degenerate within each generation . the two cancellation mechanisms have however quite different origins . the gim mechanism is due to the unitarity of the cabibbo - kobayashi - maskawa mixing matrix , whereas the cancellation observed in slepton production is due to the fact that the sum of the sm charges of all quark fields in each family vanishes identically . in contrast , what we had hoped to be only a supersymmetric background to the gluon - fusion production of slepton pairs through quartic couplings , _ i.e. _ the gluon - fusion production with the exchange of a higgs boson , turns out to give interesting results . the cross section can reach up to @xmath3fb if the exchanged higgs boson becomes resonant . this value is obtained at the peak of the resonance , for a slepton mass of @xmath209gev , for vanishing chirality - mixing terms in the squark sector . in this case , the dominant contribution comes from the @xmath159-quark loop that of the @xmath135-squark is subdominant , and the @xmath136-squark contribution is practically negligible . the same spectrum gives a cross section up to @xmath210fb for the production of a pair of opposite - chirality @xmath4 s even for @xmath211 and chirality mixing terms in the @xmath4 sector practically negligible . notice with the same type of slepton spectrum , only same - chirality sleptons can be produced through the drell - yan production mechanism @xcite . we have not given explicit values of the cross sections when chirality - mixing terms in the squarks and slepton sectors are nonvanishing , and for large values of @xmath53 , but we have argued why we expect the cross section to increase in this case , at least for sleptons still relatively light . a full - fledged analysis should be performed with more realistic squark spectra than those used here , which were devised to maximize the gluon - fusion cross sections from quartic - couplings . moreover , when @xmath53 is large , the tree - level process @xmath212 mediated by higgs boson exchange , should also be added , incoherently , to the previous processes , in order to give a correct assessment of the values that the gluon - initiated cross section can reach . this is because the @xmath207 quark is contained in the proton via a gluon and the @xmath213-initiated cross section is of the same order in @xmath22 than those studied here . finally , where the gluon - fusion mechanism of slepton production stands with respect to the drell yan production depends also strongly on the role played by the qcd corrections to the two types of processes . higher - order qcd corrections should indeed be included , together with susy - qcd corrections . the qcd corrections for the gluon - fusion mechanism are as those for the production of higgs bosons @xcite . they are known to be large , more than the corrections that qcd induces in the case of the drell yan production @xcite . the susy - qcd corrections are also well known @xcite . at large @xmath53 those to the higgs boson vertex may be well approximated @xcite by the usual yukawa coupling corrections . it is widely believed that , if at all , sleptons will be discovered at the lhc through cascade decays . the production of a pair of sleptons , if kinematically allowed , will provide their direct mass measurement . not much more than this can be learned from the drell - yan production . the gluon - fusion production mechanism , in this respect is potentially much more interesting since it can provide valuable information about @xmath5 , @xmath6 , and @xmath7 , and confirmations about the higgs spectrum . both mechanisms contribute to the pair production of sleptons , and the total cross section may be larger than previously thought , at least for resonant higgs exchange . more detailed studies on this subject are clearly needed . going back to the issue of quartic couplings , an extended gauge structure , for example with an anomalous u(1)@xmath40 in which the cancellation mechanism can be avoided , may help with their detection . it is however possible that the hadron collider environment is not the best suited for this purpose . the possibilities that a linear collider may offer in this respect should be explored . in the case of its photon - collider option , for example , it is easy to find a situation that parallels very closely the one described here of a pair of sleptons , which , in spite of not being sensitive to strong interactions , can nevertheless couple at the one - loop level to a pair of gluons , thanks to their quartic couplings to squark pairs . the situation that immediately comes to mind is that of a pair of neutral scalars that can be coupled at the one - loop level to two photons thanks to their quartic couplings to a pair of charged scalars . if the two neutral particles are supersymmetric higgs bosons , the situation can actually be much better than that of slepton production discussed here . indeed , the cancellation mechanism above described , which does work if the charged particles exchanged in the loop are squarks , is violated when these scalars are sleptons . this is simply due to the fact that the neutral components of the su(2)@xmath31 slepton doublets can not couple to photons . the production of a pair of supersymmetric higgs bosons in a photon collider was already considered in several papers @xcite . the charged slepton contribution to the loop , is mentioned only in ref . it is , however , difficult to understand from this analysis the relevance of this contribution . it may remain buried under possibly larger contributions , not only those due to the exchange of higgs bosons , but also those coming from box diagrams , that have no correspondent in our case of slepton production . a check of the relative size of all these contributions would probably be worthwhile . + the authors thank e. accomando , b. allanach , k. fujii , j.s . lee , k. odagiri , n. okamura , n. polonsky , and k. yokoya for stimulating discussions and inputs , d.w . jung for initial collaboration , and m. spira for a critical reading of the manuscript . f. b. acknowledges the ncts - kek exchange program , which has made this collaboration possible and the hospitality extended to her by the yukawa institute in kyoto , during the ykis 2009 . the work of f.b . was partially supported by the excellency research project of national taiwan university , taiwan : `` mass generation , heavy flavours , neutrinos at the particle physics frontier '' , grant no 97r0066 - 60 . is supported in part by the grant - in - aid for scientific research ( no . 20340064 ) from mext , japan . h. murayama , _ confusing signals of supersymmetry _ , international linear collider workshop ( lcws2000 ) , oct . 24 - 28 , 2000 , fermilab , + http:@xmath214conferences.fnal.gov/lcws2000/web/p3_murayama/index.html . papers cited in ref @xcite and in : + f. boudjema and r. k. singh , _ a model independent spin analysis of fundamental particles using azimuthal asymmetries _ , jhep * 0907 * ( 2009 ) 028 , + s. y. choi , k. hagiwara , h. u. martyn , k. mawatari and p. m. zerwas , _ spin analysis of supersymmetric particles _ j. c * 51 * ( 2007 ) 753 e. eichten , i. hinchliffe , k. d. lane and c. quigg , _ super collider physics _ , rev . * 56 * ( 1984 ) 579 [ addendum - ibid . * 58 * ( 1986 ) 1065 ] , + s. dawson , e. eichten and c. quigg , _ search for supersymmetric particles in hadron - hadron collisions _ , d * 31 * ( 1985 ) 1581 , + p. chiappetta , j. soffer and p. taxil , _ spin asymmetries for scalar leptons from w and z decay in p anti - p collisions _ , phys . b * 162 * ( 1985 ) 192 , + h. baer , c. h. chen , f. paige and x. tata , _ detecting sleptons at hadron colliders and supercolliders _ , phys . d * 49 * ( 1994 ) 3283 m. claudson , l. j. hall and i. hinchliffe , _ low - energy supergravity : false vacua and vacuous predictions _ , nucl . b * 228 * ( 1983 ) 501 , + j. a. casas , a. lleyda and c. munoz , _ strong constraints on the parameter space of the mssm from charge and color breaking minima _ b * 471 * ( 1996 ) 3 a. kusenko , p. langacker and g. segre , _ phase transitions and vacuum tunneling into charge and color breaking minima in the mssm _ , phys . d * 54 * ( 1996 ) 5824 , + u. sarid , _ tools for tunneling _ d * 58 * ( 1998 ) 085017 [ arxiv : hep - ph/9804308 ] , + f. borzumati , g. r. farrar , n. polonsky and s. d. thomas , _ soft yukawa couplings in supersymmetric theories _ , b * 555 * ( 1999 ) 53 s. kawabata , _ a new monte carlo event generator for high - energy physics _ , comput . commun . * 41 * ( 1986 ) 127 and _ a new version of the multidimensional integration and event generation package bases / spring _ , comput . commun . * 88 * ( 1995 ) 309 a. djouadi , _ the anatomy of electro - weak symmetry breaking . ii : the higgs bosons in the minimal supersymmetric model _ , phys . rept . * 459 * ( 2008 ) 1 , + m. s. carena _ et al . _ [ higgs working group collaboration ] , _ report of the tevatron higgs working group _ , arxiv : hep - ph/0010338 j. s. lee , a. pilaftsis , m. s. carena , s. y. choi , m. drees , j. r. ellis and c. e. m. wagner , _ cpsuperh : a computational tool for higgs phenomenology in the minimal supersymmetric standard model with explicit cp violation _ , comput . commun . * 156 * ( 2004 ) 283 d. graudenz , m. spira and p. m. zerwas , _ qcd corrections to higgs boson production at proton proton colliders _ , phys . lett . * 70 * ( 1993 ) 1372 , + m. spira , a. djouadi , d. graudenz and p. m. zerwas , _ higgs boson production at the lhc _ , nucl . b * 453 * ( 1995 ) 17 , + r. v. harlander and k. j. ozeren , _ top mass effects in higgs production at next - to - next - to - leading order qcd : virtual corrections _ , phys . b * 679 * ( 2009 ) 467 , + a. pak , m. rogal and m. steinhauser , _ virtual three - loop corrections to higgs boson production in gluon fusion for finite top quark mass _ , phys . b * 679 * ( 2009 ) 473 , + s. catani , d. de florian , m. grazzini and p. nason , _ soft - gluon resummation for higgs boson production at hadron colliders _ , jhep * 0307 * ( 2003 ) 028 s. dawson , a. djouadi and m. spira , _ qcd corrections to susy higgs production : the role of squark loops _ , * 77 * ( 1996 ) 16 , + m. muhlleitner and m. spira , _ higgs boson production via gluon fusion : squark loops at nlo qcd _ , b * 790 * ( 2008 ) 1 , + r. bonciani , g. degrassi and a. vicini , _ scalar particle contribution to higgs production via gluon fusion at nlo _ , jhep * 0711 * ( 2007 ) 095 , + m. muhlleitner , h. rzehak and m. spira , _ mssm higgs boson production via gluon fusion : the large gluino mass limit _ , jhep * 0904 * ( 2009 ) 023 w. beenakker , m. klasen , m. kramer , t. plehn , m. spira and p. m. zerwas , _ the production of charginos / neutralinos and sleptons at hadron colliders _ , phys . rev . * 83 * ( 1999 ) 3780 [ erratum - ibid . * 100 * ( 2008 ) 029901 ] , + m. grazzini , _ the drell - yan process in nnlo qcd _ , arxiv:0908.1336 [ hep - ph ] r. v. harlander and m. steinhauser , _ supersymmetric higgs production in gluon fusion at next - to - leading order _ , jhep * 0409 * ( 2004 ) 066 , + r. v. harlander and f. hofmann , _ pseudo - scalar higgs production at next - to - leading order susy - qcd _ , jhep * 0603 * ( 2006 ) 050 , + g. degrassi and p. slavich , _ on the radiative corrections to the neutral higgs boson masses in the nmssm _ , nucl . b * 825 * ( 2010 ) 119 , + c. anastasiou , s. beerli and a. daleo , _ the two - loop qcd amplitude gg @xmath215 h , h in the minimal supersymmetric standard model _ * 100 * ( 2008 ) 241806 m. s. carena , d. garcia , u. nierste and c. e. m. wagner , _ effective lagrangian for the @xmath216 interaction in the mssm and charged higgs phenomenology _ , nucl . b * 577 * ( 2000 ) 88 , + j. guasch , p. hafliger and m. spira , _ mssm higgs decays to bottom quark pairs revisited _ , phys . d * 68 * ( 2003 ) 115001 , + f. borzumati , c. greub and y. yamada , _ beyond leading - order corrections to @xmath217 at large tan(beta ) : the charged - higgs contribution _ , phys . d * 69 * ( 2004 ) 055005 , and _ towards an exact evaluation of the supersymmetric o(alpha(s ) tan(beta ) ) corrections to @xmath218 _ , arxiv : hep - ph/0305063 + d. noth and m. spira , _ higgs boson couplings to bottom quarks : two - loop supersymmetry - qcd corrections _ , phys . rev . * 101 * ( 2008 ) 181801 s. h. zhu , _ pseudoscalar higgs boson pair production in photon photon collisions _ , j. phys . g * 24 * ( 1998 ) 1703 , + s. h. zhu , c. s. li and c. s. gao , _ lightest neutral higgs pair production in photon photon collisions in the minimal supersymmetric standard model _ , phys . d * 58 * ( 1998 ) 015006 , + g. j. gounaris and p. i. porfyriadis , _ the gamma gamma @xmath215 a0 a0 process at a gamma gamma collider _ , eur . j. c * 18 * ( 2000 ) 181 y. j. zhou , w. g. ma , h. s. hou , r. y. zhang , p. j. zhou and y. b. sun , _ neutral higgs boson pair production via gamma gamma collision in the minimal supersymmetric standard model at linear colliders _ , phys . d * 68 * ( 2003 ) 093004

in this article we study the asymptotics of large directed graphs with constrained densities of edges and outward directed @xmath0-stars . large graphs are often modeled by probabilistic ensembles with one or more adjustable parameters ; see e.g. fienberg @xcite , lovsz @xcite and newman @xcite . the _ exponential random graph models _ ( ergms ) , in which parameters are used to tune the densities of edges and other subgraphs , are one such class of models ; see e.g. besag @xcite , frank and strauss @xcite , holland and leinhardt @xcite , newman @xcite , rinaldo et al . @xcite , robins et al . @xcite , snijders et al . @xcite , strauss @xcite , and wasserman and faust @xcite . it has been shown that in many ergms the subgraph densities actually _ can not _ be tuned . for example , for the class of ergms parametrized by edges and @xmath0-stars , large graphs are essentially erds - rnyi for all values of the parameters . ( see chatterjee and diaconis @xcite for more complete and precise statements . ) an alternative to ergms was introduced by radin and sadun @xcite , where instead of using parameters to control subgraph counts , the subgraph densities are controlled directly ; see also radin et al . @xcite , radin and sadun @xcite and kenyon et al . this is the approach we take in this article , with edges and ( outward ) @xmath0-stars as the subgraphs controlled directly . we also consider models which split the difference between the two approaches : edges ( resp . @xmath0-stars ) are controlled directly , while @xmath0-stars ( resp . edges ) are tuned with a parameter . see @xcite for the undirected graph version of the latter . we find that , in all our models , graphs have either uniform or bipodal structure as the number of nodes becomes infinite . our approach , following refs . @xcite , is to study maximizers of the entropy or free energy of the model as the number of nodes becomes infinite . when we constrain both edge and @xmath0-star densities , we find only bipodal structure ( except when the @xmath0-star density is fixed to be exactly equal to the @xmath0th power of the edge density ) . when we constrain either edge or @xmath0-star densities ( but not both ) , we find both uniform and bipodal structures , with a sharp change at the interface . this is in contrast with the situation in the ergm version of the model , in which one finds only uniform structures , albeit with sharp changes in their densities along a certain curve in parameter space ; see aristoff and zhu @xcite . these sharp changes along interfaces are called _ phase transitions_. phase transitions have recently been proved rigorously for ergms ; see e.g. yin @xcite and especially radin and yin @xcite for a precise definition of the term . some earlier works using mean - field analysis and other approximations include hggstrm and jonasson @xcite and park and newman @xcite . the terminology is apt , in that ergms and our models are inspired by certain counterparts in statistical physics : respectively , the _ grand canonical ensemble _ , and the _ microcanonical _ and _ canonical ensembles_. see the discussion in section [ models ] . our directed graph models are simpler than their undirected counterparts . in particular , we can rigorously identify the asymptotic structures at all parameters , while in analogous models for undirected graphs , only partial results are known @xcite . our analysis , however , does not easily extend to directed random graph models where other subgraphs , like triangles , cycles , etc . , instead of outward directed @xmath0-stars , are constrained . this article is organized as follows . in section [ models ] , we describe our models and compare them with their statistical physics counterparts . in section [ results ] , we state our main results . in section [ ldp ] , we prove a large deviations principle for edge and @xmath0-star densities . we use the large deviations principle to give proofs , in section [ proofs ] , of our main results . a directed graph on @xmath1 nodes will be represented by a matrix @xmath2 , where @xmath3 if there is a directed edge from node @xmath4 to node @xmath5 , and @xmath6 otherwise . for simplicity , we allow for @xmath7 , though this will not affect our results . let @xmath8 ( resp . @xmath9 ) be the directed edge and outward @xmath0-star homomorphism densities of @xmath10 : @xmath11 here , @xmath0 is an integer @xmath12 . the reason for the term homomorphism density is as follows . for a given graph @xmath10 , if @xmath13 ( resp . @xmath14 ) are the number of homomorphisms edge - preserving maps from a directed edge ( resp . outward @xmath0-star ) into @xmath10 , then @xmath15 with the denominators giving the total number of maps from a directed edge ( resp . outward @xmath0-star ) into @xmath10 . let @xmath16 be the uniform probability measure on the set of directed graphs on @xmath1 nodes . thus , @xmath16 is the uniform probability measure on the set of @xmath17 matrices with entries in @xmath18 . for @xmath19 $ ] and @xmath20 , define @xmath21 throughout , @xmath22 is the natural logarithm , and we use the convention @xmath23 . we are interested in the limit @xmath24 the function in will be called the _ limiting entropy density_. this is the directed graph version of the quantity studied in @xcite . see also @xcite for related work where triangles are constrained instead of @xmath0-stars . for @xmath25 , @xmath19 $ ] and @xmath20 , define @xmath26,\\ & \psi_n^{\delta}(\beta_1,s ) = \frac{1}{n^2}\log { \mathbb e}_n\left[e^{n^{2}\beta_1 e(x)}\,1_{\{s(x ) \in ( s-\delta , s+\delta)\}}\right].\\\end{aligned}\ ] ] we will also be interested in the limits @xmath27 the quantities in will be called _ limiting free energy densities_. they are the directed graph versions of the quantities studied in @xcite . we abuse notation by using the same symbols @xmath28 and @xmath29 to represent different functions in and ( and in below ) , but the meaning will be clear from the arguments of these functions , which will be written @xmath30 , @xmath31 , @xmath32 ( or @xmath33 as below ) . when it is clear which function we refer to , we may simply write @xmath28 without any arguments . we show that the limits in and exist by appealing to the variational principles in theorem [ varpsimicro ] and theorem [ varpsicanon ] below . maximizers of these variational problems are associated with the asymptotic structure of the associated constrained directed graphs . more precisely , as @xmath34 a typical sample @xmath10 from the uniform measure @xmath16 conditioned on the event @xmath35 , @xmath36 looks like the maximizer of the variational formula for in theorem [ varpsimicro ] . similarly , as @xmath34 , a typical sample from the probability measure which gives weight @xmath37 $ ] ( resp . $ ] ) to graphs @xmath10 on @xmath1 nodes with @xmath36 ( resp . @xmath35 ) , when conditioned on the event @xmath35 ( resp . @xmath36 ) , looks like the maximizer of the variational formula for in theorem [ varpsicanon ] . thus , in we have constrained @xmath8 and @xmath9 directly , while in we control one of @xmath8 or @xmath9 with a parameter @xmath39 or @xmath40 . see @xcite for discussions and related work in the undirected graph setting . closely related to and is the limit @xmath41,\ ] ] which was studied extensively in aristoff and zhu @xcite . in this _ exponential random graph model _ ( ergm ) both @xmath8 and @xmath9 are controlled by parameters @xmath39 and @xmath40 . there it was shown that @xmath42 is analytic except along a certain phase transition curve @xmath43 , corresponding to an interface across which the edge density changes sharply . see radin and yin @xcite for similar results , which use a similar variational characterization , in the undirected graph setting . the results in @xcite and @xcite will be useful in our analysis . this is because the euler - lagrange equations associated with the variational formulas for , are the same as the analogous variational formula for . the crucial difference in the optimization problems is that for the ergm , @xmath39 and @xmath40 are fixed parameters , while in our models one or both of @xmath44 is a free variable ( i.e. , a lagrange multiplier ) . thus , some solutions to the ergm variational problem may have no relevance to our models . however , many of the calculations can be carried over , which is why we can make extensive use of the results in @xcite and @xcite . this article focuses on computing the limiting entropy density and free energy densities in and , along with the corresponding maximizers in the variational formulas of theorem [ varpsimicro ] and theorem [ varpsicanon ] . in statistical physics modeling there is a hierarchy analogous to - - , with ( for example ) particle density and energy density in place of @xmath8 and @xmath9 , and temperature and chemical potential in place of @xmath39 and @xmath40 . the statistical physics versions of - - correspond to the _ microcanonical _ , _ canonical _ and _ grand canonical ensembles _ , respectively . in that setting , there are curves like @xmath45 along which the free energy densities are not analytic , and these correspond to physical phase transitions , for example the familiar solid / liquid and liquid gas transitions ; see gallavotti @xcite . ( there is no proof of this statement , though it is widely believed ; see however lebowitz et al . we find singularities in @xmath46 and @xmath47 , but not in @xmath48 . ( in @xcite we have shown @xmath42 has singularities as well . ) see radin @xcite and radin and sadun @xcite for discussions of the relationship between these statistical physics models and some random graph models that closely parallel ours . to state our results we need the following . for @xmath49 $ ] , define @xmath50 with the understanding that @xmath51 . of course @xmath52 depends on @xmath39 and @xmath40 , but we omit this to simplify notation . clearly , @xmath52 and @xmath53 are analytic in @xmath54 and continuous on @xmath55 $ ] . the function @xmath52 is essential to understanding the ergm limiting free energy density @xcite . [ trans_curve ] for each @xmath33 the function @xmath52 has either one or two local maximizers . there is a curve @xmath43 , @xmath56 , with the endpoint @xmath57 such that off the curve and at the endpoint , @xmath52 has a unique global maximizer , while on the curve away from the endpoint , @xmath52 has two global maximizers @xmath58 . we consider @xmath59 and @xmath60 as functions of @xmath39 ( or @xmath40 ) for @xmath61 ( or @xmath62 ) ; @xmath59 ( resp . @xmath60 ) is increasing ( resp . decreasing ) in @xmath39 , with @xmath63 [ trans_curve_2 ] the curve @xmath45 in theorem [ trans_curve ] is continuous , decreasing , convex , and analytic for @xmath61 , with @xmath64 moreover , @xmath59 and @xmath60 are analytic in @xmath39 and @xmath40 . the curve in theorem [ trans_curve ] will be called the _ phase transition curve _ , and its endpoint the _ critical point_. theorem [ trans_curve ] and theorem [ trans_curve_2 ] will be used extensively in most of our proofs ; because of this , we will often not refer to it explicitly . ( we comment that the last statement of theorem [ trans_curve ] , not made explicit in @xcite and @xcite , is proved in proposition [ prop ] in section [ proofs ] below . ) we will sometimes write @xmath65 for the local maximizer of @xmath52 at the critical point . note that the last part of the theorem implies @xmath66 our first main result concerns the limiting free energy density @xmath48 . [ psimicro ] the limiting entropy density @xmath67 is analytic in @xmath68 and equals @xmath69 outside @xmath70 . with @xmath71 and @xmath72 we have the semi - explicit formula @xmath73 where @xmath74 are the unique global maximizers of @xmath52 along the phase transition curve @xmath45 which satisfy @xmath75 moreover , @xmath28 is continuous on @xmath70 , and the first order partial derivatives of @xmath28 are continuous on @xmath76 but diverge on @xmath77 . when @xmath78 , we have the explicit formula @xmath79 in theorem [ psimicro ] , the formula is only semi - explicit because @xmath59 and @xmath60 depend on @xmath80 and @xmath81 . note that the formula depends _ only _ on @xmath80 and @xmath81 . ( to understand the dependence better , see the subsection _ uniqueness of the optimizer : geometric proof _ in the proof of theorem [ psimicro ] . ) note that we have not found any interesting singular behavior of @xmath28 , in contrast with the recent results of kenyon et . @xcite for the undirected version of the model . on the other hand , we are able to prove that our graphs are bipodal , whereas in @xcite it was proved only that the graphs are multipodal ( though simulation evidence from the paper suggests bipodal structure ) . [ fig1 ] -165pt for @xmath78 . the non - shaded portion represents the complement of @xmath70 ( where by convention @xmath28 equals @xmath69 ) . here , @xmath82.,title="fig : " ] -175pt our next main result concerns @xmath46 and @xmath47 . [ psicanon ] ( i ) there is a u - shaped region @xmath83 whose closure has lowest point @xmath84 such that the limiting free energy density @xmath85 is analytic outside @xmath86 . the limiting free energy density has the formula @xmath87 \\ \qquad\qquad -\left[\frac{x_2-e}{x_{2}-x_{1}}i(x_{1 } ) + \frac{e - x_1}{x_{2}-x_{1}}i(x_{2})\right ] , & ( e,\beta_{2})\in u_e \end{cases}\ ] ] where @xmath88 are the global maximizers of @xmath52 at the point @xmath89 on the phase transition curve . in particular , @xmath90 has jump discontinuities across @xmath86 away from @xmath91 , and @xmath92 is discontinuous at @xmath91 . \(ii ) there is a u - shaped region @xmath93 whose closure has rightmost point @xmath94 such that the limiting free energy density @xmath95 is analytic outside @xmath96 . the limiting free energy density has the formula @xmath97 \\ \qquad\qquad -\left[\frac{x_{2}^{p}-s}{x_{2}^{p}-x_{1}^{p}}i(x_{1 } ) + \frac{s - x_{1}^{p}}{x_{2}^{p}-x_{1}^{p}}i(x_{2})\right ] , & ( \beta_{1},s)\in u_s \end{cases}\ ] ] where @xmath88 are the global maximizers of @xmath52 at the point @xmath98 on the phase transition curve . in particular , @xmath99 has jump discontinuities across @xmath86 away from @xmath100 , and @xmath101 is discontinuous at @xmath100 . the sharp change in theorem [ psicanon ] is called a _ phase transition _ ( because of the singularity in the derivatives of @xmath28 ) . the phase transition corresponds to a sharp change from uniform to bipodal structure of the optimizers of the variational problem for ; see the discussion below theorem [ varpsicanon ] . the bipodal structure in the regions @xmath102 and @xmath103 is sometimes called a _ replica symmetry breaking phase_. in contrast with theorem [ psimicro ] , the @xmath59 and @xmath60 in theorem [ psicanon ] are functions of @xmath39 or @xmath40 only and do not depend on @xmath80 or @xmath81 . see figure 1 for the region @xmath104 from theorem [ psimicro ] , and figure 2 for the u - shaped regions @xmath102 and @xmath103 in theorem [ psicanon ] . [ fig2 ] -160pt ( left ) and @xmath105 ( right ) for @xmath106 . contour lines are included to emphasize features . the boundaries of the u - shaped regions , @xmath86 and @xmath96 , are outlined . here , @xmath107 and @xmath108.,title="fig : " ] -165pt we will need the following terminology before proceeding . a sequence @xmath109 of probability measures on a topological space @xmath10 is said to satisfy a _ large deviation principle _ with speed @xmath110 and rate function @xmath111 if @xmath112 is non - negative and lower semicontinuous , and for any measurable set @xmath113 , @xmath114 here , @xmath115 is the interior of @xmath113 and @xmath116 is its closure . see e.g. dembo and zeitouni @xcite or varadhan @xcite . we will equip the set @xmath117 of measurable functions @xmath55 \to [ 0,1]$ ] with the _ cut norm _ , written @xmath118 and defined by @xmath119 where the supremum is taken over measurable subsets @xmath113 of @xmath55 $ ] . [ ldpthm ] the sequence of probability measures @xmath120 satisfies a large deviation principle on the space @xmath55 ^ 2 $ ] with speed @xmath121 and rate function @xmath122 where @xmath123 is the set of measurable functions @xmath124 \to [ 0,1]$ ] such that @xmath125 by convention , @xmath126 if @xmath127 $ ] , and the infimum is @xmath128 if @xmath129 is empty . theorem [ psimicro ] and theorem [ ldpthm ] together imply that for large number of nodes , a typical sample @xmath10 from @xmath16 conditioned on the event @xmath130 and @xmath131 has the following behavior . approximately @xmath132 of the nodes of @xmath10 each have on average @xmath133 outward pointing edges , while the other approximately @xmath134 nodes each have on average @xmath135 outward pointing edges . when @xmath136 we call this structure _ bipodal _ ; otherwise we call it _ uniform_. the following is an immediate consequence of theorem [ ldpthm ] . [ varpsimicro ] for any @xmath19 $ ] , @xmath137.\ ] ] the next theorem is an easy consequence of theorem [ ldpthm ] and varadhan s lemma ( see e.g. @xcite ) . [ varpsicanon ] for any @xmath19 $ ] and @xmath25 , @xmath138\\ & \psi(\beta_1,s ) = \sup_{{\mathcal g}_{\cdot , s } } \left[\beta_1 \int_0 ^ 1 g(x)\,dx - \int_0 ^ 1 i(g(x))\,dx\right],\end{aligned}\ ] ] where @xmath139 ( resp . @xmath140 ) is the set of measurable functions @xmath124\to [ 0,1]$ ] satisfying @xmath141 from theorem [ ldpthm ] , the proofs of theorem [ varpsimicro ] and theorem [ varpsicanon ] are standard , so we omit them . chatterjee and varadhan @xcite established large deviations for undirected random graphs on the space of graphons ( see also lovasz @xcite ) . szemerdi s lemma was needed in order to establish the compactness needed for large deviations . since our model consists of directed graphs , the results in @xcite do not apply directly . theorem [ ldpthm ] avoids these technical difficulties : it is large deviations principle only on the space @xmath55 ^ 2 $ ] of edge and star densities , instead of on the ( quotient ) function space of graphons . our proof relies on the simplicity of our edge / directed @xmath0-star model ; it can not be easily extended to the case where edges and directed triangles ( or other more complicated directed subgraphs ) are constrained . we expect that these models can be handled by adapting the results of @xcite to the directed case . in aristoff and zhu @xcite , it was proved that @xmath142 = \sup_{0\leq x\leq 1}\left(\beta_{1}x+\beta_{2}x^{p}-i(x)\right).\ ] ] observe that the grtner - ellis theorem can not be used to obtain the large deviations principle in theorem [ ldpthm ] above , due to the fact , first observed in radin and yin @xcite , that the right hand side of is not differentiable . instead , we used mogulskii s theorem and the contraction principle in the proof of theorem [ ldpthm ] . on the other hand , once we have established the large deviations principle in theorem [ ldpthm ] , we can use varadhan s lemma to obtain an alternative expression for the limiting free energy , @xmath143 = \sup_{(e , s)\in[0,1]^{2}}\left(\beta_{1}e+\beta_{2}s+\psi(e , s)\right).\ ] ] the limiting free energy in the directed and undirected models differ by only a constant factor of @xmath144 ( see chatterjee and diaconis @xcite and radin and yin @xcite for the undirected model , and aristoff and zhu @xcite for the directed model ) . on the other hand , the limiting entropy density @xmath48 we obtain here differs nontrivially from the one recently obtained in kenyon et al . @xcite for undirected graphs . we start with the proof of the large deviations principle of section [ ldp ] . then we turn to the proofs of our main results in section [ results ] . by definition , @xmath145 is the uniform probability measure on the set of directed graphs on @xmath1 nodes . recall that @xmath146 under @xmath145 , @xmath147 are i.i.d . bernoulli random variables that take the value @xmath148 with probability @xmath149 and @xmath150 with probability @xmath149 . therefore , the logarithm of the moment generating function of @xmath151 is , for @xmath152 , @xmath153=\log\left(\frac{1}{2}e^{\theta}+\frac{1}{2}\right ) = \log(e^{\theta}+1)-\log 2.\ ] ] its legendre transform is @xmath154 ) = x\log x+(1-x)\log(1-x)+\log 2=i(x),\ ] ] where by convention @xmath126 for @xmath127 $ ] . let @xmath155 be the @xmath4th entry of the vector @xmath156 then , @xmath155 are i.i.d . bernoulli random variables . mogulskii theorem ( see e.g. dembo and zeitouni @xcite ) shows that @xmath157 satisfies a sample path large deviation principle on the space @xmath158 $ ] consisting of functions on @xmath55 $ ] equipped with the supremum norm ; the rate function is given by @xmath159 $ } \\ + \infty & \text{otherwise } \end{cases},\ ] ] where @xmath160 $ ] is the set of absolutely continuous functions defined on @xmath55 $ ] such that @xmath161 and @xmath162 . the restriction @xmath162 comes from the fact that @xmath163 . on the other hand , for any @xmath164 , @xmath165 since @xmath166 therefore , @xmath167 satisfies a large deviation principle with the same space and rate function as @xmath168 to complete the proof , we need to use the contraction principle , see e.g. dembo and zeitouni @xcite or varadhan @xcite . given @xmath169 $ ] , we may write @xmath170 for a measurable function @xmath124\to[0,1]$ ] . it is easy to see that if @xmath171 $ ] and @xmath172 in the supremum norm , then @xmath173 in the cut norm . hence , if @xmath172 in the supremum norm , then @xmath174 moreover , for any @xmath175 , @xmath176 -\left[\int_{g_{n}<g}(g(x)^{p}-g_{n}(x)^{p})dx\right],\ ] ] and since @xmath173 in the cut norm and @xmath177 , the mean value theorem shows that @xmath178 and @xmath179 therefore , the maps @xmath180 are continuous from @xmath158\cap\mathcal{ac}_{0}[0,1]$ ] to @xmath55 $ ] and thus the map @xmath181 is continuous from @xmath158\cap\mathcal{ac}_{0}[0,1]$ ] to @xmath55^{2}$ ] . by the contraction principle , we conclude that @xmath120 satisfies a large deviation principle on the space @xmath55 ^ 2 $ ] with speed @xmath121 and rate function @xmath182 the following proofs are for our main results in section [ results ] . by theorem [ varpsimicro ] , @xmath183\ ] ] where we recall @xmath123 is the set of measurable functions @xmath124\to [ 0,1]$ ] satisfying @xmath184 _ domain of @xmath48 . _ by jensen s inequality , @xmath185 on the other hand , since @xmath186 takes values in @xmath55 $ ] and @xmath187 , @xmath188 hence by convention @xmath48 is infinite outside @xmath70 . we will show later in the proof that @xmath48 is finite in @xmath70 . _ optimizing over a quotient space . _ so that we can use standard variational techniques , we show that it suffices to optimize over a compact quotient space of @xmath123 . recall that @xmath189 is the set of measurable functions @xmath55 \to [ 0,1]$ ] endowed with the cut norm . let @xmath190 be the quotient space obtained by the following equivalence relation : @xmath191 if and only if there is a measure - preserving bijection @xmath192 \to [ 0,1]$ ] such that @xmath193 . note that for any such @xmath194 we have @xmath195 , and so @xmath196 defines a metric on @xmath190 . with this metric @xmath190 is compact @xcite . observe that @xmath197 are all unchanged when @xmath186 is replaced by @xmath198 . thus , @xmath199 = \sup_{{\tilde g } \in { \tilde { \mathcal g}}_{e , s } } \left[-\int_0 ^ 1 i(g(x))\,dx\right],\ ] ] where @xmath200 is the projection of @xmath123 in the quotient space , and on the right hand side @xmath186 is any element of the equivalence class of @xmath201 . next we check that the functionals @xmath202 defined on @xmath203 , are all continuous . let @xmath204 be a sequence in @xmath190 converging to @xmath201 , and let @xmath205 be representatives for @xmath206 , respectively . we will show that @xmath207 for any uniformly continuous function @xmath208 defined on @xmath55 $ ] . for each @xmath1 , choose @xmath209 such that @xmath210 for any @xmath20 , define @xmath211 as the cut norm is equivalent to the @xmath212 norm ( this is true only in one dimension ; see @xcite ) , we have for any @xmath20 , @xmath213 where @xmath214 is the lebesgue measure of @xmath215 . now , observe that @xmath216}|f(x)| + \left|\int_{[0,1]\setminus a_{n,\delta } } [ f(g(x ) ) -f(g_n\circ \sigma_n(x))]\,dx \right| . \end{split}\end{aligned}\ ] ] let @xmath217 . using uniform continuity of @xmath208 , let @xmath218 be such that @xmath219 implies @xmath220 . then shows that @xmath221}|f(x)| + \epsilon.\ ] ] now establishes , as desired . these arguments show that there exist ( global ) maximizers @xmath186 of @xmath222.\ ] ] _ bipodal structure of the optimizers . _ if @xmath223 , then for any @xmath224 , @xmath225 suppose then that @xmath226 . since @xmath227 as @xmath228 and @xmath229 as @xmath230 , it is not hard to see that , given @xmath231 , there exists @xmath217 such that any optimizers of must be in the following set : @xmath232 \hbox { for a.e . } x \in [ 0,1]\}.\ ] ] to see this , note that if @xmath233 , then the values of @xmath186 can be adjusted to be @xmath234 distance away from @xmath150 and @xmath148 so that still @xmath235 and @xmath186 attains a larger value for the integral in . then , optimizing over @xmath236 , standard results in variational calculus ( see @xcite , theorem 9.1 , pg . 178 ) show that optimizers @xmath186 satisfy , for all @xmath237 , @xmath238 where @xmath239 , @xmath25 are lagrange multipliers , and at least one of @xmath240 , @xmath39 , @xmath40 is nonzero . the integrals in are the frchet derivatives of @xmath241 evaluated at @xmath237 ; a simplified version of the arguments in - shows the frchet derivatives are continuous in @xmath242 . ( to apply the results mentioned in @xcite , we must switch to the uniform topology induced by the sup norm @xmath243 on the banach space of bounded measurable functions @xmath55 \to { \mathbb r}$ ] . the statement about continuity of frechet derivatives then follows from the simple fact that @xmath244 implies @xmath245 . ) when @xmath246 ( called the _ normal _ case ) , we see that for some @xmath39 , @xmath40 , @xmath247.\ ] ] when @xmath248 ( called the _ abnormal _ case ) , we find that @xmath249 for a.e . @xmath49 $ ] , so that @xmath186 is constant a.e . from and jensen s inequality , this occurs precisely when @xmath250 . let @xmath251 be a maximizer of . we have shown that : @xmath252;\\ & ( b ) \hbox { if } ( e , s ) \in d , \hbox { then } \ell'(g(x ) ) = 0 \hbox { for a.e . } x \in [ 0,1 ] ; \hbox { or}\\ & ( c ) \hbox { if } s \in { \partial d}\setminus d^{up } , \hbox { then } g \hbox { is constant a.e.}\end{aligned}\ ] ] theorem [ trans_curve ] shows that , for each @xmath33 , either @xmath253 at a unique @xmath254 or @xmath253 at exactly two points @xmath255 . thus , to maximize it suffices to maximize @xmath256 over the set of functions @xmath235 of the form @xmath257 where @xmath258 $ ] and @xmath259 . observe that for such @xmath186 , @xmath260 it is therefore enough to maximize @xmath261 subject to the constraints @xmath262 in case _ ( a ) _ , we must have @xmath263 , @xmath264 and @xmath265 . in case _ ( c ) _ , we must take @xmath266 . this establishes formula for @xmath267 . it remains to consider _ ( b)_. in this case @xmath268 we have seen that @xmath269 . moreover if @xmath270 or @xmath271 then can not be satisfied , so @xmath272 and @xmath273 . introducing lagrange multipliers @xmath39 , @xmath40 , we see that @xmath274 + \beta_1(y_1-y_2 ) + \beta_2(y_1^p - y_2^p ) = 0.\end{aligned}\ ] ] thus , @xmath275 this means that @xmath276 are global maximizers of @xmath52 on the phase transition curve , away from the critical point . _ uniqueness of the optimizer ( geometric proof ) . _ we must prove uniqueness for @xmath277 . consider the class of lines @xmath278 where @xmath279 are global maximizers of @xmath52 on the phase transition curve , away from the critical point . since @xmath59 ( resp . @xmath60 ) is strictly increasing ( resp . strictly decreasing ) in @xmath39 with @xmath280 , no two distinct lines of this class can intersect . continuity of @xmath59 , @xmath60 in @xmath39 along with show that the union of all the lines equals @xmath281 . thus , given @xmath277 , there is a unique optimizer @xmath186 of the form , obtained by locating the unique line from which contains the point @xmath30 and choosing @xmath282 satisfying the constraint . solving for @xmath282 in the first constraint , we get @xmath283 . putting this into the second constraint gives @xmath284 this establishes formula for @xmath268 . _ uniqueness of the optimizer ( algebraic proof ) . _ fix @xmath268 . from , @xmath285 and so @xmath286 we must show that has a unique solution . define @xmath287 note that @xmath288 and @xmath289 also define @xmath290 then @xmath291 and @xmath292 . moreover , @xmath293 is positive if @xmath294 and it is negative if @xmath295 . finally , @xmath296 therefore , @xmath297 now by the intermediate value theorem , there exists @xmath298 such that @xmath299 . next we prove @xmath300 is unique . observe that @xmath301\frac{\partial x_{2}}{\partial\beta_{1 } } + q''(\beta_{1})(s - x_{2}^{p}).\ ] ] since @xmath302 , we have @xmath303 . we also know that @xmath304 and @xmath305 . moreover , since @xmath306 and @xmath307 , we have @xmath308 hence , we conclude that @xmath309 for @xmath310 and therefore @xmath300 is unique . _ regularity of @xmath48 . _ we turn now to the claimed regularity of @xmath48 , starting with analyticity . fix @xmath311 . then each @xmath312 satisfies , for some @xmath313 , @xmath314 we claim that defines @xmath39 implicitly as an analytic function of @xmath81 for @xmath312 . by differentiating the left hand side of with respect to @xmath39 , we find the expression @xmath315 by the mean value theorem , there is @xmath316 such that this expression becomes @xmath317 + \left[\frac{e - x_2}{x_2-x_1}\frac{\partial x_1}{\partial \beta_1 } ( py^{p-1}-px_1^{p-1})\right].\ ] ] since @xmath318 and @xmath319 , each of the terms in brackets is negative , so this expression is nonzero . by the analytic implicit function theorem @xcite , we conclude that @xmath39 is an analytic function of @xmath81 inside @xmath104 . similar arguments show that @xmath39 is an analytic function of @xmath80 inside @xmath104 . ( by theorem 1 , this means @xmath40 must also be an analytic function of @xmath80 and @xmath81 inside @xmath104 . ) since @xmath59 and @xmath60 are analytic functions of @xmath39 , we see that @xmath59 and @xmath60 are analytic functions of @xmath80 and @xmath81 inside @xmath104 . inspecting , we conclude that @xmath67 is analytic in @xmath281 . next we show that @xmath28 is continuous on @xmath70 . we have already shown that @xmath28 is analytic , hence continuous , in @xmath281 . it is easy to check that @xmath28 is continuous along each line @xmath320 and that the restriction of @xmath28 to the lower boundary @xmath321 is continuous . this is enough to conclude that @xmath28 is continuous on @xmath76 . finally for @xmath322 , we have @xmath323 next we prove continuity of the first order derivatives on @xmath76 . it suffices to show that : ( i ) the first order partial derivatives of @xmath28 are continuous in @xmath104 ; and ( ii ) the limits of the first order partial derivatives of @xmath28 at the lower boundary @xmath324 exist . with @xmath325 and using , we see that for @xmath268 , @xmath326 using the fact that @xmath327 , @xmath328 it is straightforward to compute that @xmath329 thus , @xmath330.\end{aligned}\ ] ] differentiating @xmath331 with respect to @xmath81 and using , we find that @xmath332 and so @xmath333 we have already seen that @xmath40 is an analytic , hence continuous , function of @xmath30 inside @xmath104 . since @xmath334 we see that for @xmath335 , @xmath336 where @xmath337 if @xmath338 and otherwise @xmath339 . also , for any @xmath340 $ ] , @xmath341 proofs of regularity for @xmath342 are similar so we omit them . _ explicit formula when @xmath106 . _ when @xmath78 , we can explicitly solve @xmath343 to obtain @xmath344 and @xmath345 this yields @xmath346 where the last line uses the fact that @xmath53 is symmetric around @xmath144 . _ proof of part ( i ) . _ by theorem [ varpsicanon ] , @xmath347,\ ] ] where we recall @xmath139 is the set of measurable functions @xmath124\to [ 0,1]$ ] satisfying @xmath348 arguments similar to those in the proof of theorem [ psimicro ] show that optimizers of exist and have the form @xmath349 where @xmath258 $ ] and @xmath259 . for such functions @xmath186 , we have @xmath350- \left[\lambda i(y_1 ) + ( 1-\lambda ) i(y_2)\right],\\ \int_0 ^ 1 g(x)\,dx & = \lambda y_1 + ( 1-\lambda ) y_2.\end{aligned}\ ] ] it is therefore enough to maximize @xmath351 -\lambda i(y_1 ) - ( 1-\lambda)i(y_2)\ ] ] subject to the constraint @xmath352 if @xmath271 or @xmath353 , then shows that @xmath354 . so assume that @xmath355 . it is easy to see in this case that @xmath356 . moreover if @xmath270 then again @xmath354 , so assume @xmath272 . introducing the lagrange multiplier @xmath39 , we find that @xmath274 + \beta_1(y_1-y_2 ) + \beta_2(y_1^p - y_2^p ) = 0,\end{aligned}\ ] ] and so @xmath357 if @xmath358 , then there are no solutions to , while if @xmath62 , solutions occur precisely when @xmath276 are the global maximizers of @xmath52 at the point @xmath359 along @xmath45 . in the latter case , implies @xmath360 if @xmath361 or @xmath362 , then @xmath354 . since for @xmath363 we have ( by definition ) @xmath364 , we have established that is valid in @xmath365 . next , for @xmath366 define @xmath325 and @xmath367 -\left[\lambda i(x_{1 } ) + ( 1-\lambda ) i(x_{2})\right]\right).\ ] ] to establish that is valid in @xmath102 , it suffices to show that @xmath368 for all @xmath369 . it is easy to check that @xmath370 and @xmath371 thus , @xmath372 . finally , @xmath373 from the proof of proposition 11 in @xcite , we know @xmath374 , @xmath375 , and moreover there exists @xmath376 such that @xmath377 for @xmath378 while @xmath379 for @xmath380 . the result follows . we note that the star density can be computed easily as @xmath381 it is clear that the star density @xmath382 is continuous everywhere . _ regularity of @xmath46_. from the formula it is easy to see that @xmath46 is analytic away from @xmath86 . write @xmath85 and fix @xmath62 . in the interior of @xmath383 , @xmath384 while the interior of @xmath102 , @xmath385 observe that @xmath386 so @xmath342 is continuous across @xmath86 when @xmath387 . note that @xmath388 proposition 11 of @xcite gives @xmath389 thus , @xmath390 has a jump discontinuity across @xmath86 when @xmath387 . analogous statements hold true for @xmath391 . now consider the situation at the point @xmath392 . proposition 11 of @xcite gives @xmath393 from we see that @xmath394 is discontinuous at @xmath392 . _ proof of part ( ii ) . _ by theorem [ varpsicanon ] , @xmath395\ ] ] where we recall @xmath140 is the set of measurable functions @xmath124\to [ 0,1]$ ] satisfying @xmath396 arguments analogous to those in the proof of part _ ( i ) _ show that @xmath47 has the formula . we note that the limiting edge density has the formula @xmath397 it is easy to see that @xmath398 is continuous everywhere . _ regularity of @xmath47_. from the formula it is easy to see that @xmath47 is analytic away from @xmath96 . write @xmath95 and fix @xmath61 . in the interior of @xmath399 , @xmath400 and in the interior of @xmath103 , @xmath401 note that @xmath402 so @xmath403 is continuous across @xmath96 when @xmath404 . note that @xmath405 in the computations below , let @xmath406 and @xmath407 . note that @xmath408\frac{\partial t}{\partial s } = [ \ell'(t)-\beta_{2}pt^{p-1}]\frac{\partial t}{\partial s},\ ] ] and @xmath409\frac{\partial^{2}t}{\partial s^{2 } } + \left[\ell''(t)-\beta_{2}p(p-1)t^{p-2}\right ] \left(\frac{\partial t}{\partial s}\right)^{2 } \\ & = \ell'(t)\frac{\partial^{2}t}{\partial s^{2 } } + \ell''(t)\left(\frac{\partial t}{\partial s}\right)^{2}.\end{split}\end{aligned}\ ] ] using and proposition 11 of @xcite , we get @xmath410 comparing with , we see that @xmath99 has a jump discontinuity across @xmath96 for @xmath404 . analogous results hold for @xmath411 . now consider the situation at the point @xmath412 . from , @xmath413 and @xmath414 as @xmath415 , @xmath416 and since @xmath417 , @xmath418 , we have @xmath419 while proposition 11 of @xcite gives @xmath420 the next result concerns the curve @xmath43 and the shapes of @xmath102 and @xmath103 . [ prop ] ( i ) the curve @xmath43 is analytic in @xmath61 . \(ii ) for any @xmath310 , @xmath421 and @xmath422 . moreover , @xmath423 for any @xmath424 , @xmath425 and @xmath426 . moreover , @xmath427 first we show that @xmath45 is analytic . there is an open v - shaped set containing the phase transition curve @xmath43 except the critical point @xmath428 , inside which @xmath52 has exactly two local maximizers , @xmath272 . ( see @xcite and @xcite . ) it can be seen from the proof of proposition 11 of @xcite that @xmath429 and @xmath430 inside the v - shaped region . the analytic implicit function theorem @xcite then shows that @xmath431 and @xmath432 are analytic functions of @xmath39 and @xmath40 inside this region . note that @xmath45 is defined implicitly by the equation @xmath433 differentiating the left hand side of this equation w.r.t . @xmath40 gives @xmath434\\ & = y_1^p - y_2^p < 0.\end{aligned}\ ] ] another application of the analytic implicit function theorem implies @xmath43 is analytic for @xmath435 . now we turn to the statements involving @xmath59 and @xmath60 . along @xmath43 , we have @xmath436 differentiating with respect to @xmath437 , we get @xmath438\frac{\partial x_{1}}{\partial\beta_{1}}=0.\ ] ] therefore , @xmath439 since @xmath440 = \lim_{\beta_{1}\rightarrow\beta_{1}^{c}}[-\ell''(x_{1})]=0 $ ] , by lhpital s rule , @xmath441 which implies that @xmath442 since @xmath443 as @xmath444 , it is easy to see that @xmath445 the results for @xmath446 can be proved using the similar methods . finally , notice that @xmath447 therefore the results involving @xmath448 also hold . david aristoff was supported in part by afosr award fa9550 - 12 - 1 - 0187 . lingjiong zhu is grateful to his colleague tzu - wei yang for helpful discussions . the authors are grateful for useful comments from the anonymous referees and associate editor .

trapped ultra - cold atoms have become model systems of choice for simulating physical effects from condensed matter @xcite to cosmology @xcite . the recently achieved ability to create synthetic vector potentials @xcite acting on neutral atoms has increased the versatility of the atomic - physics simulation toolkit even further . it is now possible to simulate magnetic fields by inducing spatially varying @xmath0 ( i.e. , scalar ) gauge potentials @xcite , and pseudo - spin splittings can be created in spinor gases @xcite using spatially constant vector potentials having a ( possibly non - abelian ) matrix structure . these advances have stimulated a host of theoretical works studying , e.g. , the effect of uniform @xmath1 gauge potentials on the behavior of quantum particles subject to uniform ordinary magnetic fields @xcite , or proposing the use of staggered effective spin - dependent magnetic fields in optical lattices @xcite to simulate a new class of materials called topological insulators @xcite that exhibit the quantum spin hall effect @xcite . very recently , the non - quantized intrinsic spin hall effect @xcite has been realized experimentally in a quantum gas @xcite , and the authors of this paper outline the way forward to reaching conditions where the quantum spin hall effect could be observed . furthermore , newly demonstrated methods to simulate strong - enough magnetic fields to probe ultra - cold atom gases in the ordinary quantum - hall regime @xcite are expected to be adaptable for the purpose of generating spin - dependent quantizing magnetic fields @xcite , which opens up another avenue towards the exploration of quantum - spin - hall physics . part of the motivation for our present theoretical work arises from these rapid developments of experimental capabilities . the ordinary quantum hall ( qh ) effect @xcite occurs because particles confined to move in two spatial dimensions and subject to a strong perpendicular magnetic field develop incompressibilities at integer , and certain fractional , values of the landau - level filling factor @xcite . while the landau quantization of single - particle energies is the origin of the integer qh effect , incompressibility at fractional filling factors is caused by the discrete spectrum of interaction energies for two particles occupying states from the same landau level @xcite . in the conceptually simplest realization of the quantum _ spin _ hall ( qsh ) effect @xcite , particles exhibit an integer qh effect due to a spin - dependent perpendicular magnetic field that points in opposite directions for the two opposite - spin components . it seems then quite straightforward to conjecture @xcite that a fractional version of the qsh effect should exist that mirrors features of the ordinary fractional qh effect in multi - component systems @xcite . following this line of thought , some previous discussions of a putative fractional qsh physics @xcite have been based on an _ ad hoc _ adaptation of trial wave functions first proposed in ref . however , unless only particles with the same spin interact , such an approach is fraught with difficulty @xcite . here we revisit the question of how a fractional qsh effect can arise in an interacting ( pseudo-)spin-1/2 system that experiences a spin - dependent quantizing magnetic field . in particular , we elucidate the effect of interactions between particles having opposite spin . we find that such inter - species interactions significantly alter the expected qsh physics , but they also open up new opportunities for tailoring the properties of quantum many - particle states . our study is complementary to recent investigations of fractional qsh phases @xcite that arise in materials with exotic topological band structures @xcite or strained graphene @xcite . the results obtained here are relevant for electronic systems as well as for ultra - cold bosonic or fermionic atoms . particular attention is paid to trapped bosons . the article is organized as follows . in sec . [ sec : hamiltonian ] , we introduce the basic model description of an interacting system of ( pseudo-)spin-1/2 particles that are subject to a spin - dependent magnetic field . the single - particle states are given in the representation of spin - dependent guiding - center and landau - level quantum numbers . the eigenvalue problem of two interacting particles is solved for both cases of equal and opposite - spin particles in the subsequent sec . [ sec : twopart ] . when the two interacting particles have opposite spin , important differences arise with respect to the classic results obtained @xcite for spinless ( or same - spin ) particles . we explore the ramifications of this fact by numerical exact - diagonalization studies with up to 6 bosons for which results are presented in sec . [ sec : numfew ] . our conclusions are summarized in sec . [ sec : conc ] . we consider a gas of particles ( e.g. , atoms ) that carry a ( pseudo-)spin-1/2 degree of freedom and are confined to move in the @xmath2 plane . a spin - dependent vector potential @xmath3 is presumed to be generated ( e.g. , by optical means in an atom gas @xcite ) . here @xmath4 denotes the diagonal pauli matrix , and the @xmath5 are cartesian unit vectors in real space . the vector potential ( [ eq : acgauge ] ) is abelian and gives rise to a spin - dependent magnetic field perpendicular to the @xmath2 plane : @xmath6 . this situation of opposite - spin particles being subjected to oppositely directed magnetic fields corresponds directly to setups considered for a semiconductor heterostructure @xcite and in neutral - atom systems @xcite . a somewhat related study in the context of cold bosonic gases was given in ref . @xcite , only that there the two spin components also experience a large zeeman - like energy shift and , therefore , this work focused only on the dynamics of a single component . without loss of generality , we will assume @xmath7 from now on . also note that , with unit conventions chosen in this article , the ` magnetic - field ' magnitude @xmath8 is related to a fundamental ( ` magnetic ' ) length scale @xmath9 . the most general form of the many - body hamiltonian that describes our system of interest is @xmath10 , where @xmath11 ^ 2 + { \mathcal v}({{\ensuremath{\bm{\mathrm{r } } } } } ) -\mu \ , \hat 1 \right\ } \hat\psi({{\ensuremath{\bm{\mathrm{r } } } } } ) \quad , \\ \label{eq : model_int } { \mathcal h}_{\mathrm{int } } & = & \int d^2 r \,\ , \sum_{\sigma , \sigma^\prime = \pm}\ , \frac{g_{\sigma\sigma^\prime}}{2 } \ , \hat\psi_{\sigma}^\dagger({{\ensuremath{\bm{\mathrm{r } } } } } ) \hat\psi_{\sigma^\prime}^\dagger({{\ensuremath{\bm{\mathrm{r } } } } } ) \hat\psi_{\sigma^\prime}({{\ensuremath{\bm{\mathrm{r}}}}})\hat\psi_{\sigma}({{\ensuremath{\bm{\mathrm{r } } } } } ) \quad .\end{aligned}\ ] ] here @xmath12 denotes the @xmath13 unit matrix , and @xmath14 is a two - spinor of ( bosonic or fermionic ) annihilation operators for a particle located at position @xmath15 and having a definite pseudo - spin-1/2 projection along the @xmath16 axis . furthermore , the relation @xmath17 is implicit in the formalism . as it is often useful , we also give an equivalent expression for the interaction part of the hamiltonian , @xmath18 with @xmath19 , @xmath20 , and @xmath21 . depending on whether the asymmetric interaction couplings @xmath22 , @xmath23 are positive or negative , different physical regimes may be reached @xcite.)([eq : ints_2 ] ) can be related to that which is often adopted in the atom - gas literature @xcite by setting @xmath24 , @xmath25 , and @xmath26 . see also ref . note , however , the different parameterization used in ref . @xcite where @xmath27 are interaction constants associated with the atomic spin-1 degree of freedom from which the pseudo - spin-1/2 components are derived . our notation is related to theirs via @xmath28 , @xmath29 , and @xmath30 . ] in the case where @xmath31 , the system reduces to two independent two - dimensional ( electron or atom ) gases that are each subject to a perpendicular magnetic field . known phenomena associated with the fractional qh effect @xcite will then be exhibited by the individual systems . however , as seen from our study presented in secs . [ sec : twopart ] and [ sec : numfew ] below , the behavior of the system with @xmath32 departs from the previously considered @xcite two - component fractional - qh physics because of the very different type of constraints that is placed on the orbital motion of particles subject to oppositely directed magnetic fields . we now consider single - particle states associated with spin component @xmath33 . the kinetic momentum is @xmath34 , and straightforward calculation establishes the commutation relations @xmath35=\sigma\ , i \,\frac{\hbar^2}{l_{\mathcal b}^2 } \ , \delta_{\sigma\sigma'}\ , \epsilon_{\alpha\alpha ' } \quad , \ ] ] where @xmath36 . operators for the guiding - center locations can then be defined in the usual manner @xcite , @xmath37/\hbar$ ] , and their components satisfy the commutation relations @xmath38=-\sigma\ , i \ , l_{\mathcal b}^2 \ , \delta_{\sigma\sigma'}\,\epsilon_{\alpha\alpha ' } \quad .\ ] ] moreover , we find @xmath39=0 $ ] . following the familiar approach @xcite , we define the harmonic - oscillator landau - level ladder operator for states with spin @xmath33 via @xmath40 similarly , the ladder operator operating within a landau level for spin component @xmath33 is @xmath41 we can express the kinetic energy and the @xmath42 component of angular momentum in terms of the ladder operators [ @xmath43 : @xmath44 ^ 2}{2 m } & = & \hbar\omega_{\mathrm{c } } \left ( a_\sigma^\dagger a_\sigma + \frac{1}{2 } \right ) \quad , \\ \label{eq : angmom } l_z & \equiv & x\ , p_y - y\ , p_x = \hbar\ , \sigma \left ( b_\sigma^\dagger b_\sigma - a_\sigma^\dagger a_\sigma \right ) \quad .\end{aligned}\ ] ] landau - level eigenstates are generated via @xmath45 { \left | \mathrm{vac } \right \rangle } \,\ , , \ ] ] where @xmath46 and @xmath47 is the state that is annihilated by all ladder operators @xmath48 and @xmath49 . in the following , we will focus on the case where all particles are in the lowest landau level , i.e. , when @xmath50 . before presenting a formal analysis of the interacting two - particle system subject to a strong spin - dependent magnetic field in the following subsection , we provide a heuristic argument for how the cases where the two particles feel the same and opposite magnetic fields differ . consider two particles , located at @xmath51 and @xmath52 , respectively , that interact via a generic potential @xmath53 . the corresponding first - quantized two - particle hamiltonian reads @xmath54 ^ 2}{2 m } + \frac{\left [ { { \ensuremath{\bm{\mathrm{p}}}}}_2 - { { \ensuremath{\bm{\mathrm{\cal{a}}}}}}^{(\sigma_2)}({{\ensuremath{\bm{\mathrm{r}}}}}_{2})\right]^2}{2 m } + v({{\ensuremath{\bm{\mathrm{r}}}}}_1 - { { \ensuremath{\bm{\mathrm{r}}}}}_2 ) \quad , \ ] ] with the spin - dependent vector potentials from eq . ( [ eq : acgauge ] ) . it is straightforward to show that the sum of kinetic - energy contributions for each particle can be re - arranged in terms of the linear combinations @xmath55 yielding the expression @xmath56 ^ 2}{2 m } + \frac{\left[{{\ensuremath{\bm{\mathrm{p}}}}}_{\sigma_1 \sigma_2 } - 2 { { \ensuremath{\bm{\mathrm{\cal{a}}}}}}^{(+ ) } ( { { \ensuremath{\bm{\mathrm{r}}}}}_{\sigma_1\sigma_2})\right]^2}{2 m}+v({{\ensuremath{\bm{\mathrm{r}}}}}_1 - { { \ensuremath{\bm{\mathrm{r}}}}}_2 ) . \nonumber \\\end{aligned}\ ] ] the two - particle problem for particles with the same spin reduces to two independent single - particle problems in the center - of - mass ( com ) and relative - coordinate degrees of freedom ( @xmath57 and @xmath58 , respectively ) because @xmath59 . in that case , only the relative - coordinate degree of freedom feels the interaction potential @xmath60 , and it can be minimized by placing two particles away from each other . in the specific case of @xmath61 , the zero - energy states of the two - particle system are of the form @xmath62 , where @xmath63 is a commonly used complex notation for the position of particle @xmath64 @xcite . the non - negative integers @xmath65 and @xmath66 correspond to the quantized values of com angular momentum and relative angular momentum , respectively @xcite . quite a different situation arises for opposite - spin particles . the kinetic energy of the two - particle system decouples in the coordinates @xmath67 and @xmath68 , motivating the proposal of trial wave functions @xcite @xmath69 . however , @xmath70 still couples the two - particle coordinates @xmath67 and @xmath68 and , as a result , the proposed wave function is energetically not favorable for interacting particles @xcite . rigorous examination of the interacting two - particle system in the opposite - spin configuration ( see below ) shows that energy eigenstates are not eigenstates of com angular momentum or relative angular momentum and , furthermore , have an unusual distribution . to gain a deeper understanding of the effect of two - particle interactions , we follow the basic approach employed by previous studies of the fractional qh effect @xcite and find the interaction potential in the representation of lowest - landau - level states . the starting point of such an analysis is the fourier decomposition of a spin - dependent interaction potential given by @xmath71 because its matrix elements can then be directly related to the corresponding matrix elements of the exponential in the integrand of ( [ eq : fourierpot ] ) . to find the latter for states in the lowest landau level , we use the relation @xmath72 and the straightforwardly obtained expressions @xmath73 \ , , \\ & & \frac{l_{\mathcal b}^2}{\hbar}\left({{\ensuremath{\bm{\mathrm{q } } } } } \times \hat{{{\ensuremath{\bm{\mathrm{z}}}}}}\right)\cdot \left ( \sigma_1\ , { { \ensuremath{\bm{\mathrm{\pi}}}}}_{1}^{(\sigma_1 ) } - \sigma_2\ , { { \ensuremath{\bm{\mathrm{\pi}}}}}_{2}^{(\sigma_2)}\right ) = \frac{l_{\mathcal b}}{\sqrt{2 } } \left [ q_x \left ( a_{\sigma_1 } - a_{\sigma_2 } + a_{\sigma_1}^\dagger - a_{\sigma_2}^\dagger\right ) \right . \nonumber \\ & & \hspace{4.5 cm } \left . + \sigma_1 i \ , q_y \left ( a_{\sigma_1 } - \sigma_1\sigma_2\ , a_{\sigma_2 } - a_{\sigma_1}^\dagger + \sigma_1 \sigma_2 \ , a_{\sigma_2}^\dagger\right ) \right ] . \end{aligned}\ ] ] introducing spin - resolved ladder operators for com and relative angular momentum , @xmath74 and analogous ladder operators for com energy and relative - motion energy , @xmath75 we find using @xmath76 @xmath77 \times\left\ { \begin{array}{cl } { \ensuremath{\mathrm{e}}}^{i \bar q_\sigma l_{\mathcal b}\ , b^{\mathrm{(r)}\dagger}_{\sigma\sigma}}\,\ , { \ensuremath{\mathrm{e}}}^{i q_\sigma l_{\mathcal b}\ , b^{\mathrm{(r)}}_{\sigma\sigma } } & \mbox{for}\ , \sigma_1 = \sigma_2 \equiv \sigma \\ { \ensuremath{\mathrm{e}}}^{i q_x l_{\mathcal b}\ , b^{\mathrm{(r)}\dagger}_{\sigma , -\sigma}}\,\ , { \ensuremath{\mathrm{e}}}^{i q_x l_{\mathcal b } \ , b^{\mathrm{(r)}}_{\sigma,-\sigma}}\,\ , { \ensuremath{\mathrm{e}}}^{\sigma q_y l_{\mathcal b}\ , b^{\mathrm{(c)}\dagger}_{\sigma,-\sigma}}\,\ , { \ensuremath{\mathrm{e}}}^{-\sigma q_y l_{\mathcal b}\ , b^{\mathrm{(c)}}_{\sigma,-\sigma } } & \mbox{for } \ , \sigma_1= -\sigma_2 \equiv \sigma \end{array } \right . , \\[0.1 cm ] { \ensuremath{\mathrm{e}}}^{i\frac{l_{\mathcal b}^2}{\hbar}\left({{\ensuremath{\bm{\mathrm{q } } } } } \times \hat{{{\ensuremath{\bm{\mathrm{z}}}}}}\right)\cdot \left ( \sigma_1\ , { { \ensuremath{\bm{\mathrm{\pi}}}}}_{1}^{(\sigma_1 ) } - \sigma_2\ , { { \ensuremath{\bm{\mathrm{\pi}}}}}_{2}^{(\sigma_2)}\right ) } = { \ensuremath{\mathrm{e}}}^{-\frac{q^2\ , l_{\mathcal b}^2}{2 } } \nonumber\\[0.2 cm ] \times \left\ { \begin{array}{cl } { \ensuremath{\mathrm{e}}}^{i \bar q_\sigma l_{\mathcal b}\ , a^{\mathrm{(r)}\dagger}_{\sigma\sigma}}\,\ , { \ensuremath{\mathrm{e}}}^{i q_\sigma l_{\mathcal b}\ , a^{\mathrm{(r)}}_{\sigma\sigma } } & \mbox{for}\ , \sigma_1=\sigma_2 \equiv \sigma \\ { \ensuremath{\mathrm{e}}}^{i q_x l_{\mathcal b}\ , a^{\mathrm{(r)}\dagger}_{\sigma,-\sigma}}\,\ , { \ensuremath{\mathrm{e}}}^{i q_x l_{\mathcal b } \ , a^{\mathrm{(r)}}_{\sigma,-\sigma}}\,\ , { \ensuremath{\mathrm{e}}}^{\sigma q_y l_{\mathcal b}\ , a^{\mathrm{(c)}\dagger}_{\sigma,-\sigma}}\,\ , { \ensuremath{\mathrm{e}}}^{-\sigma q_y l_{\mathcal b}\ , a^{\mathrm{(c)}}_{\sigma,-\sigma } } & \mbox{for } \ , \sigma_1= -\sigma_2 \equiv \sigma \end{array } \right . . \label{eq : spinexp2}\end{aligned}\ ] ] inspection of eqs . ( [ eq : spinexp1])([eq : spinexp2 ] ) reveals a very important formal difference between cases when the interacting particles have equal or opposite spin . for a pair of particles with the same spin , the interaction only couples to the relative - motion landau - level degrees of freedom . in contrast , for particles with opposite spin , the interaction involves both the relative - motion _ and the com _ degrees of freedom . this latter situation is unlike any other encountered previously in the context of fractional qh physics . with the expressions ( [ eq : spinexp1])([eq : spinexp2 ] ) , we are now able to express the interaction potential for a pair of particles having spin @xmath78 and @xmath79 , respectively , in the basis of com - angular - momentum and relative - angular - momentum eigenstates from the lowest landau level given by @xmath80^{m_\mathrm{c}}}{\sqrt{m_\mathrm{c } ! } } \ , \frac{\left [ b^{\mathrm{(r)}\dagger}_{\sigma_1\sigma_2 } \right]^{m_\mathrm{r}}}{\sqrt{m_\mathrm{r } ! } } \ , { \left | 0 \right \rangle}_1\otimes { \left | 0 \right \rangle}_2 \quad .\ ] ] for clarity , the cases where the interacting particles have equal or opposite spin will be discussed separately . the fact that only the relative - angular - momentum operators enter the expression ( [ eq : spinexp1 ] ) for @xmath81 implies that the interaction matrix is diagonal in com space . straightforward calculation yields @xmath82 where @xmath83 in terms of the generalized laguerre polynomial @xmath84 . using the result ( [ eq : spinlike ] ) and the relation ( [ eq : fourierpot ] ) for a contact interaction where @xmath85 yields the well - known expression @xcite @xmath86 for the interaction matrix elements . the remarkable result ( [ eq : usualpp ] ) underpins the basic description of fractional - qh physics @xcite . it implies that the two - particle eigenstates are also eigenstates of com and relative angular momentum . furthermore , the energy spectrum of two particles with spin @xmath33 from the lowest landau level is two - valued : states with @xmath87 and arbitrary @xmath65 have energy @xmath88 , and all other states have energy @xmath89 . using ( [ eq : spinexp1 ] ) for the case @xmath90 , we find @xmath91 the contact - interaction matrix element for opposite - spin particles is then calculated as @xmath92 with @xmath93}{2\sqrt{m!\ , m ' ! } } \,\ , \gamma\left ( \frac{m+m'+1}{2}\right)\ ] ] in terms of the euler gamma function @xmath94 . thus we find that the interaction matrix for two particles from the lowest landau level with opposite spin is nondiagonal in the com - angular - momentum and relative - angular - momentum spaces . this is markedly different from the case of same - spin particles . panel a : eigenvalues @xmath95 of the opposite - spin two - particle interaction matrix [ cf.eq . ( [ eq : oppspin ] ) ] in units of @xmath96 , sorted by magnitude . data are shown for various values of the angular - momentum cutoff @xmath97 ( blue ) , 20 ( red ) , 30 ( green ) , and @xmath98 . the straight line is a plot of @xmath99 with @xmath100 . panel b : energy spectrum obtained for a system of two particles with opposite spin by exact diagonalization . a finite system size is imposed by limiting the number of modes available in angular - momentum space for each particle to @xmath101 . note the @xmath101-dependence of the obtained values . the data for @xmath102 are also shown as the magenta data points in panel a and exhibit excellent agreement with the power - law - type distribution predicted from the solution in com and relative angular - momentum space . green stars show the energy calculated for two - particle versions of trial states @xcite @xmath69 with @xmath103 and @xmath104 , @xmath105 , @xmath106 . higher @xmath66 is observed to correlate with lower energy , but there are many states even lower in energy than the trial state with largest @xmath66 that is compatible with the finite systems size . panel c : comparison of two - particle densities of states for same - spin case ( blue arrows indicating delta functions ) and for opposite - spin case ( red curve).,title="fig : " ] panel a : eigenvalues @xmath95 of the opposite - spin two - particle interaction matrix [ cf.eq . ( [ eq : oppspin ] ) ] in units of @xmath96 , sorted by magnitude . data are shown for various values of the angular - momentum cutoff @xmath97 ( blue ) , 20 ( red ) , 30 ( green ) , and @xmath98 . the straight line is a plot of @xmath99 with @xmath100 . panel b : energy spectrum obtained for a system of two particles with opposite spin by exact diagonalization . a finite system size is imposed by limiting the number of modes available in angular - momentum space for each particle to @xmath101 . note the @xmath101-dependence of the obtained values . the data for @xmath102 are also shown as the magenta data points in panel a and exhibit excellent agreement with the power - law - type distribution predicted from the solution in com and relative angular - momentum space . green stars show the energy calculated for two - particle versions of trial states @xcite @xmath69 with @xmath103 and @xmath104 , @xmath105 , @xmath106 . higher @xmath66 is observed to correlate with lower energy , but there are many states even lower in energy than the trial state with largest @xmath66 that is compatible with the finite systems size . panel c : comparison of two - particle densities of states for same - spin case ( blue arrows indicating delta functions ) and for opposite - spin case ( red curve).,title="fig : " ] + [ 0.2 cm ] panel a : eigenvalues @xmath95 of the opposite - spin two - particle interaction matrix [ cf.eq . ( [ eq : oppspin ] ) ] in units of @xmath96 , sorted by magnitude . data are shown for various values of the angular - momentum cutoff @xmath97 ( blue ) , 20 ( red ) , 30 ( green ) , and @xmath98 . the straight line is a plot of @xmath99 with @xmath100 . panel b : energy spectrum obtained for a system of two particles with opposite spin by exact diagonalization . a finite system size is imposed by limiting the number of modes available in angular - momentum space for each particle to @xmath101 . note the @xmath101-dependence of the obtained values . the data for @xmath102 are also shown as the magenta data points in panel a and exhibit excellent agreement with the power - law - type distribution predicted from the solution in com and relative angular - momentum space . green stars show the energy calculated for two - particle versions of trial states @xcite @xmath69 with @xmath103 and @xmath104 , @xmath105 , @xmath106 . higher @xmath66 is observed to correlate with lower energy , but there are many states even lower in energy than the trial state with largest @xmath66 that is compatible with the finite systems size . panel c : comparison of two - particle densities of states for same - spin case ( blue arrows indicating delta functions ) and for opposite - spin case ( red curve).,title="fig : " ] straightforward diagonalization of the matrix ( [ eq : oppspin ] ) yields the two - particle eigenenergies @xmath107 when both particles have opposite spin . figure [ 2particlestheo]a shows a logarithmic plot of the @xmath107 , ordered by decreasing magnitude , for different values @xmath108 of the cut - off value for com and relative angular momentum . we observe an exponential dependence of the sorted eigenvalues as a function of the scaled index @xmath109 , which translates into a power - law density of states @xmath110 here @xmath111 is the area corresponding to the cut - off in com and relative angular momentum , and @xmath112 has been determined numerically . the numerical data deviate from eq . ( [ eq : dos ] ) close to the maximum energy @xmath113 , where the density of states reaches zero , and for small energy where it becomes cutoff dependent . none of the individual eigenvalues is strictly independent of the cutoff , which indicates that there are no compact eigenstates . figure [ 2particlestheo]c illustrates the different density - of - states behavior for interacting two - particle systems for the two cases of particles having the same and opposite spin , respectively . it is also useful to look at the distribution of eigenvalues over total angular momentum . we do this with a different numerical scheme using exact diagonalization of the two - particle hilbert space on a disk , as it preserves the @xmath42 component of angular momentum as a good quantum number . ( details are given in the following section . ) the spectrum for @xmath114 is shown in fig . [ 2particlestheo]b . note the dependence of the eigenvalues on the systems size ( ie , the cut - off in angular momentum of available landau - level states ) . any systematic difference between the results given in figs . [ 2particlestheo]a and [ 2particlestheo]b is probably at least in part due to the fact that the representation using the com and relative angular - momentum basis assumes an infinite number of single - particle angular - momentum modes to be available to the particles . nevertheless , when the energy eigenvalues obtained for the finite system are plotted alongside the results for the analytical model ( see magenta data points in fig . [ 2particlestheo]a ) , both are seen to exhibit the same exponential behavior . for comparision , the energies calculated for proposed trial states @xcite are also shown in fig . [ 2particlestheo]b as green stars . while the interacting two - particle problem has lent itself to analytical study , the behavior of systems with three or more interacting particles either requires approximate , e.g. , variational , approaches or must be done numerically . as the complications encountered already for the case of two interacting particles with opposite spin stymie progress for the variational option , we follow the numerical route here . we start by representing the schrdinger field operator for a particle at position @xmath15 with spin @xmath33 projected onto the lowest spin - related landau level , @xmath115 where @xmath116 creates a particle in component @xmath33 with angular momentum @xmath117 in the state @xmath118 . substituting this into eq . ( [ eq : model_int ] ) , we get @xmath119 where @xmath120^\ast \left[\phi^{(\sigma')}_{0 , m_2}({{\ensuremath{\bm{\mathrm{r } } } } } ) \right]^\ast \phi^{(\sigma')}_{0,m_3}({{\ensuremath{\bm{\mathrm{r}}}}})\ , \phi^{(\sigma)}_{0 , m_4}({{\ensuremath{\bm{\mathrm{r } } } } } ) \quad .\ ] ] for same - spin particles , i.e. , @xmath121 , we obtain @xmath122 in contrast , for the matrix element involving opposite - spin particles ( @xmath123 ) , we find @xmath124 the way indices are distributed in the arguments of the @xmath125-functions in eqs . ( [ eq : sameint ] ) and ( [ eq : oppositeint ] ) implies that the system s total angular momentum @xmath126 [ cf . ( [ eq : angmom ] ) for the definition of @xmath127 is a conserved quantity in the presence of interactions . cold - atom systems are usually studied while trapped by an external potential of tunable strength . to model this situation , we introduce the second - quantized form of a parabolic potential in the representation of lowest - landau - level states , @xmath128 where @xmath129 in terms of the harmonic - trap frequency @xmath130 . switching on the trap will lift degeneracies of few - particle states and serve to identify the most compact ground states of our systems of interest . spectrum for various four - particle systems ( i.e. , @xmath131 ) . energies are given in units of the intra - species haldane - pseudopotential energy scale @xmath132 . a : single - component system with @xmath133 , @xmath134 . the four - particle laughlin state is the zero - energy state with the smallest total angular momentum @xmath135 . b : system with @xmath136 and @xmath137 , @xmath138 ( no interspecies interaction ) . c : same situation as for b but with a finite trapping potential ( @xmath139 ) switched on in addition , revealing the energy degeneracies in b. the lowest - energy state is a superposition of two - particle laughlin states in each component . d : same situation as for b but with finite interspecies interaction @xmath140 in addition.,width=453,height=340 ] we calculate the few - particle energy spectra and associated eigenstates for @xmath141 in the fock basis of lowest - landau - level states for the two spin components . we focus here on the case of bosonic particles to be directly applicable to currently studied ultra - cold atom systems , but our general conclusions apply to systems of fermionic particles as well . results obtained for systems with @xmath131 are shown in fig . [ fig : n2n2spec ] . panel a shows the situation where only particles from a single component are present , which is analogous to the previously considered case of spinless bosons @xcite . the zero - energy state at lowest total angular momentum has @xmath142 and corresponds to the filling - factor-@xmath143 laughlin state @xcite . zero - energy eigenstates at higher magnitudes of total angular momentum correspond to edge excitations of the laughlin state @xcite . the @xmath144 state has an energy of @xmath145 , where @xmath146 . when particles occupy states in both components , the situation becomes complex . without interaction between the different spin species , states of each component will be the ones that are obtained by diagonalising the interacting hamiltonian within that component . the entire system is then essentially an independent superposition of eigenstates for the individual spin species . however , in contrast to ordinary multi - component qh states discussed , e.g. , in refs . @xcite , the total angular momenta for states from different components have opposite sign . therefore , e.g. , the combination of laughlin states in each component with the same number of particles has zero total angular momentum . this case is illustrated in fig . [ fig : n2n2spec]b . however , the superpositions of edge excitations with same magnitude of excess angular momentum for the opposite - spin laughlin states will also be zero - energy , zero - angular - momentum eigenstates . to reveal the associated degeneracies of the spectrum shown in fig . [ fig : n2n2spec]b , we obtained the energy eigenvalues in the presence of a parabolic confinement . see fig . [ fig : n2n2spec]c . notice the band of low - lying energy levels separated by a gap from higher - energy states . the lowest - energy @xmath144 state is the superposition of the two - particle laughlin states for the two spin species . the other states in the low - energy band correspond to edge excitations of this configuration . figure [ fig : n2n2spec]d illustrates the dramatic effect of interactions between opposite - spin particles . the spectrum seen there has to be compared with that given in panel b where only particles with the same spin interact . note the disappearance of energy gaps and accumulation of states at low energy , reflecting the characteristic features of the opposite - spin two - particle interaction spectrum shown in fig . [ 2particlestheo]b . clearly , the system is not incompressible anymore , and no qh - related physics can be expected to occur . low - lying energy levels for a system with @xmath147 in the sector of total angular momentum @xmath144 . in the calculation , lowest - landau - level states with @xmath148 have been included . different panels correspond to different interspecies - interaction strengths . a : no inter - species interactions ( @xmath138 ) . a finite trapping potential lifts the energy degeneracies seen at @xmath149 and singles out a unique lowest - energy state . for small @xmath150 , the latter turns out to be the superposition of laughlin states for each individual component . at @xmath151 it becomes an incompressible state with a single laughlin quasi - particle in each component . finally , at @xmath152 both components are bose - condensed in the lowest landau level . modest interspecies - interaction strengths ( @xmath153 in panel b and @xmath154 in panel c ) cause avoided crossings but preserve the incompressible nature of the states seen in panel a. a stronger interspecies interaction ( @xmath155 in panel d ) washes out that picture completely.,width=453,height=340 ] in figure [ fig : energy3 ] , the interplay between interactions and confinement is elucidated . panels a d show the evolution of low - lying few - particle eigenstates as the confinement strength is varied for situations with different magnitude of interaction strength between opposite - spin particles . panel a corresponds to the case with @xmath138 . due to the occurrence of level crossings , the character of the lowest - energy ( ground ) state is found to be different for regimes associated with weak , intermediate , and strong confinement . analogous behavior has been discussed previously for ordinary ( spinless ) few - boson fractional qh systems @xcite . in our case depicted in fig . [ fig : energy3]a , the ground state in the weak - confinement regime corresponds to a superposition of three - particle laughlin states for filling factor @xmath143 in the individual pseudospin components . after the first level crossing , each component turns out to be in the laughlin - quasiparticle state @xcite and , after another level crossing , each spin component has its three particles occupying the lowest state defined by the parabolic confinement potential . ( our description of the ground states found in the three different regimes is supported by the analysis of real - space density and angular - momentum distribution functions . see the following subsection for details . ) switching on interactions between opposite - spin particles turns crossings into anti - crossings . figures . [ fig : energy3]b and [ fig : energy3]c depict situations where interactions between same - spin particles are still dominant . the existence of anticrossings enables smooth transitions between the different ground states that would not be possible in the case of simple crossings as seen , e.g. , in panel a. independent tuning of interactions between opposite - spin particles can therefore be used to enable engineering of quantum many - particle states in ways not anticipated in previous work @xcite . strong interactions between opposite - spin particles are again seen to fundamentally alter the character of the system s ground and excited states . in fig . [ fig : energy3]d , the strengths of interactions between same - spin and opposite - spin particles are equal . the variation of few - particle states as a function of confinement strength is seen to be almost uniform , again pointing to the loss of distinctiveness for few - particle states in the presence of inter - species interactions . furthermore , energy differences between low - lying states are much reduced as compared to the situation depicted in panel a of the same figure , which is a reflection of the unusual distribution of energy eigenvalues found for the interacting opposite - spin two - particle system . the one - particle density profiles in coordinate space and in angular - momentum space are useful quantities to enable greater understanding of the properties of specific many - body quantum states @xcite . in the basis of lowest - landau - level states from the two spin components , the single - particle density matrix of a many - particle state @xmath156 has matrix elements @xmath157 in terms of this quantity , we can define the angular - momentum distribution for each spin component , @xmath158 and also the spin - resolved single - particle density profile in real space , @xmath159^\ast \phi^{(\sigma)}_{0 , m'}({{\ensuremath{\bm{\mathrm{r } } } } } ) \quad .\ ] ] in the following , we focus on the properties of the lowest - energy ( ground ) state in the different regimes associated with small , intermediate , and strong confinement strength for the systems whose energy spectra are shown in fig . [ fig : energy3 ] . cross - sectional pseudo - spin @xmath160 density profiles of the few - particle ground state associated with the lowest - lying energy level shown in the corresponding panels a d of fig . [ fig : energy3 ] , aggregated as a function of the confinement - potential strength @xmath150 . in panel a ( only particles with same spin interact ) , sharp transitions occur between the fqh ( laughlin ) state in the regime of small @xmath150 , a laughlin - quasiparticle - type state for intermediate @xmath150 , and the gaussian bose - einstein - condensed state at high @xmath150 . for moderate interaction strength between opposite - spin components ( repulsive in panel b , attractive in panel c ) , transitions become smooth crossovers associated with anticrossings in fig . [ fig : energy3 ] . stronger interactions strengths between the spin components significantly change the character of the few - particle state at small @xmath150 ( panel d).,title="fig : " ] cross - sectional pseudo - spin @xmath160 density profiles of the few - particle ground state associated with the lowest - lying energy level shown in the corresponding panels a d of fig . [ fig : energy3 ] , aggregated as a function of the confinement - potential strength @xmath150 . in panel a ( only particles with same spin interact ) , sharp transitions occur between the fqh ( laughlin ) state in the regime of small @xmath150 , a laughlin - quasiparticle - type state for intermediate @xmath150 , and the gaussian bose - einstein - condensed state at high @xmath150 . for moderate interaction strength between opposite - spin components ( repulsive in panel b , attractive in panel c ) , transitions become smooth crossovers associated with anticrossings in fig . [ fig : energy3 ] . stronger interactions strengths between the spin components significantly change the character of the few - particle state at small @xmath150 ( panel d).,title="fig : " ] + cross - sectional pseudo - spin @xmath160 density profiles of the few - particle ground state associated with the lowest - lying energy level shown in the corresponding panels a d of fig . [ fig : energy3 ] , aggregated as a function of the confinement - potential strength @xmath150 . in panel a ( only particles with same spin interact ) , sharp transitions occur between the fqh ( laughlin ) state in the regime of small @xmath150 , a laughlin - quasiparticle - type state for intermediate @xmath150 , and the gaussian bose - einstein - condensed state at high @xmath150 . for moderate interaction strength between opposite - spin components ( repulsive in panel b , attractive in panel c ) , transitions become smooth crossovers associated with anticrossings in fig . [ fig : energy3 ] . stronger interactions strengths between the spin components significantly change the character of the few - particle state at small @xmath150 ( panel d).,title="fig : " ] cross - sectional pseudo - spin @xmath160 density profiles of the few - particle ground state associated with the lowest - lying energy level shown in the corresponding panels a d of fig . [ fig : energy3 ] , aggregated as a function of the confinement - potential strength @xmath150 . in panel a ( only particles with same spin interact ) , sharp transitions occur between the fqh ( laughlin ) state in the regime of small @xmath150 , a laughlin - quasiparticle - type state for intermediate @xmath150 , and the gaussian bose - einstein - condensed state at high @xmath150 . for moderate interaction strength between opposite - spin components ( repulsive in panel b , attractive in panel c ) , transitions become smooth crossovers associated with anticrossings in fig . [ fig : energy3 ] . stronger interactions strengths between the spin components significantly change the character of the few - particle state at small @xmath150 ( panel d).,title="fig : " ] figure [ fig : densities3 ] shows the real - space profile of @xmath161 across a diameter of the disk - shaped three - particle systems associated with the ground - state levels shown in fig . [ fig : energy3 ] . in the absence of interactions between opposite - spin particles , each component realizes correlated few - particle states of the type that have been found in previous work @xcite . the few - particle filling - factor-1/2 fqh state is the ground state for a weak confinement potential . increasing the trapping - potential strength favors more compact correlated states , hence , at a critical value of @xmath150 , a transition occurs to a three - particle version of the laughlin - quasiparticle excited state . at even higher @xmath150 , the system transitions to the gaussian bose - einstein - condensate state . the sharpness of the transitions reflects the existence of level crossings in fig . [ fig : energy3]a . practically , simple variation of @xmath150 would not lead to any such transitions because there is no mechanism for the system to switch between different many - particle states . to make adiabatic passage between different many - particle states possible , some symmetry of the system needs to broken , and previous work has proposed scenarios for achieving this in the single - component case @xcite . one - particle angular - momentum distribution for pseudo - spin @xmath160 particles for the ground states of systems whose energy spectra are shown in fig . [ fig : energy3 ] . compare also with the real - space density profiles shown in fig . [ fig : densities3 ] . note that the single - particle angular momentum cut - off at @xmath162 defines the sample size for vanishing @xmath150 in situations where opposite - spin particles interact ( panels b d).,title="fig : " ] one - particle angular - momentum distribution for pseudo - spin @xmath160 particles for the ground states of systems whose energy spectra are shown in fig . [ fig : energy3 ] . compare also with the real - space density profiles shown in fig . [ fig : densities3 ] . note that the single - particle angular momentum cut - off at @xmath162 defines the sample size for vanishing @xmath150 in situations where opposite - spin particles interact ( panels b d).,title="fig : " ] + one - particle angular - momentum distribution for pseudo - spin @xmath160 particles for the ground states of systems whose energy spectra are shown in fig . [ fig : energy3 ] . compare also with the real - space density profiles shown in fig . [ fig : densities3 ] . note that the single - particle angular momentum cut - off at @xmath162 defines the sample size for vanishing @xmath150 in situations where opposite - spin particles interact ( panels b d).,title="fig : " ] one - particle angular - momentum distribution for pseudo - spin @xmath160 particles for the ground states of systems whose energy spectra are shown in fig . [ fig : energy3 ] . compare also with the real - space density profiles shown in fig . [ fig : densities3 ] . note that the single - particle angular momentum cut - off at @xmath162 defines the sample size for vanishing @xmath150 in situations where opposite - spin particles interact ( panels b d).,title="fig : " ] for our system of interest , an additional possibility arises from the ability to tune the interaction strength between the two spin components . as seen in panels b and c of fig . [ fig : energy3 ] , a moderate value of @xmath163 turns the crossings occurring in panel a into anti - crossings , thus , different many - particle states are now adiabatically connected . concomitantly , there is a continuous evolution of the spin - resolved one - particle density profile as a function of the confinement strength seen in figs . [ fig : densities3]b and [ fig : densities3]c . when interactions between same - spin and opposite - spin particles have the same magnitude , the density profile changes significantly ( see fig . [ fig : densities3]d ) , which indicates that the character of many - particle ground states is very different from a fractional - qsh state . investigation of the one - particle angular - momentum - state distribution for the few - particle ground states discussed so far further solidifies our conclusions . see figure [ fig : momentum3 ] . in the absence of interactions between opposite - spin particles , the characteristic distributions for few - particle versions of the laughlin and laughlin - quasiparticle states emerge at low and intermediate values of @xmath150 . in the limit of strong trapping potential , the system condenses into the @xmath164 state . switching on moderate repulsive ( attractive ) interaction strength between opposite - spin particles smoothens the transitions and also shifts the critical values of @xmath150 to larger ( smaller ) values . when interactions among same - spin and between opposite - spin particles have equal magnitude , the one - particle momentum distribution of the ground state differs markedly from that associated with a fractional - qh state . it appears that strong inter - component interactions favor a state with increased occupation of high - angular - momentum states , spreading out the particles more evenly across the accessible sample size and leading to an accumulation at the system s boundary . in the limit of vanishingly small trapping - potential strength @xmath150 , the latter is defined by the cut - off for single - particle angular momentum applied in our calculations . we have considered the interplay of landau quantization and spin - dependent interactions in systems where particles with same spin feel the same strong magnetic field whereas particles with opposite spin are subject to magnetic fields with the same magnitude but opposite direction . it has been expected @xcite that such systems exhibit the fractional qsh effect , but we find that interactions between particles with opposite spin weaken or destroy features associated with fractional - qsh physics . similar behavior has been seen in numerical studies of lattice realizations of fractional - qsh systems @xcite . we have elucidated how behavior that is very different from ordinary two - component fractional - qh systems is rooted in the drastically different spectral properties of two - particle interactions for particles feeling the same versus opposite magnetic - field directions . thus any feasible route towards realizing the fractional qsh effect using a spin - dependent uniform magnetic field @xcite should strive to eliminate interactions between the opposite - spin components . if the opposite - spin interaction strength is weak , adiabatic passage between different correlated many - particle states is facilitated by adjusting the strength of a trapping potential . our conclusions are supported by numerically obtained real - space - density profiles and angular - momentum - state occupation distributions for few - particle systems . the latter could also be utilized as blueprints for classifying images of correlated ultra - cold atom states . part of the motivation for this project came about from stimulating conversations that one of us ( uz ) had with j. j. heremans and r. winkler at the 2011 gordon godfrey workshop on spins and strong correlations ( sydney , australia , 24 28 october 2011 ) . we would also like to thank m. fleischhauer and a. h. macdonald for useful discussions . of was supported by the marsden fund council from government funding ( contract no . mau1205 ) , administered by the royal society of new zealand . 10 url # 1#1urlprefix[2][]#2 bloch i , dalibard j and nascimbene s 2012 _ nat . phys . _ * 8 * 267276

understanding of fully developed turbulence is a complex , rich , and challenging problem . among the most important features of such behavior are cascades of energy and intermittency . the former one brings energy from large scales , responsible for a creation of turbulence , toward smaller scales , in which viscosity begins to play a major role and dissipation effects dominate , whereas the latter one , being itself an irregular alternation of phases of some dynamics , in fact means that very rare configurations of a system give the most significant contributions to statistical distributions . for turbulence this fact is revealed in anomalous scaling , which implies singular behavior of various statistical quantities as functions of the integral turbulence scales @xcite . another very interesting problem is a turbulent advection of an impurity field . both natural experiments and numerical simulations suggest that deviations from the classical kolmogorov theory are even more strongly pronounced for passively advected fields than for the velocity field itself @xcite . a turbulent environment in such models may be introduced by some `` synthetic '' velocity field with prescribed statistics or by stochastic navier - stokes equation @xcite . models of the former type are more tractable from a mathematical point of view , whereas the latter models bear a closer resemblance to the real world . a very powerful theoretical framework , which has worked very well in statistical physics in order to investigate different models of critical phenomena , characterized by scaling behavior , is the field theoretic renormalization group ( rg ) and the operator product expansion ( ope ) ; it has become an indispensable tool when dealing with systems where no clear separation of scales occurs , see the monographs @xcite . in the rg+ope scenario , an anomalous scaling emerges as a consequence of the existence of some composite fields ( `` composite operators '' in the quantum - field terminology ) in the model with negative scaling dimensions @xcite . the main advantage of this approach with respect to the turbulence is that it is based on a microscopic model and allows one to construct a systematic perturbation expansion for the anomalous exponents , similar to the famous @xmath4-expansion in theory of critical phenomena . the procedure has been successful in verifying kolmogorov scaling and provides an efficient theoretical tool for calculation of universal quantities @xcite . in a number of papers the rg+ope approach has been applied to the case of passive advection by kraichnan s ensemble ( velocity field is taken isotropic , gaussian , not correlated in time , having a power - like correlation function , and fluid assumed to be incompressible ) , see @xcite , and by its numerous generalizations : large - scale anisotropy , helicity , compressibility , finite correlation time , non - gaussianity , and a more general form of the nonlinearity @xcite . a similar approach can be generalized as well to the case of non - gaussian velocity field , governed by the stochastic navier - stokes equation both for the scaling behavior of velocity field itself , and for an advection of passive fields by this ensemble @xcite . up to now a majority of works on fully developed turbulence have been concerned with an incompressible fluid . nevertheless , several results for compressible fluids have also been obtained @xcite . all of them hints at large influence of compressibility both to velocity field itself and passively advected quantities . in particular , the transition from a turbulent to a certain purely chaotic state may occur at large degrees of compressibility @xcite . in other papers corrections in the mach number to the incompressible scaling regime were studied @xcite . thus , obtained corrections become arbitrarily large and destroy the incompressible scaling regime for fixed mach numbers and large distances , what can be explained by an existence of crossover to another , yet unknown regime . the works @xcite were devoted to a compressible fluid itself . the results are rather controversial , particularly the model , considered in @xcite , appears to be in fact unrenormalizable . from a general point of view further investigations of compressibility are therefore called for . in this paper we present an application of the field theoretic renormalization group onto the scaling regimes of a compressible fluid , whose behavior is governed by a proper generalization of the stochastic navier stokes equation @xcite . a resulting stationary scaling regime in this approach is associated with an infrared ( ir ) attractive fixed point of the corresponding multiplicatively renormalizable field theoretic model . in a similar fashion as has been done for an incompressible case @xcite , the double expansion scheme is employed . here the formal expansion parameters are @xmath2 , which describes the scaling behavior of a random force , and @xmath5 , i.e. , a deviation from the space dimension @xmath0 . following a procedure of previous works @xcite , first the stochastic navier - stokes equation is discussed . after establishing an existence of necessary fixed points of navier - stokes equation , the advection of scalar fields is explored . since rg functions of the parameters , entering into navier - stokes equation , do not depend on the parameters , connected with advection - diffusion equation , this is a possible and probably the easiest way . the paper is organized as follows . in sec . [ sec : model ] a detailed description of the stochastic navier - stokes equation for a compressible fluid is given . [ sec : qft ] is devoted to field theoretic formulation of the model and the corresponding diagrammatic technique . in particular , possible types of divergent green functions at @xmath1 and @xmath0 and the necessity of introducing a new coupling constant at @xmath0 are discussed . in sec . [ sec : renorm ] the renormalizability of the model is established and one - loop explicit expressions for the renormalization constants are derived . much attention is paid here to technical aspects of performed calculations . in sec . [ sec : rc ] the rg functions ( anomalous dimensions and @xmath6 functions ) are derived and examined . ir asymptotic behavior , obtained as a result of solution of rg equations , is discussed . it is shown that , depending on two exponents @xmath2 and @xmath4 , the rg equations possess an ir attractive fixed point , which implies existence of a scaling regime in the inertial range . the corresponding scaling dimensions of all fields and parameters of the model are presented . in sec . [ sec : ops1 ] advection of a passive scalar ( density ) field by compressible velocity field , which obeys navier - stokes equation , is analyzed . field theoretic formulation of the full model is presented . it is shown that the full model is multiplicatively renormalizable ; the existence of a scaling regime in the ir range is established . the renormalization of composite operators is carried out . an inertial - range behavior of various correlation functions is studied by means of the ope . it is shown that leading terms of the inertial - range behavior are determined by the contributions of the operators built solely of the scalar fields itself . as a result , the ir behavior of the pair correlation functions of the composite operators is power - like with negative critical dimensions the situation , called anomalous scaling , occurs . sec . [ sec : conc ] is reserved for conclusion . the navier - stokes equation for a viscid compressible fluid can be written in the following form @xcite : @xmath7 v_{k } + \mu_0 \partial_{i}\partial_{k } v_{k } - \partial_{i } p + \eta_{i } , \nonumber \\ { } \label{ns}\end{aligned}\ ] ] where the differential operator on the right hand side @xmath8 is the lagrangian ( convective ) derivative , @xmath9 is a fluid density field , @xmath10 is a velocity field , @xmath11 , @xmath12 , @xmath13 is the laplace operator , @xmath14 is a pressure field , and @xmath15 is a density of an external force per unit volume . the fields @xmath10 , @xmath16 , @xmath9 and @xmath14 depend on @xmath17 with @xmath18 , where @xmath19 is a dimensionality of space . the constants @xmath20 and @xmath21 are two independent molecular viscosity coefficients @xcite ; in we have explicitly separated the transverse and longitudinal components of the viscous term . summations over repeated vector indices are always implied in this work . to get a closed system of equations , the model ( [ ns ] ) must be augmented by two additional equations , namely a continuity equation and an equation of state between deviations @xmath22 and @xmath23 from the equilibrium values . they explicitly read @xmath24 and @xmath25 in order to obtain the renormalizable field , a theoretic model expression ( [ ns ] ) has to be divided by @xmath9 , and fluctuations in viscous terms have to be neglected @xcite . further , by using expressions and the problem can be recast in the form of two coupled equations : @xmath26 v_{k}\ ! + \ ! \mu_0 \partial_{i}\partial_{k } v_{k } -\ ! \partial_{i } \phi\ ! + \ ! f_{i } , \label{anu } \\ \nabla_{t } \phi & = & -c_{0}^{2 } \partial_{i}v_{i}. \label{anu1}\end{aligned}\ ] ] here a new scalar field @xmath27 is related to the density fluctuations via the relation @xmath28 . a parameter @xmath29 is an adiabatic speed of sound , @xmath30 denotes the mean value of @xmath9 , and @xmath31 is a density of the external force per unit mass . in the stochastic formulation of the problem the turbulence is modeled by an external force it is assumed to be a random variable , which mimics the input of energy into the system from the outer large scale @xmath32 . its precise form is believed to be unimportant and is usually considered to be a random gaussian variable with zero mean and prescribed correlation function @xcite . for the use of the standard rg technique this correlator must exhibit a power law asymptotic behavior at large wave numbers @xcite . in the case of compressible fluid it should be naturally augmented with a non - transverse component , hence the simplest way is to choose it in the form @xcite @xmath33 where the argument is given by @xmath34 here , @xmath35 and @xmath36 are the transverse and longitudinal projectors , @xmath37 , amplitude @xmath38 is a free parameter , an amplitude @xmath39 is a coupling constant ( formal expansion parameter in the ordinary perturbation theory ) ; the relation @xmath40 sets in the typical ultraviolet ( uv ) momentum scale @xmath41 , which is a reciprocal of the dissipation length scale . a parameter @xmath42 provides an infrared regularization ; its precise form is unessential and the sharp cut - off is the simplest choice for calculation purposes . an exponent @xmath2 provides the uv regularization and therefore plays a role of a formally small expansion parameter @xcite . the most realistic ( physical ) value is obtained in the limit @xmath43 , when the function in ( [ power ] ) can be interpreted as power - like representation of the dirac function @xmath44 : physically it corresponds to the idealized picture of the energy input from infinitely large scales . the galilean invariance for the model ( [ ns ] ) is ensured when the function is delta - correlated in time @xcite . according to the general theorem @xcite , the stochastic problem is equivalent to the field theoretic model with a doubled set of fields @xmath45 and de dominicis - janssen action functional , written in a compact form as @xmath46v_k + u_0 \nu_0 \partial_i \partial_k v_k - \partial_i \phi \biggl\ } \nonumber \\ & + \phi'[-\partial_t \phi + v_j\partial_j \phi + v_0 \nu_0 \partial^2 \phi - c_0 ^ 2 ( \partial_i v_i ) ] , \label{eq : action } \end{aligned}\ ] ] where @xmath47 is the correlation function ( [ power ] ) . here a condensed notation has been employed , in which integrals over the spatial variable @xmath48 and the time variable @xmath49 , as well as summation over repeated indices , are implicitly assumed . moreover , we have introduced a new dimensionless parameter @xmath50 and introduced a new term @xmath51 with another positive dimensionless parameter @xmath52 , which is needed to ensure multiplicative renormalizability . the action ( [ eq : action ] ) is amenable to the standard methods of the quantum field theory , such as feynman diagrammatic technique and renormalization group procedure . in a standard approach , if quantum field methods are applied to the stochastic differential equations , the space dimension @xmath19 plays a passive role and an actual perturbative parameter is @xmath2 ; for more details see the monographs @xcite . our approach closely follows the analysis of the incompressible navier - stokes equation near space dimension @xmath53 @xcite . in this case an additional divergence appear in the green function @xmath54 . it can be absorbed by a suitable local counterterm @xmath55 , and a regular expansion in both @xmath2 and @xmath56 must be constructed . up to now the present model ( [ eq : action ] ) has been investigated in fixed space dimension @xmath1 , for which the action ( [ eq : action ] ) contains all terms that can be generated during the renormalization procedure @xcite . however , from the dimensional analysis ( see below ) it follows that in @xmath0 an additional divergence appears in a similar fashion in the green function @xmath54 . therefore , to keep the model renormalizable in @xmath0 , the kernel function @xmath57 in ( [ ff ] ) has to be generalized into the function @xmath58 , where @xmath59 a new term on the right hand side with an additional coupling constant @xmath60 absorbs divergent contributions from @xmath54 . in contrast to the two - dimensional incompressible case @xcite no momentum dependence is needed here . the field theoretic formulation ( [ eq : action ] ) means that various correlation and response functions of the original stochastic problem are represented by functional averages over the full set of fields with functional weight @xmath61 , and in this sense they can be viewed as green functions of the field theoretic model @xcite . the perturbation theory of the model can be expressed according to the usual feynman diagrammatic expansion @xcite . bare propagators are read off from the inverse matrix of the gaussian ( free ) part of the action functional , while a nonlinear part of the differential equation defines the interaction vertices . a graphical representation of the propagator functions is depicted in fig . [ fig : prop_vertex ] , and of the vertices in fig . [ fig:2 ] . from it follows that the feynman diagrammatic technique for this model contains two vertices , @xmath62 and @xmath63 . the propagator functions in the frequency - momentum representation read @xmath64 where the vector indices of the fields and the projectors are omitted . for convenience following abbreviations have been used @xmath65 and @xmath66+g_{20}\nu_0 ^ 3 . \label{energies}\end{aligned}\ ] ] in the limit @xmath67 propagator functions @xmath68 and @xmath69 become purely transverse , and all mixed propagators except @xmath70 vanish . moreover , the scalar fields @xmath71 and @xmath72 decouple from the velocity fields @xmath73 and @xmath74 it is impossible to construct a diagram with only velocity fields @xmath73 and @xmath75 as external lines , containing internal lines with fields @xmath71 or @xmath72 . thus , as it should be from physical point of view , the well - known feynman rules for the incompressible fluid @xcite are obtained . [ cols="^ " , ] canonical dimensions of the new fields and parameters of the full model ( [ fact ] ) are given in table [ table1 ] , where we have introduced a new dimensionless parameter @xmath76 with @xmath20 from ( [ ns ] ) . the expression ( [ index ] ) for the formal index of uv divergence remains valid , but the summation should now run over the full set of six fields @xmath77 . the rules ( 1 ) ( 6 ) from sec . [ sec : canon ] should be generalized and augmented as follows : 1 . all the 1-irreducible green functions without the response field @xmath78 vanish identically and require no counterterms . 2 . using integration by parts the derivative at the vertex @xmath79 can be moved onto the field @xmath78 , therefore expression should be modified as @xmath80 since the field @xmath78 can enter the counterterms only in the form of spatial derivatives , the counterterm @xmath81 to the 1-irreducible green function @xmath82 with @xmath83 , @xmath84 is forbidden . also this requires that the counterterms to the 1-irreducible function @xmath85 with @xmath86 , @xmath87 necessarily reduce to the form @xmath88 . galilean symmetry wherein allows them enter the counterterms only in the form of invariant combination @xmath89 . hence , they are also forbidden . 3 . as consequence of the linearity of the original stochastic equation ( [ density1 ] ) with respect to the field @xmath90 one obtains that for any 1-irreducible function @xmath91 , where @xmath92 is the total number of bare propagators @xmath93 entering the diagram . this fact is very important for the renormalizability of the model : without the restriction @xmath94 , we the infinity number of superficially divergent 1-irreducible functions @xmath95 would proliferate , and hence the lack of renormalizability would follow . from these rules we finally conclude that superficial divergences can be present only in 1-irreducible green function @xmath82 with the only counterterm @xmath96 . it is naturally reproduced as multiplicative renormalization of the diffusion coefficient , @xmath97 . no renormalization of the fields @xmath78 and @xmath90 is needed : @xmath98 . the renormalized analog of the action functional ( [ fact ] ) has the form @xmath99 where @xmath100 is the action ( [ raction ] ) , @xmath101 @xmath102 here stands for the covariance of stochastic force . the one - loop approximation for the 1-irreducible response function @xmath82 can be formally written as @xmath103{g2_1.eps } } , \label{dyson}\end{aligned}\ ] ] where , as earlier in expressions , @xmath104 represents an external momentum ; single solid line denotes the bare propagator @xmath105 from ( [ lines ] ) , double solid line with a slash denotes the bare propagator @xmath106 from ( [ lines3 ] ) , the slashed end corresponds to the field @xmath78 . the dots with three attached fields @xmath78 , @xmath90 and @xmath107 denote the vertex ( [ vertex1 ] ) . the constant @xmath108 is to be found from the requirement of uv finiteness of 1-irreducible green function @xmath82 . like for the original navier - stokes model , the divergent part of the considered feynman diagram is independent on @xmath109 and therefore can be calculated directly at @xmath110 ; see the discussion in sec . [ sec : c ] . the analytical expression for the diagram @xmath111 , entering into expression , is @xmath112,\end{aligned}\ ] ] where @xmath113 and @xmath114 are two vertices of the type ; scalar coefficients @xmath115 and @xmath116 of the propagator @xmath117 are defined in . since in the leading - order approximation the renormalization constant @xmath118 in the bare term of ( [ dyson ] ) is taken only in the first order in coupling constants @xmath119 and @xmath120 , during the actual calculation all other renormalization constants in the diagram @xmath111 , entering , for example , into functions @xmath115 and @xmath116 , should be replaced with unities . integration over the frequency gives @xmath121 } ; \nonumber \\ \int\frac{\drm\omega}{2\pi } \frac{b(k ) } { -i\omega+w_0\nu ( { \bf p}+{\bf k})^{2 } } & = \frac{1}{2\nu^2uk^2[uk^2+w_0({\bf p+k})^2]}. \label{tmp}\end{aligned}\ ] ] divide expression into two parts : @xmath122 } ; \nonumber \\ \check{i}_2&=-\int\frac{\drm^d{\bf k}}{(2\pi)^{d}}p_a(p+k)_c \times \frac{q_{ac}({\bf k})}{2\nu^2uk^2[uk^2+w_0({\bf p+k})^2]}.\end{aligned}\ ] ] we are interesting in the term , proportional to @xmath123 , therefore in calculation of @xmath124 one may immediately put @xmath125 and , using expressions and and notation @xmath126 , gets @xmath127 in calculation of @xmath128 the second expression of has to be expanded up to the linear term in @xmath104 ; in the end we get @xmath129 finally , collecting @xmath124 and @xmath128 yields @xmath130.\end{aligned}\ ] ] then the renormalization constant @xmath118 [ see ] should be chosen as @xmath131\frac{g_1}{y } \nonumber \\ - \frac{1}{2dw } \left [ \frac{d-1}{1+w } + \frac{u - w}{u(u+w)^{2 } } \right]\frac{g_2}{\varepsilon } , \label{zk}\end{aligned}\ ] ] while the corresponding anomalous dimension is @xmath132g_1 \nonumber \\ + \frac{1}{2dw } \left [ \frac{d-1}{1+w } + \frac{u - w}{u(u+w)^{2 } } \right]g_2 , \label{gk}\end{aligned}\ ] ] with the possible corrections coming from higher orders terms . the function @xmath133 for the new dimensionless parameter @xmath134 has the form @xmath135 see sec . [ sec : rgrg ] . now the coordinates @xmath136 of fixed points fp@xmath137 and fp@xmath138 [ see and ] are substituted into the equation @xmath139 at @xmath0 . to this end we rewrite expression for @xmath140 at @xmath141 : @xmath142 . \label{u1}\end{aligned}\ ] ] from equation it is clear that the only positive solution for both fp@xmath137 and fp@xmath138 is @xmath143 functions ( [ betagw ] ) do not depend on @xmath134 , therefore new eigenvalue of the matrix ( [ omega ] ) , corresponding to this parameter , coincides with the diagonal element @xmath144 at point @xmath145 : @xmath146}>0 \quad\text{for { fp$ { \textrm{iii}}$}}. \label{wir}\end{aligned}\ ] ] from the inequalities it follows that fixed points with the coordinates and and @xmath147 are ir attractive in the full space of couplings @xmath148 and governs the ir asymptotic behavior of the full - scale model ( [ fact ] ) . the critical dimensions of the passive fields @xmath90 and @xmath78 are obtained from the table [ table1 ] and the expression ( [ krit ] ) for @xmath149 . for fixed point fp@xmath138 they are the same as in the case @xmath1 , namely @xmath150 for another point fp@xmath137 they are @xmath151 in the following , the central role will be played by composite fields ( `` composite operators '' ) built solely of the basic fields @xmath90 : @xmath152 in general , a local composite operator is a polynomial constructed from the primary fields @xmath153 and their finite - order derivatives at a single space - time point @xmath17 . due to coincidence of the field arguments , new uv divergences arise in the green functions with such objects @xcite . the total canonical dimension of arbitrary 1-irreducible green function @xmath154 that includes one composite operator @xmath155 and arbitrary number of primary fields @xmath156 ( the formal index of uv divergence ) is given by the relation @xmath157 where @xmath158 are the numbers of the fields entering into @xmath159 , @xmath160 are their total canonical dimensions , @xmath161 is the canonical dimension of the operator , and the summation over all types of the fields is implied . during renormalization procedure operators can mix with each other , @xmath162 and @xmath163 is the renormalization matrix . but in the simplest case of operators matrix @xmath163 is diagonal , i.e. , @xmath164 . in particular this means that the critical dimension of the operator is given by the expression ( [ krit ] ) . superficial uv divergences , whose removal requires counterterms , can be present only in those functions @xmath159 for which the index of divergence @xmath165 is a non - negative integer . for the operators of the form one has @xmath166 . due to the linearity of our model in @xmath90 , the number of the fields @xmath90 in any 1-irreducible function with the operator @xmath167 can not exceed their number in the operator itself . from the analysis of expression ( [ d - comp ] ) it therefore follows that the superficial divergence can only be present in the 1-irreducible function with @xmath168 and @xmath169 for all other types of the fields @xmath156 . for this function @xmath170 and the corresponding counterterm has the form @xmath171 ; hence , operators are multiplicatively renormalizable , @xmath172 . introduce @xmath173 : the @xmath174 term of the expansion in @xmath175 of the generating functional of the 1-irreducible green functions with one composite operator @xmath167 and any number of fields @xmath90 : @xmath176 renormalization constants @xmath177 are determined by the requirement that the 1-irreducible functions be uv finite in renormalized theory . the one - loop approximation for the 1-irreducible function @xmath178 can be formally written as @xmath179{comp_density.eps } } } \label{gamma2}\end{aligned}\ ] ] the first term is the tree ( loop - less ) approximation , double solid lines with a slash denotes the propagators @xmath180 , single solid line corresponds to the propagator @xmath181 , @xmath182 is the symmetry coefficient of the given graph , and the dot with two attached lines in the top of the diagram denotes the operator vertex , i.e. , the variational derivative @xmath183 the contribution of a specific diagram into the functional for any composite operator @xmath155 is represented in the form @xmath184 where @xmath185 is the vertex factor , @xmath186 is the diagram itself , and the product @xmath187 corresponds to external tails . the divergence of the considering graph is logarithmic , hence one should set all the external frequencies and momenta equal to zero . therefore the analytical expression of the diagram is given by @xmath188,\end{aligned}\ ] ] where @xmath189 and @xmath190 are two vertices ; scalar coefficients @xmath115 and @xmath116 of the propagator @xmath117 are defined in with the replacement of original bare parameters to their renormalized counterparts . as @xmath191 , only second term in gives non - vanishing contribution . integration over the frequency gives @xmath192 contracting tensor indices , using , and collecting all the factors one obtains that expression takes form @xmath193 \biggl\ } , \label{gamma22}\end{aligned}\ ] ] in notation @xmath194 and up to a finite part and higher - order corrections . the renormalization constants @xmath195 are found from the requirement that the renormalized analog @xmath196 of the function ( [ gamma1 ] ) be uv finite in terms of renormalized parameters . in contrast to expressions , in this case renormalization constants @xmath177 belong not to some parameters , but to the green functions itself . this means that using loop expansion one finds in fact not renormalization constants @xmath177 , but inversed one @xmath197 . taking into account minus sign in the exponent , from it follows that in the ms scheme renormalization constants take form @xmath198 the corresponding anomalous dimensions are @xmath199 with higher - order corrections in @xmath119 and @xmath120 . the critical dimensions of the operators @xmath174 from the expression ( [ krit ] ) are readily derived @xmath200= n\delta_{\theta } + \gamma_{n}^{*}. \label{krop}\ ] ] substituting of fixed - point values fp@xmath137 and fp@xmath138 into ( [ gaman ] ) finally gives that for fixed point fp@xmath137 critical dimensions are @xmath201&=-n+\frac{n\varepsilon}{4}-\frac{n(n-1)}{3}\varepsilon ; \label{krop1}\end{aligned}\ ] ] for fixed point fp@xmath138 they differs from the case @xmath1 and are equal to @xmath202&=-n+\frac{ny}{6}-\frac{2n(n-1)}{3}\frac{\alpha y(y-\varepsilon)}{y(2+\alpha)-3\varepsilon}.\end{aligned}\ ] ] both expressions and suppose higher - order corrections in @xmath2 and @xmath4 . for both cases fp@xmath137 and fp@xmath138 the dimensions are negative , i.e. , `` dangerous '' in the sense of operator product expansion @xcite , and decrease as @xmath203 grows . the measurable quantities and , therefore , the objects of interest are equal - time pair correlation functions of two ( uv finite ) renormalized local composite operators @xmath204 . from the dimensionality considerations ( see table [ table1 ] ) it follows that @xmath205 where @xmath206 and @xmath161 are frequency and total canonical dimensions of the correlation function , @xmath207 , and @xmath208 is a function of dimensionless variables . if the correlation function ( [ parr0 ] ) is multiplicatively renormalizable , in the ir region it fulfills the differential equation , which describes the ir scaling behavior . that is , the behavior of the function @xmath208 for @xmath209 is determined by the ir attractive fixed points fp@xmath137 and fp@xmath138 of the rg equation . the solution of this equation leads to the following asymptotic expression : @xmath210 . \label{parr}\ ] ] here @xmath211 is the critical dimension of the correlation function , given by simple sum of the dimensions of the operators ; @xmath212 is unknown scaling function with completely ( both canonically and critically ) dimensionless arguments , and @xmath213 is invariant speed of sound . for the composite operator @xmath214 , expression ( [ parr ] ) yields @xmath215 , \label{parr2}\ ] ] where critical dimensions @xmath216 for two scaling regimes are given by expressions ( [ krop1 ] ) and ( [ krop2 ] ) . representation holds for @xmath217 and any fixed value of @xmath218 . the inertial - convective range @xmath219 corresponds to the additional condition @xmath220 . the behavior of the functions @xmath212 at @xmath221 can be studied by means of the operator product expansion ; see @xcite . according to the ope , the equal - time product @xmath222 of two renormalized operators for @xmath223 and @xmath224 has the representation @xmath225 f(t,{\bf x } ) , \label{ope}\ ] ] where @xmath226 are numerical coefficient functions analytical in @xmath218 and @xmath213 and @xmath155 are all possible renormalized local composite operators allowed by the symmetry . the correlation function ( [ parr ] ) is obtained by averaging ( [ ope ] ) with the weight @xmath227 , where @xmath228 is the renormalized action functional ( [ fact ] ) . mean values @xmath229 appear on the right hand side . their asymptotic behavior at small @xmath230 is found from the corresponding rg equations and has the form @xmath231 , \label{mean}\ ] ] with another set of scaling functions @xmath232 . since the diagrams of the perturbation theory have finite limits both for @xmath233 and @xmath234 , we may assume that the functions @xmath235 are restricted for all values of @xmath236 and can be estimated by some constants . moreover , for invariant variable @xmath213 ir asymptotic behavior together with requirement of its dimensionless gives @xmath237 where @xmath236 is renormalized speed fo sound . this means that @xmath238 . taking into account , for non - local scaling regime fp@xmath138 one obtains that for @xmath239 ( i.e. , including the most realistic case @xmath43 ) the argument @xmath240 becomes small for fixed @xmath236 and @xmath241 , and the function @xmath232 can be replaced by its finite limit value @xmath242 . for local scaling regime fp@xmath137 from it follows that at @xmath243 function @xmath232 can be replaced by its finite limit value @xmath244 . from these two remarks we finally conclude that in the ir range for both local and non - local scaling regimes up to a different constants we can write @xmath245 combining rg representation ( [ parr2 ] ) with information gained from ope ( [ ope ] ) and expression ( [ mean2 ] ) gives the desired asymptotic behavior of the scaling functions : @xmath246 \simeq \sum_{f } a_{f } [ mr , c(r)]\times ( mr)^{\delta_{f } } , \label{fin}\ ] ] where the summation runs over all galilean invariant scalar operators ( including operators with derivatives , etc . ) , with the coefficient functions @xmath247 analytical in their arguments . the leading contribution into the sum is given by the operator with the lowest ( minimal ) critical dimension ; the others can be considered as corrections . the anomalous scaling ( i.e. , singular behavior as @xmath221 ) results from the contributions of the operators with negative critical dimensions . from and it is easily seen that for both scaling regimes all the operators @xmath174 have negative dimensions , and the spectrum of their dimensions is not restricted from below . fortunately , due to the linearity of initial stochastic equation in the field @xmath90 , the number of such fields on the right hand side of expression can not exceed their number on the left hand side . this means that for a given correlation function only a finite number of those operators can contribute to the ope . for correlation functions ( [ parr2 ] ) these operators are those for which @xmath248 . the leading term of the behavior as @xmath221 is given by the operator with the maximum possible @xmath249 and without any derivatives , so the final expression has the form @xmath250 the fact that the leading term in the ope is given by the operator from the same family with the summed exponent together with inequality @xmath251 can be interpreted as the statement that the correlations of the scalar field in the model show multifractal behavior ; see @xcite . in this paper , which is a natural extension of @xcite , the stochastic navier stokes equation for a _ compressible _ fluid has been studied using the field theoretic approach . in contrast to previous work , the model is considered in special space dimension @xmath0 . in this case additional uv divergence is observed in 1-irreducible green function @xmath252 . this fact significantly affects rg analysis , regarding both technical aspects and results . in particular , renormalization group technique with double expansion scheme is needed . in the one loop approximation , depending on the exponent @xmath2 and deviation from the dimension of @xmath48 space @xmath5 , the model possesses two stable nontrivial fixed points in the ir region ( i.e. , two possible nontrivial scaling regimes ) local one , called in text fp@xmath137 , and nonlocal , called fp@xmath138 . this shows that the simple analysis around @xmath1 , which indicates existence of only one nontrivial fixed point , corresponding to non - local scaling regime , can be regard as incomplete in this case . the crossover between local and non - local regimes occurs along the line @xmath253 , which is in accordance with @xcite . the new ( local ) regime , which takes place at @xmath0 , should by continuity move into @xmath1 at @xmath243 . nevertheless , at physical values @xmath254 and @xmath255 new local regime fp@xmath137 is unstable , and non - local regime fp@xmath138 is realized , therefore the main conclusions of the previous papers @xcite about existence and ir attraction of this fixed point remain true . the model of passive scalar advection of density field by this velocity ensemble has been also analyzed . the full stochastic problem can be formulated as field theoretic model , which appear multiplicatively renormalizable . the new parameter @xmath256 does not affect to rg functions of the navier - stokes equation itself , so the critical behavior of this model is also described by two fixed points , local one and non - local one . the inertial range ( @xmath219 ) behavior of correlation functions has been studied by means of the ope technique . existence of anomalous scaling , i.e. , singular power - like dependence on the integral scale @xmath32 , has been established . the corresponding anomalous exponents have been identified with critical dimensions of certain composite operators and calculated in the leading one - loop approximation . as in the case @xmath1 , results of the present model have some similarities and some differences with simplified models , which describe compressible turbulent velocity field as some gaussian statistical ensemble @xcite . in contrast to @xcite ( gaussian ensemble with finite correlation time ) , the present model is manifestly galilean covariant , and this fact holds in all orders of the perturbation theory . as in the rapid - change ( zero correlation time ) models , critical dimensions are insensitive to the specific choice of the random force . this is a consequence of the fact that propagator @xmath257 does not enter into the relevant green functions . in principle , the anomalous exponents could depend on some dimensionless parameters like @xmath258 , @xmath259 , or @xmath260 . after the rg procedure , these parameters are replaced with the corresponding invariant variables . existence of the unique ir attractive fixed point for each pair @xmath261 shows that in the ir range these invariant variables tend to their fixed - point values @xmath262 , @xmath263 , @xmath264 irrespective of the initial values @xmath258 , @xmath259 , and @xmath260 . this is a consequence of the one - loop approximation ; in higher - loop level overlaps may occur in the picture of domains of ir stability . in contrast to the case @xmath1 , at @xmath0 anomalous dimensions of composite operators and do not grow with @xmath38 without bound . this is a consequence of eliminating of poles in @xmath4 near @xmath0 , which leads to a significant improvement of situation near @xmath1 . the situation is similar to consideration of navier - stokes equation for incompressible fluid near @xmath53 , where renormalization procedure at @xmath53 is a reason for for the kolmogorov constant s better agreement with experimental results @xcite . as in the case @xmath1 , at @xmath0 the coordinates of both nontrivial fixed points are valid for all @xmath265 , have finite limits at @xmath266 , but ceases to exist for purely potential forcing , i.e. , for @xmath267 . this fact means that under the one - loop approximation both fixed points fp@xmath137 and fp@xmath138 appears to disappear or lose its stability . corresponding scaling regimes consequently undergoes some qualitative changeover , which is possibly accompanied by phase transition to a purely chaotic state observed in @xcite for a simplified model . to clarify this problem it is necessary to go beyond the one - loop approximation and to discuss the existence , stability and the dependence on @xmath38 of fixed points at least at the two - loop level , which seems to be a difficult technical task . also it would be very interesting to investigate a scalar admixture in case of tracer field , or passively advected vector fields . another very important task is to develop compressible navier - stokes equation near @xmath53 . such analysis may show another types of ir behavior or another dependence of different parameters like @xmath38 , viscosity ratios , etc . this work is left for the future and is partly in progress . the authors are indebted to m. yu . nalimov , l. ts . adzhemyan , m. hnati , j. honkonen , and v. kultty for discussions . the work was supported by vega grant no . 1/0222/13 of the ministry of education , science , research and sport of the slovak republic , and by the russian foundation for basic research within the project 16 - 32 - 00086 . n. m. g. acknowledges the support from the saint petersburg committee of science and high school . n. m. g. and m. m. k. were also supported by the dmitry zimin s `` dynasty '' foundation . k. gawdzki and a. kupiainen , phys . lett . * 75 * , 3834 ( 1995 ) ; d. bernard , k. gawdzki , and a. kupiainen , phys . e * 54 * , 2564 ( 1996 ) ; m. chertkov , g. falkovich , i. kolokolov , and v. lebedev , phys . e * 52 * , 4924 ( 1995 ) ; m. chertkov and g. falkovich , phys . rev . lett . * 76 * , 2706 ( 1996 ) ; a. pumir , europhys . lett . * 34 * , 25 ( 1996 ) ; * 37 * , 529 ( 1997 ) ; phys . rev . e * 57 * , 2914 ( 1998 ) . l. ts . adzhemyan , n. v. antonov , v. a. barinov , yu . s. kabrits , and a. n. vasilev , phys . e * 63 * , 025303(r ) ( 2001 ) ; e * 64 * , 019901(e ) ( 2001 ) ; e * 64 * , 056306 ( 2001 ) . n. v. antonov , phys . e * 60 * , 6691 ( 1999 ) ; l. ts . adzhemyan , n. v. antonov , and j. honkonen , phys . e * 66 * , 036313 ( 2002 ) . d. ronis , phys . a * 36 * , 3322 ( 1987 ) ; l. ts . adzhemyan , j. honkonen , m. v. kompaniets , a. n. vasilev , phys . e * 68 * , 055302(r ) ( 2003 ) ; l. ts . adzhemyan , m. hnatich and j. honkonen , eur . j b * 73 * , 275 ( 2010 ) . c. de dominicis and p. c. martin , phys . a * 19 * , 419 ( 1979 ) ; p. l. sulem , j. d. fournier , and u. frisch , lecture notes in physics , * 104 * , 321 ( 1979 ) ; j. d. fournier and u. frisch , phys . a * 28 * , 1000 ( 1983 ) ; l. ts . adzhemyan , a. n. vasilev , and yu . m. pismak , theor . * 57 * , 1131 ( 1983 ) . adzhemyan , a. n. vasilev , and m. hnatich , theor . phys . * 74 * , 115 ( 1988 ) ; l. ts . adzhemyan , n. v. antonov , and t. l. kim , theor . math . phys . * 100 * , 1086 ( 1994 ) ; n. v. antonov , s. v. borisenok , and v. i. girina , theor . math . phys . * 106 * , 75 ( 1996 ) ; n. v. antonov and a. n. vasilev , theor . phys . * 110 * , 97 ( 1997 ) .

the most massive stars begin their lives in the darkest recesses of giant molecular clouds , obscured from our view by a thick veil of dust and gas . most of the time the earliest stages of massive star formation can not be observed at visual wavelengths , and in many cases even the near - infrared emission from these protostars can not escape . we owe much of our ignorance of the massive star formation process to the fact that we have a very limited range of wavelengths at our disposal with which to observe these youngest stages of massive stellar evolution . the problem is further compounded by the fact that massive ob stars ( @xmath1 8 m@xmath2 ) are relatively few in number and , like all stages of massive stellar evolution , the formation process proceeds very quickly . it is therefore difficult to catch massive stars in the act of forming . additionally , there is the problem of observing these objects with sufficient angular resolution . massive star - forming regions generally lie at kiloparsec distances , making them more than an order of magnitude farther away than the closest low - mass star - forming regions . furthermore , the percentage of binaries and multiple star systems increases as a function of mass ( larson 2001 ) , and consequently it is believed that the vast majority of massive stars do not form alone . it is this combined distance and multiplicity problem that makes it difficult to resolve and study in detail a single massive star in its earliest years of formation . observations that have been extremely helpful in the study of massive stars have been cm - wavelength radio continuum and molecular line emission observations . interferometric arrays can observe with subarcsecond angular resolution the radio continuum emission from the photoionized region in the near - stellar environment around young massive stars . these compact or ultra - compact ( uc ) regions are thought to exist @xmath310@xmath4 years after the formation of a massive star ( de pree , rodrguez , & goss 1995 ) . until recently , this was the earliest observed phase of massive star formation . in order to understand better their formation process , we need to observe massive stars at ages even younger than 10@xmath4 years . this means trying to observe sources before they develop uc regions , which in most of the cases limits our window of continuum observations to a relatively narrow wavelength regime from the mid - infrared ( @xmath33 @xmath0 m ) to the millimeter ( @xmath31 mm ) . observing sources in this `` high mass protostellar object '' ( hmpo ) phase presents a challenge since the best resolutions achievable throughout most of this wavelength range ( namely the far - infrared to the sub - millimeter ) have , until recently , been on the order of a few to tens of arcseconds . however , the bright molecular line emission from these sources would still be observable with high angular resolution radio interferometers . hot cores of compact and dense molecular gas have been known to exist in massive star forming regions for several decades now ( e.g. , wilson , downes , & bieging 1979 ) . some of these `` hot molecular cores '' ( hmcs ) are simply externally heated knots of material , but others may have stars forming at their centers . recently some hmcs have been gaining serious attention as possibly forming extremely young massive stars at their centers ( cesaroni et al . 1994 ; kurtz et al . walmsley ( 1995 ) proposed that these types of hmcs are undergoing an intense accretion phase which prevents the development of a uc region . since these types of hmcs do not possess radio continuum emission and are internally heated by massive protostars , they fit into the category of hmpos . though the compositions of these sources are beginning to be understood in detail through the molecular line studies ( e.g. , rodgers & charnley 2001 ) , continuum observations remain few because of the angular resolution problem . however , continuum observations from the mid - infrared to millimeter have the potential for revealing valuable information about the massive star formation process . models of the spectral energy distribution ( sed ) of hmcs have been developed under the assumption that they have a central massive star undergoing spherical accretion of a free - falling envelope of dust and gas ( osorio , lizano , & dalessio 1999 ) . the success of these models in fitting the observable characteristics of hmcs gave strong support to the hypothesis that some of these objects were true hmpos ( i.e. , not starless hmcs ) in a stage previous to the development of an uc region . however , for that work , only mid - infrared upper limits were available in most of the sources . since the mid - infrared range of the sed is highly sensitive to the geometry of the source ( i.e. , degree of flattening ) , in principle , these kinds of observations would allow testing of more sophisticated models , with a degree of detail similar to that of those that have been developed for low - mass protostars ( e.g. , osorio et al . 2003 ) . recently , observations by de buizer ( 2004 ) and de buizer et al . ( 2003 ) have identified four candidate hmpos that have been observed in the mid - infrared with high - angular resolution . these hmpo candidates lie in areas of high mass star formation , as evidenced by nearby uc regions . in fact in two of these four cases , high spatial resolution observations from 8 m telescopes were needed to simply isolate the mid - infrared emission of the hmpos from the nearby uc regions . this means that , though more sensitive than what is achievable from the ground , data from infrared satellites like _ msx _ ( @xmath35@xmath5 at 8 @xmath6 , @xmath313@xmath5 at 21 @xmath6 ) , and even the _ spitzer space telescope _ ( @xmath32@xmath5 at 8 @xmath6 , @xmath36@xmath5 at 24 @xmath6 ) , are in general not adequate for these studies because of their relatively poor angular resolutions . by being able to isolate emission from the hmpo alone , mid - infrared observations from ground - based 8 to 10m - class telescopes represent a invaluable tool in the study of these earliest stages of massive star formation . in this article we present new sub - arcsecond mid - infrared observations of three of the four candidate hmpos identified by de buizer ( 2004 ) and de buizer et al . the fourth hmpo candidate , g305.20 + 0.21 , will be the subject of a following article . using the gemini south telescope we imaged these three hmpo candidates through up to 11 filters , sampling their seds over the entire 225 @xmath0 m infrared atmospheric window . our motivation for performing these observations was to sample the seds of hmpos with sufficient spectral resolution and coverage to accurately test new sed models . these models use envelopes that take into account deviations from spherical geometry in their inner region due to rotation and at large scales due to natural elongation of the source . by matching these models to the data we attempt to `` back out '' of the models detailed physical parameters for these earliest stages of massive stellar life . in this way , we hope to be able to learn some details about the massive star formation process during these highly obscured stages , as well as the strength and limitations of the non - spherical sed models . observations were carried out at gemini south over the time period between 2003 september and 2004 november . imaging was performed with the thermal - region camera and spectrograph ( t - recs ) . the instrument employs a raytheon 320@xmath7240 pixel si : as ibc array which is optimized for use in the 726 @xmath0 m wavelength range , but can perform with modest sensitivity in the 25 @xmath0 m wavelength range as well . the pixel scale is 0.089@xmath5/pixel , yielding a field of view of 28@xmath88@xmath721@xmath86 . sky and telescope subtraction were achieved through the standard chop - nod technique . images were performed with _ k _ , _ l _ , and _ m _ filters , as well as six narrowband silicate filters encompassing the full 10 @xmath0 m _ n_-band spectral window , and two narrowband filters within the 20 @xmath0 m _ q_-band spectral window . effective central wavelengths for these filters and their spectral range are given in table 1 . for g11.94 - 0.62 and g45.07 + 0.13 we also have images from de buizer et al . ( 2003 ) through the _ q3 _ ( @xmath9=20.81 @xmath0 m , @xmath10=19.9921.64 @xmath0 m ) filter from the nasa infrared telescope facility ( irtf ) using the instrument mirlin . table 1 also lists the on - source integration times through each of the t - recs filters for each hmpo candidate field . in addition , this table lists the standard star used to flux calibrate each image , and the assumed flux density for that standard through the given filter . these assumed standard star flux densities were found by convolving the spectral irradiance templates of the stars from cohen et al . ( 1999 ) with the given t - recs filter transmission profile . flux densities were derived for each filter for the hmpo candidate as well as any other mid - infrared sources on the field . tables 2 - 4 list the flux densities for all sources . these flux densities are quoted with their 1-@xmath11 total error , which is a quadrature addition of the statistical variation from the aperture photometry ( due to the standard deviation of the background array noise ) and the flux calibration error . usually the flux calibration error is found from the variation of the standard star flux throughout the course of the night , however our observations were taken piece - wise on many nights throughout several semesters . flux calibration was therefore achieved by observing a standard star at a similar airmass just before or after the observation of the target science field . therefore , we do not have calibration variability statistics for each filter on each night the data were taken . the calibration factor [ ratio of accepted flux in janskys to analog - to - digital converter units ( adus ) per second per pixel ] derived from the standard star observations through each filter varied little throughout 14 months of data collection , and so we use here the standard deviation of the calibration factor over this 14 month period as the flux calibration error . this can be considered an extremely conservative estimate of the errors , taking into account a wide variety of atmospheric and observing conditions that affect flux calibration . most filters have modest flux calibration errors ( for mid - infrared observations ) with standard deviations between 2 and 10% . flux calibration through certain filters is more difficult due to the presence of various atmospheric absorption lines contaminating the filter bandpass , some of which can be highly variable . those filters most affected are the 7.7 @xmath0 m ( 21% ) , 12.3 @xmath0 m ( 19% ) , 18.3 @xmath0 m ( 15% ) , and 24.6 @xmath0 m ( 23% ) filters . furthermore , the continuum observed through these filters may be affected by the presence of certain spectral features common to young stellar objects . in particular , the 12.3 @xmath0 m continuum flux densities for the targets may be skewed to higher values due to the presence of bright [ neii ] lines at 12.8 @xmath0 m ( see for instance , faison et al . the 18.3 @xmath0 m and 20.8 @xmath0 m observations can also be skewed high or low depending on the presence in either emission or absorption of the 18 @xmath0 m silicate feature . the three sources below were chosen because they are among the only hmpo candidates known to be bright in the mid - infrared ( de buizer 2004 , de buizer et al . 2003 ) and are sufficiently far enough away from nearby regions to be studied in detail without confusion . all three sources are thought to be hmpos primarily because they are sources found in regions of known high mass star formation yet have no uc emission of their own , and all are coincident with molecular maser emission which is thought to trace massive star formation and its processes . each source has further unique evidence indicating that it is a hmpo , which we discuss in detail below . a summary of the observations for each object will also be discussed in this context . of the four sources identified as mid - infrared bright hmpo candidates by de buizer ( 2004 ) and de buizer et al . ( 2003 ) , the status of the source in g11.94 - 0.62:drt03 1 as a genuine hmpo is the most uncertain . however the hmpo candidate in g11.94 - 0.62 satisfies , at the very least , the minimum criteria outlined at the beginning of this section . drt03 1 was found as an unresolved mid - infrared source in the survey of de buizer et al . ( 2003 ) located at the site of a water maser clump offset @xmath310@xmath5 from a cometary uc region . drt03 1 has no detected cm radio continuum emission of its own . however , g11.94 - 0.62 has not been observed with high spatial resolution molecular line imaging . some molecular tracers have been discovered toward the region , like cs ( bronfman et al . 1996 ) and nh@xmath12 ( cesaroni , walmsley , & churchwell 1992 ) , but it is not known if this emission is related to the nearby uc region , the hmpo candidate , or both . while it is encouraging that some molecules have been found in the region , ch@xmath12cn ( a tracer of denser and hotter molecular material and indicator of the presence of hmcs ) has been searched for in the region and was not detected ( watt & mundy 1999 ) . the uc region was also imaged at 2.7 mm by watt & mundy ( 1999 ) , but there was no detection at the site of drt03 1 . de buizer et al . ( 2003 ) derive a lower limit to the bolometric luminosity for drt03 1 , finding that it is at least as bright as a b9 star . though this spectral type is derived from an extreme lower limit to the bolometric luminosity , it does not appear that this source is a massive ( m@xmath1310@xmath14 ) protostar , but more likely an intermediate mass protostellar object . if this is true , given its observational similarities to its higher mass counterparts , it will be interesting to study the similarities and differences in detail to learn more about the interrelation of the intermediate and high mass star formation processes . it would also have interesting consequences for the hypotheses regarding excitation of water masers by lower mass ( non - ionizing ) stars ( forster & caswell 2000 ) . our sensitive gemini mid - infrared observations of the region around g11.94 - 0.62 have revealed extended dust structure , the likes of which are not seen at any other spectral regime ( figure 1 ) . these images at the sub - arcsecond angular resolution of gemini confirm the results of the lower resolution observations of de buizer et al . ( 2003 ) that the mid - infrared morphology of the uc region is not cometary , as is seen in cm radio continuum emission ( wood & churchwell 1989 ) . the hmpo candidate , drt03 1 , is clearly detected at shorter mid - infrared wavelengths but is only a 3-@xmath11 and 4-@xmath11 detection at 20.8 and 24.6 @xmath0 m , respectively . with no information on the rayleigh - jeans side of the sed for the hmpo candidate alone , we will not be able to constrain the models at the longer wavelengths . table 2 lists the observed flux densities for the hmpo candidate as well as for the other mid - infrared sources in the field . it is not clear if each of these mid - infrared sources house a stellar component or if they are simply knots of dust in an extensive region . the hmpo candidate in g29.96@xmath150.02 is sufficiently well - studied to be considered a prototype . the hot molecular component of this source was first discovered in the observations of cesaroni et al . ( 1994 ) . in this work , a hot ( t@xmath16@xmath13 50 k ) core of molecular material was detected , traced by nh@xmath12(4,4 ) line emission near to , but offset @xmath32@xmath5 from a cometary - shaped region . this hot molecular core was found to be coincident with a cluster of water masers . cesaroni et al . ( 1994 ) conjectured that this source , like the other hmcs in their survey , is internally heated by a developing massive protostar . however , as was mentioned in the introduction , some hmcs could be externally heated by the nearby massive stars powering the uc regions . in the case of g29.96 - 0.02 , the millimeter line study of gibb , wyrowski , & mundy ( 2003 ) provided strong evidence that the hmc is internally heated . therefore , the hmc in g29.96 - 0.02 appears to be a genuine hmpo . the hmpo in g29.96 - 0.02 was directly imaged for the first time in the mid - infrared by de buizer et al . ( 2002 ) on the gemini north 8-m telescope . the hmpo lies @xmath33@xmath5 away from the radio continuum peak of the bright region and the emission from the uc region peak can not be resolved from the hmpo on 4-m class telescopes ( watson et al . however , the large aperture of gemini has sufficient resolving power in the mid - infrared ( @xmath30@xmath85 ) to pull in the peak of the dust emission from the region to allow direct observation of the hmpo . even so , the resolved thermal dust emission from the region is still extensive , and therefore we can not avoid observing the hmpo overlaid on a diffuse background of emission ( figure 2 ) . following the technique outlined by de buizer et al . ( 2002 ) , one can fit this background emission and subtract it off , effectively leaving only the mid - infrared emission of the hmpo . this technique was employed for the images obtained in the work presented here . in table 3 we present the flux densities for the hmpo candidates and the uc , as well as the flux density of the hmpo candidate without subtracting off the background due to the extended uc emission . this was done so that it is transparent to the reader how much flux has been subtracted off , and what the absolute upper limits are for the flux densities of the hmpo candidate . the g29.96 - 0.02 hmpo has been studied in great detail in molecular line studies ( e.g. , cesaroni et al . 1998 , wyrowski , schilke , & walmsley 1999 ; maxia et al . 2001 . ) however , other than the mid - infrared observations of de buizer et al . ( 2002 ) and those presented here , only a few continuum observations at other wavelengths exist . this source has been observed at 1 and 3 mm by maxia et al . however , as pointed out by de buizer et al . ( 2002 ) , there appears to be problems with these flux densities since they lead to a negative value ( see figure [ fig6 ] ) for the index of dust opacity , @xmath17 , whereas most models assume 1@xmath18@xmath17@xmath182 for @xmath19@xmath1200 @xmath0 m . another observation at 3 mm was carried out by olmi el al . they did not resolve the hmpo , and therefore there might be possible contamination from the nearby uc region . they try to obtain the flux density of the hmpo at 3 mm , making a subtraction of the free - free contamination estimated from the 2 cm image . this measurement together with the detection of the hmpo at 1.4 mm carried out by wyrowski et al . ( 2005 , personal communication ) indicate that g29.96 - 0.02 hmpo is a strong emitter at millimeter wavelengths . listed as a `` known hot core '' in the review article of kurtz et al . ( 2000 ) , the hmpo in g45.07 + 0.13 gained attention mostly through the work of hunter , phillips , & menten ( 1997 ) . however most of the observations by hunter , phillips , & menten ( 1997 ) are low spatial resolution continuum and molecular line maps observed towards uc regions . in the case of g45.07 + 0.13 they found several molecular tracers , some believed to be in outflow , emanating from the location of the uc region . hunter , phillips , & menten ( 1997 ) contest that there is a single star here , embedded in a molecular core , at a very early stage of development such that it has just begun to develop a uc region . for these reasons it has been labeled a hot core . the observations of de buizer et al . ( 2003 ) showed that there are three mid - infrared sources at this location , all within 8@xmath5 of each other . the brightest mid - infrared source on the field is coincident with the uc region . however , at a location @xmath32@xmath5 north of the uc region there is a group of water masers associated with a mid - infrared source that has no cm radio continuum emission ( figure 3 ) . de buizer et al . ( 2003 ) showed that this northern mid - infrared source , drt03 3 , has a very bright mid - infrared luminosity of @xmath34@xmath710@xmath20 l@xmath21 . assuming that the mid - infrared luminosity is equal to the bolometric luminosity ( a gross underestimate ) , drt03 3 has the equivalent luminosity of a b0 star . we therefore have a source luminous enough to be a massive protostar , situated at the location of a group of water masers , yet this source has no cm radio continuum emission of its own . for these reasons , de buizer et al . ( 2003 ) claim that drt03 3 is the true hmpo in this field . although there are multiple sources of thermal dust emission present here , it does appear from the high spatial resolution cs maps of hunter , phillips , and menten ( 1997 ) that the stellar source powering the uc region is the only viable source of the outflow in the region ( contrary to what was stated by de buizer et al . since no such tracers are seen coming from the hmpo candidate , this perhaps confirms the youth of the object , placing it in a stage of formation before the onset of an outflow . while the resolution of the 800 and 450 @xmath0 m maps of hunter , phillips , & menten ( 1997 ) are too coarse to ascertain which mid - infrared source may be dominating the emission at these longer wavelengths , their higher resolution ( @xmath32@xmath5 ) 3 mm map shows no emission from the location of drt03 3 , though emission is detected towards the other two mid - infrared sources on the field . our new observations of this field with gemini have revealed the presence of four more faint mid - infrared sources present ( figure 3 ) in addition to those already found by de buizer et al . it appears that this may be a small cluster of star formation centered on the o9 star powering the uc region . drt03 3 is seen at all 11 mid - infrared wavelengths , as is the uc region ( drt03 2 ) . table 4 lists the flux densities for all mid - infrared sources on the field . unfortunately , we have no detections of the hmpo candidate alone at any wavelength on the rayleigh - jeans side of the sed , and have to rely upon integrated flux density measurements from the whole region ( su et al . 2004 ) as upper limits to constrain the models . we will calculate the sed of a hmpo by modeling it as an envelope of gas and dust that is freely falling onto a recently formed massive central star , which is responsible for the heating . osorio , lizano & dalessio ( 1999 ) modeled the sed of a sample of such objects by assuming that the envelope has spherical symmetry . for that work , only upper limits were available for the mid - infrared flux densities of most of the sources . because this wavelength range is very sensitive to the geometry of the source , here we extend that work by including rotation and flattening of the envelope in the models , so that the results can be tested with the new mid - infrared data presented in this paper . for this purpose , we have followed procedures similar to those already developed in low - mass star models ( osorio et al . 2003 ) . envelopes with a density distribution such as that given by terebey , shu , & cassen ( 1984 , hereafter tsc envelopes ) assume that the rotation of the infalling material becomes important only in the vicinity of the centrifugal radius , @xmath22 , where @xmath23 is the angular velocity ( assumed to be constant and small ) at a distant reference radius @xmath24 , and @xmath25 is the mass of the central protostar . the tsc envelopes are flattened only in the inner region , being essentially spherical in the outer region . the inner regions ( @xmath26 ) of these envelopes are described by the solution of cassen & moosman ( 1981 ) and ulrich ( 1976 ) ( hereafter referred as cmu ) . the solution is determined by specifying the velocities and densities at the ( distant ) reference radius @xmath24 , where the velocity is nearly equal to the radial free - fall value , with only a small azimuthal component corresponding to a constant angular velocity , and the density is set by the required mass infall rate @xmath27 . with these assumptions , the density of the inner envelope ( @xmath26 ) is given by @xmath28 @xmath29 where @xmath30 and @xmath31 are the position angles with respect to the rotational symmetry axis of the system at radial distances from the central mass @xmath32 and @xmath24 , respectively . at large distances ( @xmath33 ) the motions are radial ( @xmath34 ) and the density tends to the density distribution for radial free - fall at constant mass infall rate . for @xmath35 , the motions become significantly non - radial , and the angular momentum of the infalling material causes it to land on a disk ( @xmath36 ) at distances @xmath37 . therefore , @xmath38 is the largest radius on the equatorial plane that receives the infalling material . a more complex description can be given in terms of intrinsically flattened envelopes , such as those with the density distribution resulting from the gravitational collapse of a sheet initially in hydrostatic equilibrium ( hartmann et al . 1994 ; hartmann , calvet , & boss 1996 ) . these envelopes are flattened not only in their inner region , because of rotation ( as the tsc envelopes ) , but also at large scales due to the natural elongation of the cloud . if the rotational axis is perpendicular to the plane of the self - gravitating layer , the density distribution of the infalling material will be axisymmetric and can be written as : @xmath39 where @xmath40 is the cmu density distribution for the same @xmath41 and @xmath38 , and @xmath42 is a measure of the degree of flattening ( hartmann et al . the degree of flattening in these envelopes is measured by the parameter @xmath43 , where @xmath44 is the outer radius of the envelope and @xmath45 is the scale height . hereafter , we will designate them as @xmath42-envelopes . the scale of this density distribution can be set by using the parameter @xmath46 , defined as @xmath47 where @xmath48 is a reference radius . this reference density corresponds to the density that a spherically symmetric free - falling envelope with the same mass accretion rate , @xmath27 , would have at the reference radius @xmath48 . in our modeling , we will use @xmath49 au , and we will designate the reference density as @xmath50 . the remaining parameters that define the density distribution of the @xmath42-envelopes are the centrifugal radius , @xmath38 , and the inclination of the polar ( rotational ) axis to the line of sight , @xmath51 . since @xmath42-envelopes are flattened not only in their inner region but also at large scales most of their material is accumulated over the equatorial plane , having an extinction lower than a tsc envelope of the same mass provided the line of sight is not oriented too close to the edge - on position . therefore , a flattened @xmath42-envelope model predicts stronger near and mid - infrared emission and less deep absorption features than those predicted by a tsc envelope model of the same mass . given that our sources are strong mid - infrared emitters , we will preferentially use the @xmath42-envelope density distribution to model our sources . the temperature distribution along the envelope , as well as the sed from 2 @xmath0 m to 2 cm , are determined following the procedures described by kenyon , calvet , & hartmann ( 1993 , hereafter kch93 ) , calvet et al . ( 1994 ) , hartmann et al . ( 1996 ) , and osorio et al . ( 2003 ) . for a given luminosity of the central star , @xmath52 , which is assumed to be the only heating source ( i.e. , neglecting other possible sources of heating such as accretion energy from a disk or the envelope ) , the temperature structure in the envelope is calculated from the condition of radiative equilibrium , using the approximation of an angle - averaged density distribution . nevertheless , the emergent flux density is calculated using the exact density distribution . because we want to compare our results with high spectral resolution mid - infrared data , our model spectra wavelengths were finely sampled ( @xmath53log@xmath19 = 0.02 ) in the range 3 - 25 @xmath0 m in order to define with high spectral resolution the 10 @xmath0 m silicate absorption feature . outside this range , wavelengths were sampled at @xmath53log@xmath19 = 0.2 - 0.3 . additionally , in order to make a proper comparison with the observations , we convolved our results with the t - recs filter transmission profile . given that the spectral response of the filters is fairly constant , we do not find large variations ( only @xmath35% ) between the averaged and original spectrum . in our models we use a recently improved dust opacity law , corresponding to a mixture of compounds that was adjusted by matching the observations of the well known low - mass protostar l1551 irs5 ( osorio et al . 2003 ) . this mixture includes graphite , astronomical silicates , troilite and water ice , with the standard grain - size distribution of the interstellar medium , @xmath54 , with a minimum size of 0.005 @xmath0 m and a maximum size of 0.3 @xmath0 m . the details of the dust opacity calculations are described by dalessio , calvet , & hartmann ( 2001 ) . the inner radius of the envelope , @xmath55 , is taken to be the dust destruction radius , that is assumed to occur at a temperature of 1200 k , corresponding to the sublimation temperature of silicates at low densities ( dalessio 1996 ) . the centrifugal radius is assumed to be considerably larger than the dust destruction radius in order to have a significant degree of flattening due to rotation in the inner envelope . the outer radius of the envelope , @xmath44 , is obtained from our 11.7 @xmath0 m images . in summary , once the dust opacity law and the inner and outer radius of the envelope are defined , the free parameters of our model are @xmath42 , @xmath50 , @xmath38 , @xmath51 , and @xmath52 . in the following , we will discuss on the contribution of these parameters to the resulting sed . in order to illustrate the effect of the flattening of the envelope , we have calculated the resulting sed for a tsc envelope and compared it with an @xmath56 envelope of the same mass . the results are shown in figure 4 , where it can be seen that the flattened @xmath56 envelope predicts stronger near and mid - infrared emission , and less deep absorption features than the tsc envelope , both for low ( @xmath57 ) and moderately high ( @xmath58 ) inclination angles . this behavior is a consequence of the flatter distribution of material in the @xmath56 envelope , resulting in an extinction lower than the tsc envelope over a wide range of viewing angles ( except for a nearly edge - on view , @xmath59 , where the behavior would be the opposite ) . thus , in general , the flattened @xmath42-envelopes allow the escape of a larger amount of near and mid - infrared radiation , and consequently they predict stronger flux densities than the tsc envelopes in this wavelength range . since our sources are strong infrared emitters , in our models we will use @xmath42-envelopes with a significant degree of flattening , @xmath42 , of the order of 2 . the observed seds of our sources are well sampled in the near and mid - infrared wavelength range . however , this wavelength coverage in insufficient for the models to determine the value of the bolometric luminosity ( that we take to coincide with the stellar luminosity , @xmath52 ) . the bolometric luminosity is well constrained by the far - infrared data ( where the peak of the sed occurs ) , and to a less extent by the submillimeter or millimeter data . unfortunately , the available far - infrared data for our sources do not have the appropriate angular resolution to avoid contamination from other nearby sources , and should be considered as upper limits , while millimeter data are only available for one of our sources . therefore , in general , we constrain the value of the luminosity by fitting the mid - infrared data closer to the sed peak , that are helpful to set a lower limit for the luminosity . on the other hand , the depth and shape of the silicate absorption feature ( which is very well delineated by our mid - infrared data ) determines the extinction along the line - of - sight , that in turn is related to the inclination angle , @xmath51 , the density scale , @xmath46 , and the centrifugal radius , @xmath38 . the depth of the 10 @xmath0 m silicate absorption increases not only with the density scale of the envelope , but also with the inclination angle , since the extinction is much smaller along the rotational pole ( @xmath60 ) than along the equator ( @xmath59 ) ( see osorio et al . thus , the behavior of the parameters @xmath46 and @xmath51 is somewhat interchangeable in the mid - infrared ( although such behavior is different at millimeter wavelengths ) . for instance , an envelope with a combination of low density and high inclination could have an extinction and mid - infrared emission similar to an envelope with a higher density but viewed pole - on . therefore , for the sources where millimeter data are not available , it is difficult to constrain @xmath51 and @xmath46 simultaneously , and we will search for fits both for high as well as for low inclination angles . the behavior of the sed with the centrifugal radius , @xmath38 , is more complex . increasing the value of @xmath38 will widen the region of evacuated material , producing an overall decrease of the extinction along the line of sight , but , at the same time , it will result in a decrease of the temperature of the envelope , because the material tends to pile up at larger distances from the source of luminosity . thus , depending on the wavelength range , one of the two effects will be the dominant . at short wavelengths , where the dust opacity is higher , it is expected that the lower extinction of the envelopes with large values of @xmath38 will result in an increase of the observed emission . at longer wavelengths , where the opacity is low enough ( except in the silicate absorption feature ) , the dominant effect is expected to be due to the decrease in the temperature , resulting in a decrease of the observed emission for larger values of @xmath38 . this is illustrated in figure 4 , where it is shown that the emission of the envelope with @xmath61 au dominates in the near - infrared range , while the emission of the @xmath62 au envelope dominates at longer wavelengths . as shown in the figure , this result is valid both for low and relatively high inclination angles , although we note that the point where the small @xmath38 envelope becomes dominant shifts slightly to longer wavelengths when increasing the inclination angle ( from @xmath63 m for @xmath57 to @xmath64 m for @xmath58 ) . this shift occurs because envelopes seen at higher inclination angles are more opaque , thus requiring a longer wavelength ( where the opacity is lower ) for the small @xmath38 envelope to become dominant . since we expect the value of @xmath38 to be of the order of the radius of a possible circumstellar disk , in our modeling we will explore a range of values for @xmath38 from tens to hundreds of aus , depending on the luminosity of the source . the upper limit of this range is suggested by the highest angular resolution observations of disks reported towards massive protostars ( e.g. , shepherd , claussen , & kurtz 2001 ) , while the lower limit is set by the smallest values recently reported for protoplanetary disks around low - mass ysos ( rodrguez et al . 2005 ) . from the values of the parameters obtained from our fits to the observed seds we can derive other physical parameters , such as the total mass of the envelope , @xmath65 , obtained by integration of the density distribution , and the mass of the central star , @xmath25 , derived from its luminosity , @xmath52 . once the mass of the star is known , the mass accretion rate in the envelope , @xmath27 , can be derived from the value of @xmath46 . yorke ( 1984 ) and walmsley ( 1995 ) proposed that some objects could be accreting material at a rate high enough to quench the development of an incipient hii region . these objects would lack detectable free - free emission , being good candidates to be the precursors of the uchii regions . in the case of spherical accretion , it is easy to show that the critical mass accretion rate required to reduce the hii region to a small volume close to the stellar surface is @xmath66 , where @xmath67 is the hydrogen mass , and @xmath68 is the recombination coefficient ( excluding captures to the @xmath69 level ) . we will adopt the value , @xmath70 , as a critical measure of the mass accretion rate required to `` choke off '' the possible development of an uc region . since we are assuming departures from spherical symmetry in our envelopes , in reality a more complex calculation would be required to accurately derive the corresponding critical mass rate . however , for simplicity we will assume that if the ratio between the mass accretion rate in our envelope and the critical spherical mass accretion rate @xmath71 , it would indicate that photoionized emission is not expected to be detectable in the source . in this section , we estimate the physical parameters of the three hmpo candidates by fitting our models to the observed seds , using also the available information on size and morphology inferred from the images . given that spherical envelopes do not appear to be able to account for the observed properties of the seds in the mid - infrared , we have modeled the sources using @xmath42-envelopes with a significant degree of elongation ( see discussion in @xmath724.2 ) . since the temperature distribution in the envelopes is estimated using the approximation of an angle - averaged density distribution , valid for moderately elongated envelopes , we did not attempt to consider very high values of @xmath42 , that correspond to extremely elongated envelopes where the temperature calculations probably would not be accurate enough . therefore , we explored values of @xmath42 around 2 , and in all three sources we were able to get good fits for @xmath73 . we note that this value coincides with the one that provided the best fit for the low - mass protostar l1551-irs5 , after exploring a wide range of values of the @xmath42 parameter ( osorio et al . 2003 ) . given the incomplete and non - uniform coverage of the observed sed , the fitting process can not be automated and should be carried out manually on a case - by - case basis . as was already mentioned in the previous section , the luminosity is well constrained by the peak of the sed , that is expected to happen in the far - infrared . unfortunately , at present , far - infrared observations do not reach the angular resolution required to properly isolate the emission of these distant sources from contamination of other sources in their vicinity . therefore , in general , the far - infrared data are considered only as upper limits and we have to use the mid - infrared data to obtain a range of possible values for the luminosity . thus , the general strategy was to first run a set of models with increasing luminosities until we found a value that was able to reproduce the observed mid - infrared flux density near the peak of the sed and consistent with the millimeter and submillimeter data . once an approximate value of the luminosity was found , we proceeded to run a set of models to find the density scale , which is then basically determined by the millimeter and submillimeter data points . when the density scale was fixed , we ran models with different inclinations in order to find the best fit to the absorption feature in the mid - infrared , which is particularly sensitive to the value of the inclination angle . finally , we refined the modeling by testing a range of possible values of the centrifugal radius . if the data point coverage of the sed is good enough we should end this process with essentially one final model ( the `` best fit model '' ) that is able to reproduce the observed data points within the constrains imposed by the observational uncertainties . in the case that only upper limits are available in the millimeter and submillimeter domain , the density scale and inclination angle can not be constrained simultaneously , and in this case we present two fits , one for a low inclination angle , and a second one for a high inclination angle . the value of the centrifugal radius not well constrained in this case either . however , on the basis of an additional analysis of the quality of the data points and the peculiarities of each source ( see discussion of individual sources ) we attempt to favor one of the fits as the best fit model . table [ tbl5 ] lists the values of the parameters of our best fit models : the inner , outer and centrifugal radii of the envelope , the stellar luminosity , the reference density at 1 au , and the inclination angle . our best - fit model seds are shown in figures [ fig5 ] , [ fig6 ] , and [ fig7 ] . table [ tbl6 ] lists other derived parameters for each source : the mass of the envelope , its temperature at a radius of 1000 au , the mass and spectral type of the central protostar , the mass accretion rate , the rate of ionizing photons , and the critical mass accretion rate . these parameters have been derived following the procedures explained in the footnotes to the table . this hmpo has been observed through continuum measurements at millimeter ( watt & mundy 1999 ) , submillimeter ( walsh et al . 2003 ) , and mid - infrared wavelengths . the submillimeter data will be used in our modeling only as upper limits , since they have been obtained with a large beam ( @xmath74-@xmath75 ) and therefore it is not possible to separate the emission of the nearby uc region from that of the hmpo . likewise , the millimeter flux densities will be used also as upper limits , since the analysis of watt & mundy ( 1999 ) suggests that the millimeter emission observed towards this source corresponds to free - free emission of the nearby uc region with a negligible contribution from dust . in general , to constrain the physical parameters of the sources , we use the observed sed as well as the information on size and morphology given by the images . however , the g11.94@xmath150.62 hmpo source appears unresolved in the near and mid - infrared images ( see figure [ fig1 ] ) and therefore , we have only an upper limit for its size . because we considered the far - infrared , submillimeter and millimeter data only as upper limits , the luminosity of this source is constrained essentially by the data at wavelengths near 20 @xmath0 m , that indicate that this is a low luminosity source . we obtain a value of 75 @xmath76 for the luminosity of this object , corresponding to a a0 star of 5 @xmath14 . since we only have upper limits for the flux densities in the submillimeter - millimeter wavelength range , we can not constrain the density and inclination simultaneously . therefore , we searched for fits both at low ( @xmath51=20@xmath77 - 30@xmath77 ) and high inclination angles ( @xmath51=50@xmath77 - 60@xmath77 ) . at low inclination angles , we obtained a quite good fit for @xmath7830@xmath77 , @xmath79 g @xmath80 , and @xmath81 au ( dot - dashed line in figure [ fig5 ] ) , but this model predicts too much emission at 18.3 and 24.6 @xmath0 m . in order to reproduce the observational data in the 18 - 25 @xmath0 m wavelength range , we require models with a high inclination angle . our best fit model is obtained for a high inclination angle , @xmath82 , with @xmath83 2@xmath84 g @xmath80 ( @xmath85 ) , and @xmath81 au ( solid line in figure [ fig5 ] ) . we can obtain reasonable fits for @xmath86 g @xmath80 and @xmath87 au . for example , a model with @xmath88 au is also feasible ( dotted line in figure [ fig5 ] ) . however , we favor the @xmath81 au model ( solid line in figure [ fig5 ] , and tables 5 and 6 ) because it reproduces better the flux densities shortward of 8.7 @xmath0 m and the depth of the 10 @xmath0 m silicate absorption feature . for our favored model , the ratio @xmath89 ( see table 6 ) . therefore , the source could not develop a detectable region . furthermore , since the luminosity obtained in this model is quite low , free - free emission from the hmpo would be hardly detectable , even in the case of a negligible accretion rate . thus , g11.94 - 0.62 is not luminous enough to be considered as a hmpo . most likely , given the low temperature of the envelope ( @xmath90 k ) , it would be better considered as a `` warm core '' that is forming an intermediate - mass star , with an infall rate more than one order of magnitude higher than the typical infall rates of low - mass protostars ( see kch93 ) . the g29.96 - 0.02 hmpo is the only source of our sample that appears to be associated with strong millimeter emission . thus , for this source we have both infrared and millimeter data to constrain better the models . the strong millimeter emission suggests a dense or luminous envelope . in addition , the relatively strong near - infrared emission observed suggests that the envelope should be seen nearly pole - on , since in this direction there is a decrease of the density with respect to the equatorial plane that allows the escape of near - infrared radiation . a reasonable fit ( solid - line in figure [ fig6 ] ) , explaining simultaneously the strong millimeter emission as well as the near - infrared flux densities , is obtained by assuming an extremely dense ( @xmath91 g @xmath80 ) , and therefore very massive envelope ( @xmath92 ) , with a centrifugal radius of @xmath93 au , seen with a low inclination angle ( @xmath94 ) , and heated by a luminous ( @xmath95 ) b1 star ( see tables 5 and 6 ) . this value of the luminosity is a lower limit , since it is the smallest value that is consistent with the mid - infrared and millimeter data points , and we estimate that it can be higher up to a factor of three . likewise , we estimate that the uncertainty in the density scale is @xmath330% , and that the range of plausible values is @xmath96 for the inclination angle , and @xmath97 au for the centrifugal radius . note that the 9.7 @xmath0 m data point is only an upper limit , and therefore it does not set the depth of the silicate absorption feature . also , we want to point out that we have not attempted that our models reproduce the exact values of the observational data points in the millimeter range , due to the problems discussed in [ g29db ] and because errors are not available for some of these data points . our best fit model predicts that the flux density at 3.9 @xmath0 m should be lower than the value observed . we think that the excess of flux density observed at wavelengths shorter than 3.9 @xmath0 m could be due to scattered light in a cavity carved out by the outflow associated with g29.96@xmath150.02 hmpo ( gibb et al . in fact , gibb et al . ( 2004 ) suggest that scattered light in g29.96@xmath150.02 hmpo could be dominant even at 18 @xmath0 m . however , our models show that the emission from 4 - 18 @xmath0 m can be reproduced as dust emission from an infalling flattened envelope , and that scattered light may only be dominant for wavelengths shorter than @xmath34 @xmath0 m . from table 6 we see that the ratio @xmath98 . thus , we do not expect the development of an region in the g29 hmpo . this is consistent with the lack of strong free - free emission at the position of the source . in order to illustrate the relevance of the millimeter constraints , in figure [ fig6 ] we show an additional model ( dotted - line ) , obtained by fitting only the mid - infrared observations . for this model , we obtain a value of the luminosity of only 2400 @xmath76 , a density scale of @xmath99 g @xmath80 and an envelope mass of @xmath3 98 @xmath14 . this set of parameters fits very well the mid - infrared data ; however the model predicts far too little emission at long wavelengths , underestimating by almost one order of magnitude the luminosity and mass . these results illustrate that both millimeter as well as mid - infrared data are required in order to determine reliable values of the derived parameters . it is worthwhile to mention that our models can fit the observed sed with a value of @xmath100 au , considerably smaller than the size of the elongated structures observed towards the g29.96 - 0.02 hmpo ( @xmath101 10,000 au , olmi et al . 2003 ) and some other high - mass protostars ( 1000 - 30000 au , see cesaroni 2005 and references therein ) . these structures have been interpreted by some authors as tracing large - scale disks or rotating toroids . our results suggest that these structures could be naturally explained as infalling flattened envelopes ( with sizes of thousands of aus ) , rather than as large circumstellar disks . the formation of such disks is expected to occur at a scale of the order of @xmath38 , which according to our models could be of the order of hundreds of aus . in order to compare the predicted sed of the g45.07 + 0.13 hmpo with the data , we use the existing millimeter observations ( taken from su et al . 2004 ) only as upper limits , since they do not have an angular resolution high enough to separate the hmpo emission from that of the uc region . therefore , we will restrict our comparison to our mid - infrared observations . the g45.07 + 0.13 hmpo exhibits an anomalous spectrum because in spite of its deep silicate absorption ( indicative of a large extinction ) , it presents a very strong emission in the near - infrared range ( 4 - 8 @xmath0 m ) . for this source , we also have considered models for both high and low inclination angles . a model with a highly inclined envelope ( @xmath102 ) , a density scale of @xmath103 g @xmath80 , and a centrifugal radius of @xmath104270 au ( dotted line in figure [ fig7 ] ) predicts too little emission at 4.7 @xmath0 m and shorter wavelengths , and also presents a silicate absorption feature that is too weak . alternatively , an envelope with a lower inclination ( @xmath105 ) , a larger density scale ( @xmath106 g @xmath80 ) , and a centrifugal radius of @xmath104370 au ( solid line in figure [ fig7 ] and table 5 ) reproduces better the emission at 4.7 @xmath0 m as well as the depth of the silicate absorption , although it still has problems in explaining the emission at the shortest wavelengths ( @xmath63 m ) . both models require the same luminosity ( @xmath107 @xmath76 ) , implying a b1 star of 13 @xmath14 . we favor the low inclination model ( solid line in figure [ fig7 ] and tables 5 and 6 ) because it can explain reasonably well most of values of the observed flux density . as we discussed for the g29.96 - 0.02 hmpo case , the excess of flux density observed at wavelengths shorter than 3.9 @xmath0 m could be due to scattered light . by running other models with slightly different values of the parameters , and taking into account the upper limits set by the submillimeter and millimeter data , we estimate uncertainties of @xmath108 in the inclination angle , a factor of two in the density scale and the centrifugal radius , and a factor of three in the luminosity with respect to the values given in table 5 . for the values given in table 6 , a ratio @xmath109 is derived . thus , as in the previous sources , the mass accretion rate for the g45.07 + 0.13 hmpo is high enough to quench the development of an uc region , that is consistent with the lack of detected free - free emission towards this hmpo . we have obtained mid - infrared data for four hmpo candidates at gemini observatory . these data have the subarcsecond angular resolution necessary to isolate the emission of the hmpo candidates from that of nearby sources . the data are well - sampled across the entire 2 to 25 @xmath6 atmospheric windows , and particularly throughout the 10 @xmath6 absorption feature . since this wavelength range is very sensitive to the source geometry , we have constructed a grid of sed models for the physical conditions ( density , temperature , luminosity ) expected for the early stages of massive star formation , and taking into account for the first time the rotation as well as the natural flattening of the infalling envelopes . these models reach a degree of complexity comparable to those developed for low - mass protostars . using our mid - infrared data , as well as the far - infrared and millimeter constraints available in the literature , we have fit the seds of the hmpo candidates in g11.94@xmath150.62 , g29.96@xmath150.02 , and g45.07 + 0.13 , inferring physical parameters for the infalling envelopes as well as for the central stars . 1 . the hmpo candidate in g11.94 - 0.62 does not seem to be luminous enough to be considered as a hmpo . most likely , it would be better considered as a `` warm '' core that is forming an intermediate - mass star , with an infall rate more than one order of magnitude higher than the typical infall rates of low - mass protostars . 2 . the candidates g29.96@xmath150.02 and g45.07 + 0.13 appear to be genuine hmpos with high luminosities , @xmath110 , and high accretion rates , @xmath111-@xmath112 yr@xmath113 . these values of the mass accretion rate exceed by several orders of magnitude the critical value required to quench the development of an uc region . therefore , these sources appear to be in a very early stage , previous to the development of an uc region . our models are able to fit the observed seds of the hmpos with values of the centrifugal radius of a few hundred aus , which are considerably smaller than the size of the elongated structures observed towards some high - mass protostars ( 1000 - 30000 au ) . these structures have been interpreted by some authors as tracing large - scale disks or rotating toroids . our results suggest that these structures could be naturally explained as infalling flattened envelopes ( with sizes of thousands of aus ) , rather than as large circumstellar disks . the formation of such disks is expected to occur at a scale of the order of @xmath38 , which according to our models could be of the order of hundreds of aus . our detailed modeling suggests that massive star formation can proceed in a way very similar to the formation of low - mass stars . unfortunately , massive protostars are more distant than low - mass protostars , and the instrumentation currently available in the far - infrared and millimeter ranges does not reach the angular resolution necessary to separate the hmpo emission from that of other nearby sources . since these wavelength ranges are very important to constrain properly the physical parameters of the sources , such as the luminosity , in general our fit models are not unique . new data in these wavelength domains would be most valuable to determine the physical conditions in the early stages of massive star formation . based on observations obtained at the gemini observatory , which is operated by the association of universities for research in astronomy , inc . , under a cooperative agreement with the nsf on behalf of the gemini partnership : the national science foundation ( united states ) , the particle physics and astronomy research council ( united kingdom ) , the national research council ( canada ) , conicyt ( chile ) , the australian research council ( australia ) , cnpq ( brazil ) and conicet ( argentina ) . gemini program identification numbers associated with the data presented here are gs-2003b - dd-7 and gs-2004a - q-7 . m. o. acknowledges support from iau peter gruber foundation , junta de andaluca , mcyt ( aya2002 - 00376 , including feder funds ) , and aeci . we thank g. anglada for useful discussions and helpful suggestions . we also thank f. wyrowski for communicating us his 1.4 mm data on g29.96@xmath150.02 before publication . finally , we would also like to thank our referee for his / her thoughtful review of our manuscript and useful comments .

apart from their extreme biological importance , proteins are also very interesting objects from the point of view of statistical mechanics . they possess a very well defined native structure which they are able to find in a short time among a potentially huge number of competing ones , and in spite of many metastable states . how proteins reconcile the stability of the native structure with the requirement that this structure is rapidly reached constitutes the essence of the fascinating and still open protein folding problem @xcite an interesting question is whether the property of folding is a generic property of randomly assembled polypeptidic chains , regardless of their biological function , or is a special property that has evolved through natural selection . this kind of question makes the protein folding problem a bridge between theoretical biology and the statistical mechanics of disordered systems . motivated by this , numerous authors have studied simple models of random heteropolymers @xcite , see @xcite for a review . in the following , we shall discuss only the ` random bond model ' introduced independently by garel and orland @xcite and by shakhnovich and gutin @xcite . more precisely , in our numerical simulations we will study a lattice version of this model . preliminary results of this work have already been presented in @xcite . a `` protein '' with @xmath3 `` amino acids '' is represented as a self - avoiding walk @xcite of @xmath4 steps on the simple cubic lattice . each pair @xmath5 of non - bonded monomers on nearest neighbor lattice sites contributes to the total energy an amount given by an independent and identically distributed ( iid ) gaussian variable @xmath6 with mean @xmath7 and variance @xmath8 . formally , one defines the contact map of configuration @xmath9 , @xmath10 , as the matrix of binary variables @xmath11 , with @xmath12 , such that _ ij()= \ { ll 1 & + 0 & . [ contact ] the energy of the model can then be written as e(,\{b})=_i < j _ ij()b_ij , with @xmath13 . for a given realization of the interaction energies @xmath6 ( representing a protein sequence in the biological analogy ) , the partition sum @xmath14 at temperature @xmath15 can be formally computed as z_n\{b_ij } = _ c e^-e(,\{b})/k_bt , where the sum over configurations @xmath16 runs over all self - avoiding @xmath4-step walks . obviously , the above expression depends only on the variables @xmath17 and @xmath18 , i.e. we have a two - parameter phase diagram in the variables @xmath0 and @xmath19 . the main advantage in using @xmath0 as one of the independent variables instead of @xmath15 or @xmath20 is that we can pass continuously from positive ( repulsive , hydrophilic ) to negative ( hydrophobic ) @xmath0 . as usual with random models , we have to evaluate the quenched average of the free energy . this is a very difficult task , while it is rather easy to perform an annealed average over the disorder . for several models of random spin systems it is well known that such an annealed approximation becomes exact in the high temperature phase , in the thermodynamic limit and at sufficiently large dimension . the same is thought to be true for the present model . it was indeed predicted in @xcite that the annealed approximation becomes exact in 3 dimensions when the chain length tends to infinity . for this to be true it is necessary that the overlap between two randomly chosen replicas with the same realization of disorder vanishes in the limit @xmath21 . numerical tests of this prediction have been made in the past for chains of length @xmath22 , mostly by means of exact enumerations of maximally compact chains of length 27 @xcite . these authors found deviations ( replica overlap is non - zero ) which seemed to decrease with @xmath4 . a similar result even for @xmath23 was found in @xcite , where also monte carlo simulations of very short chains were used ( up to @xmath24 ) . but it is clear that tests with such short chains can hardly be significant . in the present paper we shall present monte carlo simulations for chains of length up to @xmath25 . these simulations are made with the perm algorithm developed recently by one of us @xcite , and applied successfully to a number of different polymer problems @xcite . we study the corrections to the annealed approximation using two different approaches . first , we compute them using the replica method and assuming replica symmetry , which is believed to hold for low disorder . even if a full computation was not possible , the expected behavior was well confirmed by numerical simulations . second , we notice that corrections to the annealed approximation in the weak disorder limit can be related exactly to the average overlap between pairs of homopolymers ( without any disorder ) . we give strong theoretical arguments that this overlap does not vanish in the limit @xmath21 . we also calculate it by means of monte carlo simulations . unlike in the previous case , these simulations do not involve the averaging over the disorder and thus can be applied to larger systems . the two methods agree with each other and show that the corrections to the annealed approximation are small in @xmath26 , but do not vanish in the thermodynamic limit . deviations from the annealed approximation are larger in the coil ( high - temperature ) phase and very small in the collapsed ( globular ) phase . the annealed approximation is presented in sec.2 and compared to results of monte carlo simulations . in order to explain the observed deviations , we study in sec.3 a scenario where the overlap is non - zero but replica symmetry is unbroken . we again compare theoretical predictions with simulation results . the relationship between the weak disorder limit and homopolymer overlap is discussed in sec.4 . additional thermodynamic considerations are presented in sec.5 , and our final conclusions are drawn in sec.6 . the perm algorithm used for the simulations is discussed in an appendix . in thermodynamic systems with quenched disorder we have to consider the average of the free energy per monomer over individual realizations of disorder @xmath27 which formally is given by f_n(b , ) = -1n -1n _ i < j db_ij e^-(b_ij - b)^2/2 ^ 2 z_n\{b_ij } . as for most random systems , this can not be evaluated in closed form . much easier to evaluate is the annealed approximation f_n , ann(b , ) = -1n [ annealed ] obtained by taking the disorder average before taking the log . here the gaussian integrals can be done explicitly , with the result = _ e^-(b-12 ^ 2)_i < j _ ij ( ) . since this is the partition sum for a homopolymer with pair energy = b-12 ^ 2 , [ ann2 ] we see that @xcite f_n , ann(b , ) = f_n(,0 ) . [ f - annealed ] therefore , all thermodynamic variables can be expressed in the annealed approximation in terms of an equivalent homopolymer with shifted interaction strength . this relationship is easiest for those observables whose definition does not involve a derivative with respect to temperature , such as the gyration and end - to - end radii , and the density of non - bonded nearest neighbor contacts @xmath28 . the latter is defined as the average number of nn . contacts between non - consecutive monomers divided by @xmath4 . for these observables , we have r_n , ann(b , ) = r_n(,0 ) [ r - ann ] and c_ann(b , ) = c(,0 ) , [ c - ann ] precisely as in eq.([f - annealed ] ) . for energy @xmath29 and entropy @xmath30 the relations are less simple , since these involve derivatives of the free energy with respect to @xmath15 which are changed into derivatives with respect to @xmath0 and @xmath19 by our convention of using @xmath31 . for the energy per monomer it holds u_n , ann(b , ) = b- ^2 u_n(,0)= (b- ^2 ) c , [ u - anneal ] where we used the fact that the energy for homopolymers is @xmath32 . for the specific entropy @xmath33 we use @xmath34 together with eq.([f - annealed ] ) , and obtain s_n , ann(b,)=s_n(,0 ) - ^22 u_n(,0 ) . [ s - anneal ] the number of configurations with fixed @xmath28 should increase as @xmath35 for large @xmath4 , i.e. @xmath36 is the entropy density in the fixed-@xmath4 , fixed-@xmath28 ensemble . for homopolymers , the ensemble with fixed @xmath37 becomes equivalent to the fixed-@xmath28 ensemble in the limit @xmath21 , thus @xmath28 becomes a non - fluctuating function of @xmath37 , @xmath38 , and the above formula becomes simply s_n , ann(b,)= f()-^22c n , where @xmath39 is the solution of the saddle point equation f ( ) |_c = c = . [ fprim ] the condition for thermodynamic stability is that the second derivative of @xmath40 should be negative , corresponding to @xmath41 being minimal . this is equivalent to requiring that the specific heat is positive . in fact , the specific heat for a homopolymer is given by c_v = bct =- b^2(^2fc^2)^-1 , [ cv ] which has been obtained by deriving both sides of eq.([fprim ] ) with respect to @xmath15 . homopolymers with attraction between unbonded nearest neighbors show a collapse ( `` theta '' ) transition where the specific heat diverges . thus we expect that the second derivative @xmath42 vanishes at the theta - point @xmath43 ( the precise value of the transition point depends on the lattice considered ) . the annealed approximation is supposed to be valid both above and below the theta transition . at very low temperatures and very large disorder , it has to break down since otherwise the entropy would become negative , according to eq.([s - anneal ] ) . this signals another phase transition , the so - called freezing transition . we shall not discuss this regime in this paper , but will treat it in a forthcoming publication . since the theta point is a tricritical point @xcite , its upper critical dimension is d=3 . therefore , we expect that in three dimensions the `` swelling factor '' is constant , r^2/ n const [ r - theta ] at the theta point , up to logarithmic corrections @xcite . here , @xmath44 is any measure of the size of the polymer , such as the end - to - end distance or the gyration radius . we expect that this is still true for heteropolymers , as long as we are not yet in the frozen regime . while eq.([r - theta ] ) gives the most precise numerical estimate of the collapse transition ( with @xmath45 @xcite ) , estimates with similar precision can be obtained from the convexity of the free energy @xcite , and the volume dependence of the free energy in case of periodic boundary conditions @xcite . the collapse line in the @xmath46-plot obtained from simulating chains of length @xmath47 is shown in fig.1 . here the solid line is the annealed approximation . we see that on this scale the annealed approximation seems perfect . but this is not quite true . much more precise tests can be performed by comparing directly both sides of eqs.([f - annealed ] ) to ( [ u - anneal ] ) . typical plots obtained in this way are shown in figs.[fig : u_ann ] . for each of these plots we used at least 300 realizations of disorder , and we got at least @xmath48 independent configurations for each disorder realization . similar plots were made also for several other values of @xmath0 and @xmath19 . in all these plots we see small but significant deviations . these deviations are present both in the collapsed ( globular ) and in the open ( coil ) phase . they depend only weekly on @xmath4 . therefore , even with our long chains and high statistics it is not clear whether they disappear for @xmath21 . obviously , in order to proceed we need more refined theoretical predictions to compare with , and/or a more efficient way to do the disorder average . before we do this , we should point out that deviations from the annealed approximation were also found recently in a different model by trovato _ to go beyond the annealed approximation , we will use the replica trick = _ n0 - 1n . [ rep ] alternatively , we could try a taylor expansion = - 12 ( /^2 -1 ) + . [ taylor ] this expansion is most likely divergent . it is nevertheless useful since its first term gives already a good indication of the leading corrections . also , it suggests the inequality f_n(b , ) f_n , ann(b , ) which can easily be derived exactly from convexity of the logarithm . the same inequality is expected to hold for @xmath49 . this inequality is equivalent to the existence of a negative correlation between the average energy and the partition function . to use eq.([rep ] ) , we first have to evaluate disorder averages of @xmath50 for integer @xmath51 . these are performed similarly to the average over @xmath14 , except that the gaussian integrals give formally rise to interactions between replicas @xcite , = _ c_1_n ( -_=1^n(_i < j_ij^)+ ^22_(_i < j_ij^_ij^ ) ) . [ zn ] here the greek indeces @xmath52 and @xmath53 refer to the different replicas , @xmath54 is a configuration of replica @xmath52 , @xmath55 is its contact map and @xmath56 is given by equation ( [ ann2 ] ) . the annealed approximation is equivalent to neglecting the two - replicas term . to proceed , we define the variables c_=1n_i < j_ij^ , q_=1n_i < j_ij^_ij^ , [ cq ] which are respectively the density of contacts for the contact map @xmath52 and the overlap between two contact maps @xmath52 and @xmath53 . the overlap is a measure of similarity , and it is equal to one if and only if the two contact maps coincide . furthermore , we assume that , for large @xmath4 , the number of configuration @xmath57-tuples with @xmath58 contacts and mutual overlaps @xmath59 grows as [n(_=1^n f(c_)-_k=2^n _ k(\{c_},\{q _ } ) ) ] . in other words , @xmath60 is the entropy loss per monomer when we impose that the replica @xmath61 with density of contacts @xmath62 has overlaps @xmath63 with the @xmath64 previous replicas . this quantity can be measured , for instance for @xmath65 . we can then write & & d\{c_}d\{q _ } + & & \ { n[_=1^n(f(c_)-c_)+^22 ( _ q_- _ k=2^n_k(\{c^},\{q _ } ) ) ] } + & & [-nnf_n(\{c^},\{q _ } ) ] . [ exprep]here , @xmath66 is the free energy per monomer in a system with @xmath57 replicas . to evaluate it , we approximate the integrals over @xmath62 and @xmath67 by their saddle points . we assume replica symmetry which is expected to hold for low disorder : the saddle point is assumed to be given by @xmath68 for all @xmath52 and @xmath69 for all pairs @xmath70 . now , in order to obtain the correct free energy , we have to take the limit @xmath71 . we obtain f(b,)=-f(c)+c+12 ^ 2 cq-(c , q ) . [ f - saddle ] where @xmath72 is the average entropy gain per replica due to the condition that the overlap among all replicas is equal to @xmath73 . note that this quantity is positive because , in the limit of vanishing @xmath57 , the number of terms in the sum is -1 . finally , we have to compute the values of @xmath28 and @xmath73 at which @xmath41 is evaluated by imposing two saddle point conditions : & & f(c)c + ( c , q)c = + 12 ^ 2q [ eq_c ] , + & & 12 ^ 2c = ( c , q)q . [ eq_q ] for @xmath74 ( homopolymer ) eq.([eq_q ] ) just means that the value of the overlap is the most probable one for a given @xmath28 , @xmath75 . because of the normalization , it must be @xmath76 , thus the free energy of the homopolymer is just a special case of eq.([f - saddle ] ) . moreover , since @xmath76 is an absolute minimum , also the derivative @xmath77 must vanish at that point , thus the saddle point equation for @xmath28 valid for the homopolymer is recovered for @xmath74 . it also follows from this argument that the second derivatives of @xmath78 at the point @xmath79 must be non - negative . notice that the free energy has to be maximized as a function of @xmath73 because this variable refers to a space with a negative number of dimensions in the limit @xmath71 . we thus get a first condition of thermodynamic stability @xmath80 , which , from the above consideration , is expected to be fulfilled for @xmath19 small enough . the situation is more complicated for the variable @xmath28 . it enters both into the free energy of the replica interactions which has to be maximized for @xmath71 , and into the free energy of the homopolymer which has to be minimized , at least for @xmath74 . we conjecture that the corresponding condition of thermodynamic stability is that the hessian determinant of the free energy respect to the variables @xmath28 and @xmath73 , @xmath81 , be non positive : h(c , q)=(^2 fc^2 + ^2c^2 ) ^2q^2- (^2cq-1cq)^2 0 . [ denom ] for @xmath74 we have @xmath82 as for homopolymers . in fact , at that point the first derivatives of @xmath78 vanish . the hessian determinant vanishes also , because @xmath78 stays constant at the value zero along the line @xmath83 . thus both conditions of thermodynamic stability are fulfilled at @xmath19 small enough . the energy and entropy per monomer are obtained in the same way as in the annealed approximation . we find & & u_n(b , ) = (b-^2(1-q))c , [ eq_ene ] + & & s_n(b,)= f(c)+(c , q)-^22c(1-q ) . [ eq - ent ] although this is obtained from the saddle point method which is exact only for @xmath21 , we can use eq.([eq_ene ] ) to obtain effective values of @xmath73 also for finite @xmath4 . results are shown in fig.[fig : q]a . it appears that @xmath73 decreases with system size , but its asymptotic value seems to be finite in the random coil phase . this was confirmed by similar measurements at different values of @xmath84 . thus the annealed approximation does _ not _ hold in the random coil phase . the situation is more difficult for collapsed chains . in this case it can not be excluded from fig.[fig : q]b that the overlap asymptotically vanishes , and thus the annealed approximation becomes exact in the thermodynamic limit . however , simulations of systems of even larger size that will be presented in the next section show that this is not the case and that the corrections to the annealed approximation remain finite in the thermodynamic limit in both the coil and the collapsed phases , although in the latter phase they are rather small . as seen from fig.[fig : q]b , @xmath73 does not depend very much on @xmath19 in the collapsed phase and for large systems . to obtain numerical estimates of @xmath85 , we subtract from eq.([f - saddle ] ) the analogous equation for homopolymers , and obtain f(b,)-f(,0 ) = ( c-)- [ f(c)-f()]+12 ^ 2cq -(c , q ) . expanding the difference @xmath86 in the difference @xmath87 and using @xmath88 , we get f(b,)-f(,0 ) = -12(c-)^2f()+12 ^ 2cq -(c , q ) + o((c-)^2 ) . [ df ] this can be either evaluated directly , neglecting the last term . alternatively , we can eliminate the term involving @xmath89 by subtracting from eq.([eq_c ] ) the analogous equation for homopolymers , which gives ( c-)f ( ) = 12 ^ 2q-(c , q)c + o((c-)^2 ) . [ df ] combining this with eq.([df ] ) and neglecting terms @xmath90 , we obtain finally f(b,)-f(,0 ) = c+c4 ^ 2q -12[(c , q)+(,q ) ] + o((c-)^2 ) . [ eq - chi ] in fig.[fig : chi ] we show numerical values for @xmath91/2 $ ] obtained in this way . we see that @xmath85 is very small but definitely not zero . again we see that corrections to the annealed approximations are larger in the swollen phase than in the collapsed phase . indeed , this time it seems that the deviations have a finite limit for @xmath21 in both phases . this conclusion is supported by measurements at other values of @xmath19 ( not shown here ) . basically the same conclusions are also drawn from eq.([df ] ) , showing that @xmath78 depends weakly on @xmath28 and that eq.([df ] ) is very well satisfied . in this section we will discuss the overlap of contact matrices for homopolymer chains and their relation to corrections to the annealed approximation in the weak disorder limit . consider the derivative of the free energy of a random heteropolymer with respect to @xmath1 , at @xmath74 . using eqs.([taylor],[zn ] ) and the results of sec.2 , we obtain _ = 0 = n^-1 _ = 0 = -12f_n(b,0)b + 12n _ _ 1,_2_i < j ^1_ij^2_ij . thus we have _ = 0 = 12 cq . [ pair ] this could have been obtained of course also within the replica symmetric approach , but the above derivation shows that it is indeed a rigorous result involving neither approximations nor unjustified assumptions . notice that this can not be generalized to @xmath92 , but in principle straightforward generalizations could be used to compute all higher derivatives _ = 0 , k = 1,2,3, . numerically , the rhs . of eq.([pair ] ) can be estimated by simulating pairs of chains simultaneously . for this we used a variant of the perm algorithm where we add monomers alternatively to the first and to the second chain @xcite . in this way we guarantee that both chains have exactly the same length ( after having added an even number of monomers ) , and it is straightforward to estimate their overlap . results from such simulations with chains of length up to 1400 are shown in fig.[fig : q0 ] . these data agree nicely with extrapolations of the overlaps for @xmath93 shown in the last section . they have much smaller statistical errors , since we do not have to average over any disorder explicitly . this makes the present method much faster and allows us to study larger systems . the curve for @xmath94 in fig.[fig : q0 ] shows clearly that the annealed approximation does not become exact for @xmath21 in the open coil phase . the same is true ( although a bit less clear ) exactly at the @xmath95 point , as indicated by the curve for @xmath96 . for the collapsed phase , the evidence is not so clear . the curves for @xmath97 and for @xmath98 both are much lower for large @xmath4 and continue to decrease . superficially , one might therefore conclude that the overlap disappears for @xmath21 . but both these curves show a distinct upward curvature for the largest values of @xmath4 , indicating that the decrease will level off and @xmath99 tends to a finite constant for @xmath21 . to sustain this view , we show in fig.[fig : cq ] the plots of @xmath100 as a function of chain length @xmath4 . in this case it is evident that the curves are going to a non - zero value . since the fraction of contacts is limited ( it holds @xmath101 on the cubic lattice ) , also @xmath102 should go to a non - zero value , and the decrease observed in fig.[fig : q0 ] is just a consequence of the fact that the average fraction of contacts is increasing , approaching its stationary value . this conclusion is not , after all , very surprising . it can be backed by an argument which could presumably , with some effort , be made rigorous . the density of contacts in a self - avoiding walk ( saw , corresponding to @xmath103 ) is dominated by short range contacts , i.e. by contacts @xmath5 with small @xmath104 . the contribution of such a loop with fixed @xmath105 and @xmath106 depends weakly on the configuration of the chain far away from this monomer pair . thus , if such a loop is present in one replica , it has a high chance to be present also in the other replica , even if the global structures of both replicas are entirely different . notice that this argument depends crucially on the fact that it is the overlap between _ contact matrices _ which is of relevance , not the geometric overlap between configurations . it seems that the authors of @xcite have missed this in an otherwise very similar argument which has led them to the opposite conclusion that the annealed approximation does become exact . in collapsed chains there are relatively more long range contacts , which explains why the annealed approximation is better but still not exact in that regime . for random heteropolymers with strong disorder , and even more so for proteins , this argument suggests an increased overlap because of secondary structure . for instance , an alpha helix produces an array of contacts @xmath107 , @xmath108 , where @xmath105 labels the position of the amino acid along the protein chain . this enhances the overlap . moreover , there is a finite probability that such contacts appear simultaneously even in the structures of two unrelated proteins . thus the average overlap even in a large set of unrelated protein structure appears to attain a finite limit when the length of the chains increases @xcite . the two saddle point equations for @xmath28 and @xmath73 can not be explicitly solved , because we lack an explicit expression for the functions @xmath36 and @xmath78 . nevertheless , their qualitative behavior can be studied in more detail . this is done in the present section . deriving both equations ( [ eq_c],[eq_q ] ) with respect to the thermodynamic parameters @xmath37 and @xmath19 we can compute the derivatives of @xmath28 and @xmath73 as ( c ) _ = ^2q^2 h(c , q ) 0 , & & ( c ) _ = -h(c , q ) q ( c c -q q ) . [ der ] + (q ) _ = -^2cq-1cq h(c , q ) , & & (q ) _ = h(c , q ) [ c (^2 fc^2 + ^2c^2)- q (^2cq-1cq ) ] . where @xmath81 is given by eq.([denom ] ) . from these , the specific heat can be computed as c_v & = & u / t + & = & (b-^2(1-q))[(b-^2)c+ ] -^2c[(1-q)+(b-^2)q+ ] + & = & -1h(c , q)[^2fc^2(^2c)^2- 2 ^ 2fcq(^2c)(b-^2(1-q))+ ^2fq^2(b-^2(1-q))^2 ] + & & + ^2c(1-q ) , where @xmath109 is the free energy evaluated at the saddle point . the three terms in the square brackets are the quadratic form whose determinant is expressed by @xmath81 . since @xmath110 is positive , they would give a negative contribution to the specific heat if @xmath81 were positive . thus it is justified to require that @xmath81 is negative in order to get thermodynamic stability . the behavior of the thermodynamic derivaties ( [ der ] ) can be partly studied using the fact that the function @xmath78 attains its absolute minimum value @xmath111 along the line @xmath83 . thus , assuming that it is an analytic function of @xmath73 for @xmath112 , it can be expressed in the form ( c , q)=_k=2^(q - q_0(c))^k _ k=2^(q - q_0(c))^k , [ chi_approx ] with @xmath113 . the typical overlap @xmath75 is a small quantity and it is a decreasing function of @xmath28 , or @xmath114 ( the prime indicates derivative with respect to @xmath28 ) . the coefficients @xmath115 are expected to be quantities of order @xmath116 , as it will be argued later . we also introduce the notation @xmath117 , @xmath118 , @xmath119 . we can now develop the saddle point equations for @xmath28 close to the solution @xmath120 of the annealed approximation given by @xmath88 : & & ( ^2 fc^2+^22 ^ 2 q_0c^2 ) _ c = c (c - c)+ (^22q_0c ) _ c = c+ (_k=2^ (q - q_0(c))^k)_c = c=0 + & & _ k=1^(q - q_0(c))^k=12 ^ 2 . from these expressions one sees that both @xmath121 and @xmath87 are quantities of order @xmath122 , thus corrections to the annealed approximation are small but finite for finite @xmath19 . moreover , @xmath123 is positive for small @xmath19 . from eq.([chi_approx ] ) we find , to zero - th order in @xmath124 : (c ) _ 0 & & (c ) _ + (q ) _ 0 & & (q ) _ [c ^2 fc^2+a_2(c)q_0cq_0c ] [ chi(cq ) ] @xmath81 must be computed at the first order in @xmath124 , because the zero - th order term vanishes at the theta point @xmath43 at which @xmath125 vanishes . the result reads : h(c , q)- (q - q_0(c))c(a_2(c))^2 ^ 2 q_0c^2 . [ hh ] to proceed further , we first consider the simple case where the entropy @xmath78 depends only on the product @xmath117 representing the number of conditions that we have to impose in order to fix an overlap @xmath73 : @xmath126 . this form was assumed in the work of shakhnovich and gutin @xcite . as it is easy to see , under the above hypothesis @xmath127 coincides with the value predicted by the annealed approximation and it is independent of @xmath19 at fixed @xmath37 . thus , in this case the annealed approximation would yield the correct value for the fraction of contacts . however , our numerical results contradict this prediction . the overlap @xmath73 is in this case an increasing function of @xmath19 , and its derivative with respect to @xmath37 is proportional to @xmath128 , which is expected to diverge at the theta point @xmath43 . this fact can explain why the values of @xmath73 are much smaller in the collapsed phase than in the coil phase . a better theory for the function @xmath78 was developed by plotkin , wang and wolynes @xcite . they computed the entropy loss for two replicas with density of contacts @xmath28 being at overlap @xmath73 , @xmath129 , in the mean field approximation and in the collapsed phase . although this can be different from @xmath78 , it is a good point for understanding its qualitative behavior . unfortunately , we can not use the formula obtained in @xcite because of several reasons : first , the result is valid for finite @xmath4 and becomes pathologic as @xmath21 ; second , it is not given in the form of a differentiable function ; and third , it is assumed that the most probable overlap @xmath75 is zero in the thermodynamic limit , while our calculations show that this is not the case . we shall however use the fact that @xmath78 is the sum of three contributions : the entropy loss due to imposing that @xmath130 contacts have to coincide , the loss due the fact that @xmath131 contacts have to be different , and a combinatoric factor counting the number of different choices of @xmath130 contacts among @xmath132 . the last term can be approximated by @xmath133 $ ] , even if this is an overestimate , since not all of the combinations of different common contacts can be realized . putting everything together we have : ( c , q)=(c , cq)+c[qq+(1-q)(1-q ) ] [ plot ] . in the computation by plotkin _ _ the mixed second derivative of @xmath134 with respect to @xmath117 and @xmath28 vanishes . this simplifies considerably formulas , and it will be assumed to hold for the rest of the paper . we shall thus introduce the notation @xmath135 to denote the derivative of @xmath136 with respect to @xmath117 at fixed @xmath28 . comparing eq.([plot ] ) to eq.([chi_approx ] ) , we see that @xmath137 must be positive and that @xmath138 . we also see that @xmath139 is likely to be a quantity of order @xmath140 , as it has been assumed above . for the higher order coefficients one finds @xmath141 , as anticipated . we have now to compute the derivatives of @xmath118 : = -^2/cq^2/q^2= q_01+cq_0^(cq_0)(1-q_0 ) 0 , [ q ] in qualitative agreement with fig.[fig : cq ] . inserting the above result in the formulas ( [ chi(cq ) ] ) we see that the density of contacts decreases with @xmath19 at fixed @xmath37 . this behavior is confirmed by our numerical results ( see fig.[fig : c_delta ] ) , which also show that the decrease is maximal for @xmath142 , as expected from the fact that @xmath143 vanishes at @xmath43 . the overlap @xmath73 increases with @xmath37 at fixed @xmath19,as expected from eq.([chi(cq ) ] ) ( see fig.[fig : delta ] ) , and increases with @xmath19 at fixed @xmath37 , as expected from eq.([q ] ) ( see fig.[fig : delta ] again ) . it can thus be understood why the overlap decreases with system size : as the number @xmath4 of monomers increases the importance of surface effects is reduced ( as @xmath144 ) and @xmath145 increases , thus decreasing the value of @xmath73 . we now examine the condition of thermodynamic stability , @xmath146 . as it was already observed , since at the point @xmath147 both the gradient of @xmath78 and its hessian determinant vanish , we have @xmath148 . at the theta point this quantity vanishes , and @xmath149 is given by the deviations from the annealed approximation . three situations are possible : first , @xmath149 can be positive at the leading order in @xmath150 . in this case the thermodynamic stability would be violated around the theta point , but our simulations do not show anything strange in this region . second , the leading order in @xmath124 can be negative . in this case , the specific heat would not diverge anymore at the theta point for finite @xmath19 , but it would show a peak proportional to some negative power of @xmath124 . thus the disorder would smear out the thermodynamic singularity at @xmath43 , leaving unchanged the geometric characterization of the collapsed chains in terms of the gyration radius . it is rather difficult , if not impossible , to test this scenario by means of simulations . third , @xmath149 can vanish identically at @xmath43 . it is easy to see that this condition , combined with the assumption that @xmath78 is of the form ( [ plot ] ) , is fulfilled if and only if @xmath151 is of the form ^(cq)=(1-cq / bcq / a ) , [ chi_pr ] where @xmath152 , @xmath153 and @xmath154 are two constants , and @xmath28 is not too small so that the last inequality can be fulfilled . in this case , one finds @xmath155 $ ] , and it is easy to check that all previous results are recovered , while @xmath156 vanishes for all @xmath73 and @xmath28 , including the theta point . summarizing the discussion , we find that , if @xmath78 is of the form ( [ plot ] ) , two possibilities are open : either @xmath151 is given by eq.([chi_pr ] ) , in which case @xmath157 , or @xmath151 has a different form , in which case the specific heat is not anymore divergent at the theta point @xmath43 . unfortunately , we are not able to decide among these alternatives . we conclude this section showing in fig.[fig : chi27 ] , for the sake of completeness , the entropy loss @xmath129 obtained from simulations of homopolymer chains with @xmath158 . although the system size considered is definitely too small for quantitative considerations , the qualitative behavior is interesting and confirms the above considerations . the entropy @xmath129 was computed from _ 2(c , q)=-1np_2(c , q ) , where @xmath159 is the probability density for a pair of chains , both with density of contacts @xmath28 , being at overlap @xmath73 . we have shown that the annealed approximation is very good but not exact for a particular model of random heteropolymers , and we have given simple physical arguments for it . we have also computed the thermodynamics of the model using the replica symmetric approximation , and we have shown that such approach can explain very well , at least qualitatively , the observed deviations from the annealed approximation in the high temperature phase . the replica symmetric calculation also leaves open , surprisingly , the possibility that the disorder could cancel the thermodynamic singularity at the theta point . a numerical test of this possibility is very difficult , and it has been left out . in our present study we have not addressed the most interesting aspect of the model the freezing of the system in a finite number of mesoscopic states . this transition should represent some features of the folding transition taking place for protein structures . instead , we have studied the model at higher temperatures and at smaller disorder . this should however be of interest also in the context of the freezing , since it was conjectured @xcite that freezing can be described in this model by the random energy model , a prerequisite for which is that the annealed approximation is exact . in the present simulations we have studied chains of length up to @xmath25 . deviations from the annealed approximation decrease fast for small @xmath4 , which explains why studying very short chains has mislead several authors to the conclusion that these deviations vanish for @xmath21 . but in the high - t ( open coil ) phase this decrease clearly stops , and deviations are roughly independent of @xmath4 for @xmath160 . this is less clear in the collapsed phase . but also there numerics , general arguments , and detailed calculations within a specific scenario with unbroken replica symmetry all indicate that these deviations will settle at a non - zero value for large @xmath4 . this casts doubts on the validity of the random energy picture for protein folding . there are of course a number of questions which are left open by the present study . first of all , we have deliberately left out all questions related to freezing . secondly , our treatment of sec.3 assumed that the overlap distribution is always dominated by a single peak . this is most likely not true in the frozen regime , and the distribution of overlaps is certainly a most interesting object . we shall address these questions in a forthcoming paper @xcite . finally , we have studied only one particular model , where contact energies are independent gaussian variables . several other models of random heteropolymers are studied in the recent literature @xcite , and several of them present very intersting open problems . we are indebted to many colleges for useful discussions , in particular to h. frauenkron , w. nadler , h. orland , e. shakhnovich , a. trovato and m. vendruscolo . this method which was first described in detail in @xcite is a chain growth method . it is based on rosenbluth s idea of biased sampling @xcite , but it deviates from it by deleting ( ` pruning ' ) configurations with too low weight , and copying configurations with too large weight ( ` enrichment ' ) . we remind the reader that the bias in the rosenbluth method requires each configuration to carry a non - trivial weight which builds up gradually as monomer by monomer is added . in addition to this ` rosenbluth factor ' , there is also a boltzmann factor in the case of interacting polymers . the weights which control pruning / enrichment are the product of the two . both pruning and enrichment are done while the chains grow , i.e. based on the actual ( incomplete ) weight factors . in some cases it is advantageous to use not the present weights but ( implicit ) estimates of future weights to control the algorithm @xcite , but this is not done in the present paper . instead , we use some of the special tricks used for strongly collapsed systems in @xcite . when the weight is too low , configurations are not simply killed ( this would imply systematic errors ) , but instead they are killed with probability 1/2 , and those which are not killed get their weights doubled . similarly , in the case of cloning the weight is spread uniformly among all clones . technically , cloning is performed by means of recursive subroutine calls . a pseudocode of the basic algorithm is given in the appendix of @xcite . the most important shortcoming of the rosenbluth method is that the distribution of weights can become extremely wide for large systems and at low temperatures . the only exception is for interacting homopolymers on the simple cubic lattice at the theta point , where rosenbluth and boltzmann factors nearly cancel . in less favorable cases even a very large statistical sample can be dominated by just a handful of high weight events , and statistical errors grow out of bounds . even worse , in extreme cases the events which would carry ( in average ! ) most of the weight are so rare that they are missed completely with high probability , and the free energy is underestimated systematically . pruning and enrichment guarantee that the weights of individual configurations stay within narrow bounds , and the above can not happen . but in very difficult situations ( large @xmath4 , low t , large disorder ) it may happen that due to cloning the configurations are strongly correlated , and the weights of clusters of such correlated configurations play essentially a similar role as the weights of individual configurations in the above discussion . in the following we will call such a cluster which is composed of all configurations having a common root a ` tour ' . in order to check that the weights of tours do not become too uneven , we have measured in their distribution . let us call the weights @xmath161 , and the distribution @xmath162 . we are on safe grounds if this distribution is so narrow that @xmath162 and @xmath163 have basically the same support . in particular , we should require that the maximum of @xmath163 occurs at such values where @xmath162 is still appreciable , and the distribution is well sampled . we have verified that this is the case for all data shown in the present paper . notice , however , that this is a very stringent requirement . if it is were not satisfied , this would not necessarily mean that the data are wrong , since configurations within one tour are only partially correlated . the generalization of this algorithm to pairs of chains growing simultaneously is straightforward @xcite . one just has to add monomers alternatingly . when the total number of monomers in both chains is even , the addition of the next monomer is attempted at chain 1 ; when this number is odd , the next monomer is added to chain 2 .

in the following we want to give a new explanation of the area law of black hole ( bh ) entropy and the more general and stronger holographic principle . furthermore , we provide ( in our view ) convincing arguments that an important structural ingredient of the deep structure of our quantum vacuum is a network of microscopic wormholes . in contrast to e.g. string theory and loop quantum gravity ( lqg ) , which both employ the quantum laws more or less unaltered all the way down to the remote planck scale , we regard this as an at least debatable assumption . we rather view the holographic hypothesis as a means to understand how both quantum theory and gravitation do emerge as derived and secondary theories from a more fundamental theory living on a more microscopic scale . a central role in this enterprise is played by an analysis of the microscopic structure of the quantum vacuum which leads to the key concept of wormhole spaces . this important conceptual structure makes it possible to understand the holographic aspects of quantum gravity , on the one hand , and the ( non - local ) entanglement phenomena pervading ordinary quantum physics , on the other hand , in a relatively natural way . furthermore we think that there exist links to the old ideas of e.g. sakharov and zeldovich , dubbed induced gravity ( see for example @xcite,@xcite,@xcite,@xcite ) . some words are in order regarding the relation of our investigation to the analysis of bh entropy in , say , string theory . three scenarios are in our view in principle possible . either , both approaches adress the same phenomena in different languages , or they deal with them on different scales of resolution of space - time . be that as it may , we think that our observation that the true ground state of our quantum vacuum seems to be what we call a wormhole space ( see section [ worm ] ) is an aspect which is not apparent in the original string theory approach and may be helpful to fix the proper ground state in string theory . in @xcite bekenstein remarked that the deeper meaning of black - hole entropy ( bh - entropy ) remains mysterious . he asks , is it similar to that of ordinary entropy , i.e. the log of a counting of internal bh - states , associated with a single bh - exterior ? ( @xcite,@xcite or @xcite ) . or , similarly , is it the log of the number of ways , in which the bh might be formed . or is it the log of the number of horizon quantum states ? ( @xcite,@xcite ) . does it stand for information , lost in the transcendence of the hallowed principle of unitary evolution ? ( @xcite,@xcite ) . he then claims that the usefulness of any proposed interpretation of bh - entropy depends on how well it relates to the original `` statistical '' aspect of entropy as a measure of disorder , missing information , multiplicity of microstates compatible with a given macrostate , etc . quite a few workers in the field argue that the peculiar dependence of bh - entropy on the area of the event horizon points to the fact that the degrees of freedom ( dof ) , responsible for bh - entropy , are situated near the event horizon . this seems to be further corroborated by the corresponding behavior of the so - called entanglement entropy , i.e. its ( apparent ) linear dependence on the area of the dividing surface ( cf . , just to mention a few sources , @xcite,@xcite,@xcite or the lively debate in @xcite , concerning entanglement entropy in a more general setting , @xcite,@xcite ) . this linear dependence does however not generally hold without further qualifications . it does in particular not hold for excited states ( see @xcite ) ! that is , while some particular sort of entanglement certainly plays an important role in this context , the real question is in our view the scale of resolution of space - time where this entanglement becomes effective and the nature of the quantum vacuum on this level of resolution . + remark : we want to emphasize the in our view crucial ( but frequently apparently not fully appreciated ) point that the entropy content of a bh is maximal . + we think , the usual version of entanglement , we observe on the scales of ordinary quantum theory , is only an epiphenomenon , representing rather the coarse - grained effect of a hidden structure which lives on a much more microscopic scale . i.e. , we are sceptical whether on such a microscopic scale the quantum vacuum can still be treated in the way of an ordinary quantum field theory vacuum as suggested in some of the papers cited above . we think , the maximum - entropy property of the bh - interior suggests another interpretation . we will come back to this point in more detail in section [ 4 ] ( cf . also the sceptical remarks in some of the review papers by wald , e.g. @xcite ( see in particular sect.6 , open issues ) , @xcite ( see in particular sect.4 , some unresolved issues and puzzles],@xcite ) . as bh - entropy is widely regarded as an observational window into the more hidden and primordial quantum underground of space - time , it should be expected that it can be naturally explained within the frameworks of the leading candidates of such a theory , i.e. , to mention the most prominent , string theory or lqg . for certain extreme situations string theory manages to give an explanation of the bh - entropy - area law . whether the explanation is really natural is perhaps debatable ( it relies in fact on a number of assumptions and correspondences as e.g. peculiar intersections of various classes of p - branes ) . in a sense , it is rather a correspondence between bh - behavior and the configurational entropy of certain string states . to mention some representative papers , @xcite,@xcite,@xcite,@xcite,@xcite,@xcite . in lqg , on the other hand , it is assumed from the outset ( at least as far as we can see ) that the corresponding dof are sitting at the bh - horizon . therefore the observed area dependence of bh - entropy is perhaps not so surprising ( cf . e.g. @xcite,@xcite ) . in the enumeration of the most promising candidates for a theory of quantum gravity one approach is usually left out which , we nevertheless think , has a certain potential . one may , for example , tentatively divide quantum gravity candidates into roughly three groups , the relativisation of quantum theory ( with e.g. lqg and causal set theory as members ) , the quantisation of general relativity ( string theory being a prominent candidate ) or third , theories which underlie both general relativity and quantum theory but are in fact more fundamental and structurally different from both and contain these two pillars of modern physics as derived and perhaps merely effective sub - theories , living on coarser scales ( cf . e.g. @xcite ) . in the following we want to develop such a model theory in more detail . as far as we can see , such a philosophy is also shared by t hooft who emphasized this point in quite a few papers ( see e.g. @xcite,@xcite,@xcite,@xcite ) . we quote from @xcite : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ it may still be possible that the quantum mechanical nature of the phenomenological laws of nature at the atomic scale can be attributed to an underlying law that is deterministic at the planck scale but with chaotic effects at all larger scales since , according to our philosophy , quantum states are identified with equivalence classes _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ furthermore : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ it is the author s suspicion however , that these hidden variable theories failed because they were based far too much upon notions from everyday life and ` ordinary physics ' and in particular because general relativistic effects have not been taken into account properly . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ while t hooft usually chooses his model theories from the cellular automaton ( ca ) class , we are adopting a point of view which is on the one hand more general and flexible but , on the other hand , technically more difficult and complex . instead of a relatively rigid underlying geometric substratum in the case of ca ( typically some fixed regular lattice ) on which the ca are evolving according to a given fixed ( typically local ) ca - law , we are employing quite irregular , dynamic geometric structures called by us cellular networks , the main point being that connections ( edges or links ) between the respective nodes or cells can be created or annihilated according to a dynamical law which , in addition , determines the evolution of the local node- and edge - states . to put it briefly , the ` matter distribution ' ( i.e. the global pattern of node - states ) acts on the geometry of the network ( the global pattern of active edges ) and vice versa . thus , as in general relativity , the network is supposed to find both its internal geometry and its matter - energy distribution with the help of a generalized dynamical law which intertwines the two aspects ( cf . e.g. @xcite or @xcite and further references given there ) . technically , the geometric substructure can be modelled by large , usually quite irregular ( random ) graphs . to make our point clear , this approach should not be confused with e.g. the spin network approach in lqg or various forms of ( dynamical ) triangulations . our networks are usually extremely irregular and wildly fluctuating on a microscopic scale , resembling rather wheeler s space - time foam , and smooth geometric structures ( as e.g. dimensional notions ) are hoped to emerge via some sort of a geometric renormalisation process ( in fact a very particular organized form of coarse - graining steps ) . some of the interesting deeper mathematical aspects can for example be looked up in @xcite . in our dynamical network approach to quantum space - time physics the nodes are assumed to represent cells of some microscopic size ( presumably planck size ) , the internal details of which can not be further resolved in principle or are ignored and averaged over for convenience and will be represented instead by a simple ansatz for a local ( node ) state . it can perhaps be compared with the many existing spin - models which are designed to implement certain characteristic features of complex solids . this is more or less the same philosophy as in the ca - framework . the elementary connections between the nodes ( the edges in graph theory ) are assumed to represent elementary interactions or information channels among the cells and also carry simple edge - states . we made a detailed numerical analysis of the behavior of such networks in @xcite . + remark : we would like to emphasize however , that our approach does not really rely on this particular framework . it rather serves as a means to illustrate the various steps in our analysis within a concrete model theory . the paper is organized as follows . in the next section we analyse the basic substratum , i.e. the microscopic patterns of vacuum fluctuations , in particular the negative energy fluctuations . in section [ worm ] we describe the three different roads which lead ( in our view : inevitably ) to the concept of wormhole space . the preparatory sections 2 and 3 are then amalgamated in section 4 into a detailed analysis of the microscopic distribution pattern of short - cuts or wormholes and their consequences for the number of effective dof in a volume of space . we introduce a new type of dimension , the so - called holographic dimension . furthermore , we explain the microscopic basis of the holographic principle in general and the bulk - boundary correspondence between the dof in the interior of e.g. a bh and the dof on the boundary . some apparent counter examples concerning the area - scaling property ( see e.g. @xcite ; wheeler s bag of gold-spacetimes ) are very briefly addressed . in the last section we briefly comment on a number of immediate applications of our microscopic holographic approach and ( open ) problems which can be settled with the help of our framework . a characteristic feature of the dynamical network models we investigated is their undulatory character . as a consequence of the feedback structure of the coupling between node ( cell ) states and wiring diagram of edges ( i.e. the pattern of momentary elementary connections or interactions ) the network never settles in a static , frozen final state . the network may of course end up in some attracting subset of phase space but typical are wild fluctuations on a small ( microscopic ) scale with possibly some macroscopic patterns emerging on a coarser scale forming some kind of superstructure ( see e.g. @xcite ) . it is in our view not sufficiently appreciated that , in contrast to most of the other systems being studied in physics , the quantum vacuum is in a state of eternal unrest on a microscopic scale , with , for all we know , short - lived excitations constantly popping up and being reabsorbed by the seething sea . it therefore seems reasonable to regard our above network model ( investigated in e.g. @xcite to @xcite ) as a ( toy ) model of the quantum vacuum with the energy - momentum fluctuations on short scales being associated with the fluctuations of the local node and edge states . in the following we adopt the working hypothesis of a parallelism of network behavior and microscopic behavior of the quantum vacuum . we now come to a detailed analysis of the microscopic pattern of vacuum fluctuations . in sect.4 of @xcite we made a calculation which shows that , given the huge number of roughly planck - size grains in a macroscopic piece of space and assuming that the individual grains are allowed to fluctuate almost independently , more precisely , some grain variable like e.g. the local energy , the total fluctuations in a macroscopic or mesoscopic piece of space of typical physical quantities are still so large ( i.e. macroscopic ) that they should be observable . note that with the number of nodes of roughly planck - size , @xmath0 , in a macroscopic volume , @xmath1 , being gigantic , its square root is still very large ( for the details of the argument see @xcite ) . more precisely , with @xmath2 some physical quantity belonging to a microscopic grain ( e.g. energy , momentum , some charge etc . and taking for convenience @xmath3=0 ) and @xmath4 being the observable belonging to the volume @xmath1 , the fluctuation of the latter behaves under the above assumption as @xmath5 with @xmath6 the number of grains in @xmath1 . this is a consequence of the central limit theorem . as such large integrated fluctuations in a macroscopic region of the physical vacuum are not observed ( they are in fact microscopic on macroscopic scales ) , we conclude : the individual grains or supposed elementary dof do not fluctuate approximately independently . remark : we note that this fact is also corroborated by other , independent observations . we can refine the result further ( cf . @xcite ) by assuming that the fluctuations in the individual grains are in fact not independent but correlated over a certain distance or , more precisely , are short - range correlated . in mathematical form this is expressed as integrable correlations . this allows that `` positive '' and `` negative '' deviations from the mean value can compensate each other more effectively . letting e.g. @xmath7 be the density of a certain physical observable and @xmath8 the integral over @xmath1 . in order that @xmath9 we proved in @xcite that it is necessary that @xmath10 we made a more detailed analysis in @xcite under what physical conditions property ( [ fluc ] ) can be achieved , arriving at the result : nearly vanishing fluctuations in a macroscopic volume , @xmath1 , together with short - range correlations imply that the fluctuations in the individual grains are anticorrelated in a fine - tuned and non - trivial way , i.e. positive and negative fluctuations strongly compensate each other which technically is expressed by property ( [ fluc ] ) . remark : in @xcite we extended such a vacuum fluctuation analysis and applied it to measurement instruments , being designed to detect ( possibly ) microscopic fluctuations of distances due to passing gravitational waves . we hence infer that the fluctuation pattern of e.g. energy - momentum has to be strongly anticorrelated . but under the above assumption it is possible that the underlying compensation mechanism which balances e.g. the positive and negative energy fluctuations is of short - range type , viz . , individual grain - energies may still fluctuate almost independently if their spatial distance is sufficiently large . we show in the following that the true significance of the so - called holographic principle is it , to enforce a very rigid and long - ranged anticorrelated fluctuation pattern in the quantum vacuum . as we are at the moment only interested in matters of principle , we assume the simplest case to prevail , called the space - like holographic principle ( holding in contexts like e.g. quasi - static backgrounds or asymptotic minkowski - space ; see e.g. the beautiful review @xcite ) . there exists a class of scenarios in which the maximal amount of information or entropy which can be stored in a spherical volume is proportional to the area of the bounding surface . the same holds then for the number of available dof in @xmath1 . this is the spacelike holographic principle . in a series of papers brustein et al . developed a point of view that relates typical fluctuation results of quantum mechanical observables in quantum field theory with the area - law - like behavior of entanglement entropy and bh - entropy ( cf . e.g. @xcite ) . we already made a brief remark to this approach in @xcite . we note that we arrived at related results using different methods in another context ( see e.g. @xcite and @xcite ) . as a more detailed comment would lead us too far astray , we plan to discuss this subject matter elsewhere . what we are going to show in the following is that the mechanism leading to the strange area - behavior of the entropy of an enclosed volume , @xmath1 , is considerably subtler as usually envisaged . on the one hand , we will show that the number of elementary dof contained in @xmath1 is in principle proportional to the volume . on the other hand we infer from observations on the macroscopic or mesoscopic scale that the fluctuations of e.g. the energy are strongly anticorrelated . however , as long as this compensation mechanism is short - ranged , we would still have a number of nearly independently fluctuating clusters of elementary dof which again happens to be proportional to the volume as the cluster size is roughly equal to the correlation length . so the conclusion seems to be inescapable that the patterns of vacuum fluctuations must actually be long - range correlated on a microscopic scale . but we showed in @xcite or @xcite in quite some detail that even systems , displaying long - range correlations , will usually have an entropy which is proportional to the volume . a typical example is a ( quantum ) crystal ( @xcite,@xcite ) . it is certainly correct that below a phase transition point a system of particles in the crystal phase has a smaller entropy than in the liquid or gas phase , but still the entropy happens to be an extensive quantity . the reason is in our view that the system develops , as a result of the long - range correlations , new types of collective excitations ( e.g. lattice phonons ) which serve as new collective dof . approximately they may be treated as a gas of weakly interacting elementary modes with the usual extensive entropic behavior . that is , the holographic principle entails that the elementary dof have to be long - range anticorrelated ( cf . also the remarks in @xcite or sect.7 of @xcite ) . but we see that this is only a necessary but not a sufficient property for an entropy - area law to hold . we hence arrive at the preliminary conclusion : from our preceding arguments and observations we conclude that the holographic principle implies that the fluctuation patterns in @xmath1 are long - range anticorrelated in a fine - tuned way on a microscopic scale and are essentially fixed by the state of the fluctuations on the bounding surface . the dynamical mechanism , which generates these long - range correlations must however , by necessity , have quite unusual properties ( cf . subsection [ bulkboundary ] ) . before we derive the wormhole structure of the quantum vacuum on a primordial scale in the next sections , we continue with the general analysis of the pattern of vacuum fluctuations and derive some useful properties of it . a particular role is usually played by the energy and its fluctuations furthermore , vacuum fluctuations are frequently discussed together with the so - called zero - point energies . while they are not exactly the same , they are closely related . both occur also in connection with the cosmological constant problem ( to mention only a few sources see e.g. @xcite,@xcite,@xcite,@xcite,@xcite,@xcite,@xcite ) . in the simplest examples like e.g. the quantized harmonic oscillator or the electromagnetic field we have @xmath11 and with @xmath12 in the groundstate , @xmath13 , we have @xmath14 with @xmath15 which follows from @xmath16=-i\hbar$ ] . in the same way we have in ( matter - free ) qed : @xmath17 with @xmath18 so that again @xmath19 is a sum over pure vacuum fluctuations of the non - commuting quantities @xmath20 and @xmath21 . one should however note that in the quantum field context products of fields at the same space - time point have to be wick - ordered ( in order to be well - defined ) . it is , on the other hand , frequently argued that with gravity entering the stage , these eliminated zero - point energy fluctuations have to be taken into account again . in our view , this problem is not really settled . we now come to an important point . it is our impression that in some heuristic discussions ( vacuum fluctuations as virtual particle - antiparticle pairs ) the consequences of the fact that the vacuum state is an exact eigenstate of the hamiltonian in a hilbert space representation of some quantum field theory are not fully taken into account . i.e. , we have @xmath22 ( provided the ground state energy is for convenience normalized to zero ; note however that this may be problematical in a theory containing gravity ) . eigenstates , however , have the peculiar property that the standard deviation is necessarily zero , @xmath23 according to the standard interpretation of quantum theory combined with spectral theory , @xmath24 , this implies that in each individual observation process the total energy of the vacuum which is , according to conventional wisdom , the ( hypothetical ) sum or superposition of local ( small scale ) fluctuations , happens to be exactly zero . in other words , the elementary fluctuations have to exactly compensate each other in an apparently fine - tuned way . put differently if there are positive local energy fluctuations , there have to be at the same time by necessity negative energy fluctuations of exactly the same order . that is , at each moment , the global pattern of energy fluctuations in the quantum vacuum is an array of rigidly correlated positive and negative local excitations . remark : note the similarity of this independent observation to what we have said above in connection with the holographic hypothesis . + it should be mentioned that hawking in @xcite invoked exactly this picture of a particle pair excitation near the event horizon with the virtual particle , having negative energy , falling into the bh while the one with positive energy escapes to infinity . it would be useful to get more quantitative information on the spectral properties of the local observables , in particular estimates on negative fluctuations . one could try to make an explicit spectral resolution of these quantities , e.g. of the energy , contained in a finite volume , @xmath1 , but this turns out to be difficult in general , even if one has given an explicit model theory in some hilbert space . as we prefer a more general , model independent approach ( not necessarily based on hilbert space mathematics ) , we proceed by using ( similar to bell in his papers ) a general probabilistic approach which rather exploits the statistics of individual measurement results . unfortunately , we found that the standard estimates , known to us in this context ( e.g. the markov - chebyshev - inequality ) , always go in the wrong direction ( see e.g. @xcite or @xcite ) . therefore we present in the following our own estimate . the strategy is the following . we take an observable , @xmath25 , localized in @xmath1 with , for convenience , discrete spectral values , @xmath26 , and corresponding probabilities denoted by @xmath27 . if we assume that the expectation of @xmath25 is zero ( which can always be achieved by a simple shift ) we have @xmath28 furthermore , we assume its standard deviation in e.g. the vacuum , @xmath29 , to be finite ( which is automatically the case for bounded operators , but we want to include also more general statistical variables ) @xmath30 in a first step we make the simplifying assumption ( taking e.g. a bounded function of the energy ) @xmath31 we are interested in the amount of negative ( e.g. energy ) fluctuations we will observe in measurements . a reasonable quantitative measure of it is @xmath32 with @xmath33 the negative spectral values and their corresponding probabilities . we then have ( with @xmath34 ) @xmath35 for the lhs we have @xmath36 as the expectation of @xmath37 was assumed to be zero . this yields @xmath38 i.e. @xmath39 on the other hand ( cauchy - schwartz ) @xmath40 we hence arrive at the result if the observable , @xmath37 , is bounded , so that its spectral values fulfill @xmath41 , we have the estimate @xmath42 with @xmath43 the probabilities that the negative spectral values @xmath26 occur in an observation . that is , we manage to bound a quantity , which is difficult to measure directly , by quantities , which are usually more easily accessible . we can generalize this result to situations where the @xmath26 are not exactly bounded by some @xmath44 but are bounded in at least an essential way . we assume that there exists some @xmath44 so that @xmath45 we then have @xmath46 under the above assumption of an essentially bounded @xmath37 we have @xmath47 another , rigorous , but not quantitative , argument can be derived from axiomatic quantum field theory ( see e.g. @xcite ) . it follows from the so - called reeh - schlieder theorem that there are no local observables or fields which can annihilate the vacuum ( where by local we mean that the objects commute for space - like separation ) . i.e. , we have for any local @xmath48 ( with @xmath49 ) @xmath50 we take now as local observable the energy density integrated over a certain spatial region , @xmath1 , @xmath51 one usually normalizes @xmath52 to @xmath53 the classical expression of the energy density , being derived in lagrangian field theory , is positive . the corresponding quantized expression , after a necessary wick - ordering ( see e.g. @xcite or @xcite ) is however no longer positive definite as an operator ( density ) . this can be seen as follows . if the quantized energy density were still positive , one can take the square root ( via the spectral theorem ) of e.g. the positive operator @xmath54 and get : @xmath55 hence @xmath56 as @xmath57 is also a local observable this is a contradiction due to the reeh - schlieder theorem . @xmath54 is not a positive operator , hence its spectrum contains negative spectral values . it is then easy to construct hilbert - space vectors , @xmath58 , so that the measurement of @xmath54 in @xmath58 yields negative values for the local energy . remark : we recently learned that this argument is originally attributed to epstein ( unpublished;@xcite or see @xcite ) , while the derivation which can be found in @xcite is a completely different one . the important message ( in our view ) of all this is that , perhaps in contrast to naive expectation , the quantum vacuum contains a lot of negative energy excitations which globally exactly balance the positive excitations . one may now speculate about the possibility of making use of this observation . in this section we want to describe ( very ) briefly and sketchily the three different lines of reasoning which lead us to the concept of wormhole spaces . the first line originated from our investigation of the structure and dynamical behavior of the networks we described above . in e.g. @xcite we analyzed in some quantitative detail the unfolding of the network structure and the various network epochs under the inscribed microscopic dynamical laws and developed the two - level concept of the network structure ( or , rather , a multi - scale structure ) , which , under the right conditions , is relatively smooth on a sufficiently coarse - grained level ( level 2 ) with , among other things , a distant measure ( metric ) of the more ordinary type and ( hopefully ) an integer - valued geometric dimension , while on a more microscopic scale ( level 1 ) the network structure is expected to be very erratic with possibly a lot of links ( elementary interactions or information channels ) connecting regions which may be far apart with respect to the metric on level 2 . the association of these links with microscopic wormholes thus suggests itself ( cf . in particular observation 4.27 in @xcite ) . note furthermore that our network dynamics implies that these translocal connections are dynamically switched on or off . compare this observation with the point of view expounded in e.g. @xcite _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ but if a wormhole can fluctuate out of existence when its entrances are far apart then , by the principle of microscopic reversibility , the fluctuation into existence of a wormhole having widely separated entrances ought to occur equally readily . this means that every region of space must , through the quantum principle , be potentially `` close '' to every other region , something that is certainly not obvious from the operator field equations which , like their classical counterparts , are strictly local.it is difficult to imagine any way in which widely separated regions of space can be `` potentially close '' to each other unless space - time itself is embedded in a convoluted way in a higher - dimensional manifold . additionally , a dynamical agency in that higher - dimensional manifold must exist which can transmit a sense of that closeness . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the quantitative network calculations in the mentioned papers have mainly been performed within the framework of random graphs . important mathematical tools for the network analysis in the transition from microscopic , strongly fluctuating and geometrically irregular scales to coarse - grained and , by the same token , smoother scales have been the concepts of cliques of nodes , the clique - graph of a graph and an important network parameter which we dubbed intrinsic scaling dimension ( we later learned , @xcite , that this concept plays also an important role in geometric group theory or cayley - graphs where it is called the growth degree ) . to give a better feeling what is actually implied , we give the definitions of clique , clique - graph and internal scaling dimension ( more about graph theory can e.g. be found in @xcite , @xcite , notions and properties of graph dimension were studied in e.g. @xcite ) . a simplex in a graph is a subset of vertices ( nodes ) with each pair of nodes in this subset being connected by an edge . in graph theory it is also called a complete subgraph . the maximal members in this class are called cliques . the clique graph , @xmath59 , of a graph , @xmath60 , is built in the following way . its set of nodes is given by the cliques of @xmath60 , an edge is drawn between too of its nodes if the respective cliques have a non - empty overlap with respect to their set of nodes . graphs carry a natural neighborhood structure and notion of distance . the neighborhood @xmath61 of a node @xmath62 is the set of nodes @xmath63 which can be reached , starting at @xmath62 in @xmath64 consecutive steps , i.e. there exists a path of @xmath64 consecutive edges connecting the nodes @xmath62 and @xmath63 . [ dist]the canonical network or graph metric is given by @xmath65 here @xmath66 is the number of consecutive edges of the path . the above definition fulfills all properties of a metric . thus graphs and networks are examples of metric spaces . [ dim ] let @xmath62 be an arbitrary node of @xmath60 . let @xmath67 denote the number of nodes in @xmath61.we consider the sequence of real numbers @xmath68 . we say @xmath69 is the _ lower _ and @xmath70 the _ upper internal scaling dimension _ of g starting from @xmath62 . if @xmath71 we say @xmath60 has internal scaling dimension @xmath72 starting from @xmath62 . finally , if @xmath73 @xmath74 , we simply say @xmath60 has _ internal scaling dimension @xmath75_. we proved in @xcite ( among other things ) that this quantity does not depend on the choice of the base point for most classes of graphs . it turns out that this geometric notion is a very effective characteristic of the large - scale structure of graphs and networks . this topic was further studied in greater generality in e.g. @xcite . in @xcite we developed what we called the geometric renormalization group , to extract important geometric coarse grained , that is , large scale information from the microscopically quite chaotically looking network and its dynamics . the idea is , at least in principle , similar to the block spin transformation in statistical mechanics . that is , certain characteristic properties of the system are distilled from the microscopically wildly fluctuating statistical system by means of a series of algorithmic renormalization steps ( i.e. coarse - graining plus purification ) . the central aim is it to arrive in the end at a system which resembles , on the surface , a classical space - time , or , on the other hand , to describe the criteria a network has to fulfill in order that it actually has such a classical fixed point . in the course of this analysis we observed ( cf . section viii of @xcite ) that the so - called critical network geometries , i.e. the microscopic network geometries which are expected to play a relevant role in the analysis , are necessarily in a very specific way geometrically non - local , put differently , they have to contain a very peculiar structure of non - local links , or short - cuts , that is , in other words , the kind of wormhole structure , we already described above . relations to non - commutative geometry were established and studied in @xcite . we mention in particular section 7.2 `` microscopic wormholes and wheeler s space - time foam '' and section 8 `` quantum entanglement and quantum non - locality '' . the possible relevance for quantum theory is in fact quite apparent ( as has also been emphasized in the papers by t hooft ) , as these microscopic wormholes may be the origin of the ubiquitous entanglement phenomena in quantum theory . the following figures describe pictorially the nested structure of the cliques of nodes in consecutive renormalization steps and overlapping cliques of nodes , defining the local near - order of physical points together with shortcuts which connect distant parts of the coarse - grained surface structure . a second complex of ( related ) phenomena emerges in the field of small world networks . this is a particular class of networks of apparently quite a universal character ( described and reviewed in some detail , for the first time , in @xcite ) with applications in many fields of modern science . they consist essentially of an ordinary local network with its own local notion of distance superimposed by a typically very sparse network of so - called short - cuts living on the same set of nodes and playing a structural role similar to the microscopic wormholes described above . a typical example ( with dimension of the underlying lattice @xmath76 ) is given in the following figure . some further ( in fact very few ) references , taken from quite diverse fields are e.g. @xcite,@xcite,@xcite . its , in our view , crucial characteristic is the existence of two metrics over the same network or graph . the first , @xmath77 , is defined ( cf . definition [ dist ] ) by taking into account the full set of edges ( i.e. , including the short - cuts ) and a second ( local ) metric , @xmath78 , taking into account only the edges of the underlying local network . it hence holds @xmath79 remark : the metric @xmath78 may then be associated ( after some renormalisation or coarse - graining steps ) with an ordinary macroscopic metric defined on a smooth space ( without wormholes ) like our classical space - time . @xmath77 , on the other hand , should be regarded as a microscopic distance concept which employs the existence of wormholes . while , on the surface , the origin of this concept of small world networks seems to be quite independent of the wormholes in general relativity , it is the more surprising that on a conceptual meta level various subtle ties do emerge . to mention only one ( in our view ) important observation . in @xcite it is for example shown , that a sparse network of shortcuts superimposed upon an underlying local network , has the propensity to stabilize the overall frequency pattern ( phase locking ) of so - called phase - oscillators which represent the nodes of the networks , the links representing the couplings . the oscillators are assumed to oscillate with ( to a certain degree ) independent frequencies . if we relate these local frequencies with some local notion of time ( or clocks ) , we may infer that ( microscopic ) wormholes create or stabilize some global notion of time ! we now come to the third strand , viz . the real wormholes of general relativity or quantum gravity . we mainly concentrate on the wormholes in true , i.e. lorentzian space - time . euclidean wormholes also ( may ) play an important role and have been discussed extensively in the context of the ( nearly ) vanishing value of the cosmological constant ( see e.g. @xcite,@xcite,@xcite , @xcite,@xcite,@xcite ) . of particular relevance in the lorentzian context are the so - called traversable wormholes . their study started ( as far as we know ) with two seminal papers by thorne and coworkers ( see @xcite ) . the geometric construction of such solutions is in fact not so difficult if performed by the so - called g - method . that is , one constructs a geometric wormhole , e.g. of the static type , and , in a second step , computes the energy - momentum tensor being consistent with this solution . giving a rough outline , this can be done in following way . two open balls are removed from two different pieces of e.g. approximately flat 3-space . their boundaries are glued together with the junction being smoothed . as a consequence of the smoothing process a tube emerges interpolating between the two spheres ( see e.g. @xcite ) . it is a remarkable fact that in this process the weak energy condition ( wec ) has to be violated , the latter implying that @xmath80 that is , the matter - energy density is positive in any reference system . put differently , in order to get a traversable wormhole , one has to violate the wec . the wec is always satisfied by classical matter . therefore quantum effects are needed . the kind of negative energy needed is also called exotic matter . we showed in quite some detail in the preceding section that the quantum vacuum abounds with negative energy fluctuations . therefore the speculation in section h of the first paper in @xcite does not seem to be too far - fetched . in a next step one can study networks of such traversable wormholes . in @xcite it is speculated that such a network , existing in the early universe , may solve the horizon problem . the same situation was discussed from the point of view of our network approach in section 4.1 ( the embryonic epoch ) of @xcite . all this comes already quite near the general picture we envoked in the beginning of this section . furthermore one can envisage solutions combining black and white holes . this corresponds to some of our networks where the orientation ( direction ) of the links connecting two nodes can change under the dynamics . a review of lorentzian wormholes can be found in the book by visser ( @xcite ) . some other references are e.g. @xcite and @xcite . the above picture of a hypothetical network of wormholes sitting in the deep structure of the quantum vacuum is beautifully complemented by an approach ( see e.g. @xcite,@xcite ) which investigates within a ( semi)classical approximation the energy of a quantum vacuum state containing such an array of wormholes ( or , rather , a gas of such wormholes ) and compare it with a vacuum state which in zeroth order is flat minkowski space . it comes out ( apparently being a kind of casimir effect ) that the quantum vacuum containing the wormhole gas has in this semiclassical approximation a lower energy compared to the state , being a perturbation of minkowski space . one should note however that this is a first order quantum effect ! anyhow , this observation seems to corroborate the space - time foam picture of e.g. wheeler and we conclude this section with from our analysis in this and the preceding section emerges a model of the ground state of some preliminary version of quantum gravity which contains as an essential ingredient a network of microscopic wormholes . these wormholes can be created and annihilated and are in our picture the carriers of information between distant parts of classical space - time . we call such a physical structure a wormhole space and regard our cellular or small world networks , discussed above , as models , encoding and representing the typical characteristics of such systems . the typical characteristic is the existence of two types of distance , a microscopic one and an ordinary local one , being similar to ordinary macroscopic metrics on smooth spaces . we learned in the preceding sections that two ( presumably crucial ) properties govern the behavior of the quantum vacuum on a microscopic scale . first , the vacuum fluctuations are strongly long - range anticorrelated on a microscopic scale , i.e. there exists a fine - tuned pattern of positive and negative ( energy ) fluctuations . second , a quantum mechanical stability analysis seems to show that the quantum vacuum is pervaded by a network of microscopic wormholes . we argued above that these two features are not independent phenomena but rather are the two sides of the same medal . furthermore , the presumed wormhole structure has been supported by observations coming from other fields of research like e.g. cellular or small - world networks . in this ( central ) section we will now combine these observations and show that they underlie ( among other things ) the holographic principle and the entropy - area law of bh - thermodynamics . in the following we will use ( for convenience ) the language of our networks with the nodes of the network representing microscopic grains of space ( or space - time ) of roughly planck - size . leaving out other details we treat our quantum vacuum as a wormhole space , i.e. as a ( small world ) network consisting of an ordinary local network structure being superimposed by a ( presumably ) sparse random network with edges consisting of short - cuts , i.e. links , connecting regions of space or space - time , which may be quite a distance apart with respect to the metric , belonging to the underlying local network these short - cuts represent the wormholes of ordinary space - time . the crucial characteristic , from which everything is expected to follow , is the pattern and distribution of these short - cuts being immersed in the underlying local network . that is , we randomly select a node @xmath62 in the network @xmath60 ( @xmath60 standing for graph ) and study the distribution of short - cuts connecting @xmath62 with nodes @xmath63 on spheres of radius @xmath81 around @xmath62 ( measured with respect to some macroscopic metric or the natural metric of the underlying local network ) . we expect that the precise distribution law will depend on the concrete type of space - time we are dealing with . this holds in particular if the space - time is not static . that is , our microscopic approach to holography makes it possible to understand how holography may depend on the concretely given type of space - time ( cf . e.g. the covariant entropy bound of bousso , @xcite ) . remark : we emphasize that the network or the quantum vacuum it is representing , is basically a statistical system with all local dof fluctuating . that means , most of our statements in the following are about mean values or averages over finer statistical details . one can arrive at the law , describing the distribution of short - cuts or wormholes around some arbitrary but fixed generic node ( viz . some fixed place in space - time ) in roughly two ways . one can e.g. motivate the distribution law by appealing to certain fundamental principles like e.g. scale - freeness or absence of a particular and in some sense unnatural length scale on a fundamental level . alternatively , one can show that a reasonable choice leads to far - reaching consequences and corroborates the findings and observations made on a more macroscopic level . to keep the discussion as briefly as possible we adopt in this section the second point of view . in the following we want to concentrate , for the sake of brevity , on a simple type of quantum vacuum , that is , the vacuum belonging to ordinary minkowski space or a space - time which is asymptotically flat ( e.g. a schwarzschild space - time ) . we postpone the analysis of more general space - times as they occur in general relativity . we make the following conjecture : on the average the number of short - cuts from a central node @xmath62 to nodes @xmath63 , sitting on the sphere , @xmath82 about @xmath62 is independent of @xmath81 . denoting this number by @xmath83 , we hence have @xmath84 remark : as this number is a statistical average , it need not be an integer . + the situation is depicted in the following picture . [ bb ] we will show in subsection [ bulkboundary ] in a detailed quantitative analysis that this result approximately holds as well for nodes , not sitting exactly in the center of the spheres @xmath85 ( see the following picture ) . we denote the cluster of nodes in the ball @xmath86 being connected to an @xmath62 by short - cuts by @xmath87 . we previously introduced the internal scaling dimension of a network ( see definition [ dim ] ) . it roughly describes how fast the network is growing with respect to some base node . as this growth degree is to a large degree independent of the base node ( see e.g. @xcite ) it is a global characteristic of a given network , in fact of a whole class of similar networks ( @xcite ) . it is well known that the generalization of the concept of dimension away from smooth geometric structures is not unique . the above type of dimension has the tendency to grow if additional short - cuts are inserted into a given network geometry . we now introduce another dimensional concept which catches other important network properties being more closely related to the phenomena we want to analyze in this paper . it uses in an essential way the two metrics , @xmath88 , introduced above . from the above we infer that the number of nodes in the cluster @xmath87 is approximately equal to @xmath89 . furthermore , if the network of short - cuts is very sparse , the clusters @xmath90 with @xmath91 are essentially disjoint ( the overlap is empty or very small ) . this is the phenomenon called _ spreading _ in the theory of random graphs . hence , the following concept is reasonable . we define a holographic dimension , @xmath92 , of a network in the following way . we take some ball @xmath86 with macroscopic radius @xmath81 around some fixed but arbitrary node @xmath62 with respect to the local metric @xmath93 . we then form the @xmath94-neighborhoods around the nodes @xmath95 with respect to the microscopic metric @xmath96 . we construct a minimal cover of @xmath86 by such @xmath97 , i.e. a minimal selection of such @xmath98 s.t . @xmath99 the cardinality of such a minimal set we denote by @xmath100 . we take the limit @xmath81 large or @xmath101 ( in an infinite network ) and define we call @xmath102 the holographic dimension of the graph ( network ) , provided the limit exists . in the more general situation we can , as in definition [ dim ] , define upper and lower dimensions etc . as for the previously defined graph dimension , the limit is independent of the selected base point , @xmath62 , if the network or graph is homogeneous on the average or in the large . due to the sparseness of the embedded subgraph of short - cuts , which yields the spreading property mentioned above , the number @xmath100 scales for the wormhole spaces or small - world networks as @xmath103 with @xmath104 the dimension of the local network or its coarse - grained continuum limit space . proof : the @xmath94-neighborhoods consist of nodes lying in the neighborhoods with respect to the local metric , @xmath93 , @xmath105 , plus the vertices connected by short - cuts with @xmath63 . the cardinality of @xmath105 is independent of @xmath81 and typically ( at least in our models ) a small number . for @xmath101 @xmath106 will therefore consist mainly of nodes connected to @xmath63 by short - cuts . sparseness of the short - cut graph and spreading yield the result . @xmath107 for the type of wormhole spaces or small - world networks , defined above , we then have @xmath108 that is , in this case we have the important result @xmath109 we now come to the holographic principle and the bh - entropy area law . as already mentioned , we discuss in this paper only the example of 4-dim . asymptotically flat ( minkowski ) space - time . in planck units a macroscopic ball , @xmath86 , contains approximately @xmath110 dof or grains of planck size . the typical cluster size is @xmath111 due to the mentioned spreading property the number of ( effectively ) independent cluster in the above minimal cover is approximately @xmath112 with @xmath113 denoting the area of @xmath85 . the number of effectively independent clusters , @xmath114 in @xmath86 is @xmath115 with the typical cluster size @xmath111 to show now that the number of effective dof in a generic volume ( where by generic we mean a region in space with the diameter in all directions being roughly of the same order ) is proportional to the surface area , @xmath116 , of its boundary , we employ a general observation , made e.g. in statistical mechanics . an important tool for the analysis of systems in statistical mechanics are correlation functions . correlations decay usually for large separation of the respective dof , but what is on the other hand certainly the case is , that nearest neighbors are strongly correlated ( near order versus far order ) . we expect that the dof in each of the @xmath117 are strongly correlated . we hence take it for granted , that they act effectively as a single collective dof . remark : it may be possible , that this near order in the immediate neighborhood of the grains can be finally destroyed by the insertion of a huge amount of localized energy , but this does not seem possible with present means . due to the existence of wormholes or short - cuts , distributed in space - time , the number of effective dof ( affiliated with the respective clusters @xmath114 ) in e.g. a ball @xmath86 equals @xmath100 , that is @xmath118 this is the area - law behavior of entropy or number of dof in a volume of space found in e.g. bh - entropy . we note however , that this law , in our formulation , is essentially a statement about the collective behavior of the elementary dof in ( the interior of ) a volume of space . i.e. , the respective dof are not really sitting on the boundary of @xmath1 . as to the details of the bulk - boundary correspondence see the following subsection . if we adopt the entropy - area law of bh - thermodynamics , which is , expressed in planck units , @xmath119 we have the possibility to fix our parameter @xmath120 , which gives the number of wormholes connecting a central grain of space with the grains on a surrounding sphere @xmath85 for any @xmath81 . however , entropy is not exactly identical to number of dof . to relate the two , we have to make a simple model assumption . one frequently makes the assumption of boolean dof , i.e. the dof on an elementary scale are two - valued . with this assumption we have the relation @xmath121 with @xmath122 the entropy , @xmath123 the number of dof . with the help of this identification we get @xmath124 which can in qualitative arguments be approximated by one ! that is , in planck units , there exists roughly one short - cut between a central vertex and a surrounding sphere of radius @xmath81 . this shows that on an extremely microscopic scale , the network of short - cuts is indeed very sparse . however the picture changes considerably if we go over to more accessible length scales . if we use , for example an atomic length - scale of e.g. @xmath125 , we have approximately @xmath126 grains of planck - size in a volume element of diameter @xmath127 . if we then choose , instead of a sphere @xmath85 , a spherical shell of radius @xmath81 and thickness @xmath127 we have approximately the number of wormholes or short - cuts between a central volume element of size @xmath128 and a corresponding spherical shell of radius @xmath81 is approximately @xmath129 which is quite a large number . if we choose for example @xmath130 , we see that roughly @xmath131 grains in the shell are the endpoints of about @xmath132 short - cuts coming from the central volume element of size @xmath128 . if we replace @xmath81 by the approximate diameter of the universe , i.e. @xmath133 ly , we get ( with @xmath134 ) : @xmath135 and for the number of planck - size grains in a spherical shell of this radius : @xmath136 with still @xmath132 short - cuts ending there . that is , only one in @xmath137 grains is the endpoint of a respective short - cut . but if we select a volume element of size @xmath128 in this shell , we have still the number of wormholes ( short - cuts ) between two volume elements of size @xmath128 being a distance @xmath138 apart , is still the large number @xmath139 that is , even over such a large distance there exist still a substantial number of wormholes connecting the two volume elements . but nevertheless , the network is sparse , viewed at planck - scale resolution . we now come to the last point of this section . from what we have learned above , it is intuitively clear , that the dof sitting on the boundary @xmath85 of e.g. a ball @xmath86 should fix ( or slave ) the dof in the interior . but we note that in order that this can hold , we have to verify our statement made in observation [ bb ] . furthermore , it is of tantamount importance to understand in more quantitative detail the influence of different shapes of the region under discussion and the effect of different space - time geometries . the prerequisites for this enterprise will be derived in the following . as an example we employ , as we already did above , the simple geometry of the spacelike holographic bound . for reasons of simplicity we place the center of the ball in the origin , i.e. @xmath140 . it is of great help if we can transform the problem into a problem of ordinary continuous analysis . to this end we introduce the probability that a node in the interior of @xmath86 and an arbitrary node on the boundary @xmath85 are connected by a short - cut . with @xmath141 and @xmath142 there spatial euclidean distance in three dimensions is @xmath143 the edge probability is given by @xmath144 here @xmath145 is the number of nodes ( or planck - scale grains ) on the sphere around @xmath146 with radius @xmath147.this follows directly from what we have learned in the previous sections . what we are actually doing in the following is the calculation of the average number of short - cuts between an arbitrary node @xmath146 in the interior of @xmath86 and the nodes on the boundary @xmath85 . this will be done within the framework of random graphs . the above @xmath148 is the so - called edge probability ( for the technical details see @xcite or @xcite,@xcite ) . the sample space is the space of graphs with node set comprising the node in @xmath146 and all the nodes sitting on the boundary @xmath85 and edge set all possible different sets of short - cuts connecting @xmath62 with the nodes on @xmath85 . the probability of each graph in the sample space is calculated with the help of the above elementary edge probability @xmath148 and its dual @xmath149 . we choose @xmath146 arbitrary but fixed in @xmath150 and let @xmath151 vary over the sphere @xmath152 . the integral over @xmath152 will then give the mean number of short - cuts between @xmath146 and the grains on @xmath152 . the guiding idea is that the dof in the interior are fixed by the dof on the boundary if this integral is essentially @xmath153 , as according to our philosophy , developed previously , in that case every node in the interior has on average at least one partner on the boundary as nearest neighbor with respect to the microscopic metric @xmath96 . to make the integration easier we choose , without loss of generality , @xmath154 with @xmath155 . a straightforward calculation ( using polar coordinates and appropriate variable transformations ) yields for the average number of short - cuts , @xmath156 , @xmath157 note that the integrand @xmath158 is always positive . furthermore , our choice of a coulomb - like law ( in three dimensions ) for the distribution of short - cuts in the previous subsection , i.e. @xmath159 , makes the above integral independent of @xmath81 . we can find a closed expression for the definite integral , i.e. @xmath160 note that the position of the point @xmath146 relative to the center and the boundary can be regulated by the value of the parameter @xmath161 . we have tabulated the integral for @xmath162 from @xmath163 to @xmath164 in the following table . + [ cols="<,<,<,<,<,<,<,<,<,<,<",options="header " , ] + we see that the number of short - cuts is almost constant through the whole interior of @xmath86 apart from a thin shell near the boundary . but this is not really surprising because there the main contribution comes from the near side of the boundary and is no longer of a true short - cut character . taking into account the additional prefactor , @xmath165 , in front of the integral which is @xmath166 we have the number of short - cuts from an arbitrary node @xmath146 in @xmath86 to the boundary @xmath85 is approximately @xmath167 for most of the nodes . furthermore for our coulomb - like distribution law it is independent of the radius of the sphere and is therefore consistent with the expected holographic behavior for this geometry . it is instructive to evaluate the above formula for @xmath168 , i.e. , the influence via short - cuts of the sphere @xmath85 on a dof in the exterior of @xmath85 . for @xmath162 large , the integral is dominated by the first term in the integrand , viz . for @xmath162 large we have @xmath169 for nodes , @xmath62 , lying outside of @xmath85 , the effect of the short - cut connections between @xmath62 and @xmath85 decays like a coulomb - law . that is , the dof in the exterior are no longer fixed by the dof on @xmath85 . what remains instead is a statistical influence in form of a correlation which decays with increasing distance . by the same token , there can not be an entropy - area law for the exterior of the sphere relative to its internal boundary . anyhow , this example does not really contradict the correctness of the spatial holographic principle as being presented in this paper . it would be interesting to relate our findings to the covariant holographic principle of e.g. bousso , @xcite this simple observation has an important consequence for arguments being sometimes invoked against the general nature of the spatial holographic principle ( cf . e.g. @xcite ) . while we do not intend to discuss the holographic principle for more general space - times in this paper , we mention one counter - example which one finds frequently in the literature , i.e. a universe containing a closed spatial slice , @xmath122 with a small inner subregion , @xmath170 ( see the following picture ) . the area - law in the usual form applies for the subregion @xmath170 relative to its boundary . however , according to our ( microscopic ) version of spatial holography , the dof on the inner boundary can not slave the dof in the large region @xmath171 if the inner boundary becomes too small . they only establish some kind of correlation in the exterior . the quantitative details are given by integrating our coulomb - like influence law over the inner surface . another , related , class of interesting ( but perhaps pathological ) apparent counter examples ( which we plan to address in greater detail elsewhere ) is discussed in e.g. @xcite , i.e. spacetimes which are called by marolf bag - of - gold spacetimes. an essential ingredient is some frw - spacetime hidden in the interior of a region which resembles an ordinary bh . the innner frw - universe has of course an entropy which is proportional to its volume while from the outside the whole configuration looks like a bh . this seeming contradiction can be easily understood with the help of our microsopic holographic law as the frw - spacetime is actually only weakly coupled with the exterior of the bh via wormholes . the technical arguments are the same as above . in the preceding sections we developed only the groundwork of our approach . to keep the paper within reasonable size , we had to postpone a more detailed discussion of the many consequences and immediate applications . in this final section we at least undertake to briefly comment on a number of important points . it is however obvious that a more detailed discussion of each point would require a paper of its own . + i ) the possible connections to the ubiquituous phenomenon of entanglement in ordinary quantum theory are obvious . interesting in this respect is e.g. the well - known tension in quantum theory between the locality and causality principle of special relativity and the instantaneous state reduction , accompanying the measurement process ( cf . the respective sections in e.g. @xcite ) . we think , similar to e.g. t hooft , that ( the microscopic form of ) holography ( we developed in this paper ) is the common basis which may unite quantum theory and gravitation . + ii ) the consequences of the bh - entropy being maximal , which is quite uncharacteristic for the ground state entanglement entropy in say ordinary quantum theory , should be further analysed . + iii ) the ads - cft - correspondence is regarded in string theory as the paradigm for bulk - boundary correspondence ( we mention only the review @xcite and the popular account @xcite ) . in it two , at first glance , fundamentally different theories are related to each other , the one living in the bulk , the other living on the boundary at infinity . we must however say that the concrete physical epistemology of this latter notion is not entirely clear to us . the use of boundaries at infinity is wide spread in holography and is mathematically well - defined , in particular for certain well - adapted coordinate systems being in use in hyperbolic geometry . but in general it is rather an asymptotic property and not a concrete place . note that in our approach full information about the interior of a ( spatial ) region is distributed essentially everywhere in the exterior of the region via wormholes , but usually not in the form of another field theory ! + iv ) a virulent problem ( the unitarity problem ) in bh - thermodynamics is the question whether a pure state goes over into a mixed state or not , that is , if the laws of ordinary quantum theory are possibly violated in bh - thermodynamics ( instead of the many published papers we mention only the reviews by wald , cited above ) . this is a quite intricate epistomological problem somewhat similar to the quantum measurement problem . we think , part of the problem is that frequently pure states and mixtures are regarded as complete opposites . but this is not really correct . it is here not the place to go into more details . but in some respect it lies rather in the eye of the beholder . that is , it is the problem of dealing with the complete microscopic information of a state , or rather with some coarse - grained form . note that in our approach microscopic information is widely scattered via short - cuts or wormholes over essentially the whole space . i.e. , it is not fully accessible to a local observer . we recommend the study of some older classics on the ergodic theorem in quantum statistical mechanics ( @xcite,@xcite,@xcite ) . + v ) our analysis should be extended to more general space - times where possibly different distribution laws may show up . a.d.sakharov : `` vacuum theory in curved space - time and the theory of gravitation '' , sov.phys.dok . 12(1968)1040 a.d.sakharov : `` spectral density of eigenvalues of the wave equation and vacuum polarization '' , theor.math.phys . 23(176)435 y.b.zeldovich : `` the cosmological constant and the theory of elementary particles '' , sov.phys.usp . 11(1968)381 t.jacobson : `` black - hole entropy and induced gravity '' , gr - qc/9404039 j.d.bekenstein : `` do we understand black - hole entropy ? '' , seventh marcel - grossmann meeting , stanford 1994 , gr - qc/9409015 j.d.bekenstein : `` black - holes and entropy '' , phys.rev . d7(1973)2333 j.d.bekenstein : `` statistical black - hole thermodynamics '' phys.rev . d12(1975)3077 s.w.hawking : `` black holes and thermodynamics '' , phys.rev . d13(1976)191 g.t hooft : `` the black - hole interpretation of string theory '' , nucl.phys . b335(1990)138 l.susskind,l.thorlacius,r.uglum : `` the stretched horizon and black - hole complementarity '' , phys.rev . d48(1993)3743 , hep - th/9306069 s.w.hawking : `` break down of predictability in gravitational collapse '' , phys.rev . d14(1976)2460 s.giddings : `` comments on information loss and remnants '' , phys.rev . d49(1994)4078 g.t hooft : `` on the quantum structure of a black - hole '' . b256(1985)727 r.sorkin : `` ten theses on black - hole entropy '' , in proc.eur.sci.found.conf . on philosphy and found.issues in stat.phys . , utrecht , nov.2003 , publ . spec.iss.of stud.hist.phil.mod.phys . 36(2005)291 , hep - th/0504037 r.sorkin : `` the statistical mechanics of black - hole thermodynamics '' , chandra symp . chicago dec.1996 , p.177 , ed . r.m .. wald , univ.chic.pr . 1998 ) t.jacobson,d.marolf,c.rovelli : `` black hole entropy : inside or out ? '' , int.journ.theor.phys . 44(2005)1807 , hep - th/0501103 l.bombelli,k.rabinder,k.kaul,j.lee,r.sorkin : `` quantum source of entropy for black - holes '' , phys.rev . d34(1986)373 m.srednicki : `` entropy and area '' , phys.rev.lett . 71(1993)666 m.requardt : `` entanglement - entropy for groundstates , low - lying and highly excited eigenstates of general ( lattice-)hamiltonians '' , hep - th/0605142 r.m.wald : `` the thermodynamics of black holes '' , liv.rev.relativ . 4(2001 ) , 6 [ online article ] r.m.wald : `` gravitation , thermodynamics , and quantum theory '' , cqg 16(1999)a177 , arxiv : gr - qc/9901033 r.m.wald : `` black holes and thermodynamics '' , symposium on black holes and relativistic stars , chicago , dec . 1996 , arxiv : gr - qc/9702022 a.strominger,c.vafa : `` microscopic origin of the bekenstein - hawking entropy '' , pl b 379(1996)99 , arxiv : hep - th/9601029 j.m.maldacena : `` black holes in strin theory '' , ph.d . thesis , hep - th/9607235 g.t.horowitz : `` the origin of black hole entropy in string theory '' , talk given at pacific conference of gravitation and cosmology , seoul , south korea , 1996 , arxiv : gr - qc/9604051 g.t.horowitz,j.polchinski : `` a correspondence principle for black holes and strings '' , pr d 55(1997)6189 , arxiv : hep - th/9612146 a.w.peet : `` tasi lectures on black holes in string theory '' , tasi 1999 , arxiv : hep - th 0008 s.d.mathur : `` the fuzzball proposal for black holes '' , fortschr.phys . 53(2005)793 , arxiv : hep - th/0502050 a.ashtekar,j.baez,a.goricki,k.krasnov : `` quantum geometry and black hole entropy '' , prl 80(1998)904 c.rovelli : `` loop quantum gravity and black hole physics '' , helv.phys.act . 69(1996)582 , arxiv : gr - qc/9608032 c.j.isham : `` structural issues in quantum gravity '' , lecture given at the gr14-conference , florence 1995 , arxiv : gr - qc/9510063 g.t hooft : `` how does god play dice ? , ( pre ) determinism at the planck scale '' , an essay in honor of john s. bell , arxiv : hep - th/0104219 g.t hooft : `` quantum gravity as a dissipative deterministic system '' , cqg 16(1999)3263 , arxiv : gr - qc/9903084 g.t hooft : `` quantum information and information loss in general relativity '' , lectture held at the 5th symposion on foundations of quantum mechanics , tokyo , japan , august 1995 , arxiv : gr - qc/9509050 g.t hooft : `` dimensional reduction in quantum gravity '' , essay dedicated to abdus salam , arxiv : gr - qc/9310026 m.requardt : `` ( quantum ) space - time as a statistical geometry of lumps in random networks '' , cqg 17(2000)2029 , arxiv : gr - qc/9912059 m.requardt : `` a geometric renormalisation group in discrete quantum space - time '' , journ.math.phys . 43(2002)351 , gr - qc/0110077 m.requardt : `` the continuum limit of discrete geometries '' , int.j.geom.meth.mod.phys . 3(2006)285 , math - ph/0507017 t.nowotny,m.requardt : `` emergent properties in structurally dynamic cellular automata '' , j.cell.aut . 2(2007)273 , cond - mat/0611427 m.requardt : `` planck fluctuations , measurement uncertainties and the holographic principle '' , mod.phys.lett.a 22(2007)791 , gr - qc/0505019 m.requardt : `` about the minimal resolution of space - time grains in experimental quantum gravity '' , arxiv:0807.3619 r.bousso : `` the holographic principle '' , rev.mod.phys . 74(2002)825 , hep - th/0203101 r.brustein,a.yarom : `` thermodynamics and area in minkowski space '' , p.r . d 69(2004)064013 , arxiv : hep - th/0311029 , r.brustein,a.yarom,d.h.oaknin : `` implications of area scaling of quantum fluctuations '' , p.r . d 70(2004)044043 , arxiv : hep - th/0310091 , r.brustein,a.yarom : `` area - scaling of quantum fluctuations '' , nucl.phys . b 709(2005)391 , arxiv : hep - th/0401081 , r.brustein,a.yarom : `` dimensional reduction from entanglement in minkowski space '' , jhep 0501:046 ( 2005 ) , arxiv : hep - th/0302186 m.requardt : `` symmetry conservation and integrals over local charge densities in quantum field theory '' , comm.math.phys . 50(1976)259 m.requardt : `` fluctuation operators and spontaneous symmetry breaking '' , j.math.phys . 43(2002)351 , arxiv : math - ph/0003012 l.susskind : `` the world as a hologram '' , journ.math.phys . 36(1995)6377 , hep - th/9409089 s.e.rugh,h.zinkernagel,t.y.cao : `` the casimir effect and the interpretation of the vacuum '' , stud.hist.phil.mod.phys . 30(1999)111 s.e.rugh,h.zinkernagel : `` the quantum vacuum and the cosmological constant problem '' , stud.hist.phil.mod.phys . 33(2002)663 w.nernst : `` ueber einen versuch , von quantenmechanischen betrachtungen zur annahme stetiger energieaenderungen zurueckzukehren '' , verh.deutsch.phys.ges . 18(1916)83 c.p.enz,a.thellung : `` nullpunktsenergie und anordnung nichtvertauschbarer faktoren i m hamiltonoperator '' , helv.phys.act . 33(1960)839 t.h.boyer : `` quantum zero - point energy and long - range forces '' , ann.phys . 56(1970)474 s.weinberg : `` the cosmological constant problem '' , rev.mod.phys . 61(1989)1 n.straumann : `` on the mystery of the cosmic vacuum energy density '' , eur.journ.phys . 20(1999)419 , astro - ph/0009386 s.w.hawking : `` particle creation by black holes '' , comm.math.phys . 43(1975)199 h.bauer : `` mass- und integrationstheorie '' , de gruyter , berlin 1990 w.feller : `` probability theory '' , wiley , n.y . 1957 r.f.streater,a.s.wightman : `` pct , spin & statistics and all that '' , benjamin , n.y . 1964 j.d.bjorken,s.d.drell : `` relativistic quantum fields '' , mcgraw - hill , n.y . 1965 c.itzykson,j.-b.zuber : `` quantum field theory '' , mcgraw - hill , n.y . 1985 h.reeh : private communication s.j.summers : `` yet more ado about nothing , the remarkable relativistic vacuum state '' , arxiv:0802.1854 h.epstein,v.glaser,a.jaffe : `` nonpositivity of the energy density in quantized field theories '' , nuov.cim . xxxvi ( 1965)1016 a.anderson,b.dewitt : `` does the topology of space fluctuate ? '' , found.phys . 16(1986)91 b.bollobas : `` modern graph theory '' , springer , n.y . 1998 , `` random graphs '' , cambr.univ.pr . , cambridge 2001 t.nowotny,m.requardt : `` dimension theory on graphs and networks '' , j.phys.a:math.gen . 31(1998)2447 , arxiv : hep - th/9707082 m.requardt : `` wormhole spaces , connes points speaking to each other , and the translocal structure of quantum theory '' , arxiv : hep - th/0205168 d.j.watts : `` small worlds , the dynamics of networks between order and randomness '' , princt.univ.pr . , princeton 1999 s.h.strogatz : `` sync : the emerging science of spontaneous order '' , penguin , n.y . 2003 m.s.granovetter : `` the structure of weak ties '' , sociol.th . 1(1983)203 a.lochmann,m.requardt : `` an analysis of the transition zone between the various scaling regimes in the small world model '' , j.stat.phys . 122(2006)255 , arxiv : cond - mat/0409710 e.niebuhr,h.g.schuster,d.m.kammen,c.koch : `` oscillator - phase coupling for different two - dimensional network connectivities '' , pr a44(1991)6895 s.w.hawking : `` wormholes in spacetime '' , pr d 37(1988)904 s.coleman : `` why there is nothing rahter than something '' , nucl.phys . b 310(1988)643 i.klebanov,l.susskind,t.banks : `` wormholes and the cosmological constant '' , nucl.phys . b 317(1989)665 j.preskill : `` wormholes in spacetime and the constants of nature '' , nucl.phys . b 323(1989)141 w.g.unruh : `` quantum coherence , wormholes and the cosmological constant '' , pr d 40(1989)1053 s.w.hawking : `` do wormholes fix the constants of nature '' , nucl.phys . b 335(1990)155 s.m.morris,k.s.thorne : `` wormholes in spacetime '' , am.j.phys . 56(1988)395 , s.m.morris,k.s.thorne,u.yurtsever : `` wormholes , time machines , and the weak energy condition '' , prl 61(1988)1446 s.v.krasnikov : `` toward a traversable wormhole '' , arxiv : gr - qc/0003092 d.hochberg,t.w.kephart : `` wormhole cosmology and the horizon problem '' , prl 70(1993)2665 m.visser : `` lorentzian wormholes '' , springer , berlin 1996 i.h.redmount,wai-mo suen : `` is quantum spacetime foam unstable ? '' , pr d 47(1993)r2163 , `` quantum dynamics of lorentzian spacetime foam '' , pr d 49(1994)5199 c.j.fewster,th.a.roman : `` on wormholes with arbitrarily small quantities of exotic matter '' , arxiv : gr - qc/0507013 g.preparata,s.rovelli,s.-s.xue : `` gas of wormholes : a possible ground state of quantum gravity '' , gen.rel.grav . 32(2000)1859 , arxiv : gr - qc/9806044 r.garattini : `` large n - wormhole approach to space - time foam '' , phys.lett . b 446(1999)135 , arxiv : hep - th/9811187 , `` space - time foam and vacuum energy '' , talk given at the spanish relativity meeting , mao , minorca , spain sept.2002 , arxiv : gr - qc/0212013 y.aharonov,d.rohrlich : `` quantum paradoxes '' , wiley , n.y . 2005 o.aharony,s.s.gubser,j.maldacena,h.oguri,y.oz : `` large n field theories , string theory and gravity '' , phys.rep . 323(2000)183 j.maldacena : `` the illusion of gravity '' , sci.am . nov . 2005 , 56 j.v.neumann . `` beweis des ergodensatzes und des h - theorems in der neuen mechanik '' , zeitschr.phys . 57(1929)30 w.pauli,m.fierz : `` ueber das h - theorem in der quanten mechanik '' , zeitschr.phys . 106(1937)572 n.g.vankampen : `` grundlagen der statistischen mechanik der irreversiblen prozesse '' , fortschr.phys . 4(1956)405 d.marolf : `` black holes , ads , and cfts '' , gen.rel.grav . 41:903,2009 , arxiv:0810.4886 ; s.d.h.hsu,d.reeb : `` unitarity and the hilbert space of quantum gravity '' , class.quant.grav . 25:235007,2008 , arxiv:0803.4212

the surplus of matter over anti - matter within the universe can only be explained with a thorough understanding of cp violation . the cp phase in the quark mixing matrix of the standard model , which has been confirmed by b - meson experiments @xcite , is not sufficient to understand the baryon asymmetry of the universe @xcite . however , the minimal supersymmetric standard model ( mssm ) @xcite provides new physical phases that are manifestly cp - sensitive . after absorbing non - physical phases , we chose the complex parameters to be the higgsino mass parameter @xmath2 , the u(1 ) , and su(3 ) gaugino mass parameters @xmath1 , and @xmath4 , and the trilinear scalar coupling parameters @xmath5 of the third generation sfermions @xmath6 . the corresponding phases violate cp and are generally constrained by experimental bounds on electric dipole moments ( edms ) @xcite . however , these restrictions are strongly model dependent @xcite , such that additional measurements outside the low energy edm sector are required . + many cp observables have been proposed and studied in order to measure cp violation . total cross sections @xcite , masses @xcite , and branching ratios @xcite , are cp - even quantities . for a direct evidence of cp violation , however , cp - odd ( t - odd ) observables are required . examples are rate asymmetries of either branching ratios @xcite , cross sections @xcite , or angular distributions @xcite . since these rate asymmetries require the presence of absorptive phases , they are typically small , of the order of @xmath7 , if they are not resonantly enhanced @xcite . larger cp - odd observables which already appear at tree - level are desirable . these are t - odd triple products of momenta and/or spins , from which cp - odd asymmetries can be constructed . such triple product asymmetries are highly cp - sensitive , and have been intensively studied both at lepton and hadron colliders @xcite . + third generation sfermions have a rich phenomenology at high energy colliders like the lhc @xcite or ilc @xcite due to a sizable mixing of left and right states . in addition , the cp phases of the trilinear coupling parameters @xmath5 are rather unconstrained by the edms @xcite . the phases of @xmath8 and @xmath9 have been studied in stop @xcite and sbottom @xcite decays , respectively . since these are decays of a scalar particle , the spin - spin correlations have to be taken into account . the triple product asymmetries can then be up to @xmath10 , for sizable squark mixing . similarly for probing the cp - violating phase of @xmath11 in the stau vertex , @xmath12-@xmath13-@xmath14 , it is essential to include the tau spin . only then is there a sensitivity to the phase of @xmath11 @xcite . if the spin of the tau is summed over , this crucial information is lost . triple product asymmetries including the tau polarization have been studied in neutralino decays @xmath15 @xcite , and also in chargino decays @xmath16 @xcite . it was shown that the normal tau polarization itself is cp - sensitive , and that the asymmetries are large and of the order of @xmath17 to @xmath3 . + we are thus motivated to study cp violation , including the tau polarization , in the two - body decay of a stau @xmath18 followed by the subsequent chain of two - body decays [ eq : decaychain ] @xmath19 see fig . [ feyn ] for a schematic picture of the entire stau decay . this process is kinematically open for a mass hierarchy @xmath20 where the staus are heavier than the smuons and selectrons . we thus work in mssm scenarios with heavier stau soft susy breaking parameters @xmath21 we show that the normal tau polarization , with respect to the plane spanned by the @xmath14 and @xmath22 momentum , is a triple product asymmetry which is sensitive to the phases of @xmath0 , @xmath1 , and @xmath2 in the stau - neutralino sector . for nearly degenerate stau masses , @xmath23 a strong stau mixing is obtained which results in tau polarization asymmetries of more than @xmath3 . this should be measurable at colliders . since the stau is a scalar particle , its particular production does not contribute to cp - sensitive spin - spin correlations , and can thus be considered separately . this allows a collider - independent study , where we only discuss the boost dependence of the cp asymmetries . + the paper is organized as follows . in section [ sec : formalism ] we review stau mixing and the stau - neutralino lagrangian with complex couplings . we calculate the amplitude squared for the entire stau decay in the spin - density matrix formalism @xcite . we construct the cp asymmetry from the normal tau polarization , and discuss its mssm parameter dependence , as well its boost dependence for colliders like the ilc and lhc . in section [ sec : results ] , we numerically study the phase and parameter dependence of the asymmetry , and the stau and neutralino branching ratios . we comment on the impact of the @xmath24 decay in scenarios with nearly degenerate stau masses . we summarize and conclude in section [ sec : summary ] . the appendices contain the definitions of momenta and spin vectors , the analytical expressions for the stau decay amplitudes in the spin - density matrix formalism , and formulae for the stau decay widths . in the complex mssm , the stau mixing matrix in the @xmath25-basis is @xcite @xmath26 cp violation is parameterized by the physical phase @xmath27 , \\ \lambda_{{\tilde{\tau } } } & = & a_\tau - \mu^ * \cot{\beta},\end{aligned}\ ] ] with the complex trilinear scalar coupling parameter @xmath0 , the complex higgisino mass parameter @xmath2 , and @xmath28 , the ratio of the vacuum expectation values of the two neutral higgs fields . the left and right stau masses are @xmath29 with the real soft susy breaking parameters @xmath30 , the electroweak mixing angle @xmath31 , and the masses of the @xmath32 boson @xmath33 , and of the tau lepton , @xmath34 . + in the mass basis , the stop eigenstates are @xcite @xmath35 with the diagonalization matrix @xmath36 and the stau mixing angle @xmath37 the stau mass eigenvalues are @xmath38 . \end{aligned}\ ] ] the relevant lagrangian terms for the stau decay @xmath39 are @xcite @xmath40 with @xmath41 , and the weak coupling constant @xmath42 , @xmath43 . the couplings are defined as @xcite @xmath44 the stau diagonalization matrix @xmath45 is given in eq . ( [ eq : decaychain ] ) , and @xmath46 h_{\tau i}^r \end{array}\right ) , \qquad { \cal b}_i^\tau \equiv \left(\begin{array}{ccc } h_{\tau i}^l\\[2 mm ] f_{\tau i}^r\end{array}\right ) . \label{eq : ab}\ ] ] in the photino , zino , higgsino basis ( @xmath47 ) , we have @xmath48 , \qquad \label{eq : flt}\\[2 mm ] f_{\tau i}^r & = & \sqrt{2 } \sin \theta_w \left(\tan\theta_w n_{i2}^*-n_{i1}^*\right ) , \label{eq : frt}\\[2 mm ] h_{\tau i}^l & = & ( h_{\tau i}^r)^{\ast } = -y_\tau ( n_{i3}^\ast\cos\beta+n_{i4}^\ast\sin\beta ) , \label{eq : hrlt}\\[2 mm ] y_\tau & = & \frac{m_\tau}{\sqrt{2}\,m_w \cos\beta } , \label{eq : yt}\end{aligned}\ ] ] with @xmath49 the mass of the @xmath50 boson , and @xmath51 the complex , unitary @xmath52 matrix that diagonalizes the neutralino mass matrix @xcite @xmath53 the interaction lagrangian relevant for the neutralino decay @xmath54 , for @xmath55 is @xcite @xmath56 with the couplings @xmath57 given in eqs . and . the unnormalized , @xmath58 , hermitian , @xmath14 spin density matrix for stau decay , eqs . and , reads @xmath59 with the amplitude @xmath60 , and the lorentz invariant phase space element @xmath61 , for details see appendix [ sec : phasespace ] . the @xmath14 helicities are denoted by @xmath62 and @xmath63 . in the spin density matrix formalism @xcite , the amplitude squared is given by @xmath64 & & \sum_{\lambda_i \lambda_i^\prime } \rho_d(\tilde\tau)^{\lambda_\tau\lambda_\tau^\prime } _ { \lambda_i\lambda_i^\prime } \;\rho_{d_1}({\tilde{\chi}^0_i})^{\lambda_i^\prime\lambda_i } \;d_2(\tilde\ell ) , \label{rhoampli}\end{aligned}\ ] ] with the neutralino helicities @xmath65 . the amplitude squared decomposes into the remnants of the propagators @xmath66 with mass @xmath67 , and width @xmath68 of particle @xmath69 or @xmath70 , and the unnormalized spin density matrices for stau decay @xmath71 , and neutralino decay @xmath72 . the decay matrix of the spinless slepton is a factor since the polarizations of the final lepton and lsp are not accessible . the corresponding amplitude is denoted by @xmath73 . defining a set of spin basis vectors @xmath74 for the tau , see eqs . in appendix [ sec : momenta ] , and @xmath75 for the neutralino @xcite , the spin density matrices can be expanded in terms of the pauli matrices @xmath76 @xmath77 \sigma_{\rm d}^b\,\delta^{\lambda_\tau\lambda_\tau^\prime } ( \sigma^b)_{\lambda_i\lambda_i^\prime } + \sigma_{\rm d}^{ab}\,(\sigma^a)^{\lambda_\tau\lambda_\tau^\prime } ( \sigma^b)_{\lambda_i\lambda_i^\prime } , \label{rhod}\\[2 mm ] \rho_{{\rm d}_1}({\tilde\chi}_i^0)^{\lambda_i^\prime\lambda_i}= { \rm d}_1\,\delta^{\lambda_i^\prime\lambda_i } + \sigma_{{\rm d}_1}^b\,(\sigma^b)^{\lambda_i^\prime\lambda_i } , \phantom{spacespace } \label{rhod1}\end{aligned}\ ] ] with an implicit sum over @xmath78 , respectively . the real expansion coefficients @xmath79 , @xmath80 , @xmath81 , @xmath82 , @xmath83 and @xmath84 contain the physical information of the process . @xmath79 denotes the unpolarized part of the amplitude for stau decay @xmath85 , @xmath86 denotes the unpolarized part for neutralino decay @xmath87 , respectively . @xmath88 gives the tau polarization , @xmath82 , and @xmath83 describe the contributions from the neutralino polarization , and @xmath89 is the spin - spin correlation term , which contains the cp - sensitive parts . we give the expansion coefficients explicitly in appendix [ sec : coefficients ] . + inserting the density matrices , eqs . and , into eq . , we get for the amplitude squared @xmath90{\rm d}_2,\quad \label{eq : nosum } \end{aligned}\ ] ] with an implicit sum over @xmath78 . the amplitude squared @xmath91 is now decomposed into an unpolarized part ( first summand ) , and into the part for the tau polarization ( second summand ) , in eq . . by using the completeness relations for the neutralino spin vectors , eq . ( [ eq : completeness2 ] ) , the products in eq . ( [ eq : nosum ] ) can be written , followed by @xmath92 . the signs in parentheses in eqs . ( [ eq : product1 ] ) and ( [ eq : product2 ] ) hold for the charge conjugated stau decay @xmath93 ; @xmath92 . in order to obtain the terms for the decay @xmath94 , however , followed by the neutralino decay into a positively charged slepton , @xmath95 , one has to reverse the signs of eqs . ( [ eq : product1 ] ) and ( [ eq : product2 ] ) . this is due to the sign change of @xmath96 , see eqs . ( [ eq : sigmabd1 ] ) . in appendix [ sec : coefficients ] , we also give the terms for the neutralino decay into a left slepton , @xmath97 . note that the term proportional to @xmath34 in eq . is negligible at high particle energies @xmath98 . ] , @xmath99 & & \left [ m_{\tilde\chi_i^0}^2(p_\tau\cdot p_{\ell_1 } ) -(p_{\tilde\chi_i^0}\cdot p_\tau)(p_{\ell_1}\cdot p_{\tilde\chi_i^0 } ) \right ] , \quad \label{eq : product1}\\ [ 2 mm ] \sigma^{ab}_{\rm d}~\sigma^{b}_{{\rm d}_1 } & = & \,^{\;\,+}_{(- ) } \frac{g^4}{2}\left(|a^{{\tilde{\tau}}}_{mi}|^2 + |b^{{\tilde{\tau}}}_{mi}|^2\right ) |f^{r}_{\ell i}|^2 m_\tau \times\nonumber\\[2 mm ] & & \left [ ( s^a_\tau \cdot p_{\tilde\chi_i^0})(p_{\tilde\chi_i^0}\cdot p_{\ell_1 } ) -m_{\tilde\chi_i^0}^2 ( s^a_\tau \cdot p_{\ell_1 } ) \right ] \nonumber \\ [ 2 mm ] & & \,^{\;\,+}_{(- ) } g^4 { \mathfrak{re}}\ { a^{{\tilde{\tau}}}_{mi } ( b^{{\tilde{\tau}}}_{mi})^\ast \ } |f^{r}_{\ell i}|^2 m_{\tilde\chi_i^0 } \times\nonumber\\ & & \left [ ( p_\tau \cdot p_{\tilde\chi_i^0})(s^a_\tau \cdot p_{\ell_1 } ) - ( p_\tau \cdot p_{\ell_1 } ) ( s^a_\tau \cdot p_{\tilde\chi_i^0 } ) \right ] \nonumber \\ [ 2 mm ] & & \,^{\;\,+}_{(- ) } g^4 |f^{r}_{\ell i}|^2 m_{\tilde\chi_i^0 } \times\nonumber\\[2 mm ] & & { \mathfrak{im}}\ { a^{{\tilde{\tau}}}_{mi } ( b^{{\tilde{\tau}}}_{mi})^\ast \ } [ p_{{\tilde{\tau}}},~ p_{\ell_1},~ p_{\tau } , ~s^a_\tau ] . \label{eq : product2}\end{aligned}\ ] ] the spin - spin correlation term @xmath100 , eq . , explicitly depends on the imaginary part @xmath101 of the stau - tau - neutralino couplings , eq . . thus this term is manifestly cp - sensitive , i.e. , it depends on the phases @xmath102 , @xmath103 , @xmath104 of the stau - tau - neutralino sector . the imaginary part is multiplied by the totally anti - symmetric ( epsilon ) product , @xmath105 \equiv \epsilon_{\mu\nu\rho\sigma } \ , p_{{\tilde{\tau}}}^\mu \ , p_{\ell_1}^\nu \ , p_{\tau}^\rho \ , s^{a , \sigma}_{\tau},\label{eq : epsilon}\ ] ] with the convention @xmath106 . since each of the spatial components of the four - momenta @xmath107 , or the spin vectors @xmath108 , changes sign under a time transformation , @xmath109 , the epsilon product @xmath110 is t - odd . in the stau rest frame , @xmath111 , the epsilon product reduces to the t - odd triple product @xmath112 @xmath113 = m_{{\tilde{\tau } } } \ ; \ , ( \mathbf{p}_{\ell_1 } \times \mathbf{p}_{\tau } ) \cdot \mathbf{s}^a_{\tau } \equiv m_{{\tilde{\tau } } } \ , { \mathcal t}^a . \label{eq : tripleterm}\ ] ] the task in the next section is to define an observable that projects out from the amplitude squared the part proportional to @xmath110 ( or @xmath112 ) , in order to probe the cp - sensitive coupling combination @xmath114 . the @xmath14 polarization is given by the expectation value of the pauli matrices @xmath76@xmath115 @xcite @xmath116 with the @xmath14 spin density matrix @xmath117 , as given in eq . . in our convention for the polarization vector @xmath118 @xmath119 , the components @xmath120 and @xmath121 are the transverse and longitudinal polarizations in the plane spanned by @xmath122 and @xmath123 , respectively , and @xmath124 is the polarization normal to that plane . see our definition of the tau spin basis vectors @xmath74 in appendix [ sec : momenta ] . + the normal @xmath14 polarization is equivalently defined as @xmath125 with the number of events @xmath51 with the @xmath14 spin up @xmath126 or down @xmath127 , with respect to the quantization axis @xmath128 , see eq . . the normal @xmath14 polarization can thus also be regarded as an asymmetry @xmath129 of the triple product @xmath130 where @xmath131@xmath132 is the direction of the @xmath14 spin vector for each event . the triple product @xmath133 is included in the spin - spin correlation term @xmath100 , eq . _ eq . , and the asymmetry thus probes the term which contains the cp - sensitive coupling combination @xmath114 . + since under naive time reversal , @xmath134 , the triple product @xmath133 changes sign , the tau polarization @xmath124 , eq . , is t - odd . due to cpt invariance @xcite , @xmath124 would thus be cp - odd at tree level . in general , @xmath124 also has contributions from absorptive phases , e.g. from intermediate @xmath135-state resonances or final - state interactions , which do not signal cp violation . although such absorptive contributions are a higher order effect , and thus expected to be small , they can be eliminated in the true cp asymmetry @xcite @xmath136 where @xmath137 is the normal tau polarization for the charged conjugated process @xmath93 . for our analysis at tree level , where no absorptive phases are present , we find @xmath138 , see the sign change in eqs . and , and thus @xmath139 . we study @xmath140 in the following , which is , however , equivalent to @xmath124 at tree level . + inserting now the explicit form of the density matrix @xmath117 , eq . , into eq . , together with eq . , we obtain the cp asymmetry @xmath141 where we have used the narrow width approximation for the propagators in the phase space element @xmath61 , see eq . . note that in the denominator of @xmath140 , eq . , the spin correlation terms vanish , @xmath142 , see eq . , when integrated over phase space . in the numerator only the spin - spin correlation term @xmath143 for @xmath144 contributes , which contains the t - odd epsilon product @xmath110 , see eq . . + to qualitatively understand the dependence of the asymmetry @xmath140 , eq . , on the mssm parameters , we study in some detail its dependence on the @xmath145-@xmath14-@xmath146 couplings , @xmath147 and @xmath148 , see eq . . from the explicit form of the decay terms @xmath149 eq . , and @xmath79 , @xmath150 , eqs . , , respectively , we find that the asymmetry @xmath151 \,{{\rm d}\mathscr{l}\!\textsl{ips } } } { ( p_{{\tilde\chi}_i^0}\cdot p_{\tau } ) ( p_{{\tilde\chi}_i^0}\cdot p_{\ell_1})\,\int{{\rm d}\mathscr{l}\!\textsl{ips } } } , \label{finalmathcala}\end{aligned}\ ] ] with @xmath152 , and @xmath153 , is proportional to the decay coupling factor @xmath154 with @xmath155 $ ] . we thus expect maximal asymmetries for equal moduli of left and right couplings , @xmath156 , which have a phase difference of about @xmath157 , where the coupling factor can be maximal @xmath158 , see eq . . + to study the dependence of @xmath159 on the cp phase @xmath160 of the stau sector , and the stau mixing angle @xmath161 , we expand the imaginary part of the product of @xmath145-@xmath14-@xmath146 couplings @xmath162 \big\},\end{aligned}\ ] ] in terms of the stau mixing matrix @xmath163 , the gauge couplings @xmath164 and the higgs couplings @xmath165 . in particular , for a cp - conserving neutralino sector , @xmath166 , we have @xmath167 , \nonumber\\ \label{eq : expectedeta}\end{aligned}\ ] ] for @xmath168 , and the sign in parentheses holds for @xmath169 . thus we expect a maximal @xmath159 and thus maximal asymmetries for maximal stau mixing and @xmath170 decay typically have similar magnitude but opposite sign , and thus might cancel . see the discussion at the end of the numerics in section [ sub : mlmr ] . ] , @xmath171 , and a maximal cp phase in the stau mixing matrix , @xmath172 . note that , in particular , the dependence of @xmath160 on @xmath102 is strong for @xmath173 . we will study numerically the phase and parameter dependence on @xmath140 and @xmath159 further in section [ sec : results ] . [ fig : lhcboosts ] the triple product asymmetry @xmath140 , eq . ( [ p2 ] ) , is not lorentz invariant but depends on the boost of the decaying stau , @xmath174 in fig . 2 , we show the boost dependence of the asymmetry @xmath175 , normalized by @xmath176 . the susy parameters are given in table [ tab : benchmark.scenario ] , and we have chosen three sets of different @xmath12 soft - breaking parameters @xmath177 gev ( solid , red ) ; @xmath178 gev ( dashed , green ) ; and @xmath179 gev ( dotted , blue ) . the corresponding stau masses are @xmath180 ; @xmath181 ; @xmath182 gev , respectively . the corresponding asymmetries in the stau rest frame are @xmath183 ; @xmath184 , @xmath185 . note that we have chosen nearly degenerate stau masses which lead to an enhanced stau mixing and thus to maximal asymmetries ; see also the discussion in section [ sec : results ] . for the stau masses @xmath186 gev , the staus can be produced at the ilc with @xmath187 gev , and have a fixed boost of @xmath188 . the corresponding asymmetry is then reduced to @xmath189 if the stau rest frame can not be reconstructed . typical ilc cross section for these masses are of the order of some @xmath190 fb @xcite . if the staus are produced at the lhc , they will have a distinct boost distribution depending on their mass , which typically peaks at high values @xmath191 for stau masses of the order of a few @xmath192 gev up to a @xmath193 tev , see e.g. refs @xcite . then the normal tau polarization in the laboratory frame is obtained by folding the boost dependent polarization @xmath140 with the normalized stau boost distribution @xcite , @xmath194 with the production cross section @xmath195 . the typical reduction of the normal tau polarization @xmath196 is of the order of two thirds of the asymmetry compared to that in the stau rest frame @xmath197 . however , it has been recently shown ( for similar asymmetries in stop decays at the lhc ) , that the rest frame can be partly reconstructed event by event using on - shell mass conditions , see refs . the lhc cross section for stau pair production , @xmath198 , also sensitively depends on the stau masses , e.g. , for our benchmark scenario in table [ tab : benchmark.scenario ] , we find cross sections up to @xmath199 fb at @xmath200 tev @xcite . . benchmark scenario . the mass parameters @xmath201 , @xmath202 , @xmath0 , @xmath203 , @xmath204 @xmath205 , and @xmath206 are given in gev . [ cols="^,<,^,<,^,<,^,<,^,<,^,<,^",options="header " , ] [ tab : benchmark.masses ] we quantitatively study the tau polarization asymmetry , and the branching ratios for the two - body decay chain @xmath207 for @xmath208 . the asymmetry probes the mssm phases @xmath209 , @xmath104 and @xmath102 , of the neutralino and stau sector . we center our numerical discussion around a general mssm benchmark scenario , see table [ tab : benchmark.scenario ] . we choose heavier soft breaking parameters in the stau sector than in the @xmath210 sector , to enable the mass hierarchy @xmath211 further we choose almost degenerate staus which enhances their mixing , leading to maximal asymmetries . we choose a large value of the trilinear scalar coupling parameter , @xmath173 is restricted bz the vacuum stability contidtion as @xmath212 @xcite . ] , to enhance the impact of @xmath102 in the stau sector . finally , to reduce the number of mssm parameters , we use the ( gut inspired ) relation @xmath213 @xcite for the gaugino mass parameters . the resulting masses of the staus , neutralinos and charginos are summarized in table [ tab : benchmark.masses ] . for the benchmark scenario given in table [ tab : benchmark.scenario ] , we study the phase dependence of the asymmetry @xmath140 in the stau rest frame . in fig . [ fig : phimu_m1_asym ] , we show the dependence on the cp phases in the neutralino sector , @xmath103 and @xmath104 . in fig . [ fig : phimu_atau_asym ] , we show the dependence on the phases in the stau secotor @xmath102 and @xmath104 . the asymmetry strongly depends on @xmath214 , which we expect for @xmath215 as in our benchmark scenario , see table [ tab : benchmark.scenario ] . in particular for @xmath216 in fig . [ fig : phimu_atau_asym ] , the asymmetry follows the approximation formula eq . , and attains its maximal values at @xmath217 . in fig . [ fig : a_tanbeta_asym ] , we show the @xmath218 and @xmath219 dependence of the asymmetry @xmath140 in the stau rest frame . we can observe that the asymmetry obtains its maximum , @xmath220 , where also the coupling factor is maximal , @xmath221 , see fig . [ fig : a_tanbeta_eta_pi4 ] . as discussed in subsection [ sub : analytics ] , the imaginary part of the product of the stau couplings @xmath222 is maximal for a maximal cp phase @xmath223 in the stau sector , which we show in fig . [ fig : a_tanbeta_phistau_pi4 ] . note that the location of the maximum of @xmath140 is not at maximal stau mixing , @xmath224 , since @xmath225 starts to decrease for increasing @xmath0 and @xmath219 . to study the stau mixing , we show the @xmath226@xmath227 dependence of the asymmetry @xmath140 in fig . [ fig : ml_mr_asym ] . in the entire @xmath226@xmath227 plane , the cp phase in the stau sector is almost maximal , @xmath228 . however , the asymmetry obtains its maxima in the small corridor @xmath229 , where the stau mixing is maximal , @xmath230 . we show the @xmath202@xmath201 dependence of the asymmetry @xmath140 in fig . [ fig : mu_m2_asym ] . the maxima of @xmath140 are obtained where the coupling factor @xmath159 is also maximal , see eq . ( [ etafactor ] ) . in fig . [ fig : mu_m2_br_stau ] , we show the corresponding stau branching ratio , @xmath231 , which can be as large as @xmath232% . other competing channels can reach @xmath233 , and @xmath234 . the stau decay into the chargino @xmath235 is always open since typically the second lightest neutralino and the lightest chargino are almost degenerate , @xmath236 . the neutralino branching ratio @xmath237 , summed over @xmath238 , is shown in fig . [ fig : mu_m2_br_chi ] , which reaches up to @xmath192% . the other important competing decay channels are @xmath239 , and @xmath240 , which open around @xmath241 gev and @xmath242 gev , respectively , for @xmath243 gev . note that in our benchmark scenario , see table [ tab : benchmark.scenario ] , we have @xmath244 . as we discussed in section [ sub : atautanb ] , we find large asymmetries for nearly degenerate staus , where we naturally obtain a maximal stau mixing . however , then typically the asymmetries for @xmath246 and @xmath170 decay are similar in magnitude , but opposite in sign . for example in our benchmark scenario we find @xmath247 for @xmath246 decay , but @xmath248 for the decay of @xmath170 . if the production and decay process of @xmath246 can not be experimentally disentangled from that of @xmath170 properly , the two asymmetries might cancel . we show their sum in fig . [ fig : ml - mr_asymm_k1-plus - k2 ] in the @xmath226@xmath227 plane . in fig . [ fig : ml - mr_mstau2-mstau1 ] , we show the corresponding stau mass splitting . note that also the stau branching ratios are similar in size ; for example in our benchmark scenario we have @xmath249 , and @xmath250 . for the @xmath226@xmath227 plane shown in fig . [ fig : ml_mr _ ] , the decay branching ratio @xmath251 is at least @xmath252 , and that of @xmath253 is larger by roughly a factor of @xmath254 to @xmath255 . we have analyzed the normal tau polarization and the corresponding cp asymmetry in the two - body decay chain of a stau @xmath256 the cp - sensitive parts appear only in the spin - spin correlations , which can be probed by the subsequent neutralino decay @xmath257 for @xmath208 . the t - odd tau polarization normal to the plane spanned by the @xmath14 and @xmath22 momenta , can then be used to define a cp - odd tau polarization asymmetry . it is based on a triple product , which probes the cp phases of the trilinear scalar coupling parameter @xmath0 , the higgsino mass parameter @xmath2 , and the u(1 ) gaugino mass parameter @xmath1 . + we have analyzed the analytical and numerical dependence of the asymmetry on these parameters in detail . in particular , for nearly degenerate staus where the stau mixing is strong , the asymmetry obtains its maxima and can be larger than @xmath3 . the normal tau polarization can thus be considered as an ideal cp observable to probe the cp phases in the stau and neutralino sector of the mssm . + since the cp - sensitive parts appear only in the subsequent stau decay products the stau production process can be separated . thus both , ilc , and lhc collider studies are possible . concerning the kinematical dependence , the asymmetry is not lorentz invariant , since it is based on a triple product . at the lhc , staus are produced with a distinct boost distribution . evaluated in the laboratory frame , the resulting tau polarization asymmetries get typically reduced by a factor of two thirds , compared to the stau rest frame . + we want to stress that a thorough experimental analysis , addressing background processes , detector properties , and event rate reconstruction efficiencies , will be needed in order to explore the measurability of cp phases in the stau sector at the lhc or ilc . we hope that our work motivates such a study . we thank m. drees and f. von der pahlen for enlightening discussions and helpful comments . this work has been supported by micinn project fpa.2006 - 05294 . am was supported by the _ konrad adenauer stiftung _ , bcgs , and a fellowship of bonn university . hd was supported by the hemholtz alliance `` physics at the terascale '' and bmbf `` verbundprojekt hep - theorie '' under the contract 0509pde . sk was supported by bcgs . ok acknowledges support from cpan . for the stau decay @xmath258 , we choose the coordinate frame in the laboratory ( lab ) system , such that the momentum of decaying @xmath259 points in the @xmath260-direction . @xmath261 p^{\mu}_{\tau } & = & e_\tau ( 1 , \sin\theta_{\tau } , 0 , \cos\theta_{\tau } ) , \ ] ] with the decay angle @xmath262 , and @xmath263 in the limit @xmath264 . the momenta of the leptons from the subsequent neutralino decay @xmath265 ; @xmath266 ( [ eq : staudec ] ) , can be parameterized by @xmath267 p^{\mu}_{\ell_2 } & = & e_{\ell_2 } ( 1 , \sin\theta_2\cos\phi_2 , \sin\theta_2\sin\phi_2,\cos\theta_2),\end{aligned}\ ] ] with the energies @xmath268 and the decay angles @xmath269 , @xmath270 , that is , @xmath271 \cos\theta_{d_2 } & = & \frac{({\mathbf p}_{{\tilde{\tau}}}-{\mathbf p}_{\tau}-{\mathbf p}_{\ell_1 } ) \cdot\hat{\mathbf p}_{\ell_2 } } { | { \mathbf p}_{{\tilde{\tau}}}-{\mathbf p}_{\tau}-{\mathbf p}_{\ell_1 } | } , \end{aligned}\ ] ] with the unit momentum vector @xmath272 . we define the tau spin vectors by @xmath273 s_{\tau}^{3,\mu}&=&\frac{1}{m_{\tau } } \left(|{\bf{p}}_{\tau}| , \frac{e_{\tau}}{|{\bf{p}}_{\tau}| } { \bf{p}}_{\tau}\right ) . \label{eq : spintau}\end{aligned}\ ] ] the spin vectors @xmath274 @xmath275 for the tau , and @xmath276 @xmath277 for the neutralino @xmath278 , fulfil completeness relations @xmath279 \sum_{b}s_{\tilde\chi_i^0}^{b,\,\mu } s_{\tilde\chi_i^0}^{b,\,\nu } & = & -g^{\mu\nu } + \frac{p_{\tilde\chi_i^0}^\mu p_{\tilde\chi_i^0}^\nu}{m_{\tilde\chi_i^0}^2 } , \label{eq : completeness2}\end{aligned}\ ] ] and they form orthonormal sets @xmath280 s_{{\tilde{\chi}^0_i}}^b \cdot s_{{\tilde{\chi}^0_i}}^c & = & -\delta^{bc } , \qquad s_{{\tilde{\chi}^0_i}}^b \cdot \hat{p}_{{\tilde{\chi}^0_i } } = 0,\end{aligned}\ ] ] with @xmath281 . note that the asymmetry @xmath140 , eq . ( [ p2 ] ) , does not depend on the explicit form of the neutralino spin vectors , since they are summed in the amplitude squared , see eq . ( [ eq : product1 ] ) , using the completeness relation . the lorentz invariant phase - space element for the stau decay chain , see eqs . - , can be decomposed into two - body phase - space elements @xcite @xmath282 \times { \rm d}s_{{\tilde{\chi}^0_i}}\ , { { \rm d}\mathscr{l}\!\textsl{ips}}(s_{{\tilde{\chi}^0_i}};\,p_{\ell_1},p_{\tilde\ell})\ , { \rm d}s_{\tilde\ell}\ , { { \rm d}\mathscr{l}\!\textsl{ips}}(s_{\tilde\ell};\,p_{\ell_2},p_{{\tilde\chi}_1 ^ 0}).\quad\end{aligned}\ ] ] the different contributions are @xmath283 with @xmath284 and @xmath285 . the coefficients of the stau decay matrix , eq . , are @xmath286 & & -g^2 { \mathfrak{re}}\{a_{mi}^{{\tilde{\tau } } } \,(b_{mi}^{{\tilde{\tau}}})^\ast\ } m_{{\tilde{\chi}^0_i}}m_\tau , \label{eq : d}\\ \sigma_{\rm d}^a & = & \,^{\;\,-}_{(+ ) } \frac{g^2}{2}\left(|a^{{\tilde{\tau}}}_{mi}|^2-|b^{{\tilde{\tau}}}_{mi}|^2\right ) m_\tau(p_{\tilde\chi_i^0}\cdot s^a_{\tau } ) , \label{eq : sigmaad}\\ \sigma_{\rm d}^b & = & \,^{\;\,-}_{(+ ) } \frac{g^2}{2 } \left(|a^{{\tilde{\tau}}}_{mi}|^2-|b^{{\tilde{\tau}}}_{mi}|^2\right ) m_{{\tilde{\chi}^0_i } } ( p_\tau\cdot s_{{\tilde{\chi}^0_i}}^b ) , \label{eq : sigmabd } \\ \sigma_{\rm d}^{ab } & = & \frac{g^2}{2}\left(|a^{{\tilde{\tau}}}_{mi}|^2 + |b^{{\tilde{\tau}}}_{mi}|^2\right ) ( s^a_{\tau}\cdot s^b_{{\tilde\chi_i^0}})m_\tau m_{\tilde\chi_i^0 } \nonumber \\ [ 2 mm ] & & + g^2 { \mathfrak{re}}\ { a^{{\tilde{\tau}}}_{mi } ( b^{{\tilde{\tau}}}_{mi})^\ast \}\times \nonumber \\ [ 2 mm ] & & \left [ ( s^a_{\tau}\cdot p_{\tilde\chi_i^0 } ) ( s^b_{\tilde\chi_i^0}\cdot p_\tau ) -(s^a_{\tau}\cdot s^b_{\tilde\chi_i^0 } ) ( p_{\tilde\chi_i^0 } \cdot p_\tau ) \right ] \nonumber \\ [ 2 mm ] & & -g^2 { \mathfrak{im}}\ { a^{{\tilde{\tau}}}_{mi } ( b^{{\tilde{\tau}}}_{mi})^\ast \ } [ s^{a}_{\tau},~ p_\tau,~ s^{b}_{\tilde\chi_i^0},~p_{\tilde\chi_i^0 } ] . \label{eq : sigmaabd } \ ] ] the formulas are given for the decay of a negatively charged stau , @xmath287 . the signs in parentheses hold for the charge conjugated decay @xmath93 . + note that the terms proportional to @xmath34 in eqs . , , and , are negligible at high particle energies @xmath98 , in particular @xmath88 can be neglected . + the coefficients of the @xmath288 decay matrix , eq . , are @xcite @xmath289 \sigma_{{\rm d}_1}^b&= & \,^{\;\,+}_{(- ) } g^2|f_{\ell i}^r|^2 m_{{\tilde{\chi}^0_i}}(s_{{\tilde{\chi}^0_i}}^b\cdot p_{\ell_1 } ) , \label{eq : sigmabd1}\end{aligned}\ ] ] and the selectron decay factor is @xmath290 the signs in parentheses hold for the charge conjugated processes , that is @xmath95 in eq . ( [ eq : sigmabd1 ] ) . + for the decay into a left slepton @xmath291 , eqs . ( [ eq : d1 ] ) , ( [ eq : sigmabd1 ] ) , and ( [ eq : d2 ] ) read @xcite @xmath292 \sigma^b_{{\rm d}_1 } & = & \,^{\;\,-}_{(+ ) } g^2 |f^{l}_{\ell i}|^2 m_{\tilde\chi_i^0 } ( s^b_{\tilde\chi_i^0 } \cdot p_{\ell_1 } ) , \label{eq : sigmabd1sell } \\ [ 2 mm ] { \rm d}_2&= & g^2 |f^{l}_{\ell 1}|^2 ( m_{\tilde\ell}^2-m_{\tilde\chi_1 ^ 0}^2 ) , \label{eq : d2sell}\end{aligned}\ ] ] respectively . the expressions for eqs . ( [ eq : product1 ] ) and ( [ eq : product2 ] ) have to be changed accordingly . the sign in parenthesis in eq . ( [ eq : sigmabd1sell ] ) holds for the charge conjugated process @xmath293 . the partial decay width for the decay @xmath294 in the stau rest frame is @xcite @xmath295 with the decay function @xmath296 given in eqs . ( [ eq : d ] ) , and the approximation @xmath297 . for the decay @xmath298 the width is @xcite @xmath299 with the stau - chargino - neutrino coupling @xcite @xmath300 and the stau diagonalization matrix @xmath301 , eq . ( [ eq : rstau ] ) , the yukawa coupling @xmath302 , eq . ( [ eq : yt ] ) , and the matrix @xmath303 , that diagonalizes the chargino matrix @xcite , @xmath304 the stau decay width for the entire decay chain , eqs . - , is then given by @xmath305 & = & \frac{1}{2m_{\tilde\tau } } \int\,|\mathcal{m}|^2\ , { { \rm d}\mathscr{l}\!\textsl{ips}}(s_{{\tilde{\tau}}};\,p_\tau , p_{\ell_1},p_{\ell_2},p_{\tilde\chi_1 ^ 0 } ) \\[2 mm ] & = & \gamma(\tilde\tau)\times { \rm br}(\tilde\tau\to\tau \tilde\chi_i^0 ) \times { \rm br}(\tilde\chi_i^0\to \ell_1\tilde\ell ) \nonumber\\ & & \times { \rm br}(\tilde\ell\to \ell_2\tilde\chi_1 ^ 0 ) , \\\end{aligned}\ ] ] with the phase - space element @xmath306 , as given in the appendix [ sec : momenta ] , the amplitude squared latexmath:[\[\begin{aligned } obtained from eqs . by summing the tau helicities @xmath62 , @xmath308 . the neutralino branching ratios are given , for example , in ref . @xcite , and we assume @xmath309 . we use the narrow width approximation for the propagators @xmath310 which is justified for @xmath311 , which holds in our case with @xmath312 . note , however , that in principle the naive @xmath313-expectation of the error can easily receive large off - shell corrections of an order of magnitude , and more , in particular at threshold , or due to interferences with other resonant , or non - resonant processes @xcite . + 0 belle collab . , a. abashian _ et al . _ , phys . lett . * 86 * , 2509 ( 2001 ) , hep - ex/0102018 ; + babar collab . , b. aubert _ et al . _ , phys . lett . * 86 * , 2515 ( 2001 ) , hep - ex/0102030 . a. d. sakharov , zh . pisma * 5 * , 32 ( 1967 ) , jetp lett . * 91b * , 24 ( 1967 ) . h. e. haber and g. l. kane , phys . * 117 * ( 1985 ) 75 ; + h. p. nilles , phys . * 110 * ( 1984 ) 1 ; + 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)_. t. ibrahim and p. nath , phys . d * 57 * ( 1998 ) 478 [ erratum - ibid . d * 58 * ( 1998 errat , d60,079903.1999 errat , d60,119901.1999 ) 019901 ] [ arxiv : hep - ph/9708456 ] ; phys . b * 418 * ( 1998 ) 98 [ arxiv : hep - ph/9707409 ] ; [ erratum - ibid . d * 60 * ( 1999 ) 099902 ] [ arxiv : hep - ph/9807501 ] ; phys . d * 61 * ( 2000 ) 093004 [ arxiv : hep - ph/9910553 ] ; + m. brhlik , g. j. good and g. l. kane , phys . d * 59 * ( 1999 ) 115004 [ arxiv : hep - ph/9810457 ] ; + s. yaser ayazi and y. farzan , phys . d * 74 * , 055008 ( 2006 ) [ arxiv : hep - ph/0605272 ] . see , e.g. , a. bartl , t. gajdosik , w. porod , p. stockinger and h. stremnitzer , phys . d * 60 * ( 1999 ) 073003 [ arxiv : hep - ph/9903402 ] ; + a. bartl , t. gajdosik , e. lunghi , a. masiero , w. porod , h. stremnitzer and o. vives , phys . d * 64 * ( 2001 ) 076009 [ arxiv : hep - ph/0103324 ] ; + l. mercolli and c. smith , nucl . b * 817 * , 1 ( 2009 ) [ arxiv:0902.1949 [ hep - ph ] ] ; + v. d. barger , t. falk , t. han , j. jiang , t. li and t. plehn , phys . d * 64 * ( 2001 ) 056007 [ arxiv : hep - ph/0101106 ] . a. bartl , w. majerotto , w. porod and d. wyler , phys . d * 68 * ( 2003 ) 053005 [ arxiv : hep - ph/0306050 ] ; + k. a. olive , m. pospelov , a. ritz and y. santoso , phys . d * 72 * ( 2005 ) 075001 [ arxiv : hep - ph/0506106 ] ; + j. r. ellis , j. s. lee and a. pilaftsis , jhep * 0810 * ( 2008 ) 049 [ arxiv:0808.1819 [ hep - ph ] ] ; + s. a. abel , a. dedes and h. k. dreiner , jhep * 0005 * ( 2000 ) 013 [ arxiv : hep - ph/9912429 ] . s. y. choi , a. djouadi , m. guchait , j. kalinowski , h. s. song and p. m. zerwas , eur . j. c * 14 * ( 2000 ) 535 [ arxiv : hep - ph/0002033 ] ; + j. l. kneur and g. moultaka , phys . d * 61 * ( 2000 ) 095003 , arxiv : hep - ph/9907360 . a. pilaftsis _ et al . _ , b * 435 * ( 1998 ) 88 , arxiv : hep - ph/9805373 ; + d. a. demir _ et al . _ , phys . d * 60 * ( 1999 ) 055006 , arxiv : hep - ph/9807336 ; + a. pilaftsis and c. e. m. wagner , nucl . b * 553 * ( 1999 ) 3 , arxiv : hep - ph/9902371 ; + s. y. choi , m. drees , and j. s. lee , phys . b * 481 * ( 2000 ) 57 , arxiv : hep - ph/0002287 ; + j. l. kneur and g. moultaka , phys . d * 59 * ( 1999 ) 015005 , arxiv : hep - ph/9807336 . a. bartl , s. hesselbach , k. hidaka , t. kernreiter and w. porod , phys . d * 70 * ( 2004 ) 035003 [ arxiv : hep - ph/0311338 ] ; + k. rolbiecki , j. tattersall and g. moortgat - pick , arxiv:0909.3196 [ hep - ph ] . h. eberl , t. gajdosik , w. majerotto and b. schrausser , phys . b * 618 * ( 2005 ) 171 [ arxiv : hep - ph/0502112 ] ; + e. chistova , h. eberl , w. majerotto , and s. kraml , nucl . b * 639 * ( 2002 ) 263 , arxiv : hep - ph/0205227 ; + e. chistova , h. eberl , w. majerotto , and s. kraml , jhep * 12 * ( 2002 ) 021 , arxiv : hep - ph/0211063 ; + m. frank and i. turan , phys . rev . d * 76 * ( 2007 ) 016001 , arxiv : hep - ph/0703184 ; + m. frank and i. turan , phys . d * 76 * ( 2007 ) 076008 [ arxiv:0708.0026 [ hep - ph ] ] . e. christova , h. eberl , e. ginina and w. majerotto , phys . d * 79 * , 096005 ( 2009 ) [ arxiv:0812.4392 [ hep - ph ] ] . e. christova , h. eberl , e. ginina and w. majerotto , jhep * 0702 * ( 2007 ) 075 [ arxiv : hep - ph/0612088 ] . a. pilaftsis , nucl . b * 504 * , 61 ( 1997 ) [ arxiv : hep - ph/9702393 ] ; + s. y. choi , j. kalinowski , y. liao and p. m. zerwas , eur . j. c * 40 * , 555 ( 2005 ) [ arxiv : hep - ph/0407347 ] ; + h. k. dreiner , o. kittel and f. von der pahlen , jhep * 0801 * , 017 ( 2008 ) [ arxiv:0711.2253 [ hep - ph ] ] ; + o. kittel and f. von der pahlen , jhep * 0808 * , 030 ( 2008 ) [ arxiv:0806.4534 [ hep - ph ] ] ; + m. nagashima , k. kiers , a. szynkman , d. london , j. hanchey and k. little , phys . d * 80 * , 095012 ( 2009 ) [ arxiv:0907.1063 [ hep - ph ] ] . for studies with neutralino 3-body decays at the ilc , see + y. kizukuri and n. oshimo , phys . b * 249 * ( 1990 ) 449 ; + s. y. choi , h. s. song and w. y. song , phys . d * 61 * ( 2000 ) 075004 [ arxiv : hep - ph/9907474 ] ; + for further studies with neutralino 2-body and 3-body decays at the ilc , see + a. bartl , h. fraas , o. kittel and w. majerotto , phys . d * 69 * ( 2004 ) 035007 [ arxiv : hep - ph/0308141 ] ; + a. bartl , h. fraas , o. kittel and w. majerotto , eur . j. c * 36 * ( 2004 ) 233 [ arxiv : hep - ph/0402016 ] ; + j. a. aguilar - saavedra , nucl . b * 697 * ( 2004 ) 207 [ arxiv : hep - ph/0404104 ] ; + a. bartl , h. fraas , s. hesselbach , k. hohenwarter - sodek and g. a. moortgat - pick , jhep * 0408 * ( 2004 ) 038 [ arxiv : hep - ph/0406190 ] ; + s. y. choi , b. c. chung , j. kalinowski , y. g. kim and k. rolbiecki , eur . j. c * 46 * ( 2006 ) 511 [ arxiv : hep - ph/0504122 ] ; + for studies with chargino 2-body decays at the ilc , see + a. bartl , h. fraas , o. kittel and w. majerotto , phys . b * 598 * ( 2004 ) 76 [ arxiv : hep - ph/0406309 ] ; + o. kittel , a. bartl , h. fraas and w. majerotto , phys . d * 70 * ( 2004 ) 115005 [ arxiv : hep - ph/0410054 ] ; + for studies with chargino 3-body decays at the ilc , see + y. kizukuri and n. oshimo , arxiv : hep - ph/9310224 ; + a. bartl , h. fraas , s. hesselbach , k. hohenwarter - sodek , t. kernreiter and g. moortgat - pick , eur . j. c * 51 * ( 2007 ) 149 [ arxiv : hep - ph/0608065 ] . for recent reviews see , for example , + g. moortgat - pick , k. rolbiecki , j. tattersall and p. wienemann , arxiv:0910.1371 [ hep - ph ] ; + o. kittel , arxiv:0904.3241 [ hep - ph ] . s. abdullin _ et al . _ [ cms collaboration ] , j. phys . g * 28 * ( 2002 ) 469 [ arxiv : hep - ph/9806366 ] ; + atlas collab . , _ atlas detector and physics performance . technical design report . vol . 2 _ , cern - lhcc-99 - 15 ; + g. weiglein _ et al . _ [ lhc / lc study group ] , arxiv : hep - ph/0410364 . j. brau _ et al . _ [ ilc collaboration ] , arxiv:0712.1950 [ physics.acc-ph ] ; + j. a. aguilar - saavedra _ et al . _ [ ecfa / desy lc physics working group ] , arxiv : hep - ph/0106315 ; + t. abe _ et al . _ [ american linear collider working group ] , arxiv : hep - ex/0106055 ; + k. abe _ et al . _ [ acfa linear collider working group ] , arxiv : hep - ph/0109166 ; + j. a. aguilar - saavedra _ et al . _ , eur . j. c * 46 * ( 2006 ) 43 [ arxiv : hep - ph/0511344 ] . y. k. semertzidis , nucl . * 131 * ( 2004 ) 244 [ arxiv : hep - ex/0401016 ] ; + j. r. ellis , s. ferrara and d. v. nanopoulos , phys . b * 114 * ( 1982 ) 231 ; + w. buchmuller and d. wyler , phys . b * 121 * ( 1983 ) 321 ; + f. del aguila , m. b. gavela , j. a. grifols and a. mendez , phys . b * 126 * ( 1983 ) 71 [ erratum - ibid . b * 129 * ( 1983 ) 473 ] ; + d. v. nanopoulos and m. srednicki , phys . b * 128 * ( 1983 ) 61 ; + m. dugan , b. grinstein and l. j. hall , nucl . b * 255 * ( 1985 ) 413 ; + c. s. huang and w. liao , phys . d * 62 * ( 2000 ) 016008 [ arxiv : hep - ph/0001174 ] . s. y. choi , m. drees and b. gaissmaier , phys . d * 70 * ( 2004 ) 014010 [ arxiv : hep - ph/0403054 ] . f. deppisch and o. kittel , jhep * 0909 * , 110 ( 2009 ) [ erratum - ibid . * 1003 * , 091 ( 2010 ) ] [ arxiv:0905.3088 [ hep - ph ] ] . g. moortgat - pick , k. rolbiecki , j. tattersall and p. wienemann , jhep * 1001 * , 004 ( 2010 ) [ arxiv:0908.2631 [ hep - ph ] ] ; + g. moortgat - pick , k. rolbiecki and j. tattersall , arxiv:1008.2206 [ hep - ph ] . a. bartl , e. christova , k. hohenwarter - sodek and t. kernreiter , phys . d * 70 * ( 2004 ) 095007 [ arxiv : hep - ph/0409060 ] . j. ellis , f. moortgat , g. moortgat - pick , j. m. smillie and j. tattersall , eur . j. c * 60 * , 633 ( 2009 ) [ arxiv:0809.1607 [ hep - ph ] ] . a. bartl , e. christova , k. hohenwarter - sodek and t. kernreiter , jhep * 0611 * ( 2006 ) 076 [ arxiv : hep - ph/0610234 ] . f. f. deppisch and o. kittel , jhep * 1006 * , 067 ( 2010 ) [ arxiv:1003.5186 [ hep - ph ] ] . a. bartl , h. fraas , t. kernreiter and o. kittel , eur . j. c * 33 * ( 2004 ) 433 [ arxiv : hep - ph/0306304 ] . a. bartl , t. kernreiter and o. kittel , phys . b * 578 * ( 2004 ) 341 [ arxiv : hep - ph/0309340 ] ; + o. kittel , arxiv : hep - ph/0311169 . h. dreiner , o. kittel and a. marold , arxiv:1001.4714 [ hep - ph ] . h. e. haber , proceedings of the 21st slac summer institute on particle physics , eds . l. deprocel , ch . dunwoodie , stanford 1993 , 231 . [ arxiv : hep - ph/9405376 ] . a. bartl , k. hidaka , t. kernreiter and w. porod , phys . b * 538 * ( 2002 ) 137 phys . d * 66 * ( 2002 ) 115009 [ arxiv : hep - ph/0207186 ] . o. kittel , [ arxiv : hep - ph/0504183 ] . f. m. renard , `` basics of electron positron collisions , '' _ dreux , france : editions frontieres ( 1981)_. g. luders , kong . sel . mat . fys . med . * 28n5 * ( 1954 ) 1 ; + w. pauli , `` niels bohr and the development of physics , '' _ new york , usa : mc graw - hill ( 1955 ) _ ; + r. jost , helv . acta * 30 * ( 1957 ) 409 ; + r. jost , helv . acta * 36 * ( 1963 ) 77 ; + r. f. streater and a. s. wightman , `` pct , spin and statistics , and all that , '' _ redwood city , usa : addison - wesley ( 1989 ) ( advanced book classics ) . _ j. alwall _ et al . _ , jhep * 0709 * , 028 ( 2007 ) [ arxiv:0706.2334 [ hep - ph ] ] . j. m. frere , d. r. t. jones , and s. rabi , nucl . phys . * b222 * ( 1983 ) 11 ; m. claudson , l. j. hall , and i. hinchliffe , nucl . * b228 * ( 1983 ) 501 ; c. kounnas , a. b. lahanas , d. v. nanopoulos , and m. quiros , nucl . * b236 * ( 1984 ) 438 ; j. f. gunion , h. e. haber , and m. sher , nucl . phys . * b306 * ( 1988 ) 1 . e. byckling , k. kajantie , `` particle kinematics , '' _ london , england : john wiley & sons ( 1973 ) _ ; + g. costa _ et al . _ , tepp `` kinematics and symmetries '' , bd.1 , ed . m. nicoli , _ paris , france : institut national de physique nuclaire et de physique des particules ( 1979)_. k. hagiwara _ et al . _ , phys . d * 73 * ( 2006 ) 055005 [ arxiv : hep - ph/0512260 ] ; + d. berdine , n. kauer and d. rainwater , phys . * 99 * ( 2007 ) 111601 [ arxiv : hep - ph/0703058 ; + n. kauer , phys . b * 649 * ( 2007 ) 413 [ arxiv : hep - ph/0703077 ] ; jhep * 0804 * ( 2008 ) 055 [ arxiv:0708.1161 [ hep - ph ] ] ; + c. f. uhlemann and n. kauer , nucl . b * 814 * ( 2009 ) 195 [ arxiv:0807.4112 [ hep - ph ] ] ; + m. a. gigg and p. richardson , arxiv:0805.3037 [ hep - ph ] .

in many respects , degenerate affine hecke algebras seem to be the key behind the fascinating structure of instanton partition functions of 4d @xmath0 gauge theories . the spherical hecke central ( shc ) algebra constructed in @xcite is a limit of a symmetrized double degenerate hecke algebra ( ddaha ) . it acts on instanton partition functions as the kanno - matsuo - zhang ( kmz ) transformation of the young diagrams summand , also called _ bifundamental contribution _ @xcite . shc representations contain @xmath2 subalgebras which is the main reason behind the agt correspondence @xcite . and indeed , the various proofs of agt correspondence exploit this underlying algebraic structure . for instance , a set of ( generalized ) jack polynomials have been used in the proofs @xcite based on the free field representation of cft conformal blocks @xcite . jack polynomials are known to be the eigenstates of the calogero - moser hamiltonian which is one of the shc generators in the polynomial representation relevant to the formal case of gauge groups @xmath3 with @xmath4 @xcite . the shc algebra has a hopf algebra structure , and it is possible to employ the comultiplication to define a representation of the generators relevant to the case of a higher number of colors ( @xmath5 ) . this leads to define generalized jack polynomials which diagonalize the calogero - moser hamiltonian acting in a tensor space @xcite . the first proof of agt correspondence is also based indirectly on shc . it uses a specific basis of the cft fock space formed by the aflt states . these states were defined such that the decomposition of conformal blocks coincides with the expression of nekrasov partition functions @xcite . they were later identified with the generalized jack polynomials @xcite using the free field construction presented in @xcite . ddaha , from which shc is constructed , is formed by dunkl operators , and contains an infinite number of commuting integrals of motion . it has a deep connection with the quantum inverse scattering method @xcite , which may lead to a better understanding of the instanton r - matrix proposed by smirnov @xcite and built upon the stable map introduced in @xcite . in this paper , we are interested in yet another aspect of shc . in @xcite , nekrasov and shatashvili ( ns ) have investigated a limit of the @xmath6-background in which one of the equivariant parameters tends to zero . they have proposed to express the gauge theory free energy in this limit using the solution of a non - linear integral equation ( nlie ) reminiscent of those derived from the thermodynamical bethe ansatz @xcite . this proposal was confirmed in @xcite , exploiting the mayer cluster expansion technique to perform the subtle limit . in a different study of the ns limit , bethe - like equations and tq - relations were also obtained @xcite . in both cases , the associated integrable model bare some similarities with the @xmath7 xxx spin chain , and the system of bosons with delta interaction studied by c. n. yang @xcite . the latter also describes the solution of the quantum non - linear schrdinger equation in a sector of fixed number of particles , and will be referred shortly as _ qnls_. both integrable models are related to hecke algebras , although in a different way . xxx spin chains ( with open boundaries ) are known to possess a yangian symmetry , which is a representation of the degenerate affine hecke algebra @xcite . on the other hand , the qnls hamiltonian can be diagonalized by a technique involving dunkl operators . in the case of periodic boundary conditions , these operators form with the affine weyl group a ddaha @xcite . in view of these results , it seems necessary to understand the fate of the shc algebra in the ns limit , and possibly relate it to the algebraic structure of xxx and qnls integrable models . this paper reports on the first step in this direction . we focus on a @xmath8 quiver gauge theory with @xmath9 gauge group , @xmath10 fundamental flavors at each node , and a bifundamental matter field . this theory is asymptotically conformal . we regard more precisely the bifundamental contribution to the nekrasov instanton partition function . this quantity depends on two sets of @xmath10 young tableaux , and is a building block for the instanton partition function of more general quivers . the kmz transformation represents the action of shc generators as a variation of the number of boxes in the young diagrams . in order to proceed to the ns limit , the action of these generators must be re - written in a suitable manner . this is done in the second section . in the third section , we recall the derivation of the bethe - like equations from the invariance of the young diagrams summands under a variation of boxes . it leads to identify the bethe roots with the instanton positions of the boxes on top of each column . this identification further provides an expression of the bifundamental contribution . we then construct a set of operators upon infinitesimal variations of the bethe roots , and show that they obey the shc commutation relations at first order in @xmath1 . we also recover the kmz transformation in this limit . finally , in a fourth section , we explain the connection with the outcome of the mayer cluster expansion , and tba - like nlie . technical details are gathered in the appendices . the main properties of the shc algebra can be found in @xcite which is the starting point for our rewriting process . in order to render the @xmath1 factors explicit , we will avoid the notations @xmath11 and @xmath12 , and use instead directly the @xmath6-background equivariant parameters @xmath13 and @xmath1 , together with the shortcut notation @xmath14 . these parameters are related to the previous quantities through @xmath15 and @xmath16 . let us emphasize that no limit is taken in this section , and expressions are exact in @xmath13 and @xmath1 . the building block of quiver partition functions is the so - called _ bifundamental _ contribution . it is associated to an arrow of the quiver between two nodes @xmath17 . it depends on the information at each node @xmath18 , encoded in an object @xmath19 , and a mass @xmath20 of bifundamental matter fields . here we focus on the @xmath8 quiver , with two nodes @xmath21 , and a single fundamental mass @xmath22 for which we drop the index @xmath23 . to each node @xmath18 corresponds a gauge group @xmath24 , and a vector of coulomb branch vevs @xmath25 with @xmath26 the color index . we further associate to a node @xmath18 the object denoted @xmath19 that consists in a set of @xmath27 young diagrams @xmath28 , together with the vector @xmath25 . young diagrams @xmath28 are made of @xmath29 columns @xmath30 with @xmath31 . the dual partition consists of the sequence of integers @xmath32 for @xmath33 . indices @xmath34 of @xmath30 and @xmath35 can be extended to infinity , setting the additional quantities to be zero . for simplicity , we assume that both nodes have the same gauge group , i.e. @xmath36 . the bifundamental contribution can be expressed in terms of the variables @xmath37 and the dual ones @xmath38 associated to @xmath19 , and defined as @xmath39 where @xmath40 spans the columns , and @xmath41 the rows of the diagram @xmath28 . it reads @xmath42=\dfrac{\prod_{l,(i , j)\in y_1}\prod_{l'=1}^{\nc}(\tt_{l , j}^{(1)}-t_{l',i}^{(2)}-\mu+\e_2)\prod_{l,(i , j)\in y_2}\prod_{l'=1}^{\nc}(\tt_{l , j}^{(2)}-t_{l',i}^{(1)}+\mu-\e_1)}{\prod_{a=1,2}\left[\prod_{l,(i , j)\in y_a}\prod_{l'=1}^{\nc}(\tt_{l , j}^{(a)}-t_{l',i}^{(a)}+\e_2)(\tt_{l , j}^{(a)}-t_{l',i}^{(a)}-\e_1)\right]^{1/2}},\ ] ] with @xmath43 the total number of boxes of young diagrams . this expression has been obtained from the equation ( 6 ) of @xcite under a rescaling of coulomb branch vevs and bifundamental mass : @xmath44 and @xmath45 . the variable @xmath46 corresponds to a shifted bifundamental mass , it is related to @xmath22 through @xmath47 . in the following we mostly focus on a single object @xmath48 and , to alleviate the notations , the label @xmath18 will be dropped when no ambiguity arise . the shc algebra found in @xcite is acting on states @xmath50 characterized by an object @xmath51 ( we dropped the node index @xmath18 ) . to each box @xmath52 is associated a triplet of indices @xmath53 such that @xmath54 is a box in the @xmath55th young diagram . in our reformulation of the shc algebra , a prominent role is played by the instanton position which is a map that associates to each box @xmath52 the complex number @xmath56 thus , the variables @xmath57 defined in [ def_tli ] correspond to the instanton positions of the box locations that lay on top of each column @xmath58 . similarly , @xmath59 is the instanton position of a box located directly on the right of each row . we denote @xmath60 ( resp . @xmath61 ) the set of boxes that can be added to ( resp . removed from ) the set @xmath60 is a subset of the locations with positions @xmath57 , and also of the locations with positions @xmath59 . likewise , the set @xmath61 is a subset of the locations with positions @xmath62 , and @xmath63 : @xmath64 note that for each young diagram the number of boxes one can add is always one plus the number we can remove : @xmath65 where @xmath10 is the number of young diagrams in @xmath51 . in the original paper @xcite , the kmz transformation is expressed using a decomposition of young diagrams into elementary rectangles . this rectangle decomposition is useful to characterize the coordinates of boxes in the sets @xmath60 and @xmath61 . however , it turns out that this information is not essential to the formulation of the kmz transformation . it is sufficient to use the instanton positions @xmath66 for @xmath67 or @xmath68 without specifying the coordinates of these boxes in the young diagrams . omitting this information results in simpler expressions for the shc generators and the kmz transformation . the transition between the notations of the kmz paper @xcite and ours is explained in the appendix . it may be summarized by the formulas [ transition_1 ] and [ transition_2 ] . one of the main simplifications in our formalism concerns the factors @xmath69 given by equ ( 12 ) and ( 13 ) in @xcite that can be merged into a single quantity , @xmath70 note that the term @xmath71 has to be removed from the left product if @xmath67 and the right one for @xmath68 . the expression remains valid if @xmath72 with both products complete . in the limit @xmath73 ( or @xmath74 ) , we trivially have @xmath75 . the factors @xmath76 play a major role in the definition of the shc algebra , and it is important to study their properties . they correspond to the residues at the poles of the function @xmath77 which can be decomposed as @xmath78 an infinite tower of identities can be obtained through an expansion at @xmath79 . the first identities are given in the appendix . operators of the shc algebra act on the states @xmath50 by adding or removing boxes . we denote @xmath80 and @xmath81 the object @xmath51 with a box @xmath67 ( resp @xmath68 ) added to ( removed from ) the set of young diagrams . the behavior of the coefficients @xmath76 under such a procedure is remarkably simple : @xmath82 where we introduced the function @xmath83 it is important to note that the quantity @xmath84 is independent of the young diagrams . let @xmath68 , the first formula in [ shift_l ] is valid if @xmath85 or @xmath86 . it breaks down if @xmath87 but @xmath88 which corresponds to the poles of @xmath84 for @xmath89 or @xmath1 ( boxes under @xmath90 or on the left ) . extending the definition of @xmath91 to @xmath92 , @xmath93 in [ shift_l ] has to be replaced by its residue divided by @xmath94 . when @xmath95 but @xmath96 , i.e. @xmath71 , we have to replace @xmath97 by one . the same comment is valid for @xmath67 : if @xmath98 but @xmath96 , it implies that @xmath99 or @xmath1 ( boxes above @xmath90 or on its right ) and the pole of @xmath93 has to be replaced by the residue divided by @xmath94 . for @xmath100 , we have @xmath101 . we are now ready to introduce the generators @xmath102 of the shc algebra . the index @xmath103 runs over all integers and is referred as the _ degree _ , whereas the second index @xmath22 is a positive integer called the _ order_. the action of these generators on the states @xmath50 is known in a closed form only for the operators of degrees @xmath104 and @xmath105 . for those operators we introduce the generating series @xmath106 the construction of these operators from the degenerate affine hecke algebra is done in @xcite . for @xmath107 , the generators @xmath108 in the polynomial representation are symmetric polynomials of the cherednik(-dunkl ) operators for the calogero - moser model . in particular , @xmath109 coincide with the hamiltonian of this integrable system , and the states @xmath50 can be identified with the eigenbasis of jack polynomials . the operators @xmath110 are built by exploiting the pieiri formula satisfied by jack polynomials @xcite . at @xmath111 , it is possible to exploit the hopf algebra structure of shc to define a tensor representation for @xmath108 @xcite . then , the states @xmath50 should be identified with the generalized jack polynomials which diagonalize the comultiplication of @xmath109 @xcite . the action of the generators on states @xmath50 is defined by the formulas ( 38)-(40 ) of @xcite , which can be rewritten as @xmath112 the commutation relations ( 22)-(23 ) of @xcite can be summed up to produce @xmath113=\pm\dfrac{d_{\pm1}(w)-d_{\pm1}(z)}{z - w},\quad [ d_{-1}(z),d_1(w)]=\dfrac{e(w)-e(z)}{z - w},\ ] ] where we defined another generating series : @xmath114 the specific properties of the shc algebra comes from the expressions of the generators @xmath115 as a function of degree zero operators . they are encoded in the identity @xmath116 with the functions @xmath117 the central charges @xmath118 are the miwa transformed of the coulomb branch vevs , @xmath119 it is possible to deduce the action of @xmath120 on states @xmath50 from [ rel_e_d0 ] and the action [ act_d ] of @xmath121 : ] ] @xmath122 with the function @xmath93 defined in [ def_rz ] . in the rhs , the product over @xmath123 and coulomb branch vevs reduces to the function @xmath124 as a result of the identity @xmath125 in the @xmath126 half - plane . in green , the generators forming @xmath121 and @xmath127 for which the action on the states @xmath50 is known . in blue and red , the modes of the @xmath128 current and stress - energy tensor respectively . black arrows represent the commutation relations [ def_hg].,width=340 ] operators of higher degree are constructed using the following commutation relations @xmath129 : @xmath130,\quad d_{\pm n , m}=\pm[d_{0,m+1},d_{\pm n,0}].\ ] ] from these commutation relations , the generators of degree zero and @xmath105 engenders the whole algebra . in particular , the @xmath128 and virasoro subalgebras are generated by the modes of the current @xmath131 and stress - energy tensor @xmath132 defined with generators of * order * zero and one : @xmath133 a summary of the different relations between the shc generators can be found in figure . the generators of virasoro and @xmath128 subalgebras are orthogonal to the operators composing the series @xmath121 and @xmath127 . the standard current and virasoro modes commutation relations can be recovered from [ comm_shc ] and [ def_hg ] . however , this nested calculation rapidly becomes impractical . the kmz transformation is generated by the operators @xmath134 in @xcite . here , we use instead the generating series @xmath135 in our notations , they act on the bifundamental contributions [ cz_yy ] as @xmath136&=\sum_{x\in a(y_1)}\dfrac{\l_x(y_1)}{z-\phi_x}\cz[y_1+x , y_2]-\sum_{x\in r(y_2)}\dfrac{\l_x(y_2)}{z-\mu-\phi_x}\cz[y_1,y_2-x].\\ \d_{+1,z}\cz[y_1,y_2]&=-\sum_{x\in r(y_1)}\dfrac{\l_x(y_1)}{z-\phi_x}\cz[y_1-x , y_2]+\sum_{x\in a(y_2)}\dfrac{\l_x(y_2)}{z+\e_+-\mu-\phi_x}\cz[y_1,y_2+x ] . \end{split}\end{aligned}\ ] ] the operator @xmath137 combines an action of @xmath138 on the state @xmath139 ( first term ) and an action of @xmath140 on @xmath141 ( second term ) . in @xmath142 , the actions of @xmath127 are exchanged : @xmath143 now acts on @xmath139 and @xmath144 on @xmath141 . it is possible to work at fixed box @xmath90 using a contour integral circling @xmath66 . for @xmath145 , we have @xmath146}=\pm\l_x(y_1)\cz[y_1\pm x , y_2].\ ] ] the covariance of the bifundamental contribution @xmath147 $ ] under the kmz transformation involves the functions @xmath148 note that the minus one in the rhs subtracts the pole at @xmath149 . under the kmz transformation , @xmath147 $ ] behaves as @xmath150=u_{\pm1,z}\cz[y_1,y_2].\ ] ] at fixed box @xmath151 this relation writes @xmath152=\dfrac{\prod_{y\in r(y_1)}(\phi_x-\phi_y-\e_+)\prod_{y\in a(y_2)}(\phi_x-\phi_y-\mu+\e_+)}{\prod_{{\genfrac{}{}{0pt}{}{y\in a(y_1)}{y\neq x}}}(\phi_x-\phi_y)\prod_{y\in r(y_2)}(\phi_x-\phi_y-\mu)}\cz[y_1,y_2].\ ] ] a similar relation can be obtained for @xmath153 by exchanging @xmath154 , and replacing @xmath155 . we now recall the derivation of the bethe equations in the ns limit along the lines of @xcite . the derivation exploits the covariance of summands of the partition function under variation of the number of boxes in young diagrams . we argue that the main contribution to the partition function comes from young diagrams with columns of infinite height , and multiplicity one . the profile of these diagrams is fully encoded in a set of bethe roots which will be used to express the partition function . in @xcite , nekrasov and okounkov have exposed a procedure to compute the instanton partition function in the seiberg - witten limit @xmath156 . in this limit , young diagrams become infinitely large , and are described by a continuous profile minimizing an effective action . this method was later extended to the ns limit @xmath157 in @xcite . it relies on the idea that young diagrams summations are dominated by a set of young diagrams with a specific profile . this profile is an extremum , which implies that small deformations of the profile , typically by adding or removing some boxes , vanish at first order . and indeed the requirement of invariance under adding or removing boxes leads to bethe - like equations that characterizes the ns limit @xcite . we review this method here . the instanton partition function of the @xmath8 quiver reads @xmath158=\sum_{y_1,y_2}q_1^{|y_1|}q_2^{|y_2|}\cz[m_1,y_1]\cz[y_1,y_2]\cz[y_2,m_2],\ ] ] where the sum is over all the possible realizations of @xmath159 for a fixed set of coulomb branch vevs . the objects @xmath160 encode the fundamental flavors content of the theory . they consist of @xmath161 empty young diagrams , and a vector @xmath162 with @xmath163 corresponding to the masses of the @xmath161 flavors in the fundamental representation . they are such that @xmath164 and @xmath165 . to be able to use the transformation properties of the previous subsection , we need to require @xmath166 . in addition , the bifundamental mass @xmath46 is equal to zero for both @xmath167 $ ] and @xmath168 $ ] . the variation of the summand under addition of a box @xmath151 can be computed using the kmz transformation [ cz_yy_shift1 ] of bifundamental contributions : @xmath169\cz[y_1+x , y_2]}{q_1^{|y_1|}\cz[m_1,y_1]\cz[y_1,y_2]}\\ = & -\dfrac{q_1}{\e_1\e_2}m_1(\phi_x)\dfrac{\prod_{y\in r(y_1)}(\phi_x-\phi_y)(\phi_x-\phi_y-\e_+)}{\prod_{{\genfrac{}{}{0pt}{}{y\in a(y_1)}{y\neq x}}}(\phi_x-\phi_y)(\phi_x-\phi_y+\e_+)}\dfrac{\prod_{y\in a(y_2)}(\phi_x-\phi_y-\mu+\e_+)}{\prod_{y\in r(y_2)}(\phi_x-\phi_y-\mu)}. \end{split}\end{aligned}\ ] ] the mass polynomial @xmath170 can be found in [ pot_quiver ] below , it is monic , with zeros at the value of the fundamental masses @xmath171 for @xmath163 . to recover the bethe equations from the condition @xmath172\cz[y_1+x , y_2]}{q_1^{|y_1|}\cz[m_1,y_1]\cz[y_1,y_2]}=1,\ ] ] we assume that the configurations of young diagrams that contribute in the ns limit are such that all columns have a different height . as a consequence , a box can be added or removed from any column . furthermore , the columns heights are sent to infinity , such that the products @xmath173 remain finite . we then identify the bethe roots with the instanton positions of the boxes on top of each column . these also coincide with the position of boxes that can be added and removed , up to a negligible shift in @xmath1 : @xmath174 such an identification was actually already made in @xcite . the assumption on the shape of young diagrams will be justified by the consistency observed with the results of the mayer cluster expansion . there is however a simple heuristic explanation : to keep @xmath57 finite as @xmath157 , we effectively rescaled the young diagrams columns by @xmath175 which increases the disparities between them . , which should correspond to strings solutions of the bethe equations . a priori , there are no such solutions for the bethe equations [ bethe_1 ] , but it could be interesting to perform a deeper analysis of some degenerate situations . ] there is however a small additional subtlety , since there is always one more box that can be added to a young diagram than that can be removed . for each @xmath19 , these @xmath176 extra boxes lie on the right of the diagrams , at positions @xmath177 . in [ id_sets ] , we have specified the set @xmath178 , excluding the extra boxes at position @xmath179 . if we are considering young diagrams with infinitely many columns , implying that the number of bethe roots is also infinite , these extra boxes can be neglected and @xmath180 . here , we will keep the number of columns finite , and @xmath29 will act as a cut - off , as in @xcite . to each node of the quiver is associated a set of bethe roots . for the @xmath8 quiver , we denote the two sets of bethe roots @xmath181 and @xmath182 for the nodes @xmath183 and @xmath184 respectively , with @xmath185 . from this identification , we find for @xmath186 that the rhs of [ var_cz_yy ] becomes comes from the box @xmath187 just below @xmath90 which is such that @xmath188 . ] @xmath189 the second equation is obtained from the variation of the profile of @xmath190 , it is identical to the first one with @xmath181 and @xmath182 exchanged , and @xmath191 . these two sets of equations form the so - called bethe - like equations of the @xmath8 quiver . the factors @xmath192 , given by ( @xmath193 ) @xmath194 disappear for infinitely large young diagrams . the young diagrams with infinite profiles are the natural states on which the shc generators act in the ns limit . let us consider a regular object @xmath51 ( with a finite number of boxes ) , and pick a box @xmath195 with coordinates @xmath54 . it is shown in the appendix that @xmath196 unless @xmath197 which gives @xmath198 . thus , the summation of boxes @xmath199 defining the action [ act_d ] of @xmath127 on the states @xmath50 can be replaced by a much simpler summation over the number of colors . the argument breaks down for young tableaux of infinite profile , for which the difference of height between two neighboring columns times @xmath1 becomes macroscopic . then , @xmath76 is of order @xmath200 for all boxes in @xmath201 , and the action of @xmath127 on the corresponding states involves a non - trivial summation over the whole set of bethe roots . to summarize , we have found that for any test function @xmath202 , we have for the ns limit of the object @xmath51 : @xmath203 in particular , the function @xmath124 becomes : @xmath204 the first factor tends to one if we send the cut - offs @xmath205 . in the ns limit , we rescale the generators @xmath206 to define @xmath207 with @xmath208 . it implies that the limit of the generating series [ def_d ] reads @xmath209 the commutation relations [ comm_shc ] become @xmath210=\pm\e_2\dfrac{d_{\pm1}(w)-d_{\pm1}(z)}{z - w},\quad [ d_{-1}(z),d_1(w)]=\e_2\dfrac{e(w)-e(z)}{z - w},\ ] ] with @xmath211 and @xmath212 . to obtain the relation between modes @xmath213 and @xmath214 from [ rel_e_d0 ] we need to take the limit of the functions @xmath215 and @xmath216 with @xmath217 fixed . : \end{split}\end{aligned}\ ] ] ] we deduce @xmath219d_{0,n}\right),\ ] ] with the function @xmath220 depending on the finite rescaled central charges @xmath221 : @xmath222 the current and virasoro modes scale as @xmath223 the rescaling of the liouville field by @xmath224 in the semiclassical limit indeed brings a factor @xmath1 to the virasoro modes since @xmath225 and @xmath226 . we note that most of the correlators are now vanishing at leading order in @xmath1 , including @xmath227=o(\e_2)$ ] . this is to be expected since @xmath1 plays the role of @xmath228 in the semiclassical limit of liouville theory . in this limit , a virasoro algebra with @xmath229 is still present , but the commutators must be replaced by poisson brackets @xcite . it would be extremely interesting to investigate this phenomenon more deeply , but it is out of the scope of this paper . our choice of rescaling for @xmath230 renders the kmz transformation [ kmz ] finite , @xmath231=u_{\pm1,z}\cz[y_1,y_2],\ ] ] where we absorbed the factor @xmath232 in the definition of @xmath233 , and the functions defined in [ def_u ] simplifies into @xmath234 with : @xmath235 the factors containing @xmath236 cancel each - other in the limit @xmath237 . the shc algebra is rather disappointing in the ns limit since all commutators vanish . it is however possible to obtain non - trivial commutation relations under the rescaling of @xmath238 by @xmath239 and @xmath240 by @xmath241 : @xmath242=\pm\dfrac{d_{\pm1}(w)-d_{\pm1}(z)}{z - w},\quad [ d_{-1}(z),d_1(w)]=\dfrac{e(w)-e(z)}{z - w}.\ ] ] the commutation relations for @xmath243 $ ] will also be of order one . note however that the relation [ rel_e_d0_ii ] between @xmath244 and @xmath214 generators simplifies : the rhs becomes @xmath220 which is independent of the @xmath214 generators : @xmath213 are now simple central charges . we expect this algebra to be realized by some operators of the integrable system . in that respect , it is satisfying to see that the infinite number of modes @xmath214 remain commuting , and might be identified with the conserved charges built upon dunkl operators in qnls . the previous description of the ns limit for the objects @xmath19 allows to derive the limit of the bifundamental contribution @xmath147 $ ] defined in [ cz_yy ] . this quantity depends on two sets of bethe roots , and will be denoted @xmath245 $ ] . the derivation is a bit lengthy , it is done in the appendix . the main technicality is to get rid of the dependence in the dual variables @xmath246 . the final expression is dependence since @xmath247=o(1)$ ] as @xmath157 . ] @xmath248&=\prod_{r=1}^{n_b^{(1)}}\prod_{l=1}^{\nc}\dfrac{g(u_r-\xi_l^{(2)}+m+\e_1/2)}{\sqrt{g(u_r-\xi_l^{(1)})g(u_r-\xi_l^{(1)}+\e_1)}}\prod_{r=1}^{n_b^{(2)}}\prod_{l=1}^{\nc}\dfrac{g(v_r-\xi_l^{(1)}-m+\e_1/2)}{\sqrt{g(v_r-\xi_l^{(2)})g(v_r-\xi_l^{(2)}+\e_1)}}\\ & \times \prod_{r , s=1}^{n_b^{(1)}}\left(\dfrac{g(u_r - u_s-\e_1)}{g(u_r - u_s+\e_1)}\right)^{1/4}\times\prod_{r , s=1}^{n_b^{(2)}}\left(\dfrac{g(v_r - v_s-\e_1)}{g(v_r - v_s+\e_1)}\right)^{1/4}\times\prod_{r=1}^{n_b^{(1)}}\prod_{s=1}^{n_b^{(2)}}\dfrac{g(u_r - v_s+m+\e_1/2)}{g(u_r - v_s+m-\e_1/2)}. \end{split}\end{aligned}\ ] ] this expression involves the function @xmath249 defined as @xmath250 . this function has a branch cut on the negative real axis , such that @xmath251 . it implies that the expression we provided for @xmath245 $ ] is ambiguous , and the sheet where @xmath90 lies in @xmath249 must be specified . in the next section , the logarithm of @xmath245 $ ] will be related to a yang - yang functional that involves , in addition to the two sets of bethe roots @xmath252 ( @xmath253 , @xmath254 ) , two sets of integers @xmath255 that can be associated to the sheets of the function @xmath249 . we will further justify that this ambiguity is irrelevant for the kmz transformation . the function @xmath249 satisfies the property @xmath256 at first order in @xmath1 . it allows to study the variation of @xmath257 $ ] under an infinitesimal shift of one bethe root : @xmath258 . the constant factors @xmath259 cancel each other in the ratios , and we get @xmath260}{z[u_r , v_s]}=\prod_{l=1}^{\nc}\dfrac{(u_{r'}-\xi_l^{(2)}+m+\e_1/2)^{\pm1}}{\left[(u_{r'}-\xi_l^{(1)})(u_{r'}-\xi_l^{(1)}+\e_1)\right]^{\pm\frac12}}\prod_{{\genfrac{}{}{0pt}{}{s'=1}{s'\neq { r'}}}}^{n_b^{(1)}}\left(\dfrac{u_{r'}-u_{s'}-\e_1}{u_{r'}-u_{s'}+\e_1}\right)^{\pm\frac12}\prod_{s'=1}^{n_b^{(2)}}\left(\dfrac{u_{r'}-v_{s'}+m+\e_1/2}{u_{r'}-v_{s'}+m-\e_1/2}\right)^{\pm1}.\ ] ] a similar expression is obtained by considering shifts of the bethe roots @xmath261 , exchanging the role of @xmath181 and @xmath261 and flipping the sign of the bifundamental mass . from now on , we send the cut - offs @xmath262 such that the first product in the previous expression vanishes . in order to define the shc generators , we need to introduce a decomposition of the function @xmath263 as a sum over its poles , as we did for @xmath124 in [ prop_l ] : @xmath264 the factors @xmath265 play a role equivalent to @xmath76 in the ns limit since @xmath266 . the information of the states @xmath50 is encoded in the young diagrams structure . in the ns limit , this information is rendered by the knowledge of the set of bethe roots @xmath267 . it is then natural to introduce a state @xmath268 which is an eigenstate of @xmath238 with the eigenvalue @xmath269 this definition of the action of @xmath238 will become clear in the next section . the second term in the rhs acts as a regulator to ensure @xmath270 at infinity . the generators constituting @xmath244 are also diagonal in the basis @xmath268 , with @xmath271|u>=\l(z)^2|u>\ ] ] inherited from [ vp_e0 ] . finally , the action of @xmath127 involves the addition or subtraction of boxes to the young diagrams . these variations of the columns height generates shifts of the bethe roots . using an exponential notation for shift operators , we define @xmath272 in this definition , we employed the dunkl operator @xmath273 such that @xmath274 . it satisfies with the shift operators the commutation relation @xmath275=\pm\e_2e^{\pm\e_2\p_r}$ ] . the operator @xmath276 is obtained from @xmath265 by replacing @xmath181 with @xmath273 , it satisfies @xmath277=\pm\e_2e^{\pm\e_2\p_s}\p_s\hat\l_r$ ] . an explicit calculation shows that the operators @xmath278 obey the commutation relations [ comm_d ] of shc at first order in @xmath1 . in addition , expanding over @xmath279 , one can show that the generators @xmath214 and @xmath213 are related as in [ rel_e_d0_ii ] . gives @xmath280|u>,\ ] ] with @xmath281 for @xmath282 with @xmath283 and @xmath284 . this expression can be plugged into [ rel_e_d0_ii ] , and the summation over @xmath103 performed with the help of [ form_gn ] . it reproduces the function @xmath263 . ] the action of @xmath240 on the bifundamental contributions is obtained from [ shift_z ] . in this expression , one recognize the decomposition of @xmath285 over the poles at @xmath286 . defining the kmz transformation as in [ kmz_cz ] : @xmath287 where we indicated on which set of bethe roots the operator is actually acting , we recover exactly the kmz transformation [ kmz_ns ] . it is interesting to note that in order to satisfy the shc commutation relations , it is sufficient to expand in @xmath1 the shift operators in the definition of @xmath240 , and keep only the terms of order @xmath288 . however , the full exponential is needed to obtain the kmz transformation . the instanton partition function of @xmath0 gauge theories has been originally obtained as a set of coupled contour integrals @xcite . in the case of the @xmath8 quiver , @xmath289 with @xmath290 . the integration contours over the instanton positions @xmath291 lie along the real axis . they are closed in the upper half plane but avoid possible singularities at infinity . the potential attached to a node @xmath18 is a ratio involving the mass polynomial @xmath170 and the gauge polynomial @xmath292 . these monic polynomials have zeros respectively at the value of the masses @xmath293 of fundamental matter fields , and at the coulomb branch vevs @xmath25 : @xmath294 as in the first section , we have assumed that both nodes have a gauge group @xmath295 and @xmath176 fundamental flavors . the kernel @xmath296 is associated to the nodes , and @xmath297 to the arrow @xmath298 . the latter depends on the bifundamental mass @xmath299 . the expressions of the kernels @xmath296 and @xmath297 can be found in @xcite , they are reproduced in the appendix . they both depend on the @xmath300-background equivariant parameters @xmath301 which are assumed purely imaginary . coulomb branch vevs have a positive imaginary part , and fundamental masses @xmath293 are real . the integrals can be evaluated as a sum over residues , and poles are known to be in one to one correspondence with the boxes of the set of objects @xmath19 through the map [ map_phix ] . the resulting expression for the @xmath8 quiver has been given in [ decomp_cz ] . however , the integral definition reveals more convenient for taking the ns limit . this procedure has been described in @xcite . we employ here the method of @xcite , which has been performed only in the case of a single gauge group . it is easily extended to arbitrary quivers , this is done in the appendix . in the appendix , two different expressions of the partition function in the ns limit are provided ( equ [ sgc_onshell ] and [ cluster_hadron ] ) . the first one will be studied in the next subsection , it reproduces the familiar form of the yang - yang functional @xcite . the underlying bethe equations are related to those derived in the third section , thus bringing a justification for the assumption made on the shape of young diagrams in the ns limit . the bifundamental contribution @xmath245 $ ] given in [ zuv ] can be recovered from the yang - yang functional , and we will be able to comment on its ambiguities . the second expression for the ns partition function is derived in the appendix , and will be studied in the subsection . it consists of a sum of coupled integrals reminding of a matrix model . the integration variables are interpreted as the position of hadrons in a one dimensional space . they can be formally related to the bethe roots , and we will deduce the eigenvalues [ d0_ur ] of the operator @xmath238 . in @xcite , it was claimed that the free energy at first order in @xmath1 can be expressed as an on - shell effective action . this result is derived for the @xmath8 quiver in the appendix . the action [ sgc_onshell ] depends on two fields indexed by the node @xmath18 : the density @xmath302 and the pseudo - energy @xmath303 . the equations of motion relate these two fields as @xmath304 and produce for the fields @xmath303 a set of nlie , @xmath305 with the kernels @xmath306 and @xmath307 , @xmath308 in the nlie [ nlie_quiver ] , the integration is done over the same contour as in the original definition [ cz_a2 ] of the partition function @xmath309 . the set of nlie [ nlie_quiver ] is characteristic of the thermodynamical bethe ansatz technique developed in @xcite . it is usually associated to the following system of bethe equations , @xmath310 indeed , the method elaborated in @xcite shows that the counting functions @xmath311 obey the nlie [ nlie_quiver ] with an appropriate contour of integration . are defined such that they take integer values at the bethe roots @xmath312 , for instance @xmath313 they are identified with the pseudo - energy as @xmath314 . ] this method is briefly reviewed in @xcite . in particular , it was observed that the derivative of the field @xmath302 correspond to a density of bethe roots : ] for @xmath90 inside the integration contour ( i.e. @xmath315 ) which will always be the case . ] @xmath316 with the shortcut notations @xmath317 and @xmath318 . comparing the bethe equations [ bethe_2 ] with those obtained in the second section , we observe that they match only in the formal case @xmath319 . the mismatch comes from missing factors in the node potential [ bethe_1 ] . however , if we decompose the set of bethe roots @xmath320 into a set of fixed variables @xmath321 with @xmath40 and @xmath322 , and a set of free variables @xmath323 , then we observe that the bethe equations [ bethe_1 ] for @xmath320 become the bethe equations [ bethe_2 ] for @xmath323 . thus , to each solution @xmath324 of the bethe equations [ bethe_2 ] corresponds a solution @xmath325 of [ bethe_1 ] with the additional roots given by @xmath326 . it results from this identification that the relation [ id_densities ] among densities is modified for the bethe roots @xmath252 , @xmath327 inserted in the integrals of [ nlie_quiver ] through the relation [ def_ve ] , this amended identity gives an expression of the pseudo - energy in terms of the bethe roots @xmath252 where @xmath328 is replaced with @xmath329 .. \ ] ] ] it allows to identify @xmath314 with the counting function associated to the bethe equations [ bethe_1 ] . those equations are obtained by imposing @xmath330 , i.e. @xmath331 . interestingly , the correction term in [ id_densities_ii ] reproduces the part of the density involved in the perturbative contributions to the gauge theory partition function @xcite . it allows to interpret the bethe equations [ bethe_1 ] of the second section as the extremum of the full @xmath0 partition function , including the perturbative contribution . is not present . ] on the other hand , the bethe equations [ bethe_2 ] obtained here describe the extremization of the instanton contribution only . we now examine the free energy @xmath332 derived from the on - shell action [ sgc_onshell ] . instead of the pseudo - energies @xmath303 , we prefer to work with the counting functions @xmath311 , @xmath333+\cf_{12}[\rho_1,\rho_2]+\sum_{a=1,2}\sj[\rho_a,\eta_a].\ ] ] in the rhs , we have singled out the parts that depend on @xmath334 : @xmath335=-2i\pi\int{\rho(x)\eta(x)dx}+\dfrac1{2i\pi}\int{\text{li}_2\left(e^{2i\pi\eta(x)}\right)dx},\ ] ] and decomposed the remaining term into three contributions : @xmath336&=\dfrac14\int{\rho_a(x)\rho_a(y)g(x - y)dxdy}+\int{\rho_a(x)\log\left(\dfrac{q_am_a(x)}{\sqrt{a_a(x)a_a(x+\e_1)}}\right)}dx,\\ \cf_{12}[\rho_1,\rho_2]&=\dfrac14\sum_{a=1,2}\int{\rho_a(x)\rho_a(y)g(x - y)dxdy}+\int{\rho_1(x)\rho_2(y)g_{12}(x - y)dxdy}\\ & + \sum_{a=1,2}\int{\rho_a(x)\log\left(\dfrac{a_{b\neq a}(x+m_{ab}+\e_1/2)}{\sqrt{a_a(x)a_a(x+\e_1)}}\right)}dx . \end{split}\end{aligned}\ ] ] in these expressions , the densities @xmath302 can be replaced by the bethe roots density @xmath337 using an integration by parts and the relation [ id_densities_ii ] , leading to summations over bethe roots @xmath181 and @xmath261 . in particular , the exponential of @xmath338 $ ] reproduces the expression found in [ zuv ] for @xmath245 $ ] , @xmath339}\equiv z[u , v],\ ] ] up to ambiguities of the same nature as in section three . similarly , @xmath340 $ ] reproduces the ns limit of the matter bifundamental contributions @xmath341 $ ] . it is noted that , if there are several solutions to the bethe equations , the partition function is a sum over the contributions of each solution . we have thus shown that @xmath342 and @xmath343 reproduce the partition function of the @xmath8 quiver derived in the second section . but we have here an additional term @xmath344 $ ] that remains to be treated . to do so , we repeat the trick used in @xcite : in [ cs_rho_eta ] , the second integral simplifies using an integration by parts and the relation between the density and the counting function derived from [ def_ve ] . one of the terms obtained cancel with the first term in the expression of @xmath344 $ ] , and it only remains @xmath345=2i\pi\int{x\rho'(x)\eta(x)dx}.\ ] ] for the first node , it gives from [ id_densities_ii ] in terms of the bethe roots : @xmath346=-2i\pi\sum_ru_r\eta_r^{(1)}+2i\pi\sum_{l=1}^{\nc}\sum_{i=1}^{n_l^{(1)}}\bx_{l , i}^{(1)}\eta_1(\bx_{l , i}^{(1)}).\ ] ] with @xmath347 . the second term in the rhs is independent of the bethe roots , and simply produces an overall constant factor . the first term is more interesting since it has the same form as the ambiguity arising in [ zuv ] from @xmath348 rotations of the argument of the functions @xmath249 . since @xmath349 is an integer , a shift of @xmath350 leads to a variation @xmath351 which vanishes when exponentiated with a factor @xmath175 . thus , this ambiguity does not interfere with the kmz transformation we have defined . the full expression @xmath352 $ ] reproduces the yang - yang functional of qnls , up to the potential term . thus , [ d0_ur ] and [ d1_ur ] define an action of the shc operators at first order in @xmath1 on the yang - yang functional of an integrable system . another expression of the partition function @xmath309 in the ns limit has been obtained in the appendix , @xmath353 with the integrand @xmath354 this expression is a consequence of a phenomenon of _ clustering _ or _ confinement _ of the instantons as @xmath157 . indeed , the integrations in the original expressions [ def_cza2 ] are done over the positions of @xmath355 instantons denoted @xmath291 . it is common in the study of random matrix models to see the integration variables @xmath291 as pseudo - particles of color @xmath18 , in an external potential @xmath356 , and interacting through the kernels @xmath296 and @xmath297 . in the ns limit , a number @xmath357 of instantons at positions @xmath358 ( @xmath359 ) may become very close to each other due to the infinitely strong short - range interaction of @xmath360 . these instantons , at a distance @xmath361 of each - other , can be approximated by a single particle of _ charge _ @xmath357 and _ center of mass _ @xmath362 . by analogy with quarks bound states , it will be called a _ this phenomenon is described with more details in @xcite . the factors @xmath363 in the summation [ cz_a2 ] are a consequence of the interaction between the elementary instantons composing a hadron . for the second node , hadrons positions are denoted @xmath364 and they have the charge @xmath365 . the equality [ cz_a2 ] holds at leading order in @xmath1 . both sides contain @xmath1 corrections that do not match . to extract the leading order in @xmath1 , ( generalized ) matrix model techniques can be employed @xcite . a mayer expansion can also be considered , the first order free energy being expressed as a sum over connected clusters with a tree structure , and vertices associated to hadrons . the free energy in the ns limit is equal to the legendre transform of the canonical free energy at large @xmath366 , where @xmath366 is the conjugate variable of the chemical potential @xmath367 . in this setting , only configurations @xmath368 with @xmath369 contribute to the leading order . the charges @xmath357 are supposed to be finite , and the sum is over @xmath370 . thus , we will consider that for any test function @xmath371 , the following summations are of order one : @xmath372 in the hadronic formulation , the pseudo - energy is seen as a lagrange multiplier enforcing the following identities , @xmath373 where the field @xmath302 is defined as the dressed vertex of the mayer expansion . although pertaining to different contexts , it is instructive to formally identify these densities with the fields of the previous subsection , and use [ id_densities_ii ] to provide a bridge between the expressions in terms of @xmath374 and @xmath181 variables . it is possible to give another argument in favor of the relation [ id_densities_ii ] between variables @xmath374 and @xmath181 . let @xmath371 be a test function without singularities in the upper half plane . is a polynomial in @xmath90 , or the generating function @xmath375 for @xmath376 . ] the summation [ sum_kfx ] of @xmath377 can be interpreted as an operator evaluated in the statistical average that defines the ns partition function [ cz_a2 ] . as such , it is the ns limit of the operator @xmath378 evaluated as the vev associated to the original partition function [ def_cza2 ] . this vev can also be computed as a sum over residues , and since @xmath371 has no singularities within the contour of integration , we find @xmath379 with the object @xmath380 defined in the first section , and the map @xmath381 in [ map_phix ] . thus , in the ns limit , we can formally replace @xmath382 and similarly for @xmath190 and the variables @xmath383 . the consistency with the identification of bethe roots as the position of boxes on young diagrams edges [ dico_ur ] that provides an alternative expression of the function @xmath124 as a product over the boxes in @xmath51 . as @xmath157 , the function @xmath93 defined in [ def_rz ] is expended as @xmath384 performing the ns limit of the lhs of [ rel_r_l ] as in [ dico_kx ] , and expressing the rhs with [ def_l ] in terms of bethe roots , we get @xmath385 in agreement with [ id_ur_kx ] . note that this identity has a well defined limit when @xmath386 . ] @xmath387 this is the equivalent of the relation [ id_densities_ii ] for the densities . boxes.,width=340 ] there is a simple heuristic interpretation of the relation we have observed between the variables @xmath374 and @xmath181 . it boils down to represent the young tableaux columns as made of an infinite number of hadrons . for simplicity , let us assume @xmath4 with a unique coulomb branch vev @xmath388 , and @xmath389 . the generalization to @xmath111 is straightforward . we drop the color and node indices , and consider the expression @xmath390=\sum_{r=1}^n [ f(\bx_r+\l_r\e_2)-f(\bx_r)],\ ] ] due to the identification of @xmath181 with @xmath391 , and @xmath392 . the difference of the two functions in the rhs can be replaced by an integral of the derivative , @xmath390=\sum_{r=1}^n\int_0^{\l_r}{f'(\bx_r+\e_2 t_r)\e_2dt_r}.\ ] ] in the ns limit , each column @xmath265 contains an infinite number of boxes . we divide this infinite amount of boxes into infinitely many finite sets of size @xmath393 , ranging from @xmath394 to @xmath395 : @xmath396=\sum_{r=1}^n\sum_{s=1}^\infty\int_{u_{r , s}}^{u_{r , s}+k_{r , s}}{f'(\bx_r+\e_2 t_{r , s})\e_2dt_{r , s}}=\e_2\sum_{r=1}^n\sum_{s=1}^\infty\int_{0}^{k_{r , s}}{f'(\tx_{r , s}+\e_2 t_{r , s})dt_{r , s}},\ ] ] with @xmath397 , @xmath398 and @xmath399 . notations are displayed on figure . by the mean value theorem , for each interval there exists @xmath400 such that @xmath401 the mean value @xmath402 depends on the function @xmath202 . however , for functions of order @xmath403 , the variation of @xmath402 with @xmath202 is of order @xmath288 and thus can be neglected . it would not be the case if the @xmath393 were infinite , and indeed it is not possible to assign an average instanton position to a hadron with infinite charge because of its macroscopic size . plugging [ mean_value ] into [ before_mv ] , we recover the relation [ id_ur_kx ] for any test function @xmath371 provided we identify @xmath404 . thus , the relation between bethe roots and hadronic variables corresponds to the decomposition of the young tableaux columns into finite sets of @xmath357 boxes with an average instanton position @xmath362 . the formal relation we have observed between the variables @xmath374 and the bethe roots enables the comparison between the integral expression [ cz_a2 ] and the bethe roots summation of the @xmath8 quiver partition function . the decomposition [ decomp_cz ] of the summands can be reproduced on the integrand @xmath405 , with three parts that will be the analogue of the three bifundamental contributions @xmath167 $ ] , @xmath147 $ ] and @xmath168 $ ] respectively : @xmath406 the factors @xmath407 and @xmath408 depend only on one set of variables and can be attached to the nodes @xmath21 . on the other hand , @xmath409 is related to the arrow @xmath298 and will be referred as the _ bifundamental integrand_. up to a pure mass term , the factor @xmath407 can be obtained as @xmath410 for a specific choice of variables @xmath411 such that @xmath412 these variables @xmath411 encode the information of the object @xmath413 , in the same way that @xmath368 encodes @xmath380 . similar ideas can also be found in @xcite in the language of densities . it is remarkable that performing the change of representation @xmath414 ( and @xmath415 ) , materialized by the relation [ id_densities_ii ] among densities , the bifundamental integrand @xmath409 becomes @xmath245 $ ] , the ns limit of the bifundamental contribution [ zuv ] . this formal identification may allow to define the kmz transformation on the integrand expressions . in particular , the infinitesimal shifts of the bethe roots @xmath350 correspond to a variation of the density @xmath416 which is rendered by a shift of @xmath417 for the hadron of coordinate @xmath418 . the main difficulty lies in the decomposition of the function @xmath263 over poles , since in the representation @xmath368 this function is a product of essential singularities . it is still possible to use the cauchy identity to write it as a sum over contributions at the location of the singularities , but the evaluation of the integrals is problematic . on the other hand , the definition of @xmath238 , diagonal on the states @xmath419 easily follows from the definition [ def_d ] of @xmath121 , and the limit [ dico_kx ] , which gives , taking into account the @xmath1 factor discrepancy in [ d0_series ] between @xmath420 and @xmath421 : @xmath422 this expression can be integrated and expressed in terms of bethe roots using [ id_densities_ii ] . it leads to the formula [ d0_ur ] used in the previous section . in this paper , we considered the ns limit of the kmz transformation that represents the action of the shc generators on the bifundamental contribution to the @xmath8 quiver instanton partition function . in this limit , the bifundamental contribution can be written in terms of two sets of bethe roots . we have defined a set of generators of degree @xmath423 that exploits an infinitesimal variation of the bethe roots . these generators reproduce the proper commutation relations , and generates an equivalent of the kmz transformation . we then turned to an alternative approach that uses the mayer cluster expansion to perform the ns limit . the coordinates and charges of the hadronic variables that appear in this formalism were formally related to the bethe roots . this provides a hint on the possibility to realize the kmz transformation directly on integral expressions . finally , we studied a third expression of the partition function that takes the form of a yang - yang functional . although obtained via mayer expansion , this expression coincides with the limit of the young tableaux summations under a proper identification of the bethe roots . we deduced an action of the shc generators on the functional ( at first order in @xmath1 ) . now that the limit of the algebra has been identified , it remains to relate it to the algebraic structures of the underlying integrable systems , either the yangian of xxx spin chains or the ddaha of qnls . the connection between shc and the calogero - moser hamiltonian for @xmath4 allows to identify the states @xmath50 with jack polynomials . it is thus natural to expect a connection between the states @xmath268 and the eigenfunctions of the integrable systems . such a connection could be obtained using the fact that the calogero - moser cherednik operators reduce to the dunkl operators of qnls in a specific limit @xcite . this limiting process may be the equivalent of the ns limit from the integrable point of view . a better understanding of the states @xmath268 may also arise from the study of the single gauge group case ( @xmath424 quiver ) . the connections between shc and gaiotto states @xcite has been discussed recently in @xcite . some of these considerations may survive in the ns limit , and help to identify a semi - classical version of the gaiotto states . we hope to address these issues in a near future . i would like to thank davide fioravanti for valuable comments at the early stage of this work . i am also indebted to yutaka matsuo for several fruitful discussions and many advices . it is a pleasure to acknowledge the tokyo university ( hongo ) for the kind hospitality and support during my stay there . i also want to thank sergio andraus robayo for a very interesting discussion on dunkl and intertwining operators . i acknowledge the korea ministry of education , science and technology ( mest ) for the support of the young scientist training program at the asia pacific center for theoretical physics ( apctp ) . to derive the dictionary between the notations of @xcite and ours , let us drop for a moment the color and node indices , and focus on a single young diagram . in @xcite , young diagrams were encoded as a sequence of @xmath202 rectangles characterized by the integers @xmath425 and @xmath426 . it is further assumed that @xmath427 . the transition between notations is done as follows : to each box @xmath67 with @xmath428 , corresponds an index @xmath429 such that @xmath430 and @xmath431 ( the labels @xmath55 and @xmath18 are implicit on @xmath432 ) . defining @xmath433 as in @xcite , we deduce @xmath434 where @xmath435 denotes the coulomb branch vevs in @xcite , rescaled here as @xmath436 . in a similar way , for @xmath68 , there is a @xmath437 such that @xmath438 and @xmath439 , and defining @xmath440 we find @xmath441 with the same rescaling of the coulomb branch vevs . in some computations , it is not necessary to resort to the rectangle decomposition of young tableaux . here we provide some useful formulas in this respect . considering the difference between sets of boxes that can be added or removed , we find @xmath442 where @xmath443 is the number of rectangles in the young diagram @xmath444 and @xmath445 the total number of boxes . let us also mention the identities coming from the expansion of [ prop_l ] at @xmath79 : @xmath446 the action of the commutators of two degree ( minus ) one operators on states @xmath50 may be used to define some of the higher generators : @xmath447|y>=\sum_{x\in a / r(y)}\sum_{y\in a / r(y\pm x)}\dfrac{\l_x(y)\l_y(y\pm x)(z - w)(\phi_x-\phi_y)}{(z-\phi_x)(z-\phi_y)(w-\phi_x)(w-\phi_y)}|y\pm x\pm y>.\ ] ] let @xmath67 such that @xmath428 with @xmath54 . the multiplicity of a column @xmath58 of @xmath444 is the number of columns in @xmath444 with the same height @xmath58 . there is no other @xmath448 in the same column , i.e. such that @xmath449 . but there is always a unique @xmath450 such that @xmath451 ( except for @xmath197 ) . there is a unique @xmath452 with @xmath453 , and a unique @xmath454 with @xmath455 , if and only if the column @xmath58 is of multiplicity one . thus , @xmath196 except if @xmath197 for which @xmath198 . now , we let @xmath456 . there is no other @xmath450 with @xmath449 , but there is always a unique @xmath448 such that @xmath457 . there is a unique @xmath448 such that @xmath449 if and only if @xmath58 is of multiplicity one . likewise , there is a unique @xmath450 such that @xmath451 if and only if @xmath58 is of multiplicity one , and @xmath458 . thus , if @xmath459 , we have @xmath196 . if @xmath197 , @xmath460 is of multiplicity one , and @xmath198 . in this appendix , we take the ns limit of the bifundamental contribution [ cz_yy ] . a similar calculation can be found in @xcite . the cut - offs @xmath29 are assumed to be all equal to @xmath461 which will be sent to infinity at the end of the computation . we first concentrate on a single object @xmath51 , and examine the following quantity that is involved in the denominator : @xmath462 in the ns limit , @xmath463 is identified with the bethe root @xmath181 , and the main difficulty is to treat the dual quantities @xmath246 . for this purpose , let us focus on the @xmath55th young diagram pictured in figure , and consider the first column . boxes in the set denoted @xmath464 are such that @xmath465 for @xmath466 . similarly , boxes in the set @xmath467 are mapped to @xmath468 with @xmath469 . in general , the first column should be divided into @xmath470 sets of boxes with @xmath471 , @xmath472 with @xmath473 and @xmath474 . the same applies to the next columns , but we should start at @xmath475 for the @xmath34th column . we thus have found @xmath476 plugging in the explicit expression for @xmath59 , we obtain @xmath477 the product over @xmath478 can now be replaced by a ratio of gamma functions , @xmath479}{\g[y_{ll'ir}+\l_{r+1}^{(l)}]}.\ ] ] as @xmath157 , the gamma functions arguments tend to infinity , and we can use the stirling approximation @xmath480\simeq ( x / e)^x$ ] ( the square root is subleading ) . noticing that @xmath481 and @xmath482 , and introducing the function @xmath250 , we can write @xmath483 this expression now involves only the variables @xmath57 which can be replaced by bethe roots . it is however better to slightly shift the products indices to get @xmath484 the previous result upon @xmath485 can be used to take the ns limit of the denominator factors in [ cz_yy ] . considering @xmath486 we find @xmath487 it is possible to use the property @xmath488 to symmetrize the product . an ambiguity arises in the choice of the sense of the rotation . it is a part of the overall ill - definiteness for the final result of @xmath245 $ ] . here , we make the choices that provide the simplest expressions : we rotate both numerator and denominators in the ratios in the same direction , @xmath489 . after re - arranging the diagonal terms @xmath475 , we obtain @xmath490 finally , replacing @xmath491 and @xmath492 ( @xmath493 ) , we find : @xmath494 a similar treatment can be applied to the numerator , but we now have to keep track of the node indices . this does not add much difficulty , and we find for the numerator of [ cz_yy ] the expression @xmath495 taking the ratio , we end up with the formula given in [ zuv ] . in this appendix , we study a general class of grand - canonical partition functions defined as the discrete laplace transform @xmath496 of the following canonical partition functions @xmath497 partition functions with @xmath498 were considered in @xcite , and we show here that the method extends smoothly to @xmath499 . the integration contour is along the real axis , closed in the upper half plane , and avoiding possible singularities at infinity . we study the limit @xmath500 , with the kernels defined as @xmath501 and @xmath502 a function independent of @xmath503 . note that the poles at @xmath504 that pinch the integration contour are only present in the diagonal term @xmath505 , the non - diagonal kernel @xmath506 with @xmath507 is not singular at @xmath500 . this is in agreement with quivers instanton partition functions where @xmath506 corresponds to the bifundamental contribution ( with a non - zero mass @xmath20 ) , and @xmath505 to the vector hypermultiplet . quiver partition function , we take @xmath508 and @xmath509 with @xmath510 @xcite . it gives in the ns limit with @xmath511 , @xmath512 ( with @xmath513 ) and @xmath514 . the functions @xmath515 and @xmath516 are defined in [ g11 ] . ] we will further assume that @xmath502 satisfies the property @xmath517 . this condition is necessary to be able to expand over non - oriented clusters . for the scattering amplitudes , @xmath518 , and can be obtained as a consequence of unitarity and hermitian analyticity . ] in particular , this implies that the diagonal entries @xmath519 are even functions of @xmath90 . it is possible to express the integrand of the partition function [ defzc ] with the help of double indices @xmath520 ( and @xmath521 ) for @xmath522 and @xmath523 , and with the lexicographical order @xmath524 if and only if @xmath525 or ( @xmath526 and @xmath527 ) . applying the mayer expansion technique @xcite to [ defzgc ] for @xmath528 , we find for the free energy @xmath529 : @xmath530 the connected clusters @xmath531 are made of @xmath532 vertices @xmath520 . we denote by @xmath533 the set of vertices of the cluster @xmath531 , and @xmath534 the set of links @xmath535 connecting two vertices @xmath536 and @xmath537 . each vertex bear a color label @xmath18 , and there are @xmath538 vertices of type @xmath18 . of the quiver . it has nothing to do with the gauge color index of fields in the gauge theory . ] the symmetry factor @xmath539 of the cluster is the cardinal of the group of automorphisms preserving the cluster . these automorphisms must also preserve the color labeling because the combinatorial factor in the definition of @xmath540 is @xmath541 and not @xmath542 as in the usual mayer cluster expansion . decomposing @xmath543 , the cluster expansion becomes a sum over clusters @xmath544 with two types of links : g - links with kernel @xmath545 between vertices of color @xmath18 and @xmath546 , and p - links of kernel @xmath547 between two vertices of the same color @xmath18 : @xmath548 we have @xmath549 where @xmath550 denotes the set of p - links , and @xmath551 the set of g - links . the clusters that contribute at first order in @xmath503 have no cycle involving @xmath552-links , but do have cycles of only p - links . they are of order @xmath553 so that @xmath554 defined above is of order one @xcite . the generating function of rooted vertices , also called dressed vertex , bears an index @xmath18 corresponding to the color of the root , @xmath555 where we denoted @xmath556 . the summation is over rooted clusters @xmath557 , with the root @xmath558 of color @xmath18 . the symmetry factor @xmath559 is the number of automorphisms of @xmath544 that leave the root invariant . the generating functions @xmath560 and @xmath561 are defined with the additional requirement that the root @xmath90 is linked to other vertices with only g- or p- links respectively . at first order , we have the factorization property @xmath562 as a consequence of the property that g - links do not form cycles . minimal clusters that contribute to @xmath563 can be decomposed into sub - clusters rooted by the direct descendants . the root @xmath90 , of type @xmath18 , is linked to @xmath564 vertices @xmath565 of type @xmath546 through a link @xmath545 , which leads to @xmath566 the symmetry factors @xmath567 take into account the possibility of permuting vertices of the same color . contributions of sub - clusters factorizes to give @xmath568 this is the generalization of the formula ( 2.6 ) of @xcite . the third relation is obtained by repeating the confinement argument employed in @xcite . this argument remains unchanged because p - links only relate vertices of the same color . it gives the following relation between @xmath569 and @xmath560 : @xmath570 with the function @xmath571 defined in equ ( 2.13 ) of @xcite . for @xmath572 , this function simplifies into a logarithm : @xmath573 . the fields @xmath574 and @xmath575 are introduced as @xmath576 exploiting the relations [ eom1 ] and [ eom2 ] , we obtain the tba - like nlie : @xmath577 for @xmath513 , we recover the nlie [ nlie_quiver ] with the pseudo - energies @xmath578 defined as @xmath579 . the grand - canonical ( instanton ) free energy is obtained from the technique displayed in @xcite simply by introducing an additional sum over the different colors . the expression of @xmath554 follows from the basso - sever - vieira formula , @xmath580 with @xmath581 the function @xmath582 is a primitive of @xmath571 , it reduces at @xmath572 to the dilogarithm @xmath583 . plugging in the equations of motion , and using the variable @xmath574 and @xmath575 instead of @xmath584 and @xmath563 , we recover the results of @xcite : @xmath585 $ ] on - shell with the effective action @xmath586=\hf\sum_{a , b=1}^m\int{\rho_a(x)\rho_b(y)g_{ab}(x - y)dxdy}+\sum_{a=1}^m\int{\rho_a(x)\vphi_a(x)dx}+\dfrac1{2i\pi}\sum_{a=1}^m\int{l_{\a_a}(q_aq_a(x)e^{-\vphi_a(x)})dx}.\ ] ] the equations of motion reproduce the relations [ def_rho_phi ] and [ eom_rhophi ] . for simplicity , in this section we focus on the case @xmath587 , but the generalization to higher @xmath588 is straightforward . to obtain the hadronic partition function , we mimic the treatment performed in @xcite . at first order in @xmath1 , the instanton partition function can be written as a path integral , @xmath589e^{\frac1\e\sgc[\rho_a,\vphi_a]}}.\ ] ] the function @xmath590 contains all the dependence in @xmath591 , expanding it we get @xmath592 with @xmath593 the coefficients of the taylor expansion of @xmath594 : @xmath595 plugging this expansion in the expression of @xmath596 , we find @xmath597e^{\frac1{2\e}\sum_{a , b}\int{\rho_a(x)\rho_b(y)g_{ab}(x - y)dxdy}}e^{\frac1{\e}\int{\vphi_1(x)\left(\rho_1(x)-\e\sum_ik_i\d(x - x_i)\right)dx}}\\ & \times e^{\frac1{\e}\int{\vphi_2(y)\left(\rho_2(y)-\e\sum_jl_j\d(y - y_j)\right)dy}}. \end{split}\end{aligned}\ ] ] the fields @xmath575 appear as lagrange multipliers enforcing the conditions [ cond_density ] for the densities ( with @xmath511 ) . replacing the densities , we find @xmath598 the expression given in [ cz_a2 ] is obtained after specialization of [ defzgc ] to the quiver partition function [ def_cza2 ] . it implies to set @xmath511 , @xmath599 leading to @xmath600 . the kernels @xmath601 and @xmath602 coincide with the function @xmath552 given in [ g11 ] .

tree - shaped objects are fundamental in nature , where they appear , e.g. , as delivery systems for gases and fluids @xcite , as skeletal structures , or describing hierarchies . examples encountered in image analysis and computational biology are airway trees @xcite , vascular systems @xcite , shock graphs @xcite , scale space hierarchies @xcite and phylogenetic trees @xcite . statistical methods for tree - structured data would have endless applications . for instance , one could make more consistent studies of changes in airway geometry and structure related to airway disease @xcite to improve tools for computer aided diagnosis and prognosis . due to the wide range of applications , extensive work has been done in the past @xmath0 years on comparison of trees and graphs in terms of matching @xcite , object recognition @xcite and machine learning @xcite based on inter - tree distances . however , the existing tree - distance frameworks are algorithmic rather than geometric . very few attempts @xcite have been made to build analogues of the theory for landmark point shape spaces using manifold statistics and riemannian submersions @xcite . there exists no principled approach to studying the space of tree - structured data , and as a consequence , the standard statistical properties are not well - defined . as we shall see , difficulties appear even in the basic problem of finding the average of two tree - shapes . this paper fills the gap by introducing a shape - theoretical framework for geometric trees , which is suitable for statistical analysis . most statistical measurements are based on a concept of distance . the most fundamental statistic is the _ mean _ ( or prototype ) @xmath1 for a dataset @xmath2 , which can be defined as the minimizer of the sum of squared distances to the data points : @xmath3 this definition of the mean , called _ frchet mean _ , assumes a space of tree - shapes endowed with a distance @xmath4 , and is closely connected to geodesics , or shortest paths , between tree - shapes . for example , the midpoint of a geodesic from @xmath5 to @xmath6 is a mean for the two - point dataset @xmath7 . hence , if there are multiple geodesics connecting @xmath5 to @xmath6 , with different midpoints , then there will also be multiple means . _ as a consequence , without ( local , generic ) uniqueness of geodesics , statistical properties are fundamentally ill - posed ! _ thus , geometry enters the picture , and the idea of a geodesic tree - space gives a constraint on the possible geometric structure of tree - space . in a shape space where distances are given by path length , we must be able to continuously deform any given tree - shape into any other by traveling along the shortest path that connects the two shapes . the deformation - paths are easy to describe when only branch shape is changed , while tree topology ( branch connectivity ) is fixed . such deformations take place in portions of tree - space where all trees have the same topological structure . it is more challenging to describe deformation - paths in which the tree - topological structure is changed , for instance through a collapsed internal branch as in fig . [ treetopologyviolationgeodesic ] . we model topologically intermediate trees as collapsed versions of trees with differing tree topology , and glue the portions of tree - space together along subspaces that correspond to collapsed trees , as in fig . [ tree_space_components ] . as a consequence , tree - space has self intersections and is not smooth , but has singularities ! the main theoretical contributions of this paper are the construction of a mathematical tree - shape framework along with a geometric analysis of two natural metrics on the shape space . one of these metrics is the classical tree edit distance ( ted ) , into which we gain new insight . the second metric , called quotient euclidean distance ( qed ) , is induced from a euclidean metric . using gromov s approach to metric geometry @xcite , we show that qed generically gives locally unique geodesics and means , whereas finding geodesics and means for ted is ill - posed even locally . we explain why using ted for computing average trees must always be accompanied by a carefully engineered choice of edit paths in order to give well - defined results ; choices which can yield average trees which are substantially different from the trees in the dataset @xcite . the qed approach , on the contrary , allows us to investigate statistical methods for tree - like structures which have previously not been possible , like different well - defined concepts of average tree . _ this is our motivation for studying the qed metric ! _ the paper is organized as follows : in section [ background ] we discuss related work . the tree - space is defined in section [ geometrysection ] , and the statistical properties of tree - space are analyzed in section [ curvaturesection ] . in section [ complexitysection ] , we discuss how to overcome the computational complexity of both metrics , and present a simple qed approximation . in section [ experimentalsection ] , we illustrate the properties of qed by computing geodesics and different types of average tree for synthetic planar data trees as well as by computing qed means for sets of @xmath8 airway trees from human lungs . metrics on sets of tree - structured data have been studied by different research communities for the past 20 years . the best - known approach is perhaps tree edit distance ( ted ) , which has been used extensively for shape matching and recognition based on medial axes and shock graphs @xcite . ted and , more generally , graph edit distance ( ged ) , are also popular in the pattern recognition community , and are still used for distance - based pattern recognition approaches to trees and graphs @xcite . the ted and ged metrics will nearly always have infinitely many shortest edit paths , or geodesics , between two given trees , since edit operations can be performed in different orders and increments . as a result , even the problem of finding the average of two trees is not well posed . with no kind of uniqueness of geodesics , it becomes hard to meaningfully define and compute average shapes or modes of variation . this problem can be solved to some extent by choosing a preferred edit path @xcite , but there will always be a risk that the choice has negative consequences in a given setting . trinh and kimia @xcite face this problem when they use ted for computing average medial axes using the simplest possible edit paths , leading to average shapes which can be substantially different from most of the dataset shapes . statistics on tree - shaped objects receive growing interest in the statistical community . wang and marron @xcite study metric spaces of trees and define a notion of average tree called the median - mean as well as a version of pca , which finds modes of variation in terms of _ tree - lines _ , encoding the maximum amount of structural and attributal variation . aydin et al . @xcite extend this work by finding efficient algorithms for pca . this is applied to analysis of brain blood vessels . the metric defined by wang and marron does not give a natural geodesic structure on the space of trees , as it places a large emphasis on the tree - topological structure of the trees . the metric has discontinuities in the sense that a sequence of trees with a shrinking branch will not converge to a tree that does not have that branch . such a metric is not suitable for studying trees with continuous topological variations and noise , such as anatomical tree - structures extracted from medical images , since the emphasis on topology makes the metric sensitive to structural noise . a different approach is that of jain and obermayer @xcite , who study metrics on attributed graphs , represented as incidence matrices . the space of graphs is defined as a quotient of the euclidean space of incidence matrices by the group of vertex permutations . the graph - space inherits the euclidean metric , giving it the structure of an orbifold . this graph - space construction is similar to the tree - space presented in this paper in the sense that both spaces are constructed as quotients of a euclidean space . the graph - space framework does not , however , give continuous transitions in internal graph - topological structure , which leads to large differences between the geometries of the tree- and graph - spaces . trees also appear in genetics . hillis et al . @xcite visualize large sets of phylogenetic trees using multidimensional scaling . billera et al . @xcite have invented a phylogenetic tree - space suitable for geodesic analysis of phylogenetic trees , and owen and provan @xcite have developed fast algorithms for computing geodesics in phylogenetic tree - space . nye @xcite has developed a notion of pca in phylogenetic tree - space , but is forced to make strict assumptions on possible principle components being simple lines for the sake of computability . phylogenetic trees are not geometric , and have fixed , labeled leaf sets , making the space of phylogenetic trees much simpler than the space of tree - like shapes . we have previously @xcite studied geodesics between small tree - shapes in the same type of singular shape space as studied here , but most proofs have been left out . in @xcite , we study different algorithms for computing average trees based on the qed metric . this paper extends and continues @xcite , giving proofs , in - depth explanations and more extensive examples illustrating the potential of the qed metric . let us discuss which properties are desirable for a tree - shape model . as previously discussed , we require , at the very least , local existence and uniqueness for geodesics in order to compute average trees and analyze variation in datasets . when geodesics exist , we want the topological structure of the intermediate trees along the geodesic to reflect the resemblance in structure of the trees being compared . in particular , _ a geodesic passing through the trivial one - vertex tree should indicate that the trees being compared are maximally different . _ perhaps more importantly , we would like to compare trees where the desired edge matching is inconsistent with tree topology , as in fig . [ treetopologyviolation ] . specifically , we would like to find geodesic deformations in which the tree topology changes when we have such edge matchings , for instance as in fig . [ treetopologyviolationgeodesic ] . in this paper we shall work with two different tree - spaces : a tree - space @xmath9 , which is the space of all trees of a certain size , and a subspace @xmath10 , which is a restricted space of trees , whose exact definition is flexible ( see definition [ subspacedef ] ) . the large tree - space @xmath9 is the natural space for geometric trees . however , the available mathematical tools only allow us to prove our results locally , where the locality assumptions become very strict in @xmath9 . using a set of natural assumptions , we can restrict to a tree - space @xmath11 where our results hold in larger regions of tree - space . this is discussed in detail in section [ injectivity ] . we also believe that our results hold in @xmath9 , as described in conjecture [ conjecture ] . in this paper , a tree - shape is an embedded tree in @xmath12 or @xmath13 , and consists of a series of edge embeddings , glued together according to a rooted combinatorial tree . tree - shapes are invariant to translation , but our definition of tree - shape does not remove scale and rotation . however , tree - shapes can always be aligned with respect to scale and rotation prior to comparison , if this is important . any tree - like shape is represented as a pair @xmath14 consisting of a rooted , planar tree @xmath15 with edge attributes @xmath16 . in @xmath17 , @xmath18 is the vertex set , @xmath19 is the edge set , and @xmath20 is the root vertex . the tree @xmath15 describes the tree - shape topology , and the attributes describe edge shape , as illustrated in fig . [ pickapart ] . the shape attributes , represented by a point @xmath21 in the product space @xmath22 , is a concatenation of edgewise attributes from an attribute space @xmath23 . the attributes could , e.g. , be edge length , landmark points or edge parametrizations . in this work , we mostly use open curves translated to start at the origin , described by a fixed number of landmark points along the edge . thus , throughout the paper , the attribute space @xmath23 is @xmath24 where @xmath25 or @xmath26 and @xmath27 is the number of landmark points per edge . collapsed edges are represented by a sequence of origin points . in some illustrations , we shall use scalar attributes for the sake of visualization , in which case @xmath28 . in order to compare trees of different sizes and structures , we need to represent them in a unified way . we describe all shapes using the _ same _ tree @xmath15 to encode tree topology . by choosing a sufficiently large @xmath15 , we can represent all the trees in our dataset by filling out with empty ( collapsed ) edges . we call @xmath15 _ the maximal tree_. we model tree - shapes using _ binary _ maximal trees @xmath15 . tree - shapes which are not binary are represented by the binary tree @xmath15 in a natural way by allowing constant , or collapsed , edges , represented by the zero scalar or vector attribute . in this way _ an arbitrary attributed tree can be represented as an attributed * binary * tree _ , see fig . [ constedge ] . this is geometrically very natural . binary trees are geometrically _ stable _ in the sense that small perturbations of a binary tree - shape do not change the tree - topological structure of the shape . conversely , a trifurcation or higher - order vertex can always be turned into a series of bifurcations sitting close together by an arbitrarily small perturbation . in our representation , thus , trifurcations are represented as two bifurcations sitting infinitely close together , etc . trees embedded in the plane have a natural edge order induced by the left - right order on the children of any edge . we say that a tree is _ ordered _ whenever each set of sibling edges in the tree is endowed with such a total order . conversely , an ordered combinatorial tree always has a unique , implicit embedding in the plane where siblings are ordered from left to right . for this reason we use the terms planar tree and ordered tree interchangingly . we initially study metrics on the set of ordered binary trees ; later we use them to induce distances between unordered trees by considering all possible orders . this allows us to model trees in @xmath13 . considering all orders leads to potential computational challenges , which are discussed in section [ complexitysection ] . - and @xmath29-axes are trees with a single branch . for each real number @xmath30 , the tree - shape found at @xmath31 is also represented at @xmath32 . we build the tree - shape space by gluing the different representations of the same tree - shapes ( e.g. , @xmath33 and @xmath34 ) together , obtaining the shape space shown on the right . note the path from @xmath35 to @xmath36 through @xmath37 in @xmath38 on the right ; the corresponding path in the pre - shape space @xmath9 involves a teleportation between the representations @xmath33 and @xmath39 of @xmath37 . ] in order to build a space of tree - shapes , fix an ordered maximal binary tree @xmath15 with edges @xmath40 , which encodes the connectivity of all our trees . any attributed tree @xmath33 is now represented by a point @xmath41 in @xmath42 , where the coordinate @xmath43 describes the shape of the edge @xmath44 . since we allow zero - attributed edges , as discussed above , some tree - shapes will be represented by several points in @xmath9 ( fig . [ constedge ] ) . as a result , some natural tree - deformations are not found as continuous paths in @xmath9 . in figs . [ tedmoves4 ] and [ folding ] , the paths in @xmath9 corresponding to the indicated deformations require a teleportation between two representations of the intermediate tree - shape . we tackle this by using a refined tree - shape space @xmath38 , where different representations are identified as being the same point . the original space @xmath9 is called the tree pre - shape space , analogous to kendall s terminology @xcite . we go from pre - shapes to shapes by identifying those pre - shapes which define the same shape . consider two ordered tree - shapes with collapsed edges . replace their binary representations by collapsed representations , where the zero attributed edges have been removed . the orders of the original trees induce well - defined orders on the collapsed trees . we say that two ordered tree - shapes are _ the same _ when their collapsed ordered , attributed versions are identical , as in fig . [ constedge ] . tree identifications come with an inherent bijection of subsets of the edge set @xmath40 : if we identify @xmath45 , denote by @xmath46 the sets of non - collapsed edges with non - zero attributes . the identification of @xmath16 and @xmath29 is equivalent to an order preserving bijection @xmath47 , identifying those edges that correspond to the same edge in the collapsed tree - shape . varying the attributes @xmath43 , @xmath48 spans a family of tree - shapes with fixed topology and several representatives . thus , the edge sets @xmath49 and @xmath50 induce linear subspaces @xmath51 of @xmath9 where , except for on the axes , the topological tree structure is constant . the tree - shapes represented in @xmath52 are exactly the same as those represented in @xmath53 . the bijection @xmath48 induces a bijection @xmath54 given by @xmath55 , which connects each representation @xmath56 to the representation @xmath57 of the same shape . note that the @xmath58 are spanned by axes in @xmath9 . define a map @xmath59 for each pair of identified tree - structures , and form an equivalence on @xmath9 by setting @xmath60 for all @xmath16 and @xmath59 . for each @xmath61 , @xmath62 is the equivalence class @xmath63 . the quotient space @xmath64 of equivalence classes @xmath62 is the space of ordered tree - like shapes . quotient spaces are standard constructions from topology and geometry , where they are used to glue spaces together @xcite . the geometric interpretation of the identification in the tree - space quotient is that we fold and glue the pre - shape space space along the identified subspaces ; i.e. , when @xmath65 , we glue the two points @xmath5 and @xmath6 together . see fig . [ folding ] for an illustration . given a metric @xmath4 on the euclidean pre - shape - space @xmath42 , we induce the standard quotient pseudometric @xcite @xmath66 on the quotient space @xmath67 by setting @xmath68 this corresponds to finding the optimal path from @xmath62 to @xmath69 , consisting of any number @xmath70 of concatenated euclidean lines , passing through @xmath71 identified subspaces , as shown in fig . [ folding ] . it is clear from the definition that the distance function @xmath66 is symmetric and transitive . it is , however , an infimum , giving a risk that the distance between two distinct tree - shapes is zero , as occurs with some intuitive shape distance functions @xcite . this is why @xmath66 is called a _ _ pseudo__metric , and it remains to prove that it actually is a metric ; i.e. , that @xmath72 implies @xmath73 . we prove this for two specific metrics on @xmath9 , which come from two different ways of combining individual edge distances : the metrics @xmath74 and @xmath75 on @xmath42 are the norms @xmath76 from now on , @xmath4 and @xmath66 will denote either the distance functions @xmath74 and @xmath77 , or @xmath75 and @xmath78 . we prove the following theorem in section [ theoremproof ] : [ orderedthm ] the distance function @xmath66 is a metric on @xmath38 , which is a contractible , complete , proper geodesic space . thus , given any two trees , we can always find a geodesic between them in both metrics @xmath77 and @xmath78 . on @xmath9 , the induced pseudometric @xmath66 on @xmath38 is a metric . ] we may often want to restrict to a subset of the large tree - space . [ subspacedef ] consider a subset @xmath10 , which only contains all representations of trees of certain restricted topologies , defined by collapsed subtrees of the maximal tree @xmath15 . the @xmath79 collapsed subtree of @xmath15 is characterized by a subset @xmath80 consisting of the edges in the maximal tree @xmath15 which are not collapsed . associated to it is a linear subspace @xmath81 containing representations of all the trees of this particular topology . we include all representations of each tree topology , and obtain a restricted preshape space @xmath82 containing all the trees that have of one of the considered topologies . the equivalence relation @xmath83 on @xmath9 restricts to an equivalence relation @xmath84 on @xmath11 , from which we obtain a restricted tree - shape space @xmath85 . the metric @xmath4 on @xmath9 induces a metric @xmath86 on @xmath11 which induces a quotient pseudometric @xmath87 on @xmath88 . denote by @xmath11 the space of all trees in @xmath9 with @xmath27 leaves , now @xmath88 is the space of tree - shapes with @xmath27 leaves . note that the space of tree - shapes on @xmath27 leaves in _ i ) _ is different from the billera - holmes - vogtmann ( bhv ) space @xcite of trees with vector attributes , because in the bhv space , geodesics will always deform leaves onto leaves , whereas in @xmath88 , leaves can be transformed to non - leaf branches by a geodesic . even in the restricted tree - shape space , we obtain an induced metric : [ subspacemetric ] the pseudometric @xmath89 on @xmath88 is a metric , and @xmath90 a contractible , complete , proper metric space . first , we show that the pseudometric is a metric . the pseudometric in equation defines the distance @xmath91 as the infimum of lengths of paths in @xmath38 connecting @xmath62 and @xmath69 . any path in @xmath88 is also a path in @xmath38 , so if @xmath92 , then @xmath72 as well , so @xmath73 . the proofs of the other claims follow the proof of theorem [ orderedthm ] as in section [ theoremproof ] . the world is not two - dimensional , and for most applications it is necessary to study embedded trees in @xmath13 . the main difference from the ordered case is that trees in @xmath13 have no canonical edge order . the left - right order on children of planar trees gives an implicit preference for edge matchings , and hence reduces the number of possible matches . when we no longer have this preference , we consider all orderings of the same tree and choose orders which minimize the distance . we define the space of ( unordered ) tree - like shapes in @xmath8 as the quotient @xmath93 , where @xmath94 is the group of reorderings of the maximal binary tree @xmath15 . the metric @xmath66 on @xmath38 induces a quotient pseudometric @xmath95 on @xmath96 . again , we can prove : [ unorderedthm ] for @xmath95 induced by either @xmath77 or @xmath78 , the function @xmath95 is a metric and the space @xmath97 is a contractible , complete , proper geodesic space . the same result holds for restricted tree - spaces with several restricted tree topologies : let @xmath10 be a subspace containing only trees of certain restricted topologies , as in definition [ subspacedef ] , which is saturated with respect to the reordering group is saturated if , for each tree topology appearing in @xmath11 , all reorderings of the same tree topology also appear in @xmath11 . equivalently , @xmath98 , where @xmath99 @xmath94 . for the corresponding restricted tree - space @xmath100 , the quotient pseudometric @xmath101 is a metric and the space @xmath102 is a contractible , complete , proper geodesic space . note that @xmath94 is a finite group , which means that @xmath96 is locally well - behaved almost everywhere . in particular , off fixed - points for the action of reorderings @xmath103 on @xmath38 , the projection @xmath104 is a local isometry , i.e. , it is distance preserving within a neighborhood . hence , the geometry from @xmath38 is preserved off the fixed points . geometrically , a fixed point in @xmath38 is an ordered tree - shape where a reordering @xmath105 of certain branches does not change the ordered tree - shape ; that is , some pair of sibling edges must have the same shape attributes . in particular , the fixed points are non - generic because they belong to the lower - dimensional subset of @xmath38 where two sibling edges have identical shape . theorem [ unorderedthm ] can be proved using standard results on compact transformation groups along with similar techniques as for theorem [ orderedthm ] . while considering all different possible orderings of the tree is easy from the point of view of geometric analysis , in reality this becomes a computationally impossible task when the size of the trees grow beyond a few generations . in real applications we can , however , efficiently reduce complexity using heuristics and approximations , as discussed in section [ complexitysection ] . it follows from the definition that the metrics @xmath77 and @xmath106 coincide with the classical tree edit distance ( ted ) metric for ordered and unordered trees , respectively . in this way the abstract , geometric construction of tree - space gives a new way of viewing the intuitive ted algorithm . the metrics @xmath78 and @xmath107 are descents of the euclidean metric on @xmath38 , and geodesics in this metric are concatenations of straight lines in flat regions . in section [ fundamental ] we compare the two metrics using examples . we now pass to the proof of theorem [ orderedthm ] . the rest of the article is independent of this section , and during the first read , the impatient reader may skip to section [ curvaturesection ] . however , while the proof is technical , theorem [ orderedthm ] is a fundamental building block for the shape space framework . we shall assume that the reader has a good knowledge of metric geometry or general topology @xcite . it is crucial for the proof that we are only identifying subspaces of the euclidean space @xmath9 which are spanned by euclidean axes , and these are finite in number . this induces a well - behaved projection @xmath108 , which carries many properties from @xmath9 to @xmath38 . we say that ordered tree - shapes whose ( collapsed ) ordered structure is the same , belong to the same _ combinatorial tree - shape type_. for each combinatorial type of ordered tree - shape @xmath109 ( @xmath110 ) which can be represented by collapsing edges in the maximal tree @xmath111 , there is a family @xmath112 of subsets of @xmath40 , which induce that particular type @xmath109 when we endow the edges in @xmath112 ( @xmath113 ) with nonzero attributes and leave all other edges with zero attributes . these subsets are characterized by the properties * the cardinality @xmath114 for all @xmath115 , @xmath110 . * there is a depth - first order on each @xmath112 induced by the depth - first order on @xmath40 , such that the ordered , combinatorial structure defined by any @xmath112 coincides with that defined by @xmath116 . that is , the subset @xmath112 for any @xmath117 lists the set of edges in @xmath15 which have nonzero attributes for the @xmath79 representation of any shape of type @xmath109 . corresponding to each @xmath112 is the linear subspace @xmath118 of @xmath9 given by @xmath119 and by condition b ) we can define isometries @xmath120 by forgetting the zero entries in @xmath118 and keeping the depth - first coordinate order . we generate the equivalence @xmath83 on @xmath9 by asking that @xmath121 whenever @xmath122 for some @xmath123 . we now define the space of ordered tree - like shapes as the quotient @xmath67 , and define the quotient map @xmath108 . it is clear from the definition that the distance function @xmath66 defined in is symmetric and satisfies the triangle inequality , which makes it a _ pseudometric_. [ pseudometricismetric ] let @xmath4 denote @xmath74 or @xmath75 . the pseudometric @xmath66 is a metric on @xmath38 . it suffices to show that @xmath124 implies @xmath73 ( @xmath125 ) . moreover , it is also easy to show that @xmath126 for any @xmath127 , so it suffices to show that @xmath128 implies @xmath73 . hence , from now on , write @xmath66 for @xmath78 , and assume that @xmath72 for two tree - shapes @xmath62 and @xmath69 . choose @xmath129 such that @xmath130 that is , @xmath131 is smaller than the size of any of the non - zero edges in @xmath62 and @xmath69 . we may assume that @xmath132 since otherwise we may assume by symmetry that @xmath133 and @xmath134 . denote by @xmath135 the image of @xmath118 under the quotient projection @xmath108 for any @xmath117 . we may assume that @xmath62 and @xmath69 belong to the same identified subspace ; that is , there exist @xmath136 such that @xmath137 since otherwise , @xmath138 since @xmath69 is a finite set , and @xmath139 is a closed set , ( [ emptyint ] ) implies @xmath140 in this case , the path will have to go through some @xmath141 which does not contain points equivalent to @xmath29 , and @xmath142 since in order to reach @xmath143 , we need to remove edge attributes which are nonzero in @xmath69 , and @xmath144 for all @xmath145 . thus , eq . [ inthesame ] holds , and in fact , it holds for all the intermediate path points @xmath146 from eq . [ metricdefn ] . but if the path points stay in @xmath135 , then the path consists of shifting and changing the nonzero edge attributes of the trees in question . this will only give a sum @xmath147 if the trees are identical and the path is constant . here , we prove the rest of theorem [ orderedthm ] , namely that the tree - shape space @xmath148 is a complete , proper geodesic space , and @xmath38 is contractible . first , we note that although @xmath38 is not a vector space , there is a well - defined notion of size for elements of @xmath38 , induced by the norm on @xmath9 : [ sizelemma ] note that if @xmath149 , we must have @xmath150 ; hence we can define @xmath151 . the equivalence is generated recursively from the conditions @xmath149 whenever either @xmath152 , indicating @xmath153 ; or @xmath154 , indicating @xmath155 since the @xmath156 are isometries . hence , the lemma holds by recursion . we will prove that @xmath148 is a proper geodesic space using the hopf - rinow theorem for metric spaces @xcite , which states that every complete locally compact length space is a proper geodesic space . a length space is a metric space in which the distance between two points can always be realized as the infimum of lengths of paths joining the two points . note that this is a weaker property than being a geodesic space , as the geodesic joining two points does not have to exist ; it is enough to have paths that are arbitrarily close to being a geodesic . it follows from ( * ? ? ? * lemma 5.20 ) that @xmath148 is a length space for _ any _ metric @xmath4 on @xmath9 where @xmath66 is a metric . to see that the tree - shape space is locally compact , note that the projection @xmath108 is finite - to - one , so any open subset @xmath157 of @xmath38 has as pre - image a finite union @xmath158 of open subsets of @xmath9 , such that @xmath159 and @xmath160 is compact whenever @xmath157 is bounded . we also need to prove that @xmath148 is complete : [ completenessdescends ] let @xmath66 denote either of the metrics @xmath77 and @xmath78 . the shape space @xmath148 is complete . the proof needs a lemma from general topology : _ ( * ? ? ? * , theorem 2.3 ) _ [ duglemma ] let @xmath161 be a metric space and assume that the metric @xmath4 has the following property : _ there exists @xmath129 such that for all @xmath162 the closed ball @xmath163 is compact . _ then @xmath66 is complete . using the projection @xmath108 , we can prove : [ closedandbounded ] bounded closed subsets of @xmath38 are compact . since lemma [ sizelemma ] defines a notion of size in @xmath38 , any closed , bounded subspace @xmath164 in @xmath38 is contained in a closed ball @xmath165 in @xmath38 for some @xmath166 , where @xmath167 is the image @xmath168 . since @xmath169 , it follows that @xmath170 , which is a closed and bounded ball in @xmath9 . since closed , bounded subsets of @xmath9 are compact @xmath171 is compact . by continuity of @xmath172 , @xmath173 is compact . then @xmath164 is compact too . it is now very easy to prove proposition [ completenessdescends ] : by lemma [ closedandbounded ] , all closed and bounded subsets of @xmath38 are compact , but then by lemma [ duglemma ] the metric @xmath66 must be complete . using the hopf - rinow theorem ( * , proposition 3.7 ) we thus prove that @xmath148 is a complete , proper geodesic space . we still miss contractibility : [ contractible ] let @xmath174 be a normed vector space and let @xmath83 be an equivalence on @xmath174 such that @xmath175 implies @xmath176 for all @xmath177 . then @xmath178 is contractible . define a map @xmath179 \to \bar{b}$ ] by setting @xmath180 . now @xmath181 is well defined because of the condition on @xmath83 , and @xmath182 so @xmath181 is a homotopy from @xmath183 to the constant zero map . combining the results of section [ theoremproof ] , we see that the proof of theorem [ orderedthm ] is complete . having proved theorem [ orderedthm ] , we may now pass to studying the geometry of the tree - shape space through its geodesics . uniqueness of geodesics and means is closely connected to the geometric notion of _ curvature _ , a concept which fundamentally depends on the underlying metric . using methods from metric geometry @xcite we shall investigate the curvature of the tree - shape space with the qed and ted metrics . using curvature , we obtain well - posed statistical methods for qed . the next theorem states that in the tree - shape space endowed with the qed metric , any randomly selected point has a corresponding neighborhood within which the tree - space has non - positive curvature . we shall use this fact to show that datasets within that same neighborhood have unique averages . [ qedcurv ] * endow @xmath38 with the qed metric @xmath78 . a generic point @xmath184 has a neighborhood @xmath185 in which the curvature is non - positive . at non - generic points , the curvature of @xmath186 is @xmath187 , or unbounded from above . * endow @xmath38 with the ted metric @xmath77 . the metric space @xmath188 does not have locally unique geodesics anywhere , and the curvature of @xmath188 is @xmath187 everywhere . claims i ) and ii ) also hold in the subspace @xmath189 containing only trees of certain restricted topologies , as defined in definition [ subspacedef ] . here , local uniqueness is defined as uniqueness within a sufficiently small neighborhood . _ genericity _ is a key concept in this paper . many of our results , e.g. , uniqueness of means , do not hold in general , but they do hold for a randomly chosen dataset with respect to natural probability measures . a _ generic property _ in a metric space @xmath190 is a property which holds on an open , dense subset of @xmath9 . in the tree - spaces @xmath191 and @xmath192 , one interpretation is that generic properties hold almost surely , or with probability one , with respect to natural probability measures . thus , for a random tree - shape , we can safely assume that it satisfies generic properties , e.g. , that the tree - shape is a binary tree . non - generic property _ is a property whose not happening is generic . this is similarly interpreted as a property that may _ not _ hold for randomly selected tree - shapes . a detailed discussion of the relation between genericity and probability is found in appendix a. one common misconception is that the term generic tree refers to a particular class of trees with a particular generic property . it is important to note that many different properties , which do not necessarily happen on the same subsets of tree - space , may all be generic at the same time . however , any finite set of generic properties _ will _ all happen on an open , dense subset . [ generic ] tree - shapes that are truly binary ( i.e. their internal edges are not collapsed ) are generic in the space of all tree - like shapes . let @xmath193 be a tree - shape in @xmath38 or @xmath96 which is not truly binary , represented by a maximal binary tree @xmath15 . by adding arbitrarily small noise to the zero attributes on edges of @xmath15 , we obtain truly binary tree - shapes @xmath194 which are arbitrarily close to @xmath193 . thus , the set of full truly binary tree - shapes in @xmath38 or @xmath96 is open and dense . hence , truly binary tree - shapes are generic both in @xmath38 and in @xmath96 . goes through a tree with a trifurcation , while geodesic @xmath195 does not have internal structural transitions . ] the essence of the proposition is that binary tree - shapes are generic , but that does not mean that non - binary trees do not need to be considered ! while non - binary trees may not appear as randomly selected trees , they _ do _ appear in geodesics between randomly selected pairs of trees , as in fig . [ fundamentalfigure ] . non - binary tree - like shapes also appear as samples in real - life applications , e.g. , when studying airway trees . however , we interpret this as an artifact of resolution rather than true higher - degree vertices . for instance , airway extraction algorithms record trifurcations when the lengths of internal edges are below certain threshold values . in order to understand and prove theorem [ qedcurv ] , we need a definition of curvature in metric spaces . in spite of its simplicity and elegance , this concept from metric geometry is novel in computer vision . we shall spend a little time introducing it before proceeding to prove theorem [ qedcurv ] in section [ qedcurvproof ] . since general metric spaces can have all kinds of anomalies , the concept of curvature in such spaces is defined through a comparison with spaces that are well understood . more precisely , the metric spaces are studied using _ geodesic triangles _ , which are compared with corresponding _ comparison triangles _ in model spaces with a fixed curvature @xmath196 . the model spaces are spheres ( @xmath197 ) , the plane @xmath12 ( @xmath198 ) and hyperbolic spaces ( @xmath199 ) , and through comparison with these spaces , we can bound the curvature of the metric space by @xmath196 . in this paper we shall use comparison with planar triangles , which gives us curvature bounded from above by @xmath200 . space if , for a geodesic triangle @xmath201 and for any point @xmath16 on the triangle , the distance from @xmath16 to the opposite vertex is not longer than the corresponding distance in the planar comparison triangle @xmath202 . ] given a geodesic metric space @xmath203 , a _ geodesic triangle _ @xmath201 in @xmath203 consists of three points @xmath204 and geodesic segments joining them . a planar _ comparison triangle _ @xmath205 for the triangle @xmath201 consists of three points @xmath206 in the plane , such that the lengths of the sides in @xmath202 are the same as the lengths of the sides in @xmath201 , see fig . [ cat0 ] . a @xmath207 space is a metric space in which geodesic triangles are thinner than for their comparison triangles in the plane . that is , @xmath208 for any @xmath16 on the edge @xmath209 where @xmath210 is the unique point on the edge @xmath211 such that @xmath212 and @xmath213 . if the planar comparison triangle is replaced by a comparison triangle in the sphere or hyperbolic space of fixed curvature @xmath196 , we get a @xmath214 space . a space @xmath203 has non - positive curvature if it is locally @xmath207 , i.e. , if any point @xmath215 has a radius @xmath20 such that the ball @xmath216 is @xmath207 . similarly , define curvature bounded by @xmath196 as being locally @xmath214 . [ negcurvex ] * the space @xmath157 obtained by gluing a family euclidean spaces @xmath217 together along isomorphic affine subspaces @xmath218 is a @xmath207 space . at any point in @xmath157 which is not a glued point , the local curvature is @xmath200 , since the space is locally isomorphic to the corresponding @xmath217 . at any glued point , it can be shown that the local curvature is @xmath219 . * the gpca construction by vidal et al . @xcite defines a @xmath207 space , giving a potential use of @xmath207 spaces and metric geometry in machine learning . * the space of phylogenetic trees is a @xmath207 space @xcite . * as we are about to see , the space of tree - like shapes is locally a @xmath207 space almost everywhere . one of the main reasons why @xmath214 spaces ( and @xmath207 in particular ) are attractive , is due to the following result on existence and uniqueness of geodesics . _ * proposition ii 1.4 ) _ [ bhprop ] let @xmath220 be a @xmath214 space . if @xmath221 , then all pairs of points have a unique geodesic joining them . for @xmath197 , the same holds for pairs of points at a distance less than @xmath222 . more results on curvature in metric spaces can be found in the book by bridson and haefliger @xcite . in this section we study the curvature of tree - shape space using the theory of @xmath207 spaces . we show that at a generic tree , the shape space has bounded curvature . the results rest on the following theorem : [ unique_geo ] at a generic point @xmath184 , the shape space is locally @xmath207 , and thus , @xmath38 has locally unique geodesics in a neighborhood of @xmath62 . recall from section [ singularspace ] how @xmath38 was formed by identifying subspaces @xmath223 , defined in eq . [ vdef1 ] and [ vdef2 ] . these identified subspaces corresponded to different representations in @xmath9 of the same shape @xmath184 . the points in @xmath38 can now be divided into three categories : * points @xmath224 which do not belong to the image of an identified subspace because they only have one representative in @xmath9 , and two classes of points in @xmath38 which have more than one representative in @xmath38 . * the first class contains points @xmath225 at which the space @xmath38 is locally homeomorphic to an intersection of linear spaces , as in example [ negcurvex ] a ) . these points are images of points @xmath226 which belong to one single identified subspace @xmath58 . * the second class contains points @xmath227 whose preimages @xmath228 in @xmath9 are at the intersection of identified subspaces @xmath229 . an example of such points is the image of the origin in fig . [ folding ] . these points correspond to trees where , infinitely close to the same tree , we can find pairs of trees in @xmath38 between which geodesics are not unique . these three classes of points correspond to local curvature @xmath200 , @xmath219 and @xmath187 . that is , the space is locally @xmath207 at points in categories _ i ) _ ; at points from _ ii ) _ it is @xmath214 for every @xmath230 , so has curvature @xmath219 ; and at points from _ iii ) _ it is not @xmath214 for any @xmath230 ; hence the curvature is @xmath187 . it thus suffices to show that the points in category _ iii ) _ are non - generic , which follows easily from the fact that these must necessarily sit in a lower - dimensional subspace of @xmath38 . the proof carries over to the subspace @xmath88 . we call a @xmath207 neighborhood @xmath231 of a point @xmath184 an _ injectivity neighborhood _ of @xmath62 . based on the above , we are now ready to prove : * the qed case : since @xmath38 is locally @xmath207 at generic points @xmath62 , the curvature of @xmath38 is non - positive in a neighborhood @xmath157 of @xmath62 . at points @xmath232 , however , we will always find pairs of points @xmath233 arbitrarily close to @xmath16 with two geodesics joining them , just as in fig . [ folding ] . * the ted case : consider a tree - shape @xmath234 , represented by a point @xmath61 . induce a second tree - shape @xmath194 represented by @xmath235 , where @xmath236 such that @xmath237 and @xmath238 have one non - zero coordinate , found in different edges , which are both nonzero edges in @xmath16 . the topology of @xmath194 is the same as of @xmath193 . for any @xmath239 , we can find @xmath27 ted geodesics @xmath240 from @xmath193 to @xmath194 , where @xmath241 can be decomposed as @xmath242 . thus , there are infinitely many ted geodesics from @xmath193 to @xmath194 , and @xmath188 does not have locally unique geodesics anywhere . as a consequence , its curvature is unbounded everywhere ( * ? ? ? * proposition ii 1.4 ) . the practical meaning of theorem [ qedcurv ] is that a ) we can use techniques from metric geometry to search for qed averages , b ) as we are about to see , for datasets contained in an injectivity neighborhood , there exist unique means , centroids and circumcenters for the qed metric , and c ) the same techniques can not be used to prove existence or uniqueness of prototypes for the ted metric , even if they were to exist . in fact , any geometric method which requires bounded curvature @xcite will fail for the ted metric . _ this motivates our study of the qed metric . _ it is easy to prove that the same results also hold for unordered tree - shapes : [ unordered_curvature ] the space @xmath243 of unordered trees with the qed metric is generically non - positively curved . with the ted metric @xmath106 , however , @xmath96 has everywhere unbounded curvature , geodesics are nowhere locally unique and neither are any of the types of average tree discussed in this paper . the same holds in the subspace @xmath244 , defined in theorem [ unorderedthm ] . in this section we use what we learned in the previous section to show that , given the qed metric on a space of tree - like shapes , we can find various forms of average shape in the space of ordered tree - like shapes , assuming that the data lie within an injectivity neighborhood . there are many competing ways of defining central elements given a subset of a metric space . we discuss several : the _ circumcenter _ considered in @xcite , the _ centroid _ considered , among other places , in @xcite , and the _ mean _ @xcite . the problem of existence and uniqueness of averages can be attacked using convex functions . recall that a function @xmath245 \to { \mathbb{r}}$ ] is convex if @xmath246 for all @xmath247 $ ] and @xmath248 $ ] . if we can replace @xmath249 with @xmath250 whenever @xmath2510 , 1[$ ] , then @xmath252 is _ strictly _ convex . convex functions have minimizers , which are unique for strictly convex functions . hence , existence and uniqueness of averages can be proven by expressing them as minimizers of strictly convex functions . we say that a function @xmath253 on a geodesic metric space @xmath9 is ( strictly ) convex if for any two points @xmath254 and any geodesic @xmath255 \to x$ ] from @xmath16 to @xmath29 , the function @xmath256 is ( strictly ) convex . we shall make use of the following standard properties of convex functions : [ convexlemma ] * if @xmath257 and @xmath258 are both convex , @xmath105 is monotonous and increasing , and @xmath105 is strictly convex , then @xmath259 is strictly convex . * if @xmath257 and @xmath258 are both convex , then @xmath260 is also convex . if either @xmath252 or @xmath105 is strictly convex , then @xmath261 is strictly convex as well . the _ mean _ of a finite subset @xmath262 in a metric space @xmath190 is defined as in eq . [ meandef ] , and is called the _ frchet mean_. local minimizers of eq . [ meandef ] are called _ karcher means_. the following result follows from a more general theorem by sturm ( * ? ? ? * proposition 4.3 ) ; we include the basic version of the proof here for completeness . [ existencemeans ] means exist and are unique in @xmath207-spaces . the function @xmath263 given by @xmath264 is convex for any fixed @xmath162 by ( * ? ? ? * proposition ii.2.2 ) , so the function @xmath265 is _ strictly _ convex by lemma [ convexlemma ] i ) . but then @xmath266 is strictly convex by lemma [ convexlemma ] _ ii ) _ , and a mean is just a minimizer of the function @xmath267 . the function @xmath267 is coercive , so the minimizer exists . since @xmath267 is strictly convex , the minimizer is unique . we also consider two other types of statistical prototypes for a dataset , namely _ circumcenters _ and _ centroids_. these are both well - known in the context of metric geometry and @xmath214 spaces . * * circumcenters . * consider a metric space @xmath220 and a bounded subset @xmath268 . there exists a unique smallest closed ball @xmath269 in @xmath203 which contains @xmath11 ; the center @xmath270 of this ball is the _ circumcenter _ of @xmath11 . * * the centroid of a finite set . * let @xmath9 be a uniquely geodesic metric space ( a metric space where any two points are joined by a unique geodesic ) . the centroid of a set @xmath271 of @xmath27 elements is defined recursively as a function of the centroids of subsets with @xmath272 elements as follows : denote the elements of @xmath273 by @xmath274 . if @xmath273 contains two elements , @xmath275 , the centroid @xmath276 of @xmath273 is the midpoint of the geodesic joining @xmath277 and @xmath278 . if @xmath279 , define @xmath280 , which is a set with @xmath27 elements . similarly , for larger @xmath70 , @xmath281 . all these sets have @xmath27 elements . if the elements of @xmath282 converge to a point @xmath283 as @xmath284 , then we say that @xmath285 is the _ centroid _ of @xmath273 in @xmath9 . based on the theory of @xmath214 spaces and our results for means , we have for the set of tree - like shapes : [ all_averages ] endow @xmath38 with the qed metric @xmath78 . a generic point @xmath184 has a neighborhood @xmath157 such that sets contained in @xmath157 have unique means , centroids and circumcenters . _ the same statistical properties also hold for the qed metric on unordered tree - shapes : _ generic points in the space of unordered tree - shapes with the qed metric @xmath243 have neighborhoods within which means , circumcenters and centroids exist and are unique . for the ted metric , these are not unique . the same results hold in the restricted tree - spaces @xmath88 and @xmath192 , defined in definition [ subspacedef ] and theorem [ unorderedthm ] . first consider @xmath38 and @xmath96 with the qed metric . by theorems [ qedcurv ] and [ unordered_curvature ] , @xmath38 and @xmath96 are both locally @xmath207 spaces , and by theorems [ orderedthm ] and [ unorderedthm ] they are both complete metric spaces . we have seen in theorem [ existencemeans ] that means exist and are unique in @xmath207 spaces , so the statement holds for means . by ( * proposition 2.7 ) , any subset @xmath203 of a complete @xmath207 space has a unique circumcenter . hence , the statement holds for circumcenters . similarly , by ( * ? ? ? * theorem 4.1 ) , finite subsets of @xmath207 spaces @xmath9 have centroids ( unique by definition ) , so the statement holds for centroids . we turn to the ted metric . by definition , for any @xmath195-point dataset , all these notions of mean reduce to finding the midpoint of a geodesic connecting the two points . we know that geodesics and midpoints are not unique in the ted metric . this ends the proof . we have shown that the local curvature is nonpositive almost everywhere in @xmath38 , which makes @xmath38 well suited for geometric definitions of statistical properties . however , our notion of almost everywhere is strongly tied to ( maximal ) dimensionality , which again is strongly tied to the topological structure of the maximal tree @xmath15 . one consequence is that the injectivity neighborhoods in @xmath38 are rather small , as we are about to see through examples . in this section we impose natural constraints on tree - space that allow us to increase the size of the injectivity neighborhoods , and make a conjecture for future expansion beyond the use of @xmath207-spaces . consider the following two examples ; we thank the anonymous reviewer for the first ! * consider the tree - shapes @xmath286 and @xmath287 shown in fig . [ ref_ex_1 ] , left . these two tree - shapes are joined by two geodesics , and thus @xmath286 and @xmath287 are not contained in the same injectivity neighborhood in @xmath38 . however , in a suitably chosen @xmath88 , they can be . * consider the tree - shape @xmath193 in the space of unordered tree - shapes with attributes in @xmath12 , spanned by the maximal tree @xmath15 as in fig . [ ref_ex_2 ] , right . arbitrarily close to @xmath193 we will find two trees @xmath286 and @xmath287 which are joined by two geodesics , as shown in the figure . . ] are two other trees @xmath286 and @xmath287 who can not be joined by a unique geodesic . ] as a illustrated by these examples , the @xmath207 injectivity neighborhoods in the shape - space @xmath38 can be very small , mainly containing trees whose topology is the same as @xmath15 . however , we can obtain much larger injectivity neighborhoods by restricting to the natural subspaces @xmath189 and @xmath244 as in definition [ subspacedef ] . as shown in theorem [ all_averages ] , @xmath88 and @xmath192 have the same nice geometric properties as @xmath38 and @xmath96 , and @xmath11 can be chosen so that the injectivity neighborhoods are bigger than in @xmath38 and @xmath96 by avoiding situations as in the examples above . radius is not a good measure for the size of an injectivity neighborhood , as a tree may contain both small branches , which do not have much room to vary , as well as large branches , which will be allowed to move more throughout such a neighborhood . however , any convex neighborhood which does not contain points of curvature @xmath288 , will be an injectivity neighborhood . all the results from section [ meansection ] hold at generic points , within an injectivity neighborhood where the @xmath207 property holds . we have seen examples of points where local @xmath207-property must fail , illustrating the limitaitons of @xmath207 techniques for analyzing general trees . however , most of the situations where the @xmath207 property fails are highly non - generic , and we conjecture that more general results can be proven : [ conjecture ] for a generic set of points @xmath289 in @xmath38 , @xmath96 , @xmath88 or @xmath192 , means exist and are unique . as we have seen in theorems [ qedcurv ] , the qed metric gives locally non - positive curvature at generic points , while the ted metric gives unbounded curvature everywhere on @xmath191 and @xmath192 . equivalently , geodesics are locally unique almost everywhere in the qed metric , while nowhere locally unique in the ted metric . as emphasized by theorem [ all_averages ] , this means that we can not imitate the classical statistical procedures on shape spaces using the ted metric , while for the qed metric , we can . note , moreover , that the qed metric is the quotient metric induced from the euclidean metric on the pre - shape space @xmath9 , making it the natural choice of metric seen from the shape space point of view . from a computational point of view , the ted metric has nice local - to - global properties . if the trees @xmath286 and @xmath287 are decomposed into subtrees @xmath290 and @xmath291 as in fig . [ tedsplitting ] , such that the geodesic from @xmath286 to @xmath287 restricts to geodesics between @xmath292 and @xmath293 as well as @xmath294 and @xmath295 , then @xmath296 . many dynamic programming algorithms for ted use this property , and the same does not hold for the qed metric , due to the square root . for a qualitative comparison of qed and ted , we compare the geodesics defined by the ted and qed metrics between small , simple trees . consider the two tree - paths in fig . [ tedqedcomparison ] , where the edges are endowed with scalar attributes @xmath297 describing edge length . computing the costs of the two different paths in both metrics , we find the shortest ( geodesic ) path . path @xmath298 indicates a matching left and right side edges @xmath299 and @xmath4 , while path @xmath195 does not make the match . the cost of path @xmath298 is @xmath300 in both metrics , while the cost of path @xmath195 is @xmath301 in the qed metric and @xmath302 in the ted metric . thus , ted will identify the @xmath299 and @xmath4 edges whenever @xmath303 , while qed makes the identification whenever @xmath304 . thus , ted will be more prone to internal structural change than qed . the same occurs in the comparison of ted and qed matching in figs . [ qedmatching ] and [ tedmatching ] . although the ted is more prone to matching trees with different tree - topological structures , the edge matching results are similar , as is expected , since the metrics are closely related . in general , computational complexity is a problem for both ted and qed . computing ted distances between unordered trees is np - complete @xcite , and we conjecture that the qed metric is np - complete as well . we can , however , often use geometry and prior knowledge ( e.g. , anatomy ) to find efficient approximations . _ trees appearing in applications are often not completely unordered , but are semi - labeled_. [ semilabel ] most airway trees have similar , but not necessarily identical , topological structure , where several branches have names and can be identified by experts . the top generations of the airway tree serve very clear purposes in terms of anatomy . the root edge is the trachea ; the second generation edges are the left and right main bronchi ; the third generation edges lead to the lung lobes . as these are easily identified , we find a semi - labeling of the airway tree , and use it to simplify computations of inter - airway geodesics in section [ pulmonarysection ] . since the decomposition strategy used for dynamic programming in ted is not available for qed , an approximation of qed is used in our experiments . many anatomical trees have a somewhat fixed overall structure . for these , it is safe to assume that the number of internal structural transitions found in a geodesic deformation is low , and that the geodesics pass through identified subspaces of low codimension . the latter assumption is equivalent to assuming that the trees appearing throughout the geodesic deformation have nodes of low degree . for instance , for the airway trees studied in section [ pulmonarysection ] below , we find empirically that it is enough to allow for one structural change in each lobar subtree , which has at most a trifurcation . recall the definition of the metric from eq . [ metricdefn ] ; the approximation consists of imposing upper bounds @xmath305 on the number @xmath70 and @xmath267 on the degrees of internal vertices appearing in eq . [ metricdefn ] , respectively , giving : @xmath306 where all @xmath307 and @xmath308 have vertex degrees at most @xmath267 . geometrically , @xmath70 is the number of euclidean segments concatenated to form the geodesic . bounding @xmath70 is equivalent to bounding the number of internal topological transitions throughout the geodesic . in fact , @xmath309 _ the number of internal topological transitions in the geodesic_. all edges are translated to start at @xmath310 , and represented by a fixed number of landmark points ( in our case @xmath311 ) evenly distributed along the edge , the first at the origin . the distance between two edge attributes @xmath312 is the euclidean distance @xmath313 . we approximate of the qed metric using algorithm [ planaralg ] ; see appendix b in the supplemental material for details . the complexity of algorithm 1 is @xmath314 where @xmath27 is the number of internal vertices , @xmath305 is the bound on @xmath70 , @xmath267 is the bound on vertex degree and @xmath315 is the optimization in line @xmath316 . for the unordered airway trees in section [ pulmonarysection ] , we combine algorithm 1 with a complete search through the set of branch orderings . computing qed distances through a complete search is not optimal , and finding more exact and efficient algorithms is a nontrivial research problem . @xmath16 , @xmath29 planar rooted depth @xmath27 binary trees @xmath317 set of ordered identified pairs @xmath318 of subspaces of @xmath9 corresponding to internal topological changes through trees of order at most @xmath267 , corresponding to a subspace @xmath273 of @xmath38 , s.t . if @xmath319 , then also @xmath320 . @xmath321 @xmath322 @xmath323 @xmath324 @xmath325 @xmath326 geodesic @xmath327 [ planaralg ] the qed metric is new , whereas the properties of the ted metric are well known @xcite . our experimental results on real and synthetic data illustrate the geometric properties of the qed metric . the experiments on airway trees in section [ pulmonarysection ] show , in particular , that it is feasible to compute geodesics between real , @xmath8 data trees . movies illustrating geodesics between planar depth @xmath26 trees , as well as a table illustrating a tree - shape matching experiment for a set of synthetic planar depth @xmath26 trees , are found on the web page http://image.diku.dk/aasa/tree_shape/planar.html . the movies and matchings show the geometrically intuitive behavior of the geodesic deformations . we see that the intermediate structures resemble the geodesic endpoint trees in a reasonable manner , and show the ability of the geodesics to handle internal topological differences . these are among the wanted properties from the model listed in the beginning of section [ geometrysection ] . moreover , computed mean and centroid trees are shown in fig . [ dataset_means ] ; these are presented in greater detail in @xcite . the qed average trees clearly represent the main common properties of the dataset trees . as proof - of - concept experiments on real data we compute means for sets of pulmonary airway trees from the exact09 airway segmentation competition @xcite . the airway trees were first segmented from low dose screening computed tomography ( ct ) scans of the chest using a voxel classification based airway tree segmentation algorithm @xcite . the centerlines were extracted from the segmented airway trees using a modified fast marching algorithm based on @xcite . the method gives a tree structure directly through connectivity of parent and children branches . for simplicity , we only consider the upper airway tree down to the lobe branches . means are computed using both the qed metric ( details found in @xcite ) and using the ted method proposed by trinh and kimia @xcite . the dataset is shown in fig . [ airway_dataset ] , and the results are seen in fig . [ airway_means ] . we clearly see how the choice of ted geodesic makes the ted mean vulnerable to noise in the form of missing branches among the dataset trees , giving a mean shape whose structure is too simple . the qed mean , on the other hand , represents the data well . airway trees from the exact09 test set . @xcite ] th data tree misses these . ] starting from a purely geometric point of view , we define a shape space for tree - like shapes . this intuitive geometric framework allows us to simultaneously take both global tree - topological structure and local edgewise geometry into account . we study two metrics on this shape space , ted and qed , which both give the shape space a geodesic structure ( theorems [ orderedthm ] and [ unorderedthm ] ) . the framework is developed for tree - shapes , but carries over to other geometric trees with vector attributes . qed is the geometrically most natural metric , and turns out to have properties which are essential for statistical shape analysis . through a geometric analysis of the tree - shape framework , we show that the qed metric gives local uniqueness of geodesics and local existence and uniqueness for three versions of average shape , namely the mean , the circumcentre and the centroid ( theorem [ all_averages ] ) . our analysis gives new insight into the metric space defined by ted , and explains why the problem of finding ted - averages is ill - posed . both metrics are generally np complete to compute for @xmath8 trees . we explain how semi - labeling schemes can be used to handle complexity problems , and illustrate this by computing qed means for sets of trees extracted from pulmonary airway trees as well as synthetic planar data trees . the qed lacks an obvious decomposition strategy for dynamic programming , and involves an optimization term which makes it more time consuming to compute than the ted metric . the two metrics have similar matching properties , and thus , for applications which do not require a unique geodesic ( e.g. , clustering ) the ted metric is just as suitable as the qed metric . however , for statistical computations , the qed metric is the only suitable metric . this is emphasized by our experiments , where computed qed and ted means illustrate that the ted means are missing important structure . future research will be centered around two points : development of nonlinear statistical methods for the singular tree - shape spaces , and finding more efficient approximations and heuristics for the qed metric using both the tree geometry and the tree - space geometry . the latter will pave the road for computing averages and modes of variation for large , real @xmath8 data trees . this is by no means trivial , due to the non - smooth structure of the tree - shape space and the complexity of computing exact distances and geodesics . the purpose of this appendix is to explain the connection between genericity , defined as a topological property , and probability . in particular , we explain why a generic property can be assumed to hold with probability one , or almost everywhere , with respect to natural probability measures . the main point of the appendix is the contents of theorem [ mainthm ] . a property in a topological space @xmath9 is said to be _ generic _ if it holds on a dense , open subset of @xmath9 . a collection @xmath328 of subsets of some set @xmath9 is a _ @xmath329-algebra _ if @xmath23 has the following properties : * @xmath330 . * if @xmath331 , then its complement is also an element : @xmath332 . * if @xmath333 for a family @xmath334 , then the union is also an element : @xmath335 . a _ measure _ on a @xmath329-algebra @xmath23 is a function @xmath336 satisfying the @xmath329-additivity property : if @xmath337 is a countable family of pairwise disjoint subsets in @xmath23 , then @xmath338 . a property is said to hold _ almost everywhere _ with respect to a probability measure @xmath339 on a space @xmath9 if it holds on a subset @xmath340 such that @xmath341 . this corresponds to holding with probability one , or _ almost surely_. assume given a metric space @xmath190 , and @xmath329-algebra @xmath23 which contains the topology induced by @xmath4 , and measure @xmath339 on the sigma - algebra @xmath23 . assume also given an equivalence class @xmath83 on @xmath9 , and denote by @xmath67 the quotient of @xmath9 with respect to the equivalence . denote by @xmath342 the projection onto the quotient . the set @xmath343 is a @xmath329-algebra on @xmath38 . * @xmath344 so @xmath345 . * if @xmath346 then @xmath347 so @xmath348 , so @xmath349 . * if @xmath350 for @xmath334 then @xmath351 for all @xmath117 , so @xmath352 so @xmath353 . the function @xmath354 given by @xmath355 is a measure on @xmath356 . if @xmath357 is a countable family of pairwise disjoint subsets in @xmath356 , then @xmath358 is a countable family of pairwise disjoint subsets in @xmath23 , and @xmath359 . there are lebesgue induced measures @xmath360 and @xmath361 on the tree - spaces @xmath38 and @xmath96 , coming from the lebesgue measure @xmath362 on @xmath9 . for these measures , subsets of positive codimension have measure zero . in particular , generic properties hold almost everywhere with respect to the lebesgue induced measure . @xmath363 recall that we also work on a restricted subspace @xmath10 which consists of trees of certain fixed topologies that are collapsed versions of the maximal tree @xmath15 . the subspace @xmath11 is a finite union @xmath364 where each @xmath365 is a @xmath4-dimensional linear subspace of @xmath9 corresponding to a certain fixed collapsed version of @xmath15 described by its nonzero edge set @xmath80 . we assume , moreover , that @xmath366 whenever @xmath367 . each @xmath365 now has a lebesgue measure @xmath368 of its own , and we obtain a lebesgue induced measure @xmath369 on @xmath11 defined as @xmath370 the measure @xmath369 gives rise to lebesgue induced measures on @xmath88 and @xmath192 as explained above . there are lebesgue induced measures @xmath371 and @xmath372 on the tree - spaces @xmath88 and @xmath192 , coming from the lebesgue induced measure @xmath369 on @xmath11 . if all @xmath365 have equal dimension , subsets of positive codimension have measure zero . in particular , generic properties hold almost everywhere with respect to the lebesgue induced measure . @xmath363 @xcite let @xmath339 and @xmath373 be two measures on a @xmath329-algebra @xmath23 . the measure @xmath373 is _ absolutely continuous _ with respect to @xmath339 if and only if @xmath374 for each @xmath375 with @xmath376 . [ mainthm ] let @xmath373 be a probability measure which is absolutely continuous with respect to the lebesgue induced measure @xmath377 on one of the tree - space @xmath38 or @xmath96 , or of @xmath88 or @xmath192 under the assumption that all @xmath365 have equal dimension . now any subset of positive codimension has zero measure with respect to @xmath373 . in particular , generic properties hold almost surely , or with probability @xmath298 , with respect to @xmath373 . @xmath363 note that the converse is not necessarily true : properties which hold almost surely are not necessarily generic . in this appendix , we extend the explanation of how the approximate qed distance and geodesic computation specified in section 4.2 in the main paper can be made in practice . we shall consider computation of distances in the space of ordered trees @xmath38 , as the complete search reduction from ordered trees to the unordered tree - space @xmath96 is not hard to implement . recall that the tree - space @xmath38 is constructed from the euclidean space @xmath378 consisting of all trees spanned by a certain maximal combinatorial tree @xmath17 . the construction is made by identifying different representations of the same tree , with the help of an equivalence relation @xmath83 . denote by @xmath62 the equivalence class of a point @xmath61 . as discussed in the article , there is a standard way of defining a distance function on @xmath67 , given by the quotient pseudometric @xmath379 which is shown to be an actual metric . the quotient space construction is extremely useful for understanding the geometry of tree - space and proving statistical properties of tree - space based on geometry . however , it is less intuitive how one should proceed in order to actually compute distances in tree - space . in this section , we shall explain how the tree - space gluings are related to the trees themselves . recall that two points in @xmath9 are equivalent under @xmath83 and glued together in the formation of @xmath38 if they represent _ the same _ ordered tree - shape . how can we determine whether two points in tree - space represent the same ordered tree - shape ? this reduces to a tree isomorphism problem . more precisely : * assume given two points @xmath381 . * these points correspond to two combinatorial trees @xmath382 , equal to or derived from @xmath15 , where some of the edges in @xmath383 may be endowed with zero attributes in @xmath307 . * create two new combinatorial trees @xmath384 by deleting the edges in @xmath385 and @xmath386 with zero attributes in @xmath5 and @xmath6 , and using the induced orders on siblings from @xmath385 , @xmath386 to induce total orders on siblings in @xmath387 , @xmath388 . * the question of whether @xmath5 and @xmath6 end up in the same identified subspace is now equivalent to the question of whether @xmath387 and @xmath388 are isomorphic as ordered , combinatorial trees . the question of whether @xmath5 and @xmath6 represent the same tree - shape , is equivalent to the question of whether a ) @xmath387 and @xmath388 are topologically isomorphic , and b ) if yes , whether their ordered sets of attributes are identical . * the observations above give us a way to judge whether two points in @xmath9 are equivalent under the equivalence relation @xmath83 on @xmath9 ; equivalently , whether they are identified in @xmath38 . this realization can be used to reformulate algorithm 1 from the main paper in terms of the trees rather than in terms of the tree - space . * a geodesic in @xmath38 is composed of a set of linear stretches in @xmath9 between equivalence classes that belong to lower - dimensional identified subspaces . each such stretch can change one endpoint tree to the other endpoint tree by i ) gradually shrinking and removing edges , ii ) adding and gradually extending edges , or iii ) gradually changing edges which are present in both trees . all these changes are made simultaneously at constant speed throughout the linear stretch . * given two points @xmath389 and @xmath390 in @xmath38 , how can we find the shortest path joining them which consists of two euclidean stretches ? the length of the shortest such path is @xmath391 . we shall look at this problem in section [ twostretches ] below . * let @xmath389 , @xmath390 be tree - shapes in @xmath38 . let @xmath69 be some tree - shape whose topological structure can be obtained as a collapsed version of both @xmath389 and @xmath390 . * the topology of the tree @xmath69 is obtained from @xmath389 ( and similarly from @xmath390 ) by collapsing two types of edges : * * internal edges ( that is , there is some edge in the tree which is further from the root in @xmath389 ( or @xmath390 ) , which is not being collapsed ) , or * * external edges ( that is , all the edges in @xmath389 ( or @xmath390 ) which are farther from the root are also collapsed ) . * denote the sets of such edges as @xmath393 , @xmath394 , @xmath395 , @xmath396 , respectively . * we are looking for a point @xmath397 which contains some such @xmath69 as a subtree , such that the shortest two - stretch path from @xmath389 to @xmath390 passes through @xmath398 , and consists of the following types of deformations : * * deformation of all edges that collapse to an edge in @xmath69 ; throughout the whole path . * * gradually deforming all external collapsing edges in @xmath389 so that when we reach @xmath6 , their attributes have become zero ; throughout the whole path . * * gradually growing all external collapsing edges in @xmath390 from zero attributes when the deformation starts in @xmath5 ; throughout the whole path * * gradually deforming all internal collapsing edges in @xmath389 so that when we reach @xmath398 , their attributes are zero ; throughout the first stretch . * * from @xmath398 , gradually grow all internal collapsing edges in @xmath390 ; throughout the second stretch . * * thus , @xmath398 contains the edges from @xmath399 as well as the edges from @xmath394 and @xmath396 . + now the distance from @xmath389 to @xmath390 via @xmath398 is : @xmath400 where @xmath401 in case @xmath44 is not an edge in @xmath146 , @xmath402 . * this is illustrated in figure [ algorithm ] . * now we can define @xmath392 : @xmath403 where @xmath398 loops through the set of all trees generated from some tree @xmath69 which is a collapsed version of both @xmath389 and @xmath390 . the simplest possible path between two tree - shapes does not visit any identified subspaces , and has length @xmath404 . we have not in an obvious way taken into account that this might in fact be the shortest path . in order to see that this is part of the computation , let us start out as above : * let @xmath389 , @xmath390 be tree - shapes in @xmath38 . let @xmath69 be some tree - shape whose topological structure can be obtained as a collapsed version of both @xmath389 and @xmath390 . * the topology of the tree @xmath69 is obtained from @xmath389 ( and similarly from @xmath390 ) by collapsing two types of edges : * * internal edges ( that is , there is some edge in the tree which is further from the root in @xmath389 ( or @xmath390 ) , which is not being collapsed ) , or * * external edges ( that is , all the edges in @xmath389 ( or @xmath390 ) which are farther from the root are also collapsed ) . * assume that it is possible to find some tree @xmath69 which is obtained by only collapsing external edges . * the shortest path passing through the corresponding @xmath398 will be a single euclidean stretch , which can be obtained as described above . * hence , the option that @xmath404 might give the optimal path is part of the computation of @xmath392 . there is a chance that the intermediate trees appearing in this algorithm are too large for @xmath38 ( for example as in fig . [ algorithm ] . we have chosen to disregard this , and cut the intermediate trees off in iterative algorithms if they are too large . this can also be controlled by cutting @xmath398 off by default . * in order to compute @xmath391 , we need to search through all different collapsed versions @xmath69 of @xmath389 and @xmath390 . * in order to reduce computation time , we approximate this by setting a bound @xmath267 on the degree of vertices of the collapsed versions . * for @xmath405 , the number of such subtrees is @xmath406 , where @xmath27 is the number of edges in the tree . for @xmath407 , the number of such subtrees is @xmath408 . in our experiments we use @xmath405 . note that if we can answer the question in vii ) in the beginning of section [ reformulation ] , then we can compute @xmath409 as well : @xmath410 where @xmath398 goes through a family of trees derived from all possible collapsed trees @xmath69 obtained from @xmath389 . t. schlathlter , c. lorenz , i. c. carlsen , s. renisch , and t. deschamps . simultaneous segmentation and tree reconstruction of the airways for virtual bronchoscopy . in _ _ , volume 4684 , pages 103113 , 2002 .

many of the most accurate heavy - quark mass determinations are related to the spectral functions of the heavy - quark vector current , which can be measured in @xmath0 collisions . the energy region near the pair production threshold is particularly sensitive to the mass value . the bottom and charm masses are usually inferred from suitable averages of the pair production cross section . looking to the future , the measurement of the top quark pair production cross section in the threshold region at the planned international linear collider ( ilc ) would lead to a very precise knowledge of the top mass directly from the energy dependence of the cross section , even though the toponium resonances are smeared out due to the large top quark width @xcite . to put this into perspective , the current top mass value from direct production at the fermilab tevatron is @xmath5gev @xcite , and @xmath6gev from the large hadron collider ( lhc ) run at @xmath7tev centre - of - mass energy @xcite . a less precise value can be obtained from the total production cross section as illustrated for example in @xcite . further reduction of the uncertainty below the one of the tevatron result is very difficult due to the complicated theoretical systematics of top jet mass reconstruction at hadron colliders . the above numbers should be compared to a precision of @xmath8mev that can be achieved experimentally @xcite from the @xmath9 threshold scan at the ilc . in particular , while it is not evident which renormalized top mass parameter is determined with the quoted accuracy at the tevatron and lhc , the threshold cross section in @xmath0 collisions provides an observable that can be unambiguously related to a particular top mass definition . aside from determining a fundamental parameter of the standard model ( sm ) , accurate top mass measurements are of interest for extrapolating the sm or its tev scale extension to higher energies , either in the context of `` electroweak precision tests '' or a theory of the yukawa couplings . furthermore , the recent discovery of a higgs boson @xcite with a mass of about 125@xmath10gev if interpreted as the sm higgs boson determines the higgs self - coupling and leads to the conclusion that the sm vacuum becomes metastable at scales above @xmath11gev @xcite . the precise value of this scale turns out to depend very sensitively on the value of the top quark mass . finally , the absence of any hint on physics beyond the sm in high - energy collisions at lhc has renewed the interest in performing precision measurements of properties of sm particles , including the top quark mass , width and yukawa coupling . measurements at the top pair production threshold are uniquely suited for this purpose . the challenge is thus to calculate the heavy - quark to zero in the top - quark expressions . ] spectral functions precisely in the threshold region . this kinematic region is characterized by two features , which make the theoretical calculation of qcd corrections rather different from standard loop calculations : the small three - velocity @xmath12 of the heavy quarks , which allows to expand feynman diagrams in @xmath12 rather than calculate them exactly , and the strong colour - coulomb force , which on the other hand requires certain diagrams to be summed to all orders in the strong coupling @xmath13 similar to bound - state calculations in quantum electrodynamics . the expansion of the cross section relative to the ultra - relativistic point - particle cross section @xmath14 is then organized as @xmath15 where the overall factor of @xmath12 arises from the phase - space of the two produced massive particles , and the order of the various terms is indicated explicitly . to perform the expansion and the required summation of perturbation theory to all orders beyond the next - to - leading order ( nlo ) , non - relativistic effective field theory @xcite and the threshold expansion of feynman diagrams @xcite are the methods of choice . the cross section is then obtained from the expression - boson exchange . see section [ sec : eft ] for the full expression . ] @xmath16 g(e ) + \ldots\right ) \right],\ ] ] where @xmath17 , @xmath18 denote certain relativistic matching corrections , @xmath19 , @xmath20 the number of colours , and @xmath21 represents a two - point function of heavy - quark currents in the non - relativistic effective theory . the purpose of this paper is to present results of the part of the third - order ( nnnlo ) qcd corrections to the heavy quark anti - quark pair production cross section near threshold related to the correlation function @xmath21 , which contains the all - order summation . since this concludes the non - relativistic third - order calculation , we also present details of the methods and calculations that have been used but not given in earlier publications . for the top pair production threshold the leading order ( lo ) and next - to - leading order ( nlo ) approximations to the cross section ( [ systematics ] ) have been examined long ago @xcite . several other aspects of the threshold such as top momentum distributions and polarization have been computed at this order @xcite . beyond nlo , matching the non - relativistic approximation to qcd is non - trivial , because the separation between relativistic and non - relativistic physics is no longer unambiguous . a consistent field theoretical approach based on non - relativistic effective qcd is now required . the second - order ( nnlo ) qcd corrections to the total pair production cross section have been computed in this framework about ten years ago @xcite and turned out to be surprisingly large even for top quarks . while some of the large corrections can be understood as being due to mass renormalization @xcite , and can be avoided by a change of renormalization convention , there remains an apparently slow convergence of successive approximations to the normalization of the cross section , which necessitates the calculation of the nnnlo term . an alternative approach that sums logarithms of @xmath12 has also been pursued , and an improvement of convergence has been found in a ( still partially incomplete ) next - to - next - to - leading logarithmic ( nnll ) approximation @xcite . nonetheless , the nnnlo non - logarithmic terms not included in the nnll approximation are required to be certain that the theoretical calculation is sufficiently accurate for the proposed mass measurement at the ilc . this is the main motivation for the present work . over the past years a significant number of results relevant to a nnnlo calculation or partial results for third - order quarkonium energy levels and wave - functions at the origin have already appeared @xcite . in the following we summarize the status of the nnnlo top cross section calculation , the contribution of the present paper , and what remains to be done to complete the nnnlo calculation . while the resummed cross section is calculated in an effective field theory ( eft ) , a number of matching calculations has to be performed to guarantee that the eft reproduces qcd to the required accuracy . this is done in two steps . hard matching ( scale @xmath22 ) yields the coefficients of the non - relativistic qcd ( nrqcd ) interactions and heavy - quark currents ; soft matching ( scale @xmath23 ) the quark - anti - quark potentials . at nnnlo the coefficients of several subleading nrqcd interactions must be determined with one - loop precision . this calculation has been performed in @xcite . however , as shall be explained , the calculation of the cross section requires the @xmath24 terms of the coefficient functions . we therefore repeated the nrqcd matching calculation , confirm the results of @xcite and provide the expressions for the @xmath24 terms in this paper ( see also @xcite ) . hard matching of the non - relativistic currents at nnnlo is needed at the one - loop level for the sub - leading currents and at the three - loop level for the leading current . the former are known @xcite , and will be rederived in the present paper , but only the fermionic contributions to the three - loop matching coefficient @xmath25 have been calculated up to now @xcite . as discussed in @xcite , the incomplete result for @xmath25 presently constitutes the most important missing nnnlo contribution to the cross section . as concerns soft matching , the potentials of order @xmath26 must be determined with one - loop precision , the @xmath27 potential with two - loop precision , and the @xmath28 coulomb potential at three loops , since @xmath29 counts as the bohr radius @xmath30 . except for the coulomb potential , the coefficients of the other potentials have been calculated in @xcite , but again these results are not sufficient for the cross section calculations , since the @xmath24 terms of all these potentials are needed . we repeated the calculation of the one - loop @xmath26 potentials , confirm the results of @xcite and provide the expressions for the @xmath24 terms in this paper ( see also @xcite ) . the @xmath24 term of the two - loop @xmath27 potential coefficient @xmath31 remains presently unknown . as concerns the three - loop coulomb potential , the fermionic contributions to the three - loop coefficient @xmath32 have been calculated first in @xcite and the full result is now also known @xcite . to summarize : the matching coefficients required for the nnnlo calculation of the heavy - quark production cross section near threshold are known except for the non - fermionic contributions to @xmath25 , and the @xmath24 terms of @xmath31 . out of these , only @xmath25 is expected to have a major impact on the final result @xcite . work on these missing contributions is in progress . after matching the heavy - quark currents and lagrangians the cross section calculation is mapped to the calculation of the imaginary part of the two - point function @xmath21 of non - relativistic currents in ( [ r1 ] ) . the leading - order colour coulomb potential is now part of the leading - order effective lagrangian , since the coulomb interaction is strong near threshold . the propagators to be used in the perturbative calculation of the two - point function are the coulomb green functions , making this part of the computation similar to qed bound - state problems . the calculation can be divided into three parts : contributions up to the third order involving only the coulomb potential , already completed in @xcite ; the ultrasoft contribution , appearing first at nnnlo , which has been computed in @xcite ; and finally , contributions involving at least one potential other than the coulomb potential ( `` non - coulomb potential contribution '' ) , which are not yet known . the main result of this paper is the missing non - coulomb potential contribution . compared to the coulomb contributions the major complication is the singular nature of the potential insertions . the ultraviolet divergences must be regulated in dimensional regularization in a scheme consistent with the calculation of the matching coefficients order by order in the strong coupling , while retaining the resummation of infinitely many coulomb gluon exchanges by the use of coulomb green functions , whose @xmath33-dimensional expression is unknown . the techniques we apply are an extension of those used in the nnlo calculation @xcite . since the method of that calculation was never written up ( though some results are scattered in @xcite ) , we devote some effort to presenting the third - order calculation in some technical detail . in addition to the dominant production of the top - quark pair though a virtual photon there is also a @xmath34-boson contribution . the contribution from the vector - coupling of the @xmath34 is trivially inferred from the photon - mediated cross section , while the axial - vector coupling contribution is suppressed near threshold and begins only at nnlo . thus , only the first - order correction to the axial - vector in non - relativistic perturbation theory is needed . some results at this order are available @xcite , but none of these results are given in dimensional regularization . we recalculated these small contributions , and will present them elsewhere , since in this paper we focus on the missing third - order terms in the vector current contribution . to summarize : together with the results of this paper , the matrix element calculation is complete to nnnlo . much less work has been done on electroweak and electromagnetic corrections . counting the electromagnetic and electroweak coupling as two powers of the strong coupling , electromagnetic corrections contribute from nlo through the electromagnetic coulomb potential . this effect is easily included and has been discussed in @xcite . electroweak contributions to the matching coefficients of nrqcd currents and production operators have been calculated in @xcite . the formalism for calculating initial - state radiation , and soft and collinear photon corrections in general , simultaneous with summing coulomb exchange has been worked out in @xcite for @xmath35-boson pair production . its application to top quarks is straightforward . it remains to be done to put all these electroweak effects together . for top quarks the sizeable decay width @xmath36 introduces further complications . in leading order the width is correctly included by evaluating the current two - point functions in pnrqcd at complex energy argument @xmath37 @xcite , where @xmath38 . we adopt this prescription as the _ definition _ of the pure qcd contributions to the cross section . beginning at nlo there exist further contributions related to the finite width . since the physical final state is @xmath39 , boson may be regarded as stable . ] there exist irreducible `` backgrounds '' related to off - shell and single or non - resonant top - quark pair production . in fact , the qcd contribution as defined above can not be unambiguously separated from electroweak effects at this order perhaps not surprisingly , since the top quark width itself is such an effect and the fact that a physical scattering cross section should refer only to stable ( or sufficiently long - lived ) particles in the final state , must be taken into account . the incompleteness of the qcd cross section is signaled explicitly by the presence of uncanceled singularities in dimensional regularization with coefficients proportional to @xmath36 starting at nnlo . the origin and consistent cancellation of these singularities is discussed in @xcite and the corresponding calculations of electroweak effects and non - resonant contributions to a physical final state such as @xmath40 can in principle be performed in the framework of unstable - particle effective field theory @xcite as already done for @xmath35 pair production @xcite . the corresponding calculation of the non - resonant nlo correction for top production has been performed in @xcite and confirmed by a different method in @xcite . rather than embarking on the rather difficult computation of non - resonant contributions up to nnnlo , however , a more promising approach suppresses them by appropriate invariant mass cuts @xcite , which will anyway be applied experimentally . calculations of top quark pair production near threshold with cuts on the final - state @xmath41 invariant masses have appeared recently @xcite in the non - relativistic qcd and unstable particle effective theory frameworks . as should be expected , the non - resonant contributions to the @xmath40 are sizeable below the nominal top pair production threshold , and hence can change the shape of the threshold cross section in the region of interest for the top - quark mass determination . in the present paper , we focus only on the qcd part of the problem . we shall , however , make the finite - width @xmath42 poles explicit , so that they can be canceled analytically with future computations of electroweak effects . indeed , the pole parts of the nnlo non - resonant contribution have recently been computed @xcite and the cancelation has been verified . to summarize : while the formalism for computing electroweak and finite - width effects consistently is in place , and bits and pieces have already been calculated , significant work remains to be done on electroweak effects at nnnlo . with the third - order qcd calculation now being nearly complete , improvements in the prediction of realistic cross sections ( including electroweak effects and cuts ) must be made . 0.3 cm since our intention to present the concepts , techniques and calculations in some detail resulted in a rather lengthy text , we have split it into two parts . part i presents the effective field theory set - up , the nrqcd and pnrqcd matching coefficients and ends with a master formula for the third - order heavy - quark pair production cross sections . this part could also be read as a review of non - relativistic effective theory in the weak - coupling regime complementary to @xcite . part ii contains the actual pnrqcd matrix element calculation together with a numerical estimate of the new contributions to the top cross section . ( a preliminary version has been presented in @xcite . ) the small p - wave contributions to the cross section from the axial - vector coupling of the @xmath34-boson will be presented separately in @xcite . the outline of the paper i is as follows : in section [ sec : eft ] we review the effective field theory framework and discuss the power - counting arguments that lead to the identification of the matching coefficients needed for the nnnlo calculation . the subsequent two sections deal with matching qcd to a sequence of two non - relativistic effective theories , nrqcd and pnrqcd . section [ sec : nrqcd ] discusses the nrqcd aspects of the calculation . in particular , we calculate the relevant one - loop matching coefficients including the new @xmath24 terms and collect all other results that feed into the cross section calculation . the second matching step from nrqcd to pnrqcd is discussed at length in section [ sec : pnrqcd ] , since a coherent summary is not yet available in the literature . among the new results of this section are the path - integral derivation of the pnrqcd lagrangian ( neglecting ultrasoft gluons ) , the @xmath24 terms of the one - loop potentials , and a discussion on the non - renormalization of currents in the nrqcd to pnrqcd matching . pnrqcd perturbation theory in the pole - mass scheme provides a poor approximation to the top - quark pair production cross section near threshold . the top quark decay width is obviously an important effect , as is the conversion from the pole mass renormalization scheme that is employed in the primary calculations to renormalization schemes that absorb large corrections into the mass counterterm , which is a prerequisite for reliable perturbative calculations @xcite . furthermore , a resummation of pnrqcd perturbation theory for the green function is necessary in the vicinity of the bound state poles despite the sizeable top quark width . these refinements will be explained in paper ii . in section [ sec : master ] we conclude paper i by providing the master formula for the computation of the third - order cross section in pnrqcd . in this section we present the relation of the pair production cross section to correlation functions of heavy quark currents together with the arguments , why this relation holds true at nnnlo . we review the scales and momentum regions relevant to the problem both of which are central to the systematics of the effective theory approach . the basic top pair production mechanisms in @xmath0 annihilation are shown in the upper part of figure [ fig : eett ] . since we work to lowest order in the electromagnetic and electroweak couplings , the optical theorem allows us to relate the total cross section @xmath43 of the process @xmath44 to the two point functions of the vector and axial - vector heavy quark current . we define @xmath45 for the vector current @xmath46 and the axial vector current @xmath47 . the cross section is then given by @xmath48,\end{aligned}\ ] ] where @xmath49 is the high - energy limit of the @xmath50 production cross section , @xmath51 the center - of - mass energy squared , and @xmath52 the @xmath34-boson mass . @xmath53 denotes the top quark electric charge in units of positron charge and @xmath54 is the electromagnetic coupling . the vector and axial - vector couplings of fermion @xmath55 to the @xmath34-boson are given by @xmath56 with @xmath57 the weak mixing angle , @xmath58 the electric charge of fermion @xmath55 and @xmath59 its third component of the weak isospin . the dominant production mechanism is through the coupling to the virtual photon . the vector coupling of the @xmath34-boson increases the photon - mediated cross section by only about 8% in the threshold region @xmath60 . the axial - vector contribution is even smaller , since the axial coupling is suppressed near threshold by the small velocity of the top quarks . @xmath61 contributes to ( [ eq : sigtt ] ) only at nnlo relative @xmath62 . ( [ eq : sigtt ] ) which relates the inclusive top cross section to the spectral functions of heavy - quark currents is not exact . there exist top production mechanisms , shown in the lower part of figure [ fig : eett ] , which are not captured by the heavy - quark current correlation functions , since the photon or @xmath34-boson couples to light quarks . vice versa , there exist cuts contributing to @xmath63 related to annihilation subdiagrams , see figure [ fig : annihilation ] below , which do not contain top quarks and hence should be excluded . we shall discuss in the next subsection that these contributions are either highly suppressed and not relevant at third order , or can easily be included . the characteristic non - relativistic energy in threshold production is much smaller than @xmath64 . we define @xmath19 and the top quark `` velocity '' @xmath65 . this is related to another often used velocity parameter @xmath66 by @xmath67 real values of @xmath68 and @xmath69 should be interpreted with a @xmath70 prescription . we will extensively use another variable @xmath71 defined by @xmath72 here and below @xmath13 without any argument denotes the strong coupling in the @xmath73 scheme @xcite at the renormalization scale @xmath74 , and @xmath75 . the value of @xmath71 determines when resummation to all orders in @xmath13 is necessary . above threshold , the variable @xmath71 is purely positive - imaginary , below threshold it is real and positive . the threshold region is characterized by an absolute value of @xmath71 of order 1 or larger . in particular , the coulomb bound state poles are found at positive integer values of @xmath71 . conventional fixed - order perturbation theory can be used only when @xmath76 . as will be discussed later the width of the top quark is accounted for by substituting @xmath77 . thus , as @xmath69 varies from @xmath78 to @xmath79 the variable @xmath71 sweeps through a curve in the complex plane that begins at the origin , moves out into the first quadrant into the direction of the positive real axis and returns to the origin from above near the imaginary axis . the absolute value of @xmath71 along this curve is always smaller than @xmath80 , which is about 1.2 for top quarks , with @xmath81 , but since it reaches order one in the threshold region , the perturbation expansion in @xmath13 breaks down and resummation is necessary . near the heavy - quark pair production threshold only a small kinetic energy @xmath82 is available to the final state . in the natural frame where @xmath83 this implies that the typical three momentum of a heavy quark is of order @xmath84 ( about 20 gev for top quarks ) , while the energy and momentum of any other nearly massless particle can at most be @xmath85 ( about 2 gev for tops ) . the presence of several small scales propagates into the loop diagrams that contribute to the spectral functions and causes a breakdown of the standard perturbation expansion in the strong coupling @xmath13 . however , since @xmath12 is small one does not have to compute the loop integrals exactly an expansion in @xmath12 suffices . this leads to a reorganized expansion as shown in ( [ systematics ] ) , in which @xmath13 and @xmath12 are expansion parameters but @xmath86 or , equivalently , @xmath71 is of order one . for a given feynman diagram the expansion in @xmath12 can be constructed without first computing the full expression using the threshold expansion @xcite . the method uses that every diagram is the sum of terms , for which each loop momentum is in one of the following four regions : @xmath87 when on - shell , only massless particles ( gluons , light quarks and ghosts ) can be ultrasoft , and only the heavy quarks can be potential . in each region , the loop integrand is expanded in the terms which are small in the corresponding region and the loop integration of the expanded integrand is carried out over the complete @xmath33-dimensional space - time volume . the expansion generates ultraviolet and infrared divergences which are regulated dimensionally ( @xmath88 ) and subtracted according to the @xmath73 prescription . however , the divergences generated by the separation of the diagram into regions cancel in the sum over all terms . the procedure just described is largely equivalent to constructing appropriate effective lagrangians within dimensional regularization , but it clarifies the correct matching procedure which is subtle in dimensional regularization if the effective theory contains more than one scale as is the case in non - relativistic qcd @xcite . the threshold expansion also synthesizes the non - relativistic velocity power counting rules developed for the different modes ( momentum regions ) in @xcite . in the construction of the non - relativistic and resummed expansion of the pair production cross section we derive effective lagrangians in two steps by integrating out the large momentum modes according to the following scheme : @xmath89 \quad & \mu > m \\[0.3 cm ] \big\downarrow & \\[0.3 cm ] { \cal l}_{\rm nrqcd}\,[q(s , p),\,g(s , p , us ) ] \quad & m v < \mu < m \\[0.3 cm ] \big\downarrow & \\[0.3 cm ] { \cal l}_{\rm pnrqcd}\,[q(p),\,g(us ) ] \quad & \mu < m v\\[0.3 cm ] \end{array}\ ] ] in square brackets we display the modes of the heavy quarks ( @xmath90 ) and massless particles ( @xmath91 ) which are still contained in the effective lagrangian ; the others are integrated out when the energy cut - off @xmath74 is lowered as indicated on the right . the first step leads to nrqcd @xcite , in which all interactions are local , since only the short - distance hard modes have been eliminated . the expansion rules of the threshold expansion _ define _ the dimensionally regularized nrqcd lagrangian . the second step whereby soft modes and potential massless modes are integrated out was suggested in @xcite in the context of the effective lagrangian and threshold expansion method . the result is the potential nrqcd ( pnrqcd ) lagrangian @xcite . the pnrqcd lagrangian is not local . it contains spatially non - local but temporally local , i.e. instantaneous interactions of the heavy quarks , since the three - momentum of the potential heavy quark field still present in pnrqcd is of the same order as the one of the modes integrated out . these interactions provide a precise definition of the concept of `` heavy - quark potentials '' . perturbation theory in pnrqcd resembles quantum - mechanical perturbation theory closely , since the leading colour - coulomb interaction is part of the unperturbed theory . thus , the propagator of pnrqcd includes the leading coulomb interaction exactly , which effects the required resummation of conventional perturbation theory to all orders . to illustrate the velocity scaling of feynman diagrams , we consider the power counting for the loop integrand . ( [ eq : scales ] ) implies that the integration measure @xmath92 scales as @xmath93 , @xmath94 , @xmath95 and @xmath96 , when @xmath97 is hard , soft , potential and ultrasoft , respectively . the denominator of a gluon ( massless ) propagator with momentum @xmath98 is approximated at leading order in a given region by : @xmath99 \displaystyle -\frac{1}{\bm{\ell}^2}+\ldots & \mbox { potential } ( v^{-2 } ) \end{array } \right.\ ] ] the velocity scaling is given in brackets . for the heavy - quark propagator with momentum @xmath100 the denominators are : @xmath101 \displaystyle \frac{1}{2 m}\,\frac{1}{\ell^0 } + \ldots & \mbox { soft } ( v^{-1 } ) \\[0.5 cm ] \displaystyle \frac{1}{2 m}\,\frac{1}{e/2+\ell^0-\bm{\ell}^2/(2 m ) } + \ldots & \mbox { potential } ( v^{-2 } ) \end{array } \right.\ ] ] ladder diagrams.,scaledwidth=90.0%,height=68 ] 0.2 cm using these scaling rules , it is easy to see why an all - order resummation of feynman diagrams is required in the threshold region . it will become clear from the later systematic derivation that the relevant diagrams are the ladder diagrams shown in figure [ fig : ladder ] , and that the dominant term in the velocity expansion arises from the loop momentum region when all loop momenta are in the potential region . adding an additional rung to the ladder adds one potential gluon ( @xmath102 ) and two potential heavy - quark propagators ( @xmath103 ) to the diagram . the numerator of the diagram contains no velocity suppression factors , hence accounting for the potential loop measure ( @xmath95 ) and strong coupling from the two additional vertices ( @xmath104 ) , we find that each rung provides a factor of order @xmath86 , which is unsuppressed in the threshold region . it will be seen below that only potential gluon exchange generates this @xmath105 enhancement , which is equivalent to the statement that only the leading coulomb interaction must be included in the unperturbed effective lagrangian . we now return to the discussion of heavy - quark production mechanisms not captured by ( [ eq : sigtt ] ) , which expresses the cross section in terms of the heavy - quark current spectral functions . in the case of heavy - quark radiation ( lower left in figure [ fig : eett ] ) the final state consists of @xmath106 , and since the available energy at threshold is limited to @xmath85 , the light quarks must be ultrasoft . in the three - loop diagram that represents the square of the heavy - quark radiation amplitude , the @xmath107 loop must be potential and the two other loops ultrasoft , which leads to a factor of @xmath108 from the loop integration measure . the intermediate gluon and light - quark propagators in the amplitude must be hard to produce the @xmath107 pair and hence do not contribute inverse powers of @xmath12 . the two potential heavy quark and the two ultrasoft light quark propagators ( @xmath109 ) supply a factor of @xmath102 each , so the heavy - quark radiation contribution to the cross section scales at least as @xmath110 which should be compared to @xmath12 for the leading term . inspection of the analytic expression @xcite confirms this result , hence this contribution can be safely neglected . in the case of singlet production ( lower right in figure [ fig : eett ] ) through the coupling of the virtual photon or @xmath34-boson to light quarks the dominant term comes from three hard loops , leading to the counting @xmath111 , which represents a third - order correction to the cross section . while not part of @xmath112 this mechanism can be included in the three - loop short - distance coefficient @xmath113 of the non - relativistic heavy - quark current , which is discussed below , although it is not known at present . note that this contribution to @xmath113 is complex , but the imaginary part should be discarded , since it corresponds to the three - gluon and light - quark cut , which is not part of the heavy - quark production cross section . a similar singlet - production diagram exists for the axial - vector coupling with only two gluons coupling to the light - quark triangle , but due to velocity suppression this contribution begins only at fourth order . we have thus argued that the production mechanisms not included in @xmath114 are either suppressed or easily included at third order . consider now figure [ fig : annihilation ] , which shows a diagram contained in @xmath112 , but whose three - gluon cut should not be part of the heavy - quark cross section . the possible loop momentum regions for this diagram are h - h - h - h , p - h - h - p , p - h - h - h and h - h - h - p , where the first and last letter refers to the left and right heavy quark loop , respectively . in the all - hard configuration only the three - gluon cut contributes to @xmath112 , so the correct prescription is to simply not include this configuration . the p - h - h - p configuration may be interpreted as heavy - quark production followed by rescattering through annihilation . annihilation is suppressed by @xmath115 relative to coulomb exchange as can be seen by counting loop integration and propagator factors , so this configuration is relevant only from fourth order as a contribution to the @xmath116 potential . the remaining two configurations with only one potential heavy quark loop are analogous to the singlet production mechanisms . that is , one drops the imaginary part of the h - h - h subdiagram , which comes from the three - gluon cut and associates its real part to @xmath117 . there exists a three - loop diagram similar to figure [ fig : annihilation ] but with two gluon lines only for the axial - vector coupling , but as in the singlet diagram with a light - quark triangle discussed above , the axial - vector coupling implies another factor of @xmath118 , so this diagram is never relevant at third order . in this section we discuss the matching of the vector current correlation function for @xmath60 to its equivalent representation in non - relativistic qcd . this amounts to integrating out the hard modes , which correspond to `` relativistic effects '' involving the scale of the heavy - quark mass . non - relativistic qcd is expressed in terms of a two - component quark field @xmath119 and the corresponding anti - quark field @xmath120 to represent the remaining soft and potential fluctuations of the original quark field . the effective gluon field @xmath121 can be soft , potential and ultrasoft . before going into the details of the lagrangian and power counting we briefly sketch the result . as will be shown below the expansion of the vector current @xmath122 in terms of the non - relativistic fields is given by @xmath123 where the hard matching coefficients @xmath17 , @xmath18 have perturbative expansions in @xmath13 . in the `` rest frame '' @xmath124 , eq . ( [ eq : pi ] ) implies @xmath125 , so @xmath126 g(e ) + \ldots,\ ] ] where the neglected terms on the right - hand side include a subtraction term that does not contribute to the imaginary part of @xmath62 as well as terms beyond the third order ( nnnlo ) . the important quantity is the two - point function of the non - relativistic current @xmath127(x)\ , [ \psi^{\dag}\sigma^i\chi](0 ) ) evaluated in non - relativistic qcd ( nrqcd ) . the terms proportional to @xmath69 in ( [ eq : pitonrqcd ] ) arise from expanding the prefactor @xmath128 and from the @xmath129 suppressed current in ( [ eq : qcdvectorcurrent ] ) , whose matrix element can be reduced to the one of the leading current by an equation - of - motion relation derived later . thus the main ingredients to the non - relativistic representation are the calculation of @xmath21 and the current matching coefficients . similar relations hold for the axial - vector contribution to the cross section ( [ eq : sigtt ] ) , which arises from @xmath34-boson exchange . the axial - vector current @xmath130 is represented in nrqcd by the expansion @xmath131\chi + \ldots,\end{aligned}\ ] ] with hard matching coefficient @xmath132 . as is the case for the vector current , only the spatial components of the current contribute to the cross section , since the lepton tensor from the @xmath0 initial state is transverse to both initial state momenta when the electron mass is neglected . only the leading term in the @xmath133 expansion is needed for nnnlo accuracy , since the derivative in the leading current implies the well - known p - wave velocity suppression . the qcd correlation function is then given by the expression @xmath134^\dagger(x)\ , [ \psi^{\dag}\gamma^i\chi](0 ) ) + \ldots,\qquad \nonumber\end{aligned}\ ] ] where @xmath135 $ ] . for the present purpose the non - relativistic effective lagrangian can be divided into five parts , @xmath136 nrqcd feynman rules for two - quark vertices up to order @xmath129 . dashed ( curly ) lines denote the @xmath137 ( @xmath138 ) gluon field . @xmath139.,title="fig:",height=748 ] 0.3 cm -0.3cmthe gluon field is contained in the gauge - covariant derivative @xmath140 , field strength tensor @xmath141 $ ] , and the chromoelectric and chromomagnetic fields defined as @xmath142 , \nonumber \\ \bm{\sigma}\cdot{{\bf { b } } } & \equiv & -\frac{1}{2}\,\sigma^{ij } g^{ij } , \label{eq : magneticfield}\end{aligned}\ ] ] where @xmath143 $ ] and @xmath144 . with these definitions the bilinear heavy - quark lagrangian is given by . the @xmath145 term is misprinted in eq . ( 8) of @xcite , where @xmath146\psi$ ] should read @xmath147 . ] @xmath148 & & + \ , \psi^{\dag}\bigg(\frac{d_2 g_s}{8 m^2}[d^i , e^i ] + i \ , \frac{d_3 g_s}{8 m^2}\sigma^{ij}\{d^i , e^j\}\bigg ) \psi \nonumber \\[0.1 cm ] \nonumber & & + \,\psi^{\dag}\bigg(-d_w g_s\frac{\{{\bf d}^2,{\bm { \sigma } } \cdot { \bf b}\}}{8m^3 } + d_a g_s^2 \frac{{\bf b}^2-{\bf e}^2}{8m^3}+ d_b g_s^2\frac{\sigma^{ij}(b^i b^j- e^i e^j)}{8m^3}\,\bigg ) \psi \nonumber\\[0.1 cm ] & & + \,o\bigg(\frac{1}{m^4}\bigg ) , \\[0.2 cm ] { \cal l}_{\chi}&= & - { \cal l}_{\psi}\quad\mbox{with $ \psi\to\chi , i d^0\to -i d^0 , e^i\to - e^i$}\,.\end{aligned}\ ] ] note that in @xmath33 dimensions we can not define the @xmath149 individually , since they do not represent the components of a @xmath150 dimensional vector . however , all we need are scalars such as ( [ eq : magneticfield ] ) and @xmath151 , @xmath152 , which can be consistently defined through the @xmath33 dimensional field strength tensor : @xmath153\,.\ ] ] the four - fermion quark - antiquark terms in the effective lagrangian read @xmath154 \psi \ , \chi^{\dag}[\sigma^i,\sigma^j ] \chi + \nonumber \\ \label{eq : nrqcdqqqq } & & + \,\frac{d_{vs}}{m^2 } \psi^{\dag } t^a \psi \ , \chi^{\dag } t^a \chi - \frac{d_{vv}}{8 m^2 } \psi^{\dag } t^a [ \sigma^i,\sigma^j ] \psi \ , \chi^{\dag } t^a [ \sigma^i,\sigma^j ] \chi+o\bigg(\frac{1}{m^3}\bigg),\end{aligned}\ ] ] where the factor @xmath155 in the definition of the spin - triplet operators has been inserted , since in four dimensions @xmath156\otimes [ \sigma^i,\sigma^j ] = -8 \,\sigma^i\otimes\sigma^i$ ] . the pure gluon lagrangian takes the form @xmath157 finally , @xmath158 is the same as the light - quark lagrangian in full qcd . the feynman rules for the @xmath133 and @xmath129 terms in @xmath159 are given in figure [ fig : nrqcd_fr ] . in the fifth row can be eliminated using the heavy - quark equation of motion . this generates @xmath160 terms not shown in the figure and modifies the @xmath129 four - point vertices in the last row of the figure as follows : drop the @xmath161 and @xmath162 terms and multiply the remaining terms by two . ] the hard matching coefficients @xmath163 equal one at tree level , while the others vanish . all coefficients obtain one - loop corrections , which must be determined by matching the qcd diagrams to the nrqcd diagrams order by order in the non - relativistic expansion . for reasons which will be explained later , the matching coefficients are needed to order @xmath164 in the dimensional regularization parameter . in writing ( [ eq : nrqcdqq ] ) we did not include @xmath160 interactions with vanishing tree - level coefficients as well as mixed heavy - light quark operators of the form @xmath165 in @xmath166 , since they are irrelevant for the nnnlo calculation , as we discuss now . the velocity scaling of the fields depends on the momentum region . heavy quark fields can be potential or soft . from the scaling rules of the propagator ( compare ( [ eq : quarkprop ] ) ) and the integration measure , it follows that in both cases the field scales as @xmath167 . the same power counting can be done with the gluon field , which is either soft , potential or ultrasoft and one finds that @xmath168 scales as @xmath167 , @xmath169 and @xmath170 respectively , which includes a factor of @xmath171 from the coupling constant @xmath172 . turning to the effective lagrangian , we first consider the bilinear terms in the heavy lagrangian @xmath166 . for potential quarks , the bilinear terms in the kinetic term @xmath173 are both of order @xmath95 . the relativistic correction @xmath174 scales as @xmath175 . being suppressed by @xmath118 relative to the leading terms it contributes from nnlo . the next term in the expansion of the relativistic energy - momentum relation would be of order @xmath176 and is beyond nnnlo . for soft quarks , the quadratic kinetic energy term is an order @xmath12 correction to the leading - order static lagrangian @xmath177 , which scales as @xmath94 . this explains why static heavy - quark propagators can be used in the calculation of the heavy - quark potentials . for soft heavy quarks the quartic kinetic energy correction is a nnnlo effect . consider now the interactions of the heavy quark with the gauge field , i.e. terms of the form @xmath178 , potentially with derivatives on the quark and gluon fields . the @xmath179 interaction that arises from @xmath180 scales as @xmath95 when all fields are potential . therefore it is not suppressed relative to the bilinear terms that persist as @xmath181 . in the potential region , this interaction has to be treated non - perturbatively ; this is why the cross section near threshold requires a summation of some loop momentum contributions to all orders in @xmath13 . when the gluon field is soft , the velocity scaling of the @xmath179 interaction is @xmath182 , but since in this case the leading term @xmath183 scales as @xmath94 , this interaction is now a perturbation . it follows that all soft interactions can be treated in conventional perturbation theory . further three - point interactions of the form above carry derivatives . each derivative gives at least a suppression of @xmath12 so interactions with up to three derivatives in @xmath184 may contribute at nnnlo . the requirement of gauge invariance restricts the possible interaction terms to the so - called chromomagnetic interaction @xmath185 at order @xmath186 ( nlo ) , and the darwin and spin - orbit interactions at order @xmath175 ( nnlo ) , multiplied by the short - distance coefficients @xmath187 , @xmath188 and @xmath145 , respectively . however , a single chromomagnetic interaction contributes only in connection with a @xmath12 suppressed quark - gluon vertex from the @xmath189 interaction , and not at all to the current correlation function , since the trace over an odd number of pauli matrices vanishes . thus the chromomagnetic , darwin and spin - orbit interaction all start to contribute at nnlo ( with two insertions of the chromomagnetic term ) . hence , at nnnlo one needs the coefficient functions @xmath187 , @xmath188 and @xmath145 in the one - loop approximation . beyond order @xmath175 only the interactions with non - vanishing tree - level coefficient functions given in ( [ eq : nrqcdqq ] ) can potentially contribute to nnnlo . none of them does , however , since single insertions of interactions with pauli matrices vanish , as discussed above , while the terms with two electric or magnetic field strengths are of the form @xmath190 with two derivatives , which is smaller than nnnlo in both , the potential and soft region . bilinear heavy quark operators in conjunction with light quarks , @xmath165 , are of order @xmath186 . for these operators to contribute to the heavy - quark current correlation function at least one interaction of the form @xmath191 is required , which costs a factor of @xmath12 or , equivalently @xmath13 . thus , a @xmath165 operator is relevant at nnnlo , if its short - distance coefficient function is of order @xmath13 . the only operator that may have a tree - level coefficient is @xmath192 , since in this case the intermediate potential gluon propagator can be cancelled , making the operator local . this operator is generated at tree level from the darwin term in the lagrangian ( [ eq : nrqcdqq ] ) by the use of the field equation for the chromo - electric field . our convention is that we do not eliminate the darwin term by the field equation , hence @xmath165 operators must be added to the lagrangian only with coefficients of order @xmath193 producing corrections to the heavy - quark current correlation function beyond nnnlo . we therefore conclude that only the terms in the first two lines of ( [ eq : nrqcdqq ] ) are needed for the nnnlo calculation , the same as in nnlo . the only difference to nnlo is that the short - distance coefficients @xmath187 , @xmath188 and @xmath145 are required at the one - loop order . we also see that only the @xmath194 interaction in the potential region is non - perturbative , and this explains why only ladder diagrams of coulomb gluons must be summed to all orders . the four - quark operator lagrangian @xmath195 ( [ eq : nrqcdqqqq ] ) is generated by hard scattering of quarks and anti - quarks , or by quark anti - quark annihilation . hard scattering with momentum exchange of order @xmath196 requires the exchange of at least two gluons , corresponding to one - loop diagrams so the coefficient functions are of order @xmath193 . the four - quark operator counts as @xmath186 . including the one - loop coefficient function gives the counting @xmath197 , which is a nnnlo effect relative to the leading - order lagrangian of order @xmath95 for the scattering of potential quarks . the annihilation contribution is present already at order @xmath13 ( tree - level ) , but the operator has the colour structure @xmath198 , which does not contribute to the current correlation function due to @xmath199 . similarly , annihilation into two gluons does not contribute to the vector correlation function , since it leads to fermion loops with three vector couplings that vanish by charge conjugation . thus , at nnnlo , we can restrict ourselves to the four - fermion operators generated by hard quark anti - quark scattering . for this reason we write the operators in the `` scattering ordering '' @xmath200 rather than the `` annihilation ordering '' @xmath201 . although the two orderings are related by a fierz transformation in four space - time dimensions , the two are inequivalent in dimensional regularization . in general , we would have to introduce the difference of the two as evanescent operators . this complication is avoided here , since there are no annihilation contributions at nnnlo . adopting the `` scattering ordering '' in the lagrangian , we do not need to perform any fierz transformations . the necessity to avoid relations that hold only in four dimensions is also the reason for introducing the definition ( [ eq : magneticfield ] ) that does not make use of the three - dimensional @xmath202 symbol , which is not defined in @xmath150 dimensions . in particular , the commutator @xmath203\ , , \ ] ] must be considered as an independent element of the @xmath33-dimensional algebra of pauli matrices . this poses no difficulty in the calculation of the vector current correlation function , since in the end all expressions can be evaluated using the @xmath33-dimensional identities @xmath204 where @xmath88 and @xmath205 . the pure gauge field lagrangian ( [ eq : nrqcdg ] ) follows from integrating out heavy - quark loops with small momentum gluon lines attached . the renormalization of the standard kinetic term by the coefficient @xmath206 is well - known to be related to the matching of the strong coupling from the @xmath207 flavour theory including the heavy quark to the theory with @xmath208 light flavours . in the following we express all results in terms of the strong coupling in the @xmath73 scheme in the @xmath208 flavour theory , which is the appropriate coupling for calculations in nrqcd , where the heavy quark short - distance fluctuations have been integrated out . after redefining the strong coupling , @xmath206 should be set to one in ( [ eq : nrqcdg ] ) , so that the kinetic term is canonically normalized . the next term , @xmath209 , in the gauge field lagrangian involves two derivatives and a coefficient function @xmath210 . therefore it must be included at nnnlo . on the other hand , the term involving three gluon field strengths can be neglected at this order . having collected the relevant terms in the effective lagrangian , we are now in the position to discuss the matching calculations . most of the results required at nnnlo are available in the literature . however , many of the matching coefficients multiply nrqcd correlation functions , which exhibit @xmath42 poles . thus , as will be explained in section [ sec : matchcurrent ] below , we also need the @xmath24 terms of the matching coefficients , which have not been calculated or presented up to now . we therefore had to repeat these matching calculations and extend them to the @xmath24 terms . the matching calculation is performed in the center - of - mass frame , so that the three - momenta of the heavy quark and anti - quark are of opposite sign . the external heavy quark spinors in qcd are given by @xmath211 \bm{\sigma}\cdot{{\bf { p}}}\,\xi \end{array}\right ) , \quad v(p ) = \frac{1}{(e_p+m)^{1/2}}\left(\begin{array}{c } \bm{\sigma}\cdot{{\bf { p}}}\,\eta \\[0.2 cm ] ( e_p+m)\ , \eta \end{array}\right),\ ] ] for external momentum @xmath212 with @xmath213 . the variables @xmath214 and @xmath215 denote the quark and anti - quark two - spinors , respectively . they are normalized according to @xmath216 . the coefficient functions of the interactions of heavy quarks with a single gauge field in @xmath166 can be deduced from the heavy quark form factors in background field gauge . the finite part of the one - loop form factors was calculated in @xcite . however , the order @xmath164 coefficients were not computed there . the on - shell form factors can be brought into the general form @xmath217u(p),\end{aligned}\ ] ] where @xmath218 . for the one - loop diagrams shown in figure [ fig : formfactors ] we obtain the following result for the expansion of the ultraviolet - renormalized form factors in @xmath13 and @xmath219 up to order @xmath220 : @xmath221 , \\ \nonumber f_2 & = & \frac{\alpha_s}{\pi}\bigg(\frac{\mu}{m}\bigg)^{2\epsilon } \frac{\gamma(\epsilon)e^{\gamma_e\epsilon}}{24(4\epsilon^2 - 1 ) } \bigg[c_a\bigg(6(2\epsilon+1)(2\epsilon^2 - 1)+\frac { q^2}{m^2}(4\epsilon^4 + 8\epsilon^3 + 5\epsilon^2 - 2\epsilon-6)\bigg ) \\ & & + \,c_f\bigg(-12\epsilon(2\epsilon+1)^2-\frac { q^2}{m^2}2\epsilon(\epsilon+1)(2\epsilon+1)^2\bigg)\bigg].\end{aligned}\ ] ] here and below we use the standard colour factors @xmath222 , @xmath223 , @xmath224 . the remaining divergences are infrared divergences of the on - shell form factors . by calculation of the corresponding form factors in the effective theory we obtain the relations between the coefficients @xmath225 and the form factors : @xmath226 where @xmath227 and @xmath228 . since @xmath229 exactly , this implies the well - known relation @xmath230 . the @xmath73 renormalized coefficient functions follow by subtracting the @xmath42 poles from the above expressions . the @xmath24 terms of the above expressions are in agreement with @xcite . one - loop form factor diagrams : wavefunction renormalization and vertex corrections.,height=113 ] to obtain the bilinear pure gauge field lagrangian @xmath231 , the gluon self energy has to be matched . in the one - loop order the relevant diagram is the heavy - quark loop , which gives : @xmath232 the results agree for @xmath233 with the ones in @xcite . as mentioned above , the operator with coefficient @xmath234 is not needed , because it can contribute to the heavy quark correlation function only with an additional loop , which is beyond nnnlo . recall that @xmath235 should be used after having normalized the fields canonically . qcd diagrams for the four - fermion operators at one loop.,height=68 ] the remaining part of the lagrangian to be matched to qcd is the one containing the four - fermion operators . as discussed in section [ sec : hqcf ] we do not need the annihilation contributions and therefore restrict ourselves to the scattering diagrams shown in figure [ fig:4quark ] . the results in @xmath233 can be obtained from the equal mass limit of the unequal mass case given in @xcite . here we present the @xmath33-dimensional matching coefficients : @xmath236 this agrees with the finite part of the unequal mass case in @xcite . the @xmath24 terms of the above expressions are in agreement with @xcite . and @xmath237 defined in @xcite are @xmath238 times those above . ] we finally need an expression for the heavy quark vector current @xmath239 to nnnlo accuracy in the effective theory . the perturbative matching coefficients of the nrqcd currents come from diagrams where the hard loop connects to one of the external current vertices . since the zero component of the vector current is irrelevant , we focus on matching the operator @xmath240 . at leading order in the velocity expansion the unique nrqcd vector current is @xmath241 with coefficient function @xmath17 as given by the first term on the right - hand side of ( [ eq : qcdvectorcurrent ] ) . the precise definition of the matching coefficient is @xcite @xmath242 where @xmath243 are the on - shell wave function renormalization constants in qcd and nrqcd , respectively . @xmath244 represents the amputated , bare electromagnetic current vertex function evaluated for on - shell heavy quarks directly at threshold , i.e. with zero relative momentum , expressed in terms of the renormalized qcd coupling and pole mass . in dimensional regularization , @xmath245 , and @xmath246 , the corresponding nrqcd vertex function , equals its tree - level expression @xmath247 , since the nrqcd integrals for zero external relative momentum are scaleless . here it is important that the threshold expansion is employed to define nrqcd in dimensional regularization . thus , @xmath17 equals the uv renormalized on - shell qcd vertex directly at threshold with infrared divergences subtracted recursively by the nrqcd renormalization factor @xmath248 . this definition is equivalent at order @xmath249 to extracting the purely hard ( h - h- -h ) momentum regions in the threshold expansion of the @xmath250-loop vertex function with external heavy quark momenta in the potential region . the coefficient @xmath17 is needed to three - loop accuracy to achieve nnnlo precision . while the two - loop expression has been known for some time @xcite , the three - loop correction is not yet completed and constitutes one of the missing pieces in the complete third - order calculation of the cross section . the three - loop diagrams involving at least one fermion loop are , however , already known ) . ] @xcite , as are the logarithmic terms related to the anomalous dimension of the current and strong coupling renormalization @xcite . defining @xmath251 the coefficients of perturbative expansions of any quantity @xmath252 in @xmath253 , through @xmath254 and the coefficients of the qcd @xmath255 function in the @xmath73 scheme with @xmath208 light flavours ( not including the heavy quark ) @xcite @xmath256 the known results for the vector current matching coefficients are given by : @xmath257 c_v^{(2)}(\mu ) & = & 2 \beta_0 l_m c_v^{(1)}(m ) + l_m c_f \pi^2 \bigg[- \frac{16}{3 } c_f-8 c_a \bigg ] + c_v^{(2)}(m ) , \\[0.2 cm ] c_v^{(3)}(\mu ) & = & \left(4\beta_0 ^ 2 l_m^2 + 2\beta_1 l_m\right)c_v^{(1)}(m ) \nonumber \\ & & + 4\beta_0 l_m \bigg\ { l_m c_f \pi^2 \bigg[- \frac{16}{3 } c_f-8 c_a \bigg ] + c_v^{(2)}(m ) \bigg\ } \nonumber \\ & & + l_m^2 c_f \pi^2 \bigg [ -40c_f^2 -\frac{188}{9 } c_f c_a + \frac{128}{3 } c_a^2 \nonumber \\ & & \hspace{2 cm } -\frac{128}{9 } c_f t_f n_f -\frac{64}{3 } c_a t_f n_f \bigg ] \nonumber \\ & & + l_m c_f \pi^2 \bigg [ \left(-72 + 192\ln{2}\right ) c_f^2 + \left(-\frac{1888}{27}-96\ln{2}\right ) c_f c_a \nonumber \\ & & \hspace{2 cm } + \left ( -\frac{256}{9}-96\ln{2}\right ) c_a^2 + \frac{800}{27 } c_f t_f n_f \nonumber \\ & & \hspace{2 cm } + \frac{296}{9 } c_a t_f n_f -\frac{32}{5 } c_f t_f \bigg ] + c_v^{(3)}(m),\end{aligned}\ ] ] where @xmath258 is the matching coefficient evaluated at @xmath259 . the two - loop non - logarithmic terms are fully known but the three - loop ones only for the fermionic contributions . they read : @xmath260 , \\[0.2 cm ] c_{v}^{(3)}(m ) & = & 64\,\bigg [ c_f t_f n_f \bigg [ c_f \left(46.692(1)\right ) + c_a \left(39.623(1)\right ) \nonumber \\ & & \hspace{1.5 cm } + t_f n_f \left(-\frac{163}{162}-\frac{4}{27}\pi^2\right ) + t_f \left(-\frac{557}{162}+\frac{26}{81}\pi^2\right ) \bigg ] \nonumber \\ & & + c_f t_f \bigg [ c_f \left(-0.840(2)\right ) + c_a \left(-0.09(2)\right ) \nonumber \\ & & \hspace{1.5 cm } + t_f \left(-\frac{427}{162}+\frac{158}{2835}\pi^2+\frac{16}{9}\zeta_3\right ) \bigg]\,\bigg ] + c_{v , g}^{(3)}(m).\end{aligned}\ ] ] where the last term @xmath261 is the unknown gluonic and singlet contribution and @xmath262 is a short - hand for the riemann zeta function value @xmath263 . we recall our convention that @xmath13 denotes the strong coupling in the @xmath73 scheme with @xmath208 light flavours . turning to the next orders in the velocity expansion , we find the operators @xmath264 which are suppressed by @xmath265 relative to the leading nrqcd current . further operators of dimension five contain the ultrasoft gauge field strength @xmath266 of order @xmath182 . thus up to nnnlo all production vertices contain only the quark - antiquark pair . the on - shell heavy quark - antiquark production vertex in full qcd can be decomposed into the expression @xmath267 v(p_2),\ ] ] where now @xmath268 , @xmath269 , and @xmath270 . inserting ( [ eq : spinors ] ) for the external spinors , we obtain the exact expression @xmath271 \xi^\dagger\sigma^i\eta -2 e_p \left[\frac{m}{e_p } \hat f_1(q^2)-\hat f_2(q^2)\right ] \frac{p^i}{m ( e_p+m ) } \,\xi^\dagger{{\bf { \sigma}}}\cdot{{\bf { p}}}\,\eta.\quad\ ] ] this shows explicitly that only quark - antiquark operators with a sigma matrix can appear as assumed in ( [ eq : currentsv2 ] ) . expanding this expression in @xmath272 we find @xmath273_{\,|q^2=4 m^2 , \,\mbox{\scriptsize hard } } , \nonumber\\[0.2 cm ] d_{va } & = & [ \hat f_1-\hat f_2]_{\,|q^2=4 m^2 , \,\mbox{\scriptsize hard } } , \nonumber\\ d_{vb } & = & ( -4 ) \,\frac{d}{d(q^2/m^2 ) } [ \hat f_1+\hat f_2]_{\,|q^2=4 m^2 , \,\mbox{\scriptsize hard}}.\end{aligned}\ ] ] the subscript `` hard '' means that only the hard regions should be included in the computation . the one - loop hard form factors can be extracted from @xcite , by dropping the non - analytic terms in the expansion in @xmath274 , which originate from the potential region . we obtain @xmath275 given in ( [ eq : cv1 ] ) above and @xmath276 + o(\alpha_s^2)\ , . \label{eq : dvab}\end{aligned}\ ] ] the logarithm in @xmath277 arises as the consequence of mixing with the leading order current through sub - leading nrqcd interactions , see ( [ eq : usdiv ] ) below . these results agree with the computation of the one - loop corrected matching coefficients of the sub - leading current operators through explicit nrqcd matching @xcite . at nnnlo the correlation functions of velocity - suppressed currents will be evaluated only with the leading and next - to - leading order coulomb potential , which is spin - independent . hence , only the traces @xmath278 appear . this allows us to combine @xmath279 such that the qcd vector current is now represented by @xmath280 as anticipated in ( [ eq : qcdvectorcurrent ] ) . from ( [ eq : dvab ] ) and ( [ eq : currentspinprojection ] ) we obtain @xmath281 + o(\alpha_s^2)\ , . \label{eq : dv1loop}\end{aligned}\ ] ] the explicit scale dependence of the matching coefficients is due to evolution of the strong coupling and the factorization of the hard scale . it must cancel when all contributions to the cross section are combined . we have checked explicitly that this is indeed the case . note that we do not need the @xmath24 terms of the coefficient functions @xmath17 and @xmath18 to compute the heavy - quark current correlation function , since they multiply the finite , renormalized nrqcd correlation function in ( [ eq : pitonrqcd ] ) . one may wonder then what is the difference between the nrqcd current and the nrqcd lagrangian matching coefficients @xmath187 , @xmath188 etc . , since for the latter we need the @xmath24 terms as stated and given above . the reason is the particular definition ( [ eq : newmatch ] ) of the current matching coefficient . imagine that we calculate the qcd and nrqcd vertex functions @xmath244 with non - vanishing external relative momentum . then the nrqcd diagrams are no longer scaleless and @xmath246 is the sum of potential , soft and ultrasoft loop momentum contributions . because of the @xmath105 factors from potential gluon exchange , the higher - dimensional nrqcd interactions contribute to the leading current matching equation at some order in perturbation theory . as a result @xmath246 will be different whether one uses the nrqcd lagrangian with four - dimensional or with @xmath33-dimensional short - distance coefficients . the difference is , however , a local term that can be absorbed into the matching coefficients of the external currents . which definition does ( [ eq : newmatch ] ) correspond to ? suppose we follow the more conventional path to define the renormalized effective lagrangian with @xmath233 short - distance coefficients . in this case @xmath246 does _ not _ represent the sum of all potential , soft and ultrasoft loop momentum regions plus the hard ones not connected to the external vertex ( encapsulated in the lagrangian matching coefficients @xmath187 etc . ) , since one misses some @xmath24 terms from hard subgraphs that multiply divergent soft , potential or ultrasoft loops . these missing local contributions can be and must be added back by adapting the external current matching coefficient . hence the matching coefficient @xmath17 defined by this prescription does not correspond to ( [ eq : newmatch ] ) . on the other hand , the nrqcd lagrangian with @xmath33-dimensional short - distance coefficients reproduces these missing terms directly , so the matching coefficient corresponding to this case is simply the contribution from the purely hard ( h - h- ... -h ) regions as it was defined in ( [ eq : newmatch ] ) . moreover , the purely hard regions can now be computed directly at zero external relative momentum , which simplifies the calculation . the same discussion applies to the matching of the potentials in pnrqcd , to which we turn next . however , while the @xmath24 terms of the nrqcd lagrangian are relevant only at nnnlo , the difference between four- and @xmath33-dimensional potentials in the lagrangian matters already at nnlo . the @xmath33-dimensional ones must be used in conjunction with ( [ eq : newmatch ] ) as was done in @xcite . due to the @xmath118 suppression of the p - wave correlation function ( [ eq : piatonrqcd ] ) relative to the s - wave case ( [ eq : g ] ) , the hard matching coefficient @xmath132 of the axial - vector current is needed only with one - loop accuracy for the nnnlo calculation of the top - quark pair production cross section near threshold . the relevant expression is @xmath282 the two - loop correction is also known @xcite . as discussed in section [ sec : momentumregions ] to perform the all - order resummation , a second matching procedure is required , by which the soft region and potential light fields ( gluons and light quarks ) are integrated out . this results in the potential nrqcd ( pnrqcd ) effective field theory @xcite . in pnrqcd the light fields are purely ultrasoft and the heavy quarks are potential , hence the terms in the effective lagrangian can be assigned a unique scaling in the velocity expansion . the effective lagrangian relevant to the third - order calculation takes the simple form @xmath283(x+{\bf r } ) \ , v_{ab;cd } ( r,{{\mbox{\boldmath $ { \partial}$}}})\ , \big[\chi^\dag_c \chi^{}_d\big](x ) \\ & & -g_s\psi^\dagger(x){{\bf { x}}}\cdot{{\bf { e}}}(t,{{\bf { 0}}})\psi(x)- g_s\chi^\dagger(x){{\bf { x}}}\cdot{{\bf { e}}}(t,{{\bf { 0}}})\chi(x ) , \label{eq : pnrqcd}\end{aligned}\ ] ] where @xmath284 with @xmath285 the tree - level colour coulomb potential . the pnrqcd lagrangian consists of kinetic terms ( first line ; including the relativistic corrections proportional to @xmath286 ) , heavy - quark potential interactions ( second line ) and an ultrasoft interaction that contributes first at third order . the heavy - quark potentials generated in the the matching to pnrqcd should be considered as short - distance coefficients of the interactions . they are split into the tree - level coulomb potential , which must be treated non - perturbatively and a remainder @xmath287 , which represents a perturbation . to achieve a homogeneous velocity scaling the position argument of ultrasoft fields should be multipole - expanded in interactions with heavy quarks @xcite , which explains the space - time argument of @xmath288 and the chromo - electric field in the ultrasoft interaction terms . as will be discussed below no further matching of the non - relativistic vector current is needed , that is @xmath289 to the required accuracy . thus , instead of ( [ eq : g ] ) , we have to calculate @xmath290(x)\ , [ \psi^{\dag}\sigma^i\chi](0 ) ) evaluated to third - order in pnrqcd perturbation theory . the dimensionally regulated pnrqcd lagrangian required for second - order calculations of heavy - quark pair production near threshold was provided in @xcite , and the explicit derivation of the ultrasoft interaction in the third line of ( [ eq : pnrqcd ] ) from nrqcd was given in @xcite . the only new piece that is needed is the third - order heavy - quark potential in @xmath287 . in the remainder of this subsection we first give the pnrqcd feynman rules ( when the ultrasoft interactions are neglected ) and then sketch several ways of deriving these rules and the form of the propagator . subsequently , we summarize the heavy - quark potentials . we also derive equation - of - motion relations that allow us to reduce the number of potential insertions to be calculated and briefly discuss the ultrasoft contribution already calculated in @xcite . we begin by summarizing the rules for calculating pnrqcd diagrams with insertions of potential interactions , but no ultrasoft interactions . since the leading - order lagrangian includes the coulomb potential @xmath291 , the propagator is the one for a heavy quark anti - quark _ pair_. we draw the propagator as in the left diagram of figure [ fig : pnrqcd ] , where the blob stands for the sum of all potential ( coulomb ) ladder diagrams , which is included in the propagator . for a pair in a colour - singlet state each propagator gives a factor @xmath292 where @xmath19 is the non - relativistic energy of the pair and @xmath161 ( @xmath293 ) the three - momentum of the in - coming ( out - going ) quark . for the colour - octet state the propagator is @xmath294 . the function @xmath295 is the solution to the @xmath33-dimensional lippmann - schwinger equation for a pair in irreducible su(3 ) colour representation @xmath296 , @xmath297 where the @xmath298 and @xmath299 for the colour - singlet and colour - octet representation , respectively . explicit expressions will be given below . the scale @xmath300^{1/2}$ ] is defined such that minimal subtraction of @xmath42 poles corresponds to the @xmath73 rather than ms scheme @xcite . the fourier transform @xmath301 satisfies in @xmath233 dimensions the schrdinger equation @xmath302 in the following , when the superscript is left out , the propagator refers to the colour - singlet representation . the vertex associated with the insertion of a perturbation potential @xmath303 in momentum space is given by @xmath304 and internal relative momenta @xmath305 are integrated over with measure @xmath306 . note that the insertion of a potential does not change the colour - state of the quark anti - quark pair , when it is in an irreducible representation , that is the colour - singlet or the colour - octet state . the reason for this is that @xmath307 and @xmath308 constitute a complete set of orthogonal projectors . if the in - coming pair is , for example , in the colour - singlet state , all propagators will be colour - singlet propagators and the potential insertions are effectively projected to the colour - singlet potential @xmath309 this is different when an ultrasoft gluon is emitted , in which case the pair changes its colour state , as explicitly seen in ( [ eq : gfcorr ] ) below . a general pnrqcd diagram with multiple insertions of perturbation potentials is therefore an expression of the form @xmath310 i{g_0}({\bf p}_1,{\bf p}_2;e ) i\delta{v}_1({\bf p}_2,{\bf p}_3 ) i{g_0}({\bf p}_3,{\bf p}_4;e)\ , i\delta{v}_2({\bf p}_4,{\bf p}_5 ) i{g_0}({\bf p}_5,{\bf p}_6;e)\ , \ldots \label{multipleinsertions}\ ] ] ( for the case of a colour - singlet state ) . for convenience of notation here and below we often leave out factors of @xmath311 required in dimensional regularization to restore the proper dimension of the given expression . in this section we sketch three derivations of the rules for pnrqcd perturbation theory : diagrammatic , quantum - mechanical , and by path - integral methods . for simplicity we assume the colour - singlet representation , but the derivation is easily generalized to an arbitrary irreducible representation . consider the amputated amplitude of the heavy - quark scattering process @xmath312 with non - relativistic external momenta @xmath313 , @xmath314 and @xmath315 , @xmath316 in the rest frame of the @xmath107 pair . the sum of all ( ladder ) diagrams with any number ( greater than zero ) of leading - order potential insertions @xmath291 is given in momentum space by @xmath317 \frac{(i g_s)^2 i}{({{\bf { k}}}_1-{{\bf { k}}}_0)^2}\ , \frac{(i g_s)^2 i}{({{\bf { k}}}_2-{{\bf { k}}}_1)^2}\ , \ldots \frac{(i g_s)^2 i}{({{\bf { k}}}_{n+1}-{{\bf { k}}}_n)^2}\ , \nonumber\\ & & \cdot \,\prod_{i=1}^n\ , \frac{i}{\frac{e}{2}+k_i^0-\frac{({{\bf { p}}}+{{\bf { k}}}_i)^2}{2m}+ i\epsilon}\ , \frac{-i}{\frac{e}{2}-k_i^0-\frac{({{\bf { p}}}+{{\bf { k}}}_i)^2}{2m}+ i\epsilon } \,,\end{aligned}\ ] ] where we define @xmath318 and @xmath319 . we perform the integrations over the loop momentum zero components @xmath320 by closing the contour in the upper half plane , and pick up the residues from the poles at @xmath321 , which results in @xmath322 \,\frac{1}{{{\bf { k}}}_1 ^ 2}\ ; \nonumber\\ & & \hspace*{0.0cm}\times \prod_{i=1}^n\ , \frac{1}{({{\bf { k}}}_{i+1}-{{\bf { k}}}_i)^2 ( e-\frac{({{\bf { p}}}+{{\bf { k}}}_i)^2}{2m}+i\epsilon)}\ , . \qquad \label{eq : hc}\end{aligned}\ ] ] the @xmath323 term in this and the previous sum is understood as @xmath324 , which is the expression for the exchange of a single potential ( coulomb ) gluon . @xmath325 sums the leading p - p- ... -p region to all orders . next we multiply the propagator factors @xmath326 for the external pairs of lines and add the zero - coulomb exchange graph . multiplying by @xmath327 this defines @xmath328 it is straightforward to show that this expression satisfies the @xmath33-dimensional lippmann - schwinger equation ( [ eq : lippmann - schwinger ] ) , and hence represents the colour - singlet coulomb green function . the summation of ladder diagrams is therefore accomplished by associating the quantity @xmath329 with the propagator of the quark anti - quark pair and the vertex @xmath330 with the interaction potentials . it is understood that the integrations over the zero - components of loop momenta are already done . note that while no closed expression for the green function is known in @xmath33 dimensions , it is important that the above expression is defined in dimensional regularization , and that it can be expanded perturbatively in @xmath104 in @xmath33 dimensions . this guarantees the consistency of the dimensional regularization procedure , which requires subtracting terms with a finite number of coulomb exchanges from @xmath331 in @xmath33 dimensions as will be seen in part ii of the paper . the pnrqcd correlation function ( [ eq : gpnrqcd ] ) describes a quark anti - quark pair created in a colour - singlet , spin - triplet state at point 0 , which propagates and is destroyed locally at point @xmath332 . in terms of the momentum - space propagator it is given by @xmath333\,,\qquad \label{eq : gexp}\end{aligned}\ ] ] where the potential refers to the colour - singlet potential ( [ eq : singlet - potential ] ) , and the ellipses to terms with multiple potential insertions . both the propagator and the potential carry spin - indices in general . however , the unperturbed pnrqcd lagrangian is spin - independent , so the propagator is diagonal in the spin indices and we drop the @xmath334 spin factor . the spin - dependence of the perturbation potential enters only at nnlo , hence up to n@xmath335lo only one of the @xmath336 insertions can carry a non - trivial spin - dependence . when this insertion appears in ( [ eq : gexp ] ) , @xmath336 is understood as @xmath337 and the trace must be carried out in @xmath33 dimensions . greek indices of the potential refer to spin ( rather than colour ) . the normalization factor and pauli matrices result from the definition of ( [ eq : gpnrqcd ] ) and correspond to the spin - triplet projection of the potential . for the insertions of spin - independent potentials the spin - factor @xmath338 is included in ( [ eq : gexp ] ) with only potential interactions the pnrqcd lagrangian can be projected onto the fock states with a single quark and a single anti - quark without loss of content , since potential interactions do not change particle number . we define the centre - of - mass wave function of a quark anti - quark state @xmath339 in the position - space representation via @xmath340^\dagger simplicity , we assume again a projection on the colour - singlet representation . by reversing the steps that lead from the schrdinger equation to a second - quantized schrdinger field theory , making use of the field equation and the canonical commutation relations , we find that @xmath341 \psi(t,{{\bf { r}}})\,.\ ] ] the green function of the schrdinger operator is given by @xmath342^{-1}|{{\bf { r}}}'\rangle$ ] with @xmath343 a quark anti - quark separation eigenstate with eigenvalue @xmath344 . in operator notation the green function is @xmath345^{-1}$ ] such that @xmath346 where @xmath347^{-1}$ ] and @xmath348 . since @xmath349 , the previous equation verifies the pnrqcd feynman rules . in this notation the correlation function ( [ eq : gpnrqcd ] ) is given by @xmath350 which is equivalent to ( [ eq : gexp ] ) upon using ( [ eq : goperatorexp ] ) and inserting complete sets of momentum eigenstates , @xmath351 leaving the spin - average being implicit . in @xcite the pnrqcd lagrangian is expressed in terms of composite colour - singlet and colour - octet fields . here we provide a path - integral derivation of this formulation . we focus on the colour - singlet field and drop the colour and spin indices of the composite field @xmath352_{x^0=y^0 } = \left [ \psi^\dag(x)\chi(y ) \right]_{x^0=y^0}\ , , \qquad \left [ s^\dag(y , x)\right]_{x^0=y^0 } = \left [ \chi^\dag(y)\psi(x ) \right]_{x^0=y^0}\,.\ ] ] the partition function @xmath353 of pnrqcd is defined as @xmath354 where we use the leading - order pnrqcd lagrangian @xmath355\psi(x ) + \chi^\dag(x)\bigg[i\partial_0 - \frac{{\mbox{\boldmath $ { \partial}$}}^2}{2m}\bigg]\chi(x ) \nonumber\\ & & -\int d^4y \left[\psi^\dag(x)\chi(y)\right ] v(x , y ) \left[\chi^\dag(y ) \psi(x)\right],\end{aligned}\ ] ] with the quark anti - quark potential @xmath356 . the derivation remains however valid with an arbitrary potential , not just the leading - order coulomb potential . the different sign of the potential term relative to ( [ eq : pnrqcd ] ) is due to the different order of fermion fields . the composite field is introduced by means of @xmath357_{x^0=y^0 } - \left[\psi^\dag(x)\chi(y)\right]_{x^0=y^0}\right ) \nonumber\\ & = & \int { \cal d}s \,{\cal d}\sigma\ , \exp\bigg\ { \,i \!\int d^4x \int d^4y \ , \sigma(x , y ) \delta(x^0-y^0 ) \left ( s(x , y)- \left[\psi^\dag(x)\chi(y)\right]\right ) \bigg\}\ , , \qquad\end{aligned}\ ] ] where the second line is the fourier representation of the functional delta - function which introduces the auxiliary field @xmath358 . with a similar formula for the complex conjugate field we can rewrite the partition function as @xmath359 where @xmath360\right ) \nonumber\\ & & + \,\sigma^\dag(y , x)\left ( s^\dag(y , x)- \left[\chi^\dag(y)\psi(x)\right ] \right)\,.\end{aligned}\ ] ] to make equations concise we introduced @xmath361 and @xmath362 our task is to integrate over @xmath363 and their conjugates to obtain the lagrangian for the @xmath252 field . in the following we do this step by step obtaining the sequence of lagrangians @xmath364 we drop field - independent factors which can be absorbed into the path - integral measure . first we integrate out @xmath119 and its conjugate by completing squares in the exponent : @xmath365_{\psi=0 } \nonumber\\ & & \qquad + \ , \left ( \psi^\dag- \chi^\dag\cdot\sigma^\dag\cdot k_q^{-1 } \right ) \cdot k_q\cdot \left(\psi- k_q^{-1 } \cdot\sigma\cdot \chi \right ) -\chi^\dag\cdot\sigma^\dag\cdot k_q^{-1 } \cdot \sigma \cdot \chi\ , . \qquad\end{aligned}\ ] ] in the second line the abbreviation @xmath366 stands for the integration of the adjoining variables , such that , for example , the last term reads explicitly @xmath367 the inverse operators are given by @xmath368 changing variables to @xmath369 ( similarly for @xmath370 ) and integrating over @xmath119 , @xmath370 we obtain @xmath371_{\psi=0 } -\chi^\dag\cdot\sigma^\dag\cdot k_q^{-1 } \cdot \sigma \cdot \chi . \qquad\end{aligned}\ ] ] the integration over @xmath372 is done analogously resulting in the partition function @xmath373 the determinant contains the @xmath358 field . to write it as a term in the lagrangian we use @xmath374 and keep only the bilinear term in the @xmath375 field in the expansion of the logarithm . we comment on the other terms below . the lagrangian after this procedure is @xmath376 now we integrate over the delta - functions in the time coordinates implicit in the definition of the @xmath375 field and obtain @xmath377 where @xmath378 represents the coordinates of the quark and anti - quark at coincident time @xmath379 . the fields with argument @xmath380 are defined as @xmath381_{x^0=y^0 } \nonumber\\ \sigma(z)&=&\left [ \sigma(x , y)\right]_{x^0=y^0 } , \nonumber\\ k_\sigma(z',z)&=&\left[k_{q}^{-1}(x',x ) k_{\bar{q}}^{-1}(y , y ' ) \right]_{t = x^0=y^0 , \,t'=x^{0'}=y^{0 ' } } \label{eq : ksig}\end{aligned}\ ] ] the last step consists of performing the gaussian integral over @xmath358 . the inverse of @xmath382 is defined by @xmath383 resulting in @xmath384 to compute @xmath385 we use ( [ eq : ksig ] ) and the definitions ( [ eq : inversekq ] ) . the integrals over the zero - components of the two momenta can be written as integrals over relative momentum @xmath386 and total momentum @xmath387 . since the exponentials are independent of @xmath387 , the @xmath387 integral can be performed by contour integration which gives @xmath388 where @xmath389 . it follows that @xmath390\,,\ ] ] and therefore @xmath391 with @xmath392 after separating the free centre - of - mass motion this represents the pnrqcd lagrangian expressed in terms of the composite quark anti - quark field . when one keeps the higher - order terms in the expansion of the logarithm in ( [ eq : expdet ] ) the path - integral over @xmath358 can no longer be done exactly . expanding the quartic and higher - order terms in the exponential , we obtain vertices involving four and more @xmath252 fields , which describe scattering of composite fields . these terms are clearly not relevant to the threshold dynamics of a single quark anti - quark pair . in four dimensions explicit solutions for the the schrdinger equation ( [ eq : schroedinger ] ) can be found , equivalent to the sum of diagrams ( [ eq : coulombfn ] ) . we quote the results for the colour - singlet green function . the general case is obtained by substituting @xmath393 everywhere . the momentum space pnrqcd propagator ( coulomb green function ) can be expressed in the form @xcite @xmath394 which closely resembles ( [ eq : coulombfn ] ) and shows that the sum from @xmath395 to infinity in ( [ eq : hc ] ) can be transformed into a remarkably simple integral . at this point we omit the @xmath70 prescription on @xmath69 and regard @xmath331 as a function of a complex energy variable , which has a cut for @xmath396 and isolated poles on the negative real axis . the variable @xmath71 equals @xmath397 as defined in ( [ eq : deflambda ] ) . the first line of ( [ eq : schwingerrep ] ) separates the zero- and one - coulomb gluon exchange terms . in practice , we find it simpler to perform the all - order summation in the position space representation , where the potential insertions take a simple multiplicative ( rather than convolutive ) form , and therefore we do not make use of the above representation in the calculation in paper ii . an integral representation for the position space coulomb green function is @xmath398^\lambda [ t(s-1)]^{-\lambda}\ , \nonumber\\ & & \times \frac{\partial^2}{\partial t\partial s } \left(\frac{ts}{|s{{\bf { r}}}-t{{\bf { r}}}^\prime| } \ , e^{-\sqrt{-me}\,((1-t)r^\prime+(s-1)r+|s{{\bf { r}}}-t{{\bf { r}}}^\prime| ) } \right),\end{aligned}\ ] ] valid for @xmath399 , where @xmath400 , @xmath401 @xcite . for @xmath402 exchange @xmath403 in the above expression . putting one of the arguments to zero , this simplifies to @xmath404 which depends only on @xmath400 . we use this form of the coulomb green function mainly for propagators connecting to the external current vertex , in which case ( [ eq : greenint ] ) applies . for the general case of a propagator in between two potential insertions the representation of the position - space green function in terms of laguerre polynomials @xmath405 @xcite turns out to be most useful . in this representation one first performs a partial wave expansion @xmath406}(r , r^\prime;e)\ , , \label{eq : gplexpand}\ ] ] where @xmath407 are the legendre polynomials . the partial - wave green functions read @xmath408}(r , r^\prime;e ) = \frac{m p}{2\pi } \ , ( 2 p r)^l ( 2 p r^\prime)^l\ , e^{-p(r+r^\prime ) } \sum_{s=0}^{\infty}\ , \frac{s!\,l_s^{(2l+1)}(2pr ) l_s^{(2l+1)}(2 p r^\prime ) } { ( s+2l+1 ) ! ( s+l+1-\lambda)}\ , , \label{eq : gpartial}\ ] ] where @xmath409 , and the laguerre polynomials are defined by @xmath410.\ ] ] since at nnnlo accuracy the potential insertions can not change the angular momentum of the quark anti - quark pair and since the production current @xmath241 creates an s - wave state , we only need the @xmath411 green function to compute the potential contributions to the pnrqcd correlation function ( [ eq : gpnrqcd ] ) . the p - wave green function is required to compute the ultrasoft contribution @xcite and the contribution ( [ eq : qcdaxialvectorcurrent ] ) from the p - wave production current . a general property of the coulomb interaction is that the ultraviolet behaviour of the ladder diagrams improves with the number of exchanges . thus , when the external current or potential insertions cause uv divergences , it is necessary to subtract only the first few terms in the sum of ladder diagrams . the divergent diagrams must be done in @xmath33 dimensions using standard methods , while for the convergent remainder one of the above expressions , properly subtracted , can be used . we therefore use the notation @xmath412 for example , the three terms in ( [ eq : schwingerrep ] ) correspond to @xmath413 . from ( [ eq : gexp ] ) it follows that the leading term in @xmath21 equals @xmath414 , which is , however , divergent as can be seen from ( [ eq : greenint ] ) . to compute @xmath415 in dimensional regularization , we note that the zero - coulomb exchange term is linearly divergent , the one - coulomb exchange logarithmically , and the remainder is convergent . we therefore compute @xmath416 from the first line of ( [ eq : schwingerrep ] ) , which yields @xmath417 . \label{eq : g01ex}\end{aligned}\ ] ] the remaining terms can be calculated from a modified version of ( [ eq : greenint ] ) with @xmath418 and integrand @xmath419 . after subtracting the first two terms of the @xmath13 expansion the result is finite as the regulator @xmath420 and gives the remaining contribution @xmath421 . the final result for the @xmath73 subtracted zero - distance green function @xcite is @xmath422\ , . \label{eq : g00msbar}\ ] ] the poles of the euler psi - function at positive integer @xmath71 correspond to the s - wave quark anti - quark bound states . near the bound - state poles the green function takes the form @xmath423 where @xmath424 is the wave - function at the origin of the @xmath250th bound state with energy @xmath425 . the imaginary part of the green function for @xmath396 is known as the sommerfeld factor @xcite . explicitly , the imaginary part below and above threshold is given by @xmath426 for real energies @xmath427 . this expression can be used in ( [ r1 ] ) to obtain the leading - order approximation to the resummed top pair production cross section in the threshold region for zero width of the top quark . we now summarize the momentum - space potentials required for the nnnlo calculation with the pnrqcd lagrangian ( [ eq : pnrqcd ] ) . only the colour - singlet projection @xmath428 of the general quark anti - quark potential @xmath429 is relevant to top - quark pair production through the electromagnetic and electroweak current , and this will be given in the following . the various potential terms can be ordered in a @xmath133 expansion , beginning with the coulomb potential of order @xmath430 . allowing for the spin - dependence from order @xmath129 , we write the singlet - potential in the general form : @xmath431\otimes[\sigma_i,\sigma_j ] \nonumber \\ & & -\,{\cal v}_{p}(\alpha_s)\frac{2\pi c_f \alpha_s({\bf p}^2+{\bf p}^{\prime 2})}{m^2{\bf{q}}^2 } + { \cal v}_{hf}(\alpha_s)\frac{\pi c_f \alpha_s}{4m^2{\bf q}^2 } [ \sigma_i,\sigma_j]q_j\otimes [ \sigma_i,\sigma_k]q_k \nonumber \\ & & - { \cal v}_{so}(\alpha_s)\frac{3\pi c_f \alpha_s}{2m^2 { \bf q}^2 } \bigg([\sigma_i,\sigma_j]q_i p_j \otimes 1 -1\otimes [ \sigma_i,\sigma_j]q_i p_j\bigg ) + \ldots . \label{eq : potentialbeforespin } \end{aligned}\ ] ] the coefficients @xmath432 of the potentials are @xmath13 dependent : @xmath433 the above representation of the potential therefore corresponds to an expansion in @xmath13 and @xmath12 , since the potential momenta @xmath161 , @xmath293 and @xmath434 are of order @xmath435 . with @xmath436 the leading term is of order @xmath437 . the ellipses denote terms of order @xmath438 , which would contribute from n@xmath439lo . a notation has been used which is valid in @xmath33 dimensions by avoiding the use of vector products or the totally antisymmetric @xmath164 tensor that would arise from using the three - dimensional identity for the commutator of pauli matrices . the coefficients of the potentials are chosen such that the leading - order coefficients are either one or zero . the tensor products @xmath440 refer to the spin matrices on the quark ( @xmath441 ) and anti - quark line ( @xmath442 ) . for the first three and the fifth terms of ( [ eq : potentialbeforespin ] ) , which are spin - independent , we omitted the trivial @xmath443 factor . the on - shell matching calculation of the potential coefficients @xmath444 , @xmath445 , @xmath446 , @xmath447 results in infrared ( ir ) divergences @xcite , which are related to ultraviolet divergences in the calculation of the ultrasoft correction . it is convenient to subtract these divergences in the results given below and add them back to the ultrasoft calculation ( see @xcite and section [ sec : ultrasoft ] ) . the subtraction term , which is _ added _ to the potential , is @xmath448 . \label{eq : counter}\end{aligned}\ ] ] in addition to potential insertions , the relativistic correction to the kinetic energy term @xmath449 in ( [ eq : pnrqcd ] ) needs to be included in pnrqcd perturbation theory . formally , this can be done by adding @xmath450 to the potential . since @xmath451 , it counts as a nnlo potential . the delta function eliminates the momentum integration that is associated with a potential insertion , but which is not present for the kinetic energy correction to a ( anti-)quark propagator . we now present the results for the potential coefficients , starting with the coulomb potential . the coefficient @xmath452 encodes the quantum corrections to the coulomb potential , which are needed up to the three - loop order . the insertions of coulomb potentials are finite , so we do not need the @xmath33-dimensional expression of the potential , as long as only coulomb potential insertions are considered . this reflects the fact that the schrdinger equation with the @xmath28 potential is non - singular and could be solved exactly , without referring to pnrqcd perturbation theory , as was done , for instance in @xcite . however , in the third - order computation of the top anti - top production cross section also the double insertion of the nlo coulomb potential together with the singular insertion of a nnlo non - coulomb potential has to be taken into account ; hence the order @xmath164 part of the one - loop coulomb potential multiplies a divergent quantity and contributes to the final result . the coefficient @xmath453 is therefore given with the full @xmath164 dependence . the first four terms in the expansion of the coulomb potential can be represented in the form @xmath454\ , \frac{\beta_0}{\epsilon}\ , + \bigg(\frac{\mu^2}{{{\bf { q}}}^2 } \bigg)^{\!\epsilon}\ , a_{1}(\epsilon ) , \label{eq : vcoulombnlo}\\ { \cal v}_{c}^{\,(2 ) } & = & a_2 + \left(2a_1\beta_0+\beta_1\right)\ln\frac{\mu^2}{{{\bf { q}}}^2 } + \beta_0 ^ 2\ln^2\frac{\mu^2}{{{\bf { q}}}^2 } , \\ { \cal v}_{c}^{\,(3 ) } & = & a_3 + \left(2a_1\beta_1+\beta_2 + 3a_2\beta_0 + 8\pi^2c_a^3\right ) \ln\frac{\mu^2}{{{\bf { q}}}^2 } \nonumber\\ & & + \left(\frac{5}{2}\beta_0\beta_1 + 3a_1\beta_0 ^ 2\right ) \ln^2\frac{\mu^2}{{{\bf { q}}}^2}+\beta_0 ^ 3\ln^3\frac{\mu^2}{{{\bf { q}}}^2 } , \label{eq : vcoulombn3lo}\end{aligned}\ ] ] with @xmath455 - 4\,t_fn_f\bigg)\ , \frac{e^{\gamma_e\epsilon}\,\gamma(1-\epsilon)\,\gamma(2-\epsilon)\ , \gamma(\epsilon)\ , } { ( 3 - 2\epsilon)\,\gamma(2 - 2\epsilon ) } -\frac{\beta_0}{\epsilon } , \\ & & a_{2 } = \left(\frac{4343}{162}+4\pi -\frac{\pi^{2}}{4 } + \frac{22}{3}\zeta_{3}\right)c_{a}^{2 } -\left(\frac{1798}{81}+\frac{56}{3}\zeta_{3}\right ) c_{a}t_{f}n_f \nonumber\\ & & \hspace*{1 cm } -\,\left(\frac{55}{3}-16\zeta_{3}\right ) c_f t_{f}n_f + \frac{400}{81}\,(t_f n_f)^2.\end{aligned}\ ] ] the one - loop correction @xmath456 has been known in @xmath233 for some time @xcite . we computed the @xmath33-dimensional expression and confirmed the result first shown in @xcite . the term @xmath457 without a momentum factor arises from the charge renormalization counterterm and the square bracket in ( [ eq : vcoulombnlo ] ) produces the associated logarithm . however , the expansion in @xmath164 can only be done after the momentum integrals of the potential insertion are performed , if these integrals are divergent . the two - loop coefficient @xmath458 has first been calculated in @xcite and correctly in @xcite . the three - loop coefficient @xmath32 is also known @xcite , though not all colour coefficients have been computed analytically : @xmath459 ( n_f t_f)^2 \nonumber \\ & & + \bigg[\ , \,\left(-709.717\right ) c_a^{\,2 } + \left(-\frac{71281}{162}+264\zeta_3 + 80\zeta_5\right ) c_a c_f \nonumber\\ & & \hspace{1 cm } + \left(\frac{286}{9}+\frac{296}{3}\zeta_3 - 160\zeta_5\right ) c_f^{\,2 } \bigg]\,n_f t_f \nonumber \\ & & + \bigg [ -56.83(1)\ , \biggl(\frac{d_f^{abcd}d_f^{abcd}}{2n_a\,t_f}\biggr)\ , \bigg ] n_f \nonumber \\ & & + \bigg[\ , 502.24(1)\ , c_a^{\,3 } + \,\left(-136.39(12)\right)\ , \biggl(\frac{d_a^{abcd}d_f^{abcd}}{2n_a\,t_f}\,\biggr ) \bigg]\ , . \label{eq : a3}\end{aligned}\ ] ] the numerical coefficients are quoted from @xcite , and the color factors @xmath460 for @xmath461 are given by @xmath462 parts of the nnnnlo coulomb potential are also known @xcite , but not needed for the third - order cross section calculation . the third - order coulomb potential has an ir divergence @xcite , which cancels against a divergence in the calculation of the nnnlo ultrasoft calculation . the corresponding @xmath42 pole is subtracted and therefore does not appear in ( [ eq : vcoulombn3lo ] ) , but the logarithmic part multiplied by @xmath463 in this equation comes from this divergence . note that the coefficient of this logarithm agrees with @xcite , but is three times larger than the one in @xcite . the reason for this is that here as in @xcite all three loops are computed in @xmath33 dimensions not just the divergent one , as is required for consistency with the ultrasoft calculation . hence the divergence related to the coulomb potential in ( [ eq : counter ] ) is multiplied by @xmath464 rather than @xmath465 . the coefficient of the @xmath466 potential is generated first at the one - loop order , where it is suppressed by @xmath467 relative to the leading coulomb potential . hence the two - loop coefficient is required for the nnnlo calculation of the cross section . the insertions of this potential cause ultraviolet divergences such that we need the one - loop coefficient to @xmath468 and the two - loop one to @xmath24 . up to the two - loop order we can represent the @xmath469 potential in the form @xmath470\left(-\frac{8}{3\epsilon}\right ) \bigg(2 c_f c_a + c_a^2 \bigg)\ , \nonumber \\ & & + \ , \bigg[\bigg(\frac{\mu^2}{{{\bf { q}}}^2}\bigg)^{\!2\epsilon } - \bigg ( \frac{{\mu } ^2}{{{\bf { q}}}^2}\bigg)^{\!\epsilon}\ , \bigg]\ , \frac{2\beta_0}{\epsilon } \ , b_{1}(\epsilon)\ , + \bigg(\frac{\mu^2}{{{\bf { q}}}^2 } \bigg)^{\ ! 2\epsilon } \ , 4 b_{2}(\epsilon)\ , , \label{eq : vb2}\end{aligned}\ ] ] with @xmath471 -c_a\,[1-\epsilon ] \right)\ , \frac { e^{\gamma_e\epsilon}\,\gamma(\frac{1}{2}-\epsilon)^{2 } \gamma(\frac{1}{2}+\epsilon)}{\pi^{\frac{3}{2}}\,\gamma(1 - 2\epsilon)}\ , , \label{b1eps}\\ b_{2}(\epsilon ) & = & \left[\frac{65}{18}-\frac{8 } { 3}\ln{2}\right ] c_ac_f -\left[\frac{101}{36}+\frac{4 } { 3}\ln{2}\right ] c_a^{2 } \nonumber \\ & & + \left[\frac{49}{36}\,c_a -\frac{2}{9 } c_f\right ] t_f n_f + \epsilon \,b_{2}^{(\epsilon)}+o(\epsilon^2)\,.\end{aligned}\ ] ] once again we subtracted the @xmath42 ir pole that remains after charge renormalization by adding the relevant part of @xmath472 from ( [ eq : counter ] ) . hence ( [ eq : vb2 ] ) is finite ; however , the expansion in @xmath164 must be performed only after the computation of the potential insertion . the one - loop expression @xmath473 has been computed in @xmath33 dimensions @xcite , and the four - dimensional value @xmath474 of the two - loop coefficient is quoted from @xcite . the @xmath24 term of @xmath475 has been parameterized above by @xmath476 . while known , it has not yet been published @xcite . in the numerical results in part ii of the paper we use the estimate @xmath477 , which is significantly larger but more realistic than our previous estimate @xmath478 ( @xmath479 ) @xcite . the coefficients of the @xmath480 potentials are generated at tree level , where they are suppressed by @xmath118 relative to the leading coulomb potential . hence the one - loop coefficients are required for the nnnlo calculation of the cross section . the tree - level coefficients are : @xmath481 as can be seen from ( [ eq : potentialbeforespin ] ) , spin - dependence arises first within the @xmath480 potentials . the insertions of these potentials are again ultraviolet divergent . we therefore need the @xmath24 term of the one - loop coefficients . these are available only from @xcite . we computed the @xmath33-dimensional expressions and confirmed the previous result . the spin - projected expression has already been given in our previous work @xcite . nrqcd tree level diagrams of order @xmath129 . dashed ( curly ) lines denote the @xmath137 ( @xmath138 ) gluon field . the number @xmath482 at the vertex refers to the nrqcd interaction with coefficient function @xmath225 , see also figure [ fig : nrqcd_fr ] and ( [ eq : nrqcdqq ] ) , ( [ eq : nrqcdqqqq ] ) , ( [ eq : nrqcdg ] ) . symmetric diagrams are not shown.,height=132 ] the complete one - loop coefficients consist of two different contributions . the first arises from the one - loop correction to the nrqcd couplings @xmath225 , already discussed in the previous section . we call this the `` hard '' contribution . it can be extracted from the tree diagrams in figure [ fig : nrqcdtree ] with one - loop corrected nrqcd vertices . the second contribution arises from explicitly integrating out soft loops . potential as discussed in section [ subsec : vectormatch ] below . however , there is no such contribution to the @xmath129 potentials at one loop . the reason for this is that in the case of the @xmath133 potential the relevant contribution arises from the box integral with all vertices of the leading - order @xmath483 type , expanded to subleading order in the potential region . the next correction is always suppressed by _ two _ powers of @xmath12 ( for instance , from replacing one of the vertices by a @xmath265 vertex from the nrqcd lagrangian ) , and hence can contribute only at order @xmath160 . ] the corresponding one - loop nrqcd diagrams are shown in figure [ fig : nrqcdloop ] . these contributions will be called `` soft '' . the total potential including tree and one - loop correction is then @xmath484 . the hard one - loop contributions can be easily calculated with the feynman rules presented in figure [ fig : nrqcd_fr ] , since all required @xmath33-dimensional nrqcd matching coefficients are already known at one - loop order from the previous section . the result reads : @xmath485 + o(\alpha_s^2 ) , \nonumber\\ { \cal v}_{p}^{(hard)}(\alpha_s ) & = & 1+o(\alpha_s^2 ) , \\ { \cal v}_{so}^{(hard)}(\alpha_s ) & = & \frac{1}{3}(2d_1+d_3)+o(\alpha_s^2 ) \\ & = & \nonumber 1+\frac{\alpha_s}{\pi}\bigg(\frac{\mu^2}{m^2}e^{\gamma_e}\bigg)^{\epsilon } \gamma(\epsilon ) \bigg[\frac{c_a(2\epsilon^2 - 1)-2c_f\epsilon(2\epsilon+1)}{6\epsilon-3}\bigg ] + o(\alpha_s^2 ) , \\ { \cal v}_{hf}^{(hard)}(\alpha_s ) & = & d_1 ^ 2+o(\alpha_s^2 ) \\ & = & \nonumber 1+\frac{\alpha_s}{\pi}\bigg(\frac{\mu^2}{m^2}e^{\gamma_e}\bigg)^{\epsilon } \gamma(\epsilon ) \bigg[\frac{c_a(2\epsilon^2 - 1)-2c_f\epsilon(2\epsilon+1)}{4\epsilon-2}\bigg ] + o(\alpha_s^2 ) , \\ { \cal v}_{s}^{(hard)}(\alpha_s ) & = & \frac{1}{2\pi c_f \alpha_s}(d_{sv}+c_f d_{vv})+o(\alpha_s^2 ) \label{eq : vm2hlast } \\ & = & \nonumber \frac{\alpha_s}{\pi}\bigg(\frac{\mu^2}{m^2}e^{\gamma_e}\bigg)^{\epsilon } \gamma(\epsilon ) \bigg[-\frac{c_a}{8}+\frac{\epsilon}{4\epsilon+2}c_f\bigg]+o(\alpha_s^2)\ , .\end{aligned}\ ] ] in this notation the tree - level value of the potential coefficient is included and assigned to the hard contribution . nrqcd one - loop diagrams of order @xmath129 ( non - vanishing diagrams shown only , symmetric diagrams not displayed).,height=377 ] the soft contributions come from diagrams of order @xmath129 shown in figure [ fig : nrqcdloop ] , from diagrams of order @xmath430 , where the denominator of a propagator has been expanded to higher orders as appropriate to the soft region , from soft self - energy insertions containing gluons and light quarks into the tree diagrams , and from charge renormalization counterterms . the final result is : @xmath486+o(\alpha_s^2 ) , \label{eq : vm2s}\\ { \cal v}_{p}^{(soft)}(\alpha_s ) & = & \frac{\alpha_s}{4\pi } \bigg[\bigg(\frac{\mu^2}{{\bf q}^2}\bigg)^{\!\epsilon } \frac{(-\epsilon)\gamma(-\epsilon)^2\gamma(\epsilon ) e^{\gamma_e\epsilon}\big(c_a(56\epsilon^2 - 121\epsilon+57 ) + 12n_f t_f(\epsilon-1)\big ) } { 6(4\epsilon^2 - 8\epsilon+3)\gamma(-2\epsilon ) } \nonumber \\ & & -\frac{\beta_0}{\epsilon}\bigg]+o(\alpha_s^2 ) , \\ { \cal v}_{so}^{(soft)}(\alpha_s ) & = & \frac{\alpha_s}{4\pi } \bigg[\bigg(\frac{\mu^2}{{\bf q}^2}\bigg)^{\!\epsilon } \frac{-\epsilon\gamma(-\epsilon)^2\gamma(\epsilon)e^{\gamma_e\epsilon } } { 6(4\epsilon^2 - 8\epsilon+3)\gamma(-2\epsilon ) } \big(c_a(d_3(8\epsilon^2 - 19\epsilon+11 ) \nonumber \\ & & + 2d_1(8\epsilon^2 - 15\epsilon+5))+ 4(2d_1+d_3)n_f t_f(\epsilon-1)\big)-\frac{2d_1+d_3}{3}\frac{\beta_0}{\epsilon}\bigg ] \nonumber \\ & & + o(\alpha_s^2 ) , \\ { \cal v}_{hf}^{(soft)}(\alpha_s ) & = & \frac{\alpha_s}{4\pi } \bigg[\bigg(\frac{\mu^2}{{\bf q}^2}\bigg)^{\!\epsilon } \frac{(-d_1 ^ 2)\epsilon(\epsilon-1 ) \gamma(-\epsilon)^2\gamma(\epsilon)e^{\gamma_e\epsilon } \big(c_a(4\epsilon-5)+ 4n_f t_f\big ) } { ( 8\epsilon^2 - 16\epsilon+6)\gamma(-2\epsilon)}- d_1 ^ 2\frac{\beta_0}{\epsilon}\bigg ] \nonumber \\ & & + o(\alpha_s^2 ) , \\ { \cal v}_{s}^{(soft)}(\alpha_s ) & = & \frac{\alpha_s}{4\pi } \bigg(\frac{\mu^2}{{\bf q}^2}\bigg)^{\!\epsilon } \frac{d_1 ^ 2\epsilon\gamma(-\epsilon)^2\gamma(\epsilon)e^{\gamma_e\epsilon } c_a } { ( 8\epsilon-4)\gamma(-2\epsilon)}+o(\alpha_s^2 ) . \label{eq : vm2slast}\end{aligned}\ ] ] here the @xmath225 can be set to their tree - level value 1 , since the entire expressions are already of order @xmath13 . an alternative method of calculating the soft contribution is to extract the soft region directly from the qcd diagrams . in this way , there are fewer diagrams than in the nrqcd calculation ( but more terms from the expansion ) . this calculation has also been performed and we checked that the results are exactly the same . note that with the second method we can not get the result in terms of the hard matching coefficients , but we obtain ( [ eq : vm2s ] ) to ( [ eq : vm2slast ] ) with @xmath487 . some remarks are in order on the pole structure of the hard and soft @xmath129 potential coefficients . we first note that we did not yet add the relevant contribution from the counterterm ( [ eq : counter ] ) , so the expressions given are still divergent . the @xmath42 pole in the hard contribution is infrared , while the soft contribution contains both infrared and ultraviolet divergences . it is instructive to separate the two : @xmath488_{\rm ir } \nonumber\\ & & + \,\bigg(\frac{\mu^2}{{\bf q}^2}\bigg)^{\!\epsilon } \left ( \left[\frac{13}{6 } c_a+\frac{2}{3 } c_f\right]_{\rm uv } + \left[\frac{8}{3 } c_a-\frac{16}{3 } c_f\right]_{\rm ir } \right ) \bigg\}+o(\alpha_s^2 ) , \\ { \cal v}_{p}(\alpha_s ) & = & 1 + \frac{\alpha_s}{4\pi \epsilon } \bigg(\frac{\mu^2}{{\bf q}^2}\bigg)^{\!\epsilon } \left[\frac{8}{3}c_a \right]_{\rm ir } + o(\alpha_s^2 ) , \\ { \cal v}_{so}(\alpha_s ) & = & 1 + \frac{\alpha_s}{4\pi \epsilon } \left\ { \bigg(\frac{\mu^2}{m^2}\bigg)^{\!\epsilon } \left[\frac{4}{3 } c_a\right]_{\rm ir } + \bigg(\frac{\mu^2}{{\bf q}^2}\bigg)^{\!\epsilon } \left[-\frac{4}{3 } c_a\right]_{\rm uv } \right\}+o(\alpha_s^2 ) , \\ { \cal v}_{hf}(\alpha_s ) & = & 1 + \frac{\alpha_s}{4\pi \epsilon } \left\ { \bigg(\frac{\mu^2}{m^2}\bigg)^{\!\epsilon } \left[2 c_a\right]_{\rm ir } + \bigg(\frac{\mu^2}{{\bf q}^2}\bigg)^{\!\epsilon } \left[-2 c_a\right]_{\rm uv } \right\}+o(\alpha_s^2 ) , \\ { \cal v}_{s}(\alpha_s ) & = & \frac{\alpha_s}{4\pi \epsilon } \left\ { \bigg(\frac{\mu^2}{m^2}\bigg)^{\!\epsilon } \left[-\frac{c_a}{2}\right]_{\rm ir } + \bigg(\frac{\mu^2}{{\bf q}^2}\bigg)^{\!\epsilon } \left[\frac{c_a}{2 } \right]_{\rm uv } \right\}+o(\alpha_s^2 ) . \label{eq : vm2polelast}\end{aligned}\ ] ] a term @xmath489 that was included in ( [ eq : vm2s ] ) to ( [ eq : vm2slast ] ) , which is related to the logarithms from charge renormalization , has been omitted from the expression in curly brackets for all potentials except the last one , which has no tree - level term . we see that the ir poles from the hard region cancel the uv poles from the soft region , as it should be , since these singularities arise from hard - soft factorization . the remaining ir singularities in the soft contribution appear only in the spin - independent potentials @xmath490 , @xmath491 . they are precisely of the form of the remaining terms in the subtraction term ( [ eq : counter ] ) and therefore cancel with uv divergences in the ultrasoft calculation . again , the structure of divergences is as required by consistency , since the ultrasoft contribution is spin - independent at nnnlo . since the spin - dependent potentials appear first at nnlo , their double insertion is of higher order than nnnlo . hence , when one computes the correlation function of the spin - triplet current ( [ eq : gpnrqcd ] ) , or the corresponding spin - singlet one , the spin - algebra can effectively be performed before the computation by working with spin - projected potentials . given a potential with spin - dependence @xmath440 , where @xmath441 ( @xmath442 ) refers to the spin - matrix on the quark ( anti - quark ) line , we replace @xmath492 note that the traces must be performed in @xmath150 space dimensions . only the spin - triplet projection is relevant to the third - order top production cross section in @xmath0 collisions . for the three spin - dependent terms in the general potential ( [ eq : potentialbeforespin ] ) , the projections result in @xmath493q_i p_j \otimes 1 -1\otimes [ \sigma_i,\sigma_j]q_i p_j\rightarrow 0 , \\ & & [ \sigma_i,\sigma_j]q_j\otimes [ \sigma_i,\sigma_k]q_k \rightarrow \frac{10 - 7d+d^2}{1-d}\ , 4{\bf q}^2 , \\ & & [ \sigma_i,\sigma_j]\otimes[\sigma_i,\sigma_j ] \rightarrow ( -4)(10 - 7d+d^2)\,,\end{aligned}\ ] ] where we have omitted the trivial @xmath443 dependence as done earlier . after the projection the four potentials @xmath494 , @xmath495 , @xmath496 , @xmath497 in ( [ eq : potentialbeforespin ] ) can be merged into a single expression @xmath498 , and we arrive at the spin - triplet , colour - singlet potential already presented in @xcite : @xmath499.\end{aligned}\ ] ] the coulomb and @xmath133 potentials are as given earlier . the two @xmath129 terms read , up to the one - loop order : @xmath500\ , \frac{1}{\epsilon}\ , \bigg(\frac{8}{3}c_a+\beta_0\bigg ) + \bigg ( \frac{\mu^2}{\bf{q}^2 } \bigg)^{\!\epsilon}\ , v^{(1)}_p(\epsilon)\ , , \label{eq : vp1 } \\[0.2 cm ] { \cal v}_{1/m^2}^{\,(0)}&= & v_0(\epsilon ) = -\frac{4-\epsilon-2\,\epsilon^2}{6 - 4\epsilon}\ , , \\ \label{eq : vc1 } { \cal v}_{1/m^2}^{(1 ) } & = & \bigg[\bigg(\frac{\mu^2}{\bf{q}^2 } \bigg)^{\!\epsilon } -1 \bigg]\ , \frac{1}{\epsilon}\ , \bigg(\,\frac{7}{3}c_f-\frac{11}{6}c_a\ , + \beta_0\ , v _ { 0}(\epsilon)\bigg ) \nonumber \\ & & + \,\bigg[\left(\frac{{\mu } ^2}{m^2 } \right)^{\!\epsilon}-1 \bigg]\ , \frac{1}{\epsilon}\ , \bigg(\,\frac{c_f}{3}+\frac{c_a}{2}\,\bigg ) + \bigg(\frac{\mu^2}{\bf{q}^2 } \bigg)^{\!\epsilon}\ , v^{(1)}_{q}(\epsilon ) + \bigg(\frac{{\mu } ^2}{m^2 } \bigg)^{\!\epsilon}\ , v^{(1)}_{m}(\epsilon)\,.\quad\end{aligned}\ ] ] the one - loop coefficients ( expanded up to @xmath24 ) are given by @xmath501 v^{(1)}_{q}(\epsilon ) & = & -\frac{c_f}{3 } -\frac{11}{27 } c_a + \frac{40}{27 } \,t_f n_f \label{eq : epsterms}\\ & & + \ , \epsilon \left\{\bigg(-\frac{419}{81}+\frac{77\pi^2}{216}\bigg)c_a + \bigg(2 -\frac{7\pi^2}{36 } \bigg)c_f+\bigg(\frac{274}{81 } -\frac{2\pi^2}{27}\bigg ) t_f n_f \right\}+o(\epsilon^2 ) \ , , \nonumber \\[0.2 cm ] v^{(1)}_{m}(\epsilon ) & = & -\frac{c_f}{3 } -\frac{29}{9 } c_a + \frac{4}{15}\,t_f + \epsilon \left\{\bigg(\frac{379}{54 } + \frac{\pi^2}{24}\bigg ) c_a+\bigg(-10 + \frac{\pi^2}{36}\bigg ) c_f \right\ } \nonumber\\ & & + \,o(\epsilon^2)\,,\end{aligned}\ ] ] where now the @xmath129 pieces of the subtraction term ( [ eq : counter ] ) have been added so that there are no @xmath42 poles . the four - dimensional expressions @xmath502 for @xmath503 agree with those obtained from @xcite , and the @xmath24 term agrees with @xcite . having determined the matching coefficients of the pnrqcd lagrangian we now return to the question whether the nrqcd spin - triplet current is renormalized when it is matched to pnrqcd . in general , we may write , in analogy with ( [ eq : qcdvectorcurrent ] ) , @xmath504 \psi^\dag\sigma^i\,{\bf d^2}\chi_{|\rm nrqcd } & = & \tilde{d}_{v2 } \,\psi^\dag\sigma^i\,{\bf d^2}\chi_{|\rm pnrqcd } + \ldots.\end{aligned}\ ] ] non - trivial ( @xmath505 , @xmath506 ) pnrqcd matching coefficients of the currents can arise from three sources : ( 1 ) soft loops not accounted in the matching of the lagrangian . this implies that the soft loop momentum must flow through the external current vertex . ( 2 ) off - shell effects . since the pnrqcd lagrangian is matched on - shell , off - shell effects that are not reproduced by the lagrangian interactions must be absorbed into a renormalization of the external currents . ( 3 ) @xmath507 terms of soft loops contributing to the matching of the lagrangian that multiply @xmath42 poles of pnrqcd loops are local and must be absorbed into a renormalization of the external currents , when the matching coefficients ( the potentials ) are defined in four dimensions . as discussed before , we choose to work with @xmath33-dimensional potentials , hence these contributions are included in the pnrqcd calculation without a modification of the external current . we shall now prove that there is also no further renormalization of the currents from ( 1 ) and ( 2 ) , that is , @xmath508 to three loops , and @xmath509 , @xmath510 at one loop . we first consider the issue ( 1 ) of soft renormalization of the nrqcd currents . the relevant vertex diagram at the one - loop order is shown in figure [ fig : vertex1loop ] with external momenta @xmath511 , @xmath512 and @xmath513 , and @xmath514 . the nrqcd expression is @xmath515 where the @xmath70 prescription for the propagators is left implicit . the unspecified polynomial factor arises from derivatives in the subleading nrqcd interactions or from the subleading external current . power - counting for soft loop momentum @xmath516 shows that this integral gives rise to a @xmath517 correction to @xmath518 . however , in the soft region we pick the pole at @xmath519 of the gluon propagator and expand the two quark propagators in @xmath69 and @xmath520 . the resulting integral @xmath521 is scaleless and vanishes in dimensional regularization . this holds to any order in the expansion in the soft region @xcite , hence there is no soft one - loop correction to the matching of the pnrqcd currents . moving to the two - loop level , we consider as an example the planar vertex diagram in figure [ fig : vertexplanar2loop ] . the momentum regions of interest are s - s and s - p , where the first letter refers to the inner vertex subgraph and loop momentum @xmath522 , the second to the box subgraph and loop momentum @xmath523 . the p - s combination is not relevant , since the soft loop does not flow through the external vertex . such contributions are included in the one - loop potentials . the inner vertex subdiagram in the s - s and s - p regions is an expression similar to ( [ vertex1loopnrqcd ] ) with @xmath524 in the quark propagator and @xmath525 in the gluon propagator . picking up the gluon propagator pole results in @xmath526 for the inner integral now , in the s - p region @xmath523 is potential , and @xmath527 must be expanded , in which case the integral is scaleless as before . if , as in the s - s region , @xmath523 is also soft , then the @xmath528-integration picks up the pole of the second gluon propagator at @xmath519 and the entire 2-loop integral is scaleless . this discussion evidently applies to all 2-loop vertex diagrams . since the vanishing of the integrals is due to the analytic structure of the propagator denominators , it generalizes to higher orders in the velocity expansion , which contains higher powers of the same propagators and numerator polynomials . hence , we conclude that there are no s - p and s - s contributions to two - loop vertex diagrams in any order in the threshold expansion , in accordance with the results of @xcite , and hence no soft renormalization of the currents at two loops . the structure of the analysis extends to higher loop orders . either one of the inner sub - graphs is scaleless , because an outer loop momentum is potential . or the entire diagram is soft and scaleless , because the external quark momenta are potential . we therefore conclude that there is no contribution to the matching of the nrqcd currents to pnrqcd from soft loops ( item ( 1 ) ) . we now turn to the discussion of off - shell effects , item ( 2 ) , and start again with the one - loop diagram shown in figure [ fig : vertex1loop ] . now , however , the loop momentum is potential , and we have to compare the nrqcd potential contribution contained in ( [ vertex1loopnrqcd ] ) with the pnrqcd expression @xmath529 which arises from the single insertion of the tree - level pnrqcd potential and the @xmath517 term of the coulomb green function . the factor @xmath530 corresponds to the coulomb potential insertion , while higher - order potentials as well as derivative factors from the external current are contained in the unspecified polynomial . the potential contribution in nrqcd is defined as the contribution from the quark - propagator pole at @xmath531 in ( [ vertex1loopnrqcd ] ) . the difference @xmath532 between @xmath533 and @xmath534 contributes to the matching of the external current and arises as follows : when the tree - level pnrqcd potential is derived from the one - gluon exchange diagram , the external quark lines are assumed on - shell , which implies @xmath535 and @xmath536 with the momentum assignment as in figure [ fig : vertex1loop ] . however , no such restrictions are imposed on the loop momentum @xmath523 in the calculation of the nrqcd diagram , figure [ fig : vertex1loop ] . thus the difference between @xmath533 and @xmath534 comes from the expansion of the gluon propagator in the potential region @xmath537 ^ 2}{{{{\bf { k}}}}^4 } + { \cal o}(v^2 ) , \label{offshell1}\ ] ] and the difference of polynomial factors , which after a short computation can be determined to be @xmath538 note that the leading contribution to @xmath533 and @xmath534 is of order @xmath86 . the terms neglected in ( [ offshell1 ] ) and ( [ offshell2 ] ) are therefore of order @xmath539 . these are fourth - order corrections to the cross section beyond the accuracy we aim at . in total , we arrive at @xmath540 ^ 2}{{{{\bf { k}}}}^2}\right\}_{\!|k^0 = \frac{e}{2}-\frac{({{{\bf { p}}}}+{{{\bf { k}}}})^2}{2 m } } \nonumber\\ & = & ( i g_s)^2 c_f\tilde\mu^{2\epsilon } \!\int \frac{d^{d-1 } { { { \bf { k}}}}}{(2\pi)^{d-1}}\ , \frac{1}{{{{\bf { k}}}}^2}\ , \frac{1}{2 k^0 } \times \left\{\frac{k^0}{m } - \frac{[k^0]^2}{{{{\bf { k}}}}^2}\right\}_{\!|k^0 = \frac{e}{2}-\frac{({{{\bf { p}}}}+{{{\bf { k}}}})^2}{2 m}}\,.\end{aligned}\ ] ] to arrive at the second line we used that the on - shell condition implies @xmath541 . the two terms in curly brackets each cancel the heavy - quark propagator denominator @xmath542 , after which the integral is scaleless and vanishes . this must be so , since a non - zero contribution to @xmath532 at this order would have scaled as @xmath467 , but there is no @xmath543 production current . that off - shell effects from the potential region _ are _ relevant in general can be seen from the calculation of the @xmath133-potential at @xmath544 . the potential region of the one - loop box diagram shown in figure [ fig : box1loop ] is not completely reproduced by pnrqcd . the difference is a contribution to the @xmath133 potential , that is , a pnrqcd matching coefficient , which is crucial to obtain the gauge - invariant result given in @xcite and ( [ b1eps ] ) . the difference between the box and the vertex diagram discussed above is that the box loop integral is not scaleless after the cancellation of the quark propagator by the off - shell terms , since there is a second gluon propagator @xmath545 . returning to the vertex diagrams , we now consider the planar two - loop diagram of figure [ fig : vertexplanar2loop ] in the p - p and p - s region , where the first letter refers again to the inner vertex subgraph . the corresponding pnrqcd diagrams are the two - loop vertex diagram with tree - level potentials and a one - loop vertex diagram with insertion of a one - loop potential , respectively . the off - shell terms of the nrqcd diagram in the p - p region are precisely the ones that contribute to the @xmath133 potential discussed in the previous paragraph ; they are correctly reproduced by the pnrqcd diagram with the insertion of the @xmath133 potential . the off - shell terms in the p - s region have a similar origin as in the one - loop vertex diagram . the soft box graph gives rise to a one - loop potential , but since the potentials are matched on - shell , the soft box graph is not completely reproduced when it appears as a subgraph in a larger diagram . since the leading p - s region is @xmath546 , if all the off - shell corrections were of order @xmath118 relative to the leading term as in ( [ offshell1 ] ) , ( [ offshell2 ] ) , we could immediately dismiss them , since there is no @xmath547 hard vertex correction into which it could be absorbed . this is indeed true for the planar diagram but not in general . as an example , we consider the non - planar nrqcd two - loop diagram shown in figure [ fig : vertexnonplanar2loop ] , which ( neglecting constant factors ) is given by the expression @xmath548 in the p - s region the @xmath522-integral for the soft crossed - box subdiagram is exactly the same that appears in the computation of the one - loop coulomb and @xmath133 potentials except for the additional @xmath528 in the last quark propagator that is absent when the quark lines of the inner vertex subgraph are on - shell ( see discussion above ) . since @xmath523 is potential , @xmath549 must be expanded relative to @xmath550 , which results in a series of corrections beginning at @xmath543 . if this off - shell correction were non - zero , it would result in a nnlo @xmath544 contribution to the coefficient function @xmath551 . however , as @xmath523 is potential , the @xmath528-integral is the contribution from the pole of the quark propagator in the first line of ( [ nonplanar2loopnrqcd ] ) . thus , in complete analogy with the discussion of the one - loop vertex diagram , the expansion in @xmath528 cancels the remaining quark propagator and renders the @xmath552-integral scaleless . this cancellation is generic for all two - loop vertex diagrams in the p - s region . the structure of the argument extends to the 3-loop order . the possible off - shell terms are either already accounted in the matching of subleading potentials ; or an inner vertex subgraph becomes scaleless due to a cancellation of the remaining quark - propagator of a potential loop . hence , we conclude that there is no contribution to the matching of the nrqcd currents to pnrqcd from off - shell terms ( item ( 2 ) ) , at least up to the nnnlo order . the integrals for pnrqcd diagrams with insertions of potentials or external currents such as ( [ multipleinsertions ] ) can sometimes be simplified by the equation of motion for the pnrqcd quark - antiquark propagator . we will make use of this to reduce the number of independent insertions that need to be calculated ( in part ii of the paper ) , and provide the relevant identities here . in the present context the equation of motion is the lippmann - schwinger equation ( [ eq : lippmann - schwinger ] ) for the colour - singlet coulomb green function written in the form @xmath553 in the remainder of this subsection we drop the energy argument of the green function , which is always @xmath69 , and set @xmath554 to simplify the notation . as our first example we consider the insertion of the subleading derivative current ( [ eq : qcdvectorcurrent2 ] ) into one of the vertices . the relevant integral is @xmath555 the second term on the right - hand side is a scaleless integral . the third one is seen to contain @xmath556 after shifting @xmath557 . hence , the insertion of @xmath558 can be replaced by the factor @xmath559 . this holds true in expressions of the form @xmath310 \,\frac{{{\bf { p}}}^2_1}{m^2}\ , i{g_0}({\bf p}_1,{\bf p}_2 ) i\delta{v}_1({\bf p}_2,{\bf p}_3 ) i{g_0}({\bf p}_3,{\bf p}_4)\ , i\delta{v}_2({\bf p}_4,{\bf p}_5 ) i{g_0}({\bf p}_5,{\bf p}_6)\ , \ldots\ ] ] that contain multiple potential insertions . this shows that the _ unrenormalized , dimensionally regulated _ matrix element of the subleading external current operator @xmath560 is @xmath561 times the unrenormalized matrix element of the leading - order current @xmath562 , which has been used in ( [ r1 ] ) . as our second example we consider the insertion of the @xmath563 potential . with @xmath564 we find @xmath565^{\frac{1}{2}+\epsilon } } \bigg ] g_0({\bf p}_3,{\bf p}_4)\,.\end{aligned}\ ] ] the final expression follows , since the second term on the left - hand side of the first equation contains the scaleless @xmath566-integral , while in the third term the integration over @xmath567 can be performed with the help of the integral @xmath568^{1+u } ( { { \bf { q}}}-{{\bf { k}}})^2 } \equiv \frac{1}{[{{\bf { q}}}^2]^{\frac{1}{2}+u+\epsilon}}\ , \frac{k(u)}{8}\end{aligned}\ ] ] with @xmath569 whose general form for @xmath570 will be needed below . this shows that the insertion of the @xmath563 potential can be eliminated in favour of the insertion of the coulomb and the @xmath33-dimensional @xmath571 potential . in the general case , after applying the spin projection , only six different types of insertions are needed for the nnnlo calculation : @xmath572 the first four come from the potentials in ( [ eq : potentialafterspin ] ) , the fifth is the kinetic energy correction and the last might be used for the conversion of the pole scheme to threshold mass scheme as discussed in paper ii . the identities given below show that the last three types of insertions can be reduced by using the equation of motion to the first three and the delta - function potential @xmath573 without factors of @xmath274 . at third order , both , single insertions and double insertions with an additional coulomb potential insertion have to be considered . for the single insertions , the equation of motion relations read : @xmath574^{\frac{1}{2}+\epsilon } } \frac{k(0)}{8 } g_0({\bf p}_3,{\bf p}_4 ) , \\ \frac{{\bf p}_2 ^ 2+{\bf p}_3 ^ 2}{2m^2 { \bf q}^2}:&&\int \prod_{i=1}^4\frac{d^{d-1}{\bf p}_i}{(2\pi)^{d-1}}g_0({\bf p}_1,{\bf p}_2 ) \frac{{\bf p}_2 ^ 2+{\bf p}_3 ^ 2}{2m^2 { \bf q}^2 } g_0({\bf p}_3,{\bf p}_4 ) \label{eompoverq } \\ & = & \int \prod_{i=1}^4\frac{d^{d-1}{\bf p}_i}{(2\pi)^{d-1}}g_0({\bf p}_1,{\bf p}_2 ) \bigg[\frac{e}{m } \frac{1}{{\bf q}^2 } + \frac{\pi c_f\alpha_s}{2 m } \frac{k(0)}{[{\bf q}^2]^{\frac{1}{2}+\epsilon } } \bigg ] g_0({\bf p}_3,{\bf p}_4).\nonumber\end{aligned}\ ] ] the delta - function potential @xmath573 appears implicitly in the first two relations in integrands of the form @xmath575 , where the delta - function has been eliminated to set two arguments equal . the equation of motion identities for the double insertions with a coulomb potential ( with @xmath576 and @xmath577 ) read : @xmath578^{1+u}}\nonumber \\ \nonumber & = & e\int \prod_{i=1,3 - 6}\frac{d^{d-1}{\bf p}_i}{(2\pi)^{d-1 } } g_0({\bf p}_1,{\bf p}_3 ) \frac{g_0({\bf p}_3,{\bf p}_4)g_0({\bf p}_5,{\bf p}_6)}{[{\bf q}_2 ^ 2]^{1+u } } \\ \nonumber & + & \int \prod_{i=1}^4\frac{d^{d-1}{\bf p}_i}{(2\pi)^{d-1 } } \frac{g_0({\bf p}_1,{\bf p}_2)g_0({\bf p}_3,{\bf p}_4)}{[{\bf q}_1 ^ 2]^{1+u } } \\ & + & 4\pi c_f \alpha_s\int \prod_{i=1}^6\frac{d^{d-1}{\bf p}_i}{(2\pi)^{d-1}}g_0({\bf p}_1,{\bf p}_2)\frac{1}{{\bf q}_1 ^ 2 } \frac{g_0({\bf p}_3,{\bf p}_4)g_0({\bf p}_5,{\bf p}_6)}{[{\bf q}_2 ^ 2]^{1+u } } , \\ \delta^{(d-1)}({\bf q}_1)\frac{{\bf p}_2 ^ 4}{m^3}:&&\int \prod_{i=1}^6\frac{d^{d-1}{\bf p}_i}{(2\pi)^{d-1}}g_0({\bf p}_1,{\bf p}_2 ) ( 2\pi)^{d-1}\delta^{(d-1)}({\bf q}_1)\frac{{\bf p}_2 ^ 4}{m^3 } \frac{g_0({\bf p}_3,{\bf p}_4)g_0({\bf p}_5,{\bf p}_6)}{[{\bf q}_2 ^ 2]^{1+u}}\nonumber \\ \nonumber & = & \frac{e^2}{m}\int \prod_{i=1,3 - 6}\frac{d^{d-1}{\bf p}_i}{(2\pi)^{d-1 } } g_0({\bf p}_1,{\bf p}_3 ) \frac{g_0({\bf p}_3,{\bf p}_4)g_0({\bf p}_5,{\bf p}_6)}{[{\bf q}_2 ^ 2]^{1+u } } \\ \nonumber&+&\int \prod_{i=1}^4\frac{d^{d-1}{\bf p}_i}{(2\pi)^{d-1}}g_0({\bf p}_1,{\bf p}_2 ) \left[\frac{2e}{m}\frac{1}{[{\bf q}_1 ^ 2]^{1+u}}+ \frac{\pi c_f \alpha_s}{2 m}\frac{k(u)}{[{\bf q}_1 ^ 2]^{\frac{1}{2}+u+\epsilon}}\right ] g_0({\bf p}_3,{\bf p}_4 ) \\ \nonumber&+&4\pi c_f \alpha_s\int \prod_{i=1}^6\frac{d^{d-1}{\bf p}_i}{(2\pi)^{d-1}}g_0({\bf p}_1,{\bf p}_2 ) \left[\frac{2e}{m}\frac{1}{{\bf q}_1 ^ 2}+ \frac{\pi c_f \alpha_s}{2m}\frac{k(0)}{[{\bf q}_1 ^ 2]^{\frac{1}{2}+\epsilon}}\right ] \\ & & \times\frac{g_0({\bf p}_3,{\bf p}_4)g_0({\bf p}_5,{\bf p}_6)}{[{\bf q}_2 ^ 2]^{1+u } } , \\ \frac{{\bf p}_2 ^ 2+{\bf p}_3 ^ 2}{2 m^2 { \bf q}_1 ^ 2}:&&\int \prod_{i=1}^6\frac{d^{d-1}{\bf p}_i}{(2\pi)^{d-1}}g_0({\bf p}_1,{\bf p}_2 ) \frac{{\bf p}_2 ^ 2+{\bf p}_3 ^ 2}{2 m^2 { \bf q}_1 ^ 2 } \frac{g_0({\bf p}_3,{\bf p}_4)g_0({\bf p}_5,{\bf p}_6)}{[{\bf q}_2 ^ 2]^{1+u}}\nonumber \\ & & \hspace*{-1 cm } = \,\int \prod_{i=1}^6\frac{d^{d-1}{\bf p}_i}{(2\pi)^{d-1}}g_0({\bf p}_1,{\bf p}_2 ) \left[\frac{e}{m } \frac{1}{{\bf q}_1 ^ 2 } + \frac{\pi c_f\alpha_s}{2 m } \frac{k(0)}{[{\bf q}_1 ^ 2]^{\frac{1}{2}+\epsilon}}\right ] \frac{g_0({\bf p}_3,{\bf p}_4)g_0({\bf p}_5,{\bf p}_6)}{[{\bf q}_2 ^ 2]^{1+u}}\nonumber \\ & & \hspace*{-1 cm } + \ , \frac{k(u)}{16m}\int\prod_{i=1}^4\frac{d^{d-1}{\bf p}}{(2\pi)^{d-1 } } \frac{g_0({\bf p}_1,{\bf p}_2)g_0({\bf p}_3,{\bf p}_4)}{[{\bf q}_1 ^ 2]^{\frac{1}{2}+u+\epsilon } } \,.\end{aligned}\ ] ] we note that we have kept the insertion of the coulomb potential in the more general form @xmath579^{1+u}$ ] , since in the double insertions we may also need the @xmath33-dimensional coulomb potential , implying @xmath580 . at third order there is for the first time a contribution from an ultrasoft loop momentum region . the ultrasoft correction to the heavy - quark correlation function has been computed separately in @xcite and will be incorporated in the results shown in paper ii . here we provide a short overview of the ultrasoft calculation and discuss some issues of the factorization , which are important to understand the splitting of various divergent parts . the relevant ultrasoft interaction terms in the pnrqcd lagrangian ( [ eq : pnrqcd ] ) are given by @xmath581\psi(x ) + g_s\chi^\dagger(x)\big[a_0(t,{{\bf { 0 } } } ) -{{\bf { x}}}\cdot { { \bf { e}}}(t,{{\bf { 0}}})\big]\chi(x ) . \label{usterms}\ ] ] the derivation of the chromoelectric dipole interaction from the multipole expansion of the nrqcd lagrangian can be found in @xcite . the interaction with @xmath582 can be removed by a field redefinition involving a time - like wilson line . this modifies the external current that creates the heavy - quark pair , as discussed in @xcite . in the present case of colour - singlet production in @xmath0 collisions the wilson lines cancel , and the @xmath582 terms in ( [ usterms ] ) can be dropped . with @xmath583 , and @xmath584 for ultrasoft gluon fields , it follows that the chromoelectric dipole interaction is suppressed by @xmath167 relative to the kinetic term in the action . two ultrasoft interaction vertices are required to form a loop , from which it follows that the leading ultrasoft contribution arises first at the third order . the ultrasoft correction can be expressed in the form @xmath585,\ , \label{eq : gfcorr}\end{aligned}\ ] ] with the understanding that one picks up only the pole at @xmath586 in the gluon propagator . here @xmath587 is the colour - singlet and @xmath588 the colour - octet coulomb green function ( [ eq : lippmann - schwinger ] ) . however , as explained in @xcite , this expression can not be used in practice , because it is ultraviolet ( uv ) divergent . the regularization and subtraction of divergences must be done consistently with the calculation of potential insertions and hard matching coefficients , which have been done in dimensional regularization . in order to apply dimensional regularization to the ultrasoft contribution , ( [ eq : gfcorr ] ) is transformed to momentum space . it also turns out to be convenient to revert the derivation of the pnrqcd ultrasoft interaction ( [ usterms ] ) and to instead use the nrqcd vertices . the reason for this is that the derivation of ( [ usterms ] ) uses the pnrqcd equation of motion , which reshuffles the loop expansion , and employs four - dimensional identities @xcite . the correspondence between uv divergences in the ultrasoft calculation and ir divergences in the potential and hard matching calculations is more directly seen at the level of nrqcd diagrams , and the correct evaluation of the finite terms requires the consistent use of dimensional regularization in every loop order . the uv divergences arise from the integral over the three - momentum @xmath589 of the ultrasoft gluon , and from the subsequent potential loop integrations . the former divergence is related to the factorization of the ultrasoft scale from the other scales , and cancels when all pieces of the calculation are added . the uv - divergent part of the ultrasoft integral has the form of a single insertion of a third - order potential and of a one - loop correction to the coefficient @xmath18 of the derivative current in ( [ eq : dv1loop ] ) . we therefore define the ultrasoft correction by adding counterterms that cancel these ultrasoft subdivergences . with these subtractions , the ultrasoft correction reads @xcite @xmath590 ^ 2\int \frac{d^{d-1}{\mbox{\boldmath $ { \ell}$}}}{(2\pi)^{d-1 } } \frac{d^{d-1}{\mbox{\boldmath $ { \ell ^\prime}$}}}{(2\pi)^{d-1 } } \bigg\{\delta d_v^{\rm div}\,(-1 ) \frac{{\mbox{\boldmath $ { \ell}$}}^2+{\mbox{\boldmath $ { \ell}$}}^{\prime\,2 } } { 6 m^2}\ , g^{(1)}_0({\mbox{\boldmath $ { \ell}$}},{\mbox{\boldmath $ { \ell}$}}^\prime;e ) \nonumber\\[0.2 cm ] & & \hspace*{-1.5 cm } + \,\big[\tilde\mu^{2\epsilon}\big]^2 \int \frac{d^{d-1}{\bf p}}{(2\pi)^{d-1 } } \frac{d^{d-1}{\bf p^\prime}}{(2\pi)^{d-1 } } \,g^{(1)}_0({\mbox{\boldmath $ { \ell}$}},{\bf p};e ) \,i\big[\delta u -\delta v_{c.t.}\big ] \,ig^{(1)}_0({\bf p^\prime},{\mbox{\boldmath $ { \ell^\prime}$}};e ) \bigg\}.\qquad \label{eq : defus}\end{aligned}\ ] ] here @xmath472 represents the potential subtraction ( [ eq : counter ] ) , and @xmath591 is the ultrasoft insertion ( containing the octet green function ) . with an opposite sign compared to @xcite . ] the first line of ( [ eq : defus ] ) is related to the renormalization of the @xmath129 suppressed vector current @xmath592 . if @xmath593_{\rm ren } = z_{1/m^2 } [ j_{1/m^2}]_{\rm bare}$ ] , the one - loop counterterm @xmath594 equals the infrared divergence @xmath595 that was subtracted to obtain the finite expression ( [ eq : dv1loop ] ) . the explicit expression is @xmath596 the remaining divergences are associated with the three - loop hard matching coefficient @xmath25 . note that this divergence structure implies that the nrqcd current mixes with the @xmath129 suppressed current through ultrasoft interactions . schematically , @xmath597^{2}\,\right)_{| \rm us } = \mbox{const}\times\frac{\alpha_s}{\epsilon } \ , \frac{1}{6 m^2}\,\psi^\dagger \sigma^i { { \bf { d^2}}}\chi + \ldots \ , , \label{eq : usdiv}\ ] ] where the ellipses denote the remaining divergences related to time - ordered products with potentials and the leading - power current itself . the subtracted expression is then simplified and reduced to a number of integrations that can mostly be done only numerically . the code that computes the ultrasoft correction was developed in conjunction with @xcite , and is implemented in our program for the third - order cross section . we have now collected all prerequisites to write down the expansion of the non - relativistic correlation function @xmath598(x)\ , [ \psi^{\dag}\sigma^i\chi](0 ) ) non - relativistic ( pnrqcd ) perturbation theory . adopting the operator notation from section [ sec : qm ] , the expansion is given by @xmath599 with @xmath600 as given in ( [ eq : g00msbar ] ) , and @xmath601 \delta_2 g(e ) & = & \langle { { \bf { 0}}}| \hat g_0(e)i \delta v_1 i\hat g_0(e ) i\delta v_1 i\hat g_0(e)|{{\bf { 0}}}\rangle + \langle { { \bf { 0}}}| \hat g_0(e)i\delta v_2 i\hat g_0(e)|{{\bf { 0}}}\rangle,\\[0.2 cm ] \delta_3 g(e ) & = & \langle { { \bf { 0}}}| \hat g_0(e)i \delta v_1 i\hat g_0(e ) i \delta v_1 i\hat g_0(e ) i\delta v_1 i\hat g_0(e)|{{\bf { 0}}}\rangle \nonumber\\ & & + \ , 2 \langle { { \bf { 0}}}| \hat g_0(e)i \delta v_1 i\hat g_0(e ) i\delta v_2 i\hat g_0(e)|{{\bf { 0}}}\rangle + \langle { { \bf { 0}}}| \hat g_0(e)i\delta v_3 i\hat g_0(e)|{{\bf { 0}}}\rangle \nonumber\\ & & + \ , \delta^{us}g(e)\ , . \label{masterthirdorder}\end{aligned}\ ] ] in momentum space these expressions are of the from of single and multiple insertions of potentials as defined in ( [ multipleinsertions ] ) plus the ultrasoft correction . @xmath602 denotes a potential correction of order @xmath250 . the first - order potential consists only of the one - loop correction to the coulomb potential : @xmath603 at second order , we have the two - loop coulomb potential , the one - loop @xmath604 potential and the @xmath118-suppressed potentials at tree - level . together with the kinetic - energy correction , we obtain @xmath605 \nonumber\\[-0.1 cm ] & & - \,2\times \frac{{{\bf { p}}}^4}{8 m^3}\,(2\pi)^{d-1}\delta^{(d-1)}({{\bf { q}}}).\end{aligned}\ ] ] the notation for the potentials is defined in ( [ eq : potentialafterspin ] ) , where also the explicit expressions are given . note that different from ( [ eq : calvexp ] ) the @xmath250-loop potential @xmath606 includes the coupling constant factor , i.e. @xmath607 . there are no new potentials appearing at third order , hence @xmath608.\end{aligned}\ ] ] note that there is no kinetic - energy term in @xmath609 , because the kinetic energy term in the lagrangian is not renormalized . these potentials appear as single insertions in @xmath610 . in addition , @xmath610 receives contributions from multiple insertions of the lower - order potentials . to complete the perturbative expansion of the third - order cross section , the expansion ( [ greenexpand ] ) of the green function is inserted into ( [ r1 ] ) , @xmath611 g(e ) + \ldots\right ) \right].\ ] ] the coefficient functions @xmath17 and @xmath18 are likewise expanded , and product terms of order higher than three in non - relativistic perturbation theory are dropped . note that @xmath612 counts as second order in this expansion . this work concerns the third - order correction @xmath613 . the triple insertion of the first - order coulomb potential is algebraically complicated but has no uv or ir divergences and can therefore be computed numerically as done in @xcite . the ultrasoft correction was obtained in @xcite . in part ii of the present work we will give the details of the calculation of the remaining single and double insertions in the second line of ( [ masterthirdorder ] ) in dimensional regularization as is necessary for a consistent combination with the matching calculations performed and summarized in part i. a more precise master formula for the third - order correction to the green function that accounts for the pole structure of the @xmath33-dimensional potentials will be given in part ii . we thank a. maier , j. piclum and m. steinhauser for comments on the text . this work has been supported by the dfg sonderforschungsbereich / transregio 9 `` computergesttzte theoretische teilchenphysik '' , the dfg graduiertenkolleg `` elementarteilchenphysik an der tev - skala '' and the dfg cluster of excellence `` origin and structure of the universe '' . i. i. y. bigi , y. l. dokshitzer , v. a. khoze , j. h. khn and p. m. zerwas , _ production and decay properties of ultraheavy quarks _ , _ phys . _ * b181 * ( 1986 ) 157 . and * d0 * collaboration , t. aaltonen _ et al . _ , _ combination of the top - quark mass measurements from the tevatron collider _ , _ phys.rev . _ * d86 * ( 2012 ) 092003 [ http://arxiv.org/abs/1207.1069[arxiv:1207.1069 ] ] . and * cms * collaboration , _ combination of atlas and cms results on the mass of the top quark using up to 4.9 fb@xmath614 of data _ , atlas - conf-2012 - 095 , cms - pas - top-12 - 001 . m. beneke , p. falgari , s. klein , j. piclum , c. schwinn _ et al . _ , _ inclusive top - pair production phenomenology with topixs _ , _ jhep _ * 1207 * ( 2012 ) 194 [ http://arxiv.org/abs/1206.2454[arxiv:1206.2454 ] ] . m. martinez and r. miquel , _ multi - parameter fits to the t anti - t threshold observables at a future @xmath0 linear collider _ , _ eur . phys . j. _ * c27 * ( 2003 ) 4955 [ http://arxiv.org/abs/hep-ph/0207315 [ hep - ph/0207315 ] ] . k. seidel , f. simon , m. tesar and s. poss , _ top quark mass measurements at and above threshold at clic _ , _ eur.phys.j . _ * c73 * ( 2013 ) 2530 [ http://arxiv.org/abs/1303.3758[arxiv:1303.3758 ] ] . t. horiguchi , a. ishikawa , t. suehara , k. fujii , y. sumino _ et al . _ , _ study of top quark pair production near threshold at the ilc _ , [ http://arxiv.org/abs/1310.0563[arxiv:1310.0563 ] ] . collaboration , g. aad _ _ , _ observation of a new particle in the search for the standard model higgs boson with the atlas detector at the lhc _ , _ phys.lett . _ * b716 * ( 2012 ) 129 [ http://arxiv.org/abs/1207.7214[arxiv:1207.7214 ] ] . collaboration , s. chatrchyan _ et al . _ , _ observation of a new boson at a mass of 125 gev with the cms experiment at the lhc _ , _ phys.lett . _ * b716 * ( 2012 ) 3061 [ http://arxiv.org/abs/1207.7235[arxiv:1207.7235 ] ] . g. degrassi , s. di vita , j. elias - miro , j. r. espinosa , g. f. giudice _ _ , _ higgs mass and vacuum stability in the standard model at nnlo _ , _ jhep _ * 1208 * ( 2012 ) 098 [ http://arxiv.org/abs/1205.6497[arxiv:1205.6497 ] ] . b. a. thacker and g. p. lepage , _ heavy quark bound states in lattice qcd _ , _ phys . * d43 * ( 1991 ) 196208 . g. p. lepage , l. magnea , c. nakhleh , u. magnea and k. hornbostel , _ improved nonrelativistic qcd for heavy quark physics _ , _ phys . * d46 * ( 1992 ) 40524067 [ http://arxiv.org/abs/hep-lat/9205007 [ hep - lat/9205007 ] ] . g. t. bodwin , e. braaten and g. p. lepage , _ rigorous qcd analysis of inclusive annihilation and production of heavy quarkonium _ , _ phys . * d51 * ( 1995 ) 11251171 [ http://arxiv.org/abs/hep-ph/9407339 [ hep - ph/9407339 ] ] . m. beneke and v. a. smirnov , _ asymptotic expansion of feynman integrals near threshold _ , _ nucl . phys . _ * b522 * ( 1998 ) 321344 [ http://arxiv.org/abs/hep-ph/9711391[hep-ph/9711391 ] ] . v. s. fadin and v. a. khoze , _ threshold behavior of heavy top production in @xmath0 collisions _ , _ jetp lett . _ * 46 * ( 1987 ) 525529 . v. s. fadin and v. a. khoze , _ production of a pair of heavy quarks in @xmath615 annihilation in the threshold region _ , _ sov . _ * 48 * ( 1988 ) 309313 . m. j. strassler and m. e. peskin , _ the heavy top quark threshold : qcd and the higgs _ , _ phys . * d43 * ( 1991 ) 15001514 . m. jezabek , j. h. khn and t. teubner , _ momentum distributions in t anti - t production and decay near threshold _ , _ z. phys . _ * c56 * ( 1992 ) 653660 . y. sumino , k. fujii , k. hagiwara , h. murayama and c. k. ng , _ top quark pair production near threshold _ , _ phys . * d47 * ( 1993 ) 5681 . k. fujii , t. matsui and y. sumino , _ physics at @xmath9 threshold in @xmath0 collisions _ , _ phys . _ * d50 * ( 1994 ) 43414362 . r. harlander , m. jezabek , j. h. khn and t. teubner , _ polarization in top quark pair production near threshold _ , _ phys . * b346 * ( 1995 ) 137142 [ http://arxiv.org/abs/hep-ph/9411395[hep-ph/9411395 ] ] . a. h. hoang and t. teubner , _ top quark pair production at threshold : complete next - to - next - to - leading order relativistic corrections _ * d58 * ( 1998 ) 114023 [ http://arxiv.org/abs/hep-ph/9801397[hep-ph/9801397 ] ] . k. melnikov and a. yelkhovsky , _ top quark production at threshold with o(@xmath616 ) accuracy _ , _ nucl . * b528 * ( 1998 ) 5972 [ http://arxiv.org/abs/hep-ph/9802379[hep-ph/9802379 ] ] . m. beneke , a. signer and v. a. smirnov , _ top quark production near threshold and the top quark mass _ , _ phys . * b454 * ( 1999 ) 137146 [ http://arxiv.org/abs/hep-ph/9903260[hep-ph/9903260 ] ] . a. h. hoang and t. teubner , _ top quark pair production close to threshold : top mass , width and momentum distribution _ , _ phys . * d60 * ( 1999 ) 114027 [ http://arxiv.org/abs/hep-ph/9904468 [ hep - ph/9904468 ] ] . o. i. yakovlev , _ top quark production near threshold : nnlo qcd correction _ lett . _ * b457 * ( 1999 ) 170176 [ http://arxiv.org/abs/hep-ph/9808463[hep-ph/9808463 ] ] . t. nagano , a. ota and y. sumino , _ @xmath617 corrections to @xmath618 total and differential cross sections near threshold _ , _ phys . rev . _ * d60 * ( 1999 ) 114014 [ http://arxiv.org/abs/hep-ph/9903498[hep-ph/9903498 ] ] . a. a. penin and a. a. pivovarov , _ analytical results for @xmath618 and @xmath619 observables near the threshold up to the next- to - next - to - leading order of nrqcd _ , _ phys . atom . * 64 * ( 2001 ) 275293 [ http://arxiv.org/abs/hep-ph/9904278[hep-ph/9904278 ] ] . m. beneke , _ a quark mass definition adequate for threshold problems _ , _ phys . lett . _ * b434 * ( 1998 ) 115125 [ http://arxiv.org/abs/hep-ph/9804241[hep-ph/9804241 ] ] . a. h. hoang , a. v. manohar , i. w. stewart and t. teubner , _ a renormalization group improved calculation of top quark production near threshold _ * 86 * ( 2001 ) 19511954 [ http://arxiv.org/abs/hep-ph/0011254[hep-ph/0011254 ] ] . a. h. hoang , a. v. manohar , i. w. stewart and t. teubner , _ the threshold @xmath9 cross section at nnll order _ , _ phys . _ * d65 * ( 2002 ) 014014 [ http://arxiv.org/abs/hep-ph/0107144[hep-ph/0107144 ] ] . a. pineda and a. signer , _ heavy quark pair production near threshold with potential non - relativistic qcd _ * b762 * ( 2007 ) 6794 [ http://arxiv.org/abs/hep-ph/0607239[hep-ph/0607239 ] ] . a. hoang and m. stahlhofen , _ the top - antitop threshold at the ilc : nnll qcd uncertainties _ , [ http://arxiv.org/abs/1309.6323[arxiv:1309.6323 ] ] . n. brambilla , a. pineda , j. soto and a. vairo , _ the infrared behaviour of the static potential in perturbative qcd _ , _ phys . rev . _ * d60 * ( 1999 ) 091502 [ http://arxiv.org/abs/hep-ph/9903355 [ hep - ph/9903355 ] ] . b. a. kniehl and a. a. penin , _ ultrasoft effects in heavy quarkonium physics _ , _ nucl . _ * b563 * ( 1999 ) 200210 [ http://arxiv.org/abs/hep-ph/9907489[hep-ph/9907489 ] ] . n. brambilla , a. pineda , j. soto and a. vairo , _ the heavy quarkonium spectrum at order @xmath620 _ , _ phys . _ * b470 * ( 1999 ) 215 [ http://arxiv.org/abs/hep-ph/9910238 [ hep - ph/9910238 ] ] . b. a. kniehl and a. a. penin , _ order @xmath621 corrections to heavy- quarkonium creation and annihilation _ , _ nucl . phys . _ * b577 * ( 2000 ) 197208 [ http://arxiv.org/abs/hep-ph/9911414[hep-ph/9911414 ] ] . a. v. manohar and i. w. stewart , _ running of the heavy quark production current and @xmath105 potential in qcd _ * d63 * ( 2001 ) 054004 [ http://arxiv.org/abs/hep-ph/0003107[hep-ph/0003107 ] ] . y. kiyo and y. sumino , _ @xmath622 quarkonium 1s spectrum in large @xmath623 approximation and renormalon cancellation _ , _ phys . lett . _ * b496 * ( 2000 ) 8388 [ http://arxiv.org/abs/hep-ph/0007251[hep-ph/0007251 ] ] . b. a. kniehl , a. a. penin , m. steinhauser and v. a. smirnov , _ nonabelian @xmath624 heavy - quark - antiquark potential _ , _ phys . * d65 * ( 2002 ) 091503 [ http://arxiv.org/abs/hep-ph/0106135 [ hep - ph/0106135 ] ] . b. a. kniehl , a. a. penin , v. a. smirnov and m. steinhauser , _ potential nrqcd and heavy - quarkonium spectrum at next - to - next - to - next - to - leading order _ , _ nucl . phys . _ * b635 * ( 2002 ) 357383 [ http://arxiv.org/abs/hep-ph/0203166[hep-ph/0203166 ] ] . a. a. penin and m. steinhauser , _ heavy quarkonium spectrum at @xmath625 and bottom / top quark mass determination _ , _ phys . _ * b538 * ( 2002 ) 335345 [ http://arxiv.org/abs/hep-ph/0204290[hep-ph/0204290 ] ] . b. a. kniehl , a. a. penin , m. steinhauser and v. a. smirnov , _ heavy - quarkonium creation and annihilation with @xmath626 accuracy . _ , _ phys . * 90 * ( 2003 ) 212001 [ http://arxiv.org/abs/hep-ph/0210161[hep-ph/0210161 ] ] . a. h. hoang , _ three - loop anomalous dimension of the heavy quark pair production current in non - relativistic qcd _ , _ phys . * d69 * ( 2004 ) 034009 [ http://arxiv.org/abs/hep-ph/0307376 [ hep - ph/0307376 ] ] . m. beneke , y. kiyo and k. schuller , _ third - order coulomb corrections to the s - wave green function , energy levels and wave functions at the origin _ , _ nucl . _ * b714 * ( 2005 ) 6790 [ http://arxiv.org/abs/hep-ph/0501289[hep-ph/0501289 ] ] . a. a. penin , v. a. smirnov and m. steinhauser , _ heavy quarkonium spectrum and production / annihilation rates to order @xmath627 _ , _ nucl . phys . _ * b716 * ( 2005 ) 303318 [ http://arxiv.org/abs/hep-ph/0501042[hep-ph/0501042 ] ] . p. marquard , j. h. piclum , d. seidel and m. steinhauser , _ fermionic corrections to the three - loop matching coefficient of the vector current _ , _ nucl . * b758 * ( 2006 ) 144160 [ http://arxiv.org/abs/hep-ph/0607168[hep-ph/0607168 ] ] . m. beneke , y. kiyo and k. schuller , _ third - order non - coulomb correction to the s - wave quarkonium wave functions at the origin _ , _ phys . _ * b658 * ( 2008 ) 222229 [ http://arxiv.org/abs/arxiv:0705.4518 [ hep - ph][arxiv:0705.4518 ] ] . m. beneke , y. kiyo and a. a. penin , _ ultrasoft contribution to quarkonium production and annihilation _ , _ phys . lett . _ * b653 * ( 2007 ) 5359 [ http://arxiv.org/abs/arxiv:0706.2733 [ hep - ph][arxiv:0706.2733 ] ] . m. beneke , y. kiyo , a. penin and k. schuller , _ nnnlo correction to the toponium and bottomonium wave - functions at the origin _ , econf * c0705302 * ( 2007 ) top01 [ http://arxiv.org/abs/arxiv:0710.4236 [ hep - ph][arxiv:0710.4236 ] ] . m. beneke , y. kiyo and k. schuller , _ nnnlo results on top - quark pair production near threshold _ , pos radcor * 2007 * ( 2007 ) 051 [ http://arxiv.org/abs/arxiv:0801.3464 [ hep - ph][arxiv:0801.3464 ] ] . m. beneke and y. kiyo , _ ultrasoft contribution to heavy - quark pair production near threshold _ , _ phys . * b668 * ( 2008 ) 143147 [ http://arxiv.org/abs/0804.4004 [ arxiv:0804.4004 ] ] . a. v. smirnov , v. a. smirnov and m. steinhauser , _ fermionic contributions to the three - loop static potential _ , _ phys . lett . _ * b668 * ( 2008 ) 293298 [ http://arxiv.org/abs/0809.1927[arxiv:0809.1927 ] ] . p. marquard , j. piclum , d. seidel and m. steinhauser , _ completely automated computation of the heavy - fermion corrections to the three - loop matching coefficient of the vector current _ , _ phys.lett . _ * b678 * ( 2009 ) 269275 [ http://arxiv.org/abs/arxiv:0904.0920 [ arxiv:0904.0920 ] ] . c. anzai , y. kiyo and y. sumino , _ static qcd potential at three - loop order _ , _ phys . * 104 * ( 2010 ) 112003 [ http://arxiv.org/abs/0911.4335[arxiv:0911.4335 ] ] . a. v. smirnov , v. a. smirnov and m. steinhauser , _ three - loop static potential _ , _ phys . * 104 * ( 2010 ) 112002 [ http://arxiv.org/abs/0911.4742[arxiv:0911.4742 ] ] . y. kiyo and y. sumino , _ perturbative heavy quarkonium spectrum at next - to - next - to - next - to - leading order _ , [ arxiv:1309.6571 ] . a. v. manohar , _ the hqet / nrqcd lagrangian to order @xmath628 _ , _ phys . * d56 * ( 1997 ) 230237 [ http://arxiv.org/abs/hep-ph/9701294[hep-ph/9701294 ] ] . s. wster , _ heavy quark potential at order @xmath629 [ in german ] _ , _ diploma thesis _ rwth aachen university ( 2003 ) . m. e. luke and m. j. savage , _ power counting in dimensionally regularized nrqcd _ , _ phys . _ * d57 * ( 1998 ) 413423 [ http://arxiv.org/abs/hep-ph/9707313[hep-ph/9707313 ] ] . a. a. penin and a. a. pivovarov , _ top quark threshold production in gamma gamma collision in the next - to - leading order _ , _ nucl . * b550 * ( 1999 ) 375396 [ http://arxiv.org/abs/hep-ph/9810496 [ hep - ph/9810496 ] ] . j. h. khn and t. teubner , _ axial contributions at the top threshold _ , _ eur . phys . j. _ * c9 * ( 1999 ) 221228 [ http://arxiv.org/abs/hep-ph/9903322[hep-ph/9903322 ] ] . a. h. hoang , c. j. reisser and p. ruiz - femenia , _ phase space matching and finite lifetime effects for top- pair production close to threshold _ * d82 * ( 2010 ) 014005 [ http://arxiv.org/abs/1002.3223[arxiv:1002.3223 ] ] . m. beneke , b. jantzen and p. ruiz - femenia , _ electroweak non - resonant nlo corrections to @xmath630 in the @xmath631 resonance region _ , _ nucl . * b840 * ( 2010 ) 186213 [ http://arxiv.org/abs/1004.2188[arxiv:1004.2188 ] ] . r. j. guth and j. h. khn , _ top quark threshold and radiative corrections _ phys . _ * b368 * ( 1992 ) 3856 . a. h. hoang and c. j. reisser , _ electroweak absorptive parts in nrqcd matching conditions _ , _ phys . * d71 * ( 2005 ) 074022 [ http://arxiv.org/abs/hep-ph/0412258[hep-ph/0412258 ] ] . a. h. hoang and c. j. reisser , _ on electroweak matching conditions for top pair production at threshold _ , _ phys . * d74 * ( 2006 ) 034002 [ http://arxiv.org/abs/hep-ph/0604104[hep-ph/0604104 ] ] . d. eiras and m. steinhauser , _ complete higgs mass dependence of top quark pair threshold production to order @xmath632 _ , _ nucl . phys . _ * b757 * ( 2006 ) 197210 [ http://arxiv.org/abs/hep-ph/0605227 [ hep - ph/0605227 ] ] . y. kiyo , d. seidel and m. steinhauser , _ @xmath633 corrections to the @xmath634 vertex at the top quark threshold _ , http://arxiv.org/abs/0810.1597[arxiv:0810.1597 ] . s. actis , m. beneke , p. falgari and c. schwinn , _ dominant nnlo corrections to four - fermion production near the w - pair production threshold _ , _ nucl . * b807 * ( 2009 ) 132 [ http://arxiv.org/abs/0807.0102[arxiv:0807.0102 ] ] . m. beneke , a. p. chapovsky , a. signer and g. zanderighi , _ effective theory approach to unstable particle production _ , _ phys . * 93 * ( 2004 ) 011602 [ http://arxiv.org/abs/hep-ph/0312331 [ hep - ph/0312331 ] ] . m. beneke , a. p. chapovsky , a. signer and g. zanderighi , _ effective theory calculation of resonant high - energy scattering _ , _ nucl . * b686 * ( 2004 ) 205247 [ http://arxiv.org/abs/hep-ph/0401002 [ hep - ph/0401002 ] ] . m. beneke , p. falgari , c. schwinn , a. signer and g. zanderighi , _ four - fermion production near the w pair production threshold _ , _ nucl . * b792 * ( 2008 ) 89135 [ http://arxiv.org/abs/0707.0773 [ arxiv:0707.0773 ] ] . a. a. penin and j. h. piclum , _ threshold production of unstable top _ , _ jhep _ * 1201 * ( 2012 ) 034 [ http://arxiv.org/abs/1110.1970 [ arxiv:1110.1970 ] ] . a. h. hoang , c. j. reisser and p. ruiz - femenia , _ implementing invariant mass cuts and finite lifetime effects in top - antitop production at threshold _ , _ nucl . * 186 * ( 2009 ) 403406 [ http://arxiv.org/abs/0810.2934[arxiv:0810.2934 ] ] . b. jantzen and p. ruiz - femenia , _ nnlo non - resonant corrections to threshold top - pair production from @xmath0 : endpoint - singular terms _ , _ phys . _ * d88 * ( 2013 ) 054011 [ http://arxiv.org/abs/1307.4337[arxiv:1307.4337 ] ] . a. pineda , _ review of heavy quarkonium at weak coupling _ , _ prog.part.nucl.phys . _ * 67 * ( 2012 ) 735785 [ http://arxiv.org/abs/1111.0165[arxiv:1111.0165 ] ] . m. beneke , j. piclum and t. rauh , _ in preparation_. m. beneke , _ new results on heavy quarks near threshold _ , in : proceedings of the 33rd rencontres de moriond : electroweak interactions and unified theories , les arcs , france , 14 - 21 mar 1998 , [ http://arxiv.org/abs/hep-ph/9806429[hep-ph/9806429 ] ] . w. a. bardeen , a. buras , d. duke and t. muta , _ deep inelastic scattering beyond the leading order in asymptotically free gauge theories _ , _ phys.rev . _ * d18 * ( 1978 ) 3998 . m. beneke , _ nonrelativistic effective theory for quarkonium production in hadron collisions _ , in : ` the strong interaction , from hadrons to partons ' , stanford 1996 , p. 549574 [ http://arxiv.org/abs/hep-ph/9703429 [ hep - ph/9703429 ] ] . p. labelle , _ effective field theories for qed bound states : extending nonrelativistic qed to study retardation effects _ , _ phys . rev . _ * d58 * ( 1998 ) 093013 [ http://arxiv.org/abs/hep-ph/9608491 [ hep - ph/9608491 ] ] . m. e. luke and a. v. manohar , _ bound states and power counting in effective field theories _ , _ phys . * d55 * ( 1997 ) 41294140 [ http://arxiv.org/abs/hep-ph/9610534[hep-ph/9610534 ] ] . b. grinstein and i. z. rothstein , _ effective field theory and matching in non - relativistic gauge theories _ , * d57 * ( 1998 ) 7882 [ http://arxiv.org/abs/hep-ph/9703298[hep-ph/9703298 ] ] . a. pineda and j. soto , _ effective field theory for ultrasoft momenta in nrqcd and nrqed _ , _ nucl . _ * 64 * ( 1998 ) 428432 [ http://arxiv.org/abs/hep-ph/9707481[hep-ph/9707481 ] ] . a. pineda and j. soto , _ the lamb shift in dimensional regularization _ lett . _ * b420 * ( 1998 ) 391396 [ http://arxiv.org/abs/hep-ph/9711292[hep-ph/9711292 ] ] . n. brambilla , a. pineda , j. soto and a. vairo , _ potential nrqcd : an effective theory for heavy quarkonium _ , _ nucl . phys . _ * b566 * ( 2000 ) 275 [ http://arxiv.org/abs/hep-ph/9907240[hep-ph/9907240 ] ] . a. h. hoang , m. jezabek , j. h. khn and t. teubner , _ radiation of heavy quarks _ lett . _ * b338 * ( 1994 ) 330335 [ http://arxiv.org/abs/hep-ph/9407338[hep-ph/9407338 ] ] . a. pineda and j. soto , _ matching at one loop for the four - quark operators in nrqcd _ , _ phys . _ * d58 * ( 1998 ) 114011 [ http://arxiv.org/abs/hep-ph/9802365[hep-ph/9802365 ] ] . m. beneke , a. signer and v. a. smirnov , _ two - loop correction to the leptonic decay of quarkonium _ , _ phys . * 80 * ( 1998 ) 25352538 [ http://arxiv.org/abs/hep-ph/9712302 [ hep - ph/9712302 ] ] . a. czarnecki and k. melnikov , _ two - loop qcd corrections to the heavy quark pair production cross section in @xmath0 annihilation near the threshold _ _ * 80 * ( 1998 ) 25312534 [ http://arxiv.org/abs/hep-ph/9712222[hep-ph/9712222 ] ] . o. v. tarasov , a. a. vladimirov and a. y. zharkov , _ the gell - mann - low function of qcd in the three loop approximation _ , _ phys . * b93 * ( 1980 ) 429432 . w. bernreuther , r. bonciani , t. gehrmann , r. heinesch , t. leineweber _ _ , _ two - loop qcd corrections to the heavy quark form - factors : the vector contributions _ , _ nucl.phys . _ * b706 * ( 2005 ) 245324 [ http://arxiv.org/abs/hep-ph/0406046[hep-ph/0406046 ] ] . b. kniehl , a. onishchenko , j. piclum and m. steinhauser , _ two - loop matching coefficients for heavy quark currents _ , _ phys.lett . _ * b638 * ( 2006 ) 209213 [ http://arxiv.org/abs/hep-ph/0604072 [ hep - ph/0604072 ] ] . m. beneke , _ perturbative heavy quark - antiquark systems _ , in : proceedings of the 8th international symposium on heavy flavor physics ( heavy flavors 8) , 25 - 29 july 1999 , southampton , england , [ http://arxiv.org/abs/hep-ph/9911490[hep-ph/9911490 ] ] . m. beneke , p. falgari and c. schwinn , _ soft radiation in heavy - particle pair production : all - order colour structure and two - loop anomalous dimension _ _ nucl.phys . _ * b828 * ( 2010 ) 69101 [ http://arxiv.org/abs/0907.1443[arxiv:0907.1443 ] ] . m. beneke , p. falgari and c. schwinn , _ threshold resummation for pair production of coloured heavy ( s)particles at hadron colliders _ , _ nucl.phys . _ * b842 * ( 2011 ) 414474 [ http://arxiv.org/abs/1007.5414[arxiv:1007.5414 ] ] . m. b. voloshin , _ nonperturbative effects in hadronic annihilation of heavy quarkonium _ , _ sov . * 40 * ( 1984 ) 662667 . m. b. voloshin , _ precoulombic asymptotics for energy levels of heavy quarkonium _ , _ sov . j. nucl . * 36 * ( 1982 ) 143 . d. eiras and j. soto , _ effective field theory approach to pionium _ , _ phys.rev . _ * d61 * ( 2000 ) 114027 [ http://arxiv.org/abs/hep-ph/9905543[hep-ph/9905543 ] ] . a. sommerfeld , _ ber die beugung und bremsung der elektronen _ , _ annalen der physik _ * 403 * ( 1931 ) 257330 . t. appelquist , m. dine and i. muzinich , _ the static limit of quantum chromodynamics _ , _ phys.rev . _ * d17 * ( 1978 ) 2074 . w. fischler , _ quark - anti - quark potential in qcd _ phys . _ * b129 * ( 1977 ) 157174 . a. billoire , _ how heavy must be quarks in order to build coulombic q anti - q bound states _ , _ phys . lett . _ * b92 * ( 1980 ) 343 . y. schrder , _ the static potential in qcd _ , _ ph.d . thesis _ ( 1999 ) . desy - thesis-1999 - 021 . m. peter , _ the static quark - antiquark potential in qcd to three loops _ * 78 * ( 1997 ) 602605 [ http://arxiv.org/abs/hep-ph/9610209[hep-ph/9610209 ] ] . m. peter , _ the static potential in qcd : a full two - loop calculation _ , _ nucl . * b501 * ( 1997 ) 471494 [ http://arxiv.org/abs/hep-ph/9702245[hep-ph/9702245 ] ] . y. schrder , _ the static potential in qcd to two loops _ , _ phys . lett . _ * b447 * ( 1999 ) 321326 [ http://arxiv.org/abs/hep-ph/9812205[hep-ph/9812205 ] ] . a. v. smirnov , v. a. smirnov and m. steinhauser , _ full result for the three - loop static quark potential _ , _ pos _ * radcor2009 * ( 2010 ) 075 [ http://arxiv.org/abs/1001.2668[arxiv:1001.2668 ] ] . n. brambilla , x. garcia i tormo , j. soto and a. vairo , _ the logarithmic contribution to the qcd static energy at nnnnlo _ , _ phys . lett . _ * b647 * ( 2007 ) 185 [ http://arxiv.org/abs/hep-ph/0610143 [ hep - ph/0610143 ] ] . m. steinhauser , _ private communication _ , _ _ ( 2010 ) .

complex systems have attracted much attention among physicists , applied mathematicians , engineers , and social scientists in recent years . in particular , agent - based models have become an important part of research on complex adaptive systems @xcite . for example , self - organized phenomena in an evolving population consisting of agents competing for a limited resource , have potential applications in areas such as engineering , economics , biology , and social sciences @xcite . the bar - attendance problem proposed by arthur @xcite constitutes an everyday example of such a system in which a population of agents decide whether to go to a popular bar having limited seating capacity . the agents are informed of the attendance in past weeks , and hence share common information , make decisions based on past experience , interact through their actions , and in turn generate this common information collectively . these ingredients are key characteristics of complex systems @xcite . an important step in the recent explosion of research in agent - based models within the physics community , has been the introduction of binary ising - like versions of models of competing populations examples include the minority game ( mg ) @xcite and the binary - agent - resource ( b - a - r ) game @xcite . for modest resource levels in which there are more losers than winners , the minority game @xcite represents a simple , yet highly non - trivial , model that captures many of the essential features of such a competing population . it has been the subject of many theoretical studies @xcite . the mg considers an odd number @xmath0 of agents . at each timestep , the agents independently decide between two options ` @xmath3 ' and ` @xmath4 ' . the winners are those who choose the minority option . the agents learn from past experience by evaluating the performance of their strategies , where each strategy maps the available global information , i.e. the record of the most recent @xmath5 winning options , to an action . one important quantity in the mg is the standard deviation @xmath6 of the number of agents making a particular choice . this quantity reflects the performance of the population as a whole in that a small @xmath6 implies on average more winners per turn , and hence a higher success rate per turn among the agents . in the mg , @xmath6 exhibits a non - monotonic dependence on the memory size @xmath5 of the agents @xcite . when @xmath5 is small , there is a significant overlap between the agents strategies . this crowd effect @xcite leads to a large @xmath6 , implying the number of losers is high . this is the crowded , or informationally efficient , phase of mg . in the informationally inefficient phase where @xmath5 is large , @xmath6 is moderately small and the agents perform better than if they were to decide their actions randomly . in this regime , information is left in the resulting bit - string patterns for a single realization of the system . in the inefficient phase , the mg can be mapped on to disordered spin systems and hence the machinery in statistical physics of disordered systems , most noticeably the replica trick , can be applied @xcite . however , the replica trick becomes ineffective in the efficient phase . the crowd - anticrowd theory @xcite gives a physically transparent , quantitative theory of the observed features of the mg in both the efficient and inefficient regimes . the crowd - anticrowd theory is based on the fact that it is the difference in the numbers of agents playing a given strategy @xmath7 and the corresponding anti - correlated strategy @xmath8 , that dictates the size of the fluctuations and hence performance of the population as a whole . in agent - based models , the non - triviality of the results comes from the actions taken by the agents which are directly related to the decision mechanism . the decision mechanism depends sensitively on how each strategy performs at the moment of decision . in the efficient phase of the mg , no strategy outperforms the others and therefore the relative performance of the strategies oscillates as the game proceeds . this anti - persistent nature of the strategy performance @xcite is crucial in arriving at a quantitative understanding of the mg s dynamics . as the systems evolves , it goes from one @xmath5-bit outcome or history bit - string to another . mathematically , the evolution can be viewed in terms of transitions in the global information ( i.e. history ) space . the @xmath9 possible history bit - strings for a given value of @xmath5 constitute the nodes in this history space . as the system evolves , it makes transitions from one node to another . _ showed that in the efficient ( i.e. low @xmath5 ) phase of the mg , an effective restoring force dominates the strategy - score dynamics yielding a eulerian trail quasi - attractor in history space @xcite . as @xmath5 increases , a competing bias term associated with the initial strategy allocation becomes increasingly important and eventually leads to instability of the eulerian trail quasi - attractor @xcite . johnson _ et al . _ @xcite subsequently introduced and studied numerically what is known as the binary - agent - resource ( b - a - r ) model , in which the winning group is not necessarily decided by the minority rule . the b - a - r model features a cutoff parameter @xmath2 , which is referred to as the global resource level @xmath2 ( @xmath10 ) . the values of @xmath2 and @xmath0 are not known to the agents . at each timestep @xmath11 , each agent decides upon two possible options : whether to access the resource or not . the winning action is decided by whether the number of agents attempting to access the resource , actually exceeds this resource level . the mg therefore corresponds to the particular case of @xmath12 in the b - a - r model . in ref.@xcite , it was found numerically that the population may unwittingly self - segregate itself into groups if @xmath2 deviates sufficiently from @xmath13 . as an example , consider a very high resource level of @xmath14 with each agent holding two strategies : it was found numerically that approximately @xmath15 agents are persistent winners while the rest are persistent losers . in the present work we analyze , both numerically and analytically , the transition from modest to high resources in this generic b - a - r model of a competing multi - agent population . surprisingly , we find that the system exhibits a set of abrupt transitions between distinct yet well - defined states as the resource level @xmath2 increases from @xmath13 to @xmath0 . in particular , both the highest success rate @xmath16 and the mean success rate @xmath17 among the agents , show abrupt transitions as @xmath2 varies . in addition , @xmath16 exhibits _ fractional _ values in the plateau regions between each transition . we show that this behavior can be understood in terms of the system s trajectory in the history space as time evolves . in particular , as @xmath2 increases , the portion of the history space that the system visits becomes increasingly restricted . we derive analytic expressions for the observed plateaux values and the values of @xmath2 at which the transitions occur . although the present analysis focuses on a non - networked population , the same elements of ( i ) strategy performance over time , and hence the dynamics of strategy scores , plus ( ii ) the system s trajectory in history space , provide the foundation for a quantitative understanding of a large class of agent - based models , including networked populations @xcite . we also note that for physicists interested in random walks , the present b - a - r system provides a fascinating laboratory for studying correlated , non - markovian diffusion on a non - trivial network ( i.e. the history space , which corresponds to a de bruijn graph ) . this non - trivial diffusion is in turn strongly coupled to the non - random temporal patterns arising in the strategy - performance dynamics . finally , we note in passing that the occurrence of abrupt transitions between plateau states , and stable fractions , are known to arise in a multi - electron quantum system as the external magnetic field is increased monotonically ( i.e. fractional quantum hall effect @xcite ) . however it is very curious to see such ` quantized ' phenomena arise in a classical multi - particle system as a function of a monotonically increasing external control parameter . the plan of the paper is as follows . in sec . ii , we define the b - a - r model . in sec . iii , we present numerical results from extensive simulations . in particular , we demonstrate the existence of different phases or states at different values of the resource level . in sec . iv , we discuss how the strategies performance evolves as the system evolves . in sec . v , we discuss the history space of the b - a - r model and present results for the statistics in the outcome bit - strings at high resource levels . in sec . vi , we explain the observed numerical features based on the idea that at high resource levels , the system restricts itself to only visit a restricted portion of the history space . we derive an expression for the highest success rate @xmath16 among the agents and discuss the critical values of resource level at which transitions occur . we summarize the results in sec . vii , together with a discussion of how the present approach can be generalized to a wider class of agent - based models . we consider the binary - agent - resource ( b - a - r ) model @xcite . the b - a - r model is a binary version of arthur s el farol bar attendance model @xcite , in which a population of agents repeatedly decide whether to go to a bar with limited seating based on the information of the crowd size in recent weeks . in the b - a - r model , there is a global resource level @xmath2 which is not announced to the agents , where @xmath0 is the total number of agents . at each timestep @xmath11 , each agent decides upon two possible options : whether to access resource @xmath2 ( action ` 1 ' ) or not ( action ` 0 ' ) . the two global outcomes at each timestep , ` resource over - used ' and ` resource not over - used ' , are denoted by ` @xmath3 ' and ` @xmath4 ' . if the number of agents @xmath18 choosing action @xmath4 exceeds @xmath2 ( i.e. resource over - used and hence global outcome ` @xmath3 ' ) then the @xmath19 abstaining agents win . by contrast if @xmath20 ( i.e. resource not over - used and hence global outcome ` @xmath4 ' ) then the @xmath18 agents win . in order to investigate the behavior of the system as @xmath2 changes , it is sufficient to study the range @xmath21 . the results for the range @xmath22 can be obtained from those in the present work by suitably interchanging the role of ` 0 ' and ` 1 ' @xcite . in the special case of @xmath23 , the b - a - r model reduces to the minority game . in the b - a - r model , each agent shares a common knowledge of the past history of the most recent @xmath5 outcomes , i.e. the winning option in the most recent @xmath5 timesteps . the full strategy space thus consists of @xmath24 strategies , as in the mg . initially , each agent randomly picks @xmath25 strategies from the pool of strategies , with repetitions allowed . the agents use these strategies throughout the game . at each timestep , each agent uses his momentarily best performing strategy with the highest virtual points . the virtual points for each strategy indicate the cumulative performance of that strategy : at each timestep , one virtual point ( vp ) is awarded ( deducted ) to ( from ) a strategy that would have predicted the correct ( incorrect ) outcome after all decisions have been made . a random coin - toss is used to break ties between strategies . in the b - a - r model , the population may or may not contain network connections . in the case of a networked population @xcite each agent has access to additional information from his connected neighbors , such as his neighbors strategies and/or performance . in the present work , we focus on the b - a - r model with a non - networked population . to evaluate the performance of an agent , one ( real ) point is awarded to each winning agent at a given timestep . a maximum of @xmath2 points per turn can therefore be awarded to the agents . an agent has a success rate @xmath26 , which is the mean number of points awarded to the agent per turn over a long time window . the mean success rate @xmath27 among the agents is then defined to be the mean number of points awarded per agent per turn , i.e. an average of @xmath26 over the agents . we are interested in investigating the details of how the success rate changes as the resource level @xmath2 varies in the efficient phase , where the number of strategies ( repetitions counted ) in play is larger than the total number of distinct strategies in the strategy space . the effects of varying @xmath2 were first reported by johnson _ et al . these authors studied numerically the dependence of the fluctuations in the number of agents taking a particular option , on the memory size @xmath5 for different values of @xmath2 . for the mg ( i.e. @xmath23 ) in the efficient phase ( i.e. small values of @xmath5 ) the number of agents making a particular choice varies from timestep to timestep , with additional stochasticity introduced via the random tie - breaking process . the corresponding period depends on the memory length @xmath5 . the underlying reason is that in the efficient phase for @xmath12 , no strategy is better overall than any other . hence there is a tendency for the system to restore itself after a finite number of timesteps , thereby preventing a given strategy s vps from running away from the others . as a result , the outcome bit - string shows the feature of anti - persistency or double periodicity @xcite . since a maximum of @xmath12 points can be awarded per turn , the mean success rate @xmath27 over a sufficiently large number of timesteps is bound from above by @xmath28 . in the b - a - r system with high resource level , the mean success rate behaves differently . taking the extreme case of @xmath29 , the winning action is obviously ` 1 ' ( i.e. access resource ) and in principle every agent could win in every timestep . the history upon which the agents decide , is persistently @xmath5-bits of ` 1 ' . however , due to the random initial strategy distribution , some agents may not hold a strategy that predicts the winning option for a history of @xmath5 @xmath4 s . therefore , there are still losers and @xmath27 is less than @xmath2 . the number of losers depends on @xmath25 , the number of strategies that each agent holds . for @xmath30 and assuming that the strategies are picked randomly , a mean number of @xmath31 agents in a large population will hold two strategies both predicting the wrong option . the mean success rate is thus given by @xmath32 . we have carried out extensive numerical simulations on the b - a - r model to investigate the dependence of the success rate on @xmath2 for @xmath21 . unless stated otherwise , we consider systems with @xmath33 agents and @xmath30 . figure [ fig : figure1](a ) shows the results of the mean success rate ( dark solid line ) as a function of @xmath2 in a typical run for @xmath34 , together with the range corresponding to one standard deviation about @xmath17 in the success rates among the @xmath0 agents ( dotted lines ) and the spread in the success rates given by the highest and the lowest success rates ( thin solid lines ) in the population . by taking a larger value of @xmath0 than most studies in the literature , we can analyze the dependence on @xmath2 and @xmath5 in great detail . in particular , these quantities all exhibit abrupt transitions ( i.e. jumps ) at particular values of @xmath2 . between the jumps , the quantities remain essentially constant and hence form steps or ` plateaux ' . we refer to these different plateaux as states or phases , since it turns out that the jump occurs when the system makes a transition from one type of state characterizing the outcome bit - string to another . for different runs , the results are almost identical . at most , there are tiny shifts in the @xmath2 values at which jumps arise due to ( i ) different initial strategy distributions among the agents in different runs , and ( ii ) different random initial history bit - strings used to start the runs . these different states are most clearly seen by monitoring the _ highest _ success rate @xmath16 among the agents for given values of @xmath2 and @xmath5 . the most striking feature in fig.[fig : figure1](a ) is that the values of the plateaux in @xmath16 are given by _ simple fractions _ , e.g. @xmath35 , @xmath36 , @xmath37 , @xmath38 , @xmath39 , etc . this feature strongly suggests that the system goes through different states with different ratios of ` @xmath4 ' and ` @xmath3 ' in the outcome bit - string as @xmath2 varies , as will be discussed in later sections . figure [ fig : figure1](b ) shows that the features in the success rates for the simpler case of @xmath40 are similar to those in fig.[fig : figure1](a ) , except that the plateaux in @xmath16 take on fewer values , i.e. @xmath4 , @xmath41 , @xmath39 as @xmath2 decreases . these values are closely related to the statistics in the outcome bit - string . for large @xmath0 and @xmath40 , the outcome bit - string shows a period of 4 bits . for values of @xmath2 with @xmath42 , it turns out that the fraction of the outcome ` @xmath4 ' in a period is exactly @xmath39 . for the range of @xmath2 corresponding to @xmath43 , there are three @xmath4 s in a period of @xmath44 , and so on . for @xmath34 , we have also carried out detailed analysis of the outcome bit - string . for later discussions , we summarize in table [ tab : otsm3s2 ] the values of @xmath16 , the range of @xmath2 corresponding to the observed value of @xmath16 , the ratio of number of occurrence of ` 1'-bits to ` 0'-bits and the period in the outcome bit - string , as obtained numerically from the data shown in fig.[fig : figure1 ] . hereafter , @xmath16 is used to label the state at a given @xmath2 . the agents decide based on the best performing strategy that they hold at the moment of their decision . a strategy s performance is evaluated by its virtual points , which vary as the game proceeds . it is most illustrative to consider the case of @xmath40 and @xmath30 ( see fig.[fig : figure1](b ) ) since one can readily follow the dynamics for different values of @xmath2 . for @xmath40 , there are only four different strategies in the whole strategy space . these strategies can be represented by ( @xmath45 ) , ( @xmath46 ) , ( @xmath47 ) , and ( @xmath48 ) , with the first ( second ) index in ( @xmath49 ) giving the action for the history bit - string of ` @xmath3 ' and ` @xmath4 ' , respectively . the virtual points ( vp ) of the four strategies can then be represented by a @xmath50 matrix vp@xmath51 , with the @xmath49-element giving the vp of the strategy ( @xmath49 ) . table [ tab : strgevom1s2 ] shows the time evolution of the vps of the strategies in a few timesteps , the number of agents @xmath52 taking the action ` 1 ' , and the outcome for @xmath40 in a population with _ uniform _ initial distribution of all possible pairs of strategies to the agents . we illustrate these ideas by considering the case of an initial history of ` 0 ' for given @xmath2 , but the same results are obtained for an initial history of ` 1 ' . for @xmath53 , the system follows the dynamics shown in the left column . in this range of @xmath2 , the vps can not run away due to the decision making process of the agents . the vps restore their values in a few timesteps , as in the mg . for systems of large @xmath0 , the outcome series shows a 4-bit periodic pattern of @xmath54 , with @xmath55 in agreement with the numerical results in fig . [ fig : figure1](b ) . this highest success rate is achieved by the agents who hold two identical strategies . the result also implies that an agent s success rate is determined by the hamming distance between the two strategies that the agent holds @xcite . one could therefore say that different ` species ' of agent emerge in the population due to the dynamics of the system , with each species characterized by its own hamming distance ` gene ' . for @xmath56 , two of the four strategies will have runaway vps , with one tending to increase without bound while another tends to decrease without bound ( see central column in table [ tab : strgevom1s2 ] ) . following the dynamics , the outcome series shows a 4-bit periodic pattern of @xmath57 , in the limit of large @xmath0 . in this range , @xmath43 as observed numerically and this value is achieved by those agents holding the strategy whose vps increase without bound . for @xmath58 , two strategies have their vps increasing ( decreasing ) without bound ( see right column in table [ tab : strgevom1s2 ] ) . the outcome series is persistently ` @xmath4 ' , i.e. it becomes effectively period-1 . in this high resource level regime , @xmath32 since three - quarters of the agents hold at least one strategy which predicts the persistently winning option . in this way , it is possible to follow the dynamics of the system and obtain the number of possible states and the range of @xmath2 for each state . table [ tab : wrm1s2 ] summarizes the theoretical results for @xmath40 and @xmath30 , by following the analysis on the dynamics as shown in table [ tab : strgevom1s2 ] . the results for the location of the transitions , the values of @xmath27 and @xmath16 are all in good agreement with numerical data ( see fig . [ fig : figure1](b ) ) . the important point is that for small @xmath5 , the system undergoes several changes of state with successively higher values of @xmath27 and @xmath16 as @xmath2 increases . the value of @xmath16 is related to the ratio of occurrences of the two possible outcomes in the outcome series , which in turn is related to the strategies performance and thus the decision - making process for a given value of @xmath2 . we have carried out similar analysis for the case of @xmath59 and @xmath34 . for @xmath34 , the fraction of ` @xmath4 ' in a period is found to take on values in the set @xmath60 , which coincides with the numerical values of @xmath16 obtained by numerical simulations ( see fig . [ fig : figure1](a ) ) and shown in table [ tab : otsm3s2 ] . however , the analysis becomes increasingly complicated for higher values of @xmath5 and/or @xmath25 . the reason is that there are @xmath61 strategies , and allowing @xmath25 strategies per agent divides the agents into @xmath62 groups according to the sequence in which an agent picks his @xmath25 strategies . this number increases rapidly with @xmath5 and @xmath25 , and the above microscopic analysis becomes hard to implement . it turns out the states are closely related to the way in which the system explores the possible histories . in what follows , we analyze the b - a - r model by an approach that focuses on the transitions between history bit - strings and hence on the path in the history space @xcite . our approach couples together consideration of the probability of the occurrence of various histories and the ranking in the performance of the strategies @xcite . the history space consists of all the possible history bit - strings for a given value of @xmath5 . for @xmath34 , it includes @xmath63 bit - strings of @xmath3 s and @xmath4 s . figure [ fig : figure2](a ) shows the history space for @xmath34 , together with the possible transitions from one history to another . each history constitutes a node in the history space . the transitions are marked by arrows together with the outcome necessary for making the transitions . it will prove convenient to group the possible history bit - strings for a given @xmath5 into columns , in the way shown in fig . [ fig : figure2](a ) . each column is labelled by a parameter @xmath64 , which is the number of s in the @xmath65-bit history ( histories ) concerned . one immediate advantage of this labelling scheme is that the different states characterized by @xmath16 turn out to involve paths in a _ restricted _ portion of the full history space . for example , the state with @xmath66 is restricted to the @xmath67 portion of the history space , i.e. the @xmath68 history bit - string leads to an outcome of ` 1 ' and hence persistent self - looping at the node @xmath68 in history space . the states with @xmath69 , @xmath36 , and @xmath37 correspond to different paths in the history space restricted to the @xmath70 and @xmath71 groups of histories , as shown in fig . [ fig : figure2](b ) . the states with @xmath72 and @xmath38 have paths extended to include @xmath73 histories . the state with @xmath74 has paths that cover the whole history space ( @xmath75 ) . in general , the deviation of @xmath2 from @xmath13 acts like a driving force in the history space that drifts the system towards an increasingly restrictive portion of the history space bounded by a smaller value of @xmath64 . one can also view this behavior as the system , in response to the global resource level @xmath2 , effectively avoiding certain nodes in the history space and hence avoiding particular patterns of historical outcomes . as the game proceeds , the system evolves from one history bit - string to another . this can be regarded as transitions between different nodes ( i.e. different histories ) in the history space . for @xmath12 in the efficient phase , it has been shown @xcite that the conditional probability of an outcome of , say , ` @xmath4 ' following a given history is the same for all histories . for @xmath76 , the result still holds for states characterized by @xmath74 . note that a history bit - string can only make transitions to history bit - strings that differ by the most recent outcome , e.g. 111 can only be make transitions to either 110 or 111 , and thus many transitions between two nodes in the history space are forbidden . in addition , these allowed transitions do not in general occur with equal probabilities . this leads to specific outcome ( and history ) bit - string statistics for a state characterized by @xmath16 . we have carried out detailed analysis of the statistics of the outcomes following a given history bit - string for @xmath34 , and for each of the possible states over the whole range of @xmath2 . table [ tab : tmm3 ] gives the relative numbers of occurrences of each outcome for every history bit - string . for the state with @xmath77 , for example , the outcomes ` @xmath3 ' and ` @xmath4 ' occur with equal probability for every history bit - string , as in the mg . for the other states , the results reveal several striking features . it turns out that @xmath16 is given by the _ relative frequency _ of an outcome of ` @xmath4 ' in the outcome bit - string , which in turn is governed by the resource level @xmath2 . for example , a ` 0 ' to ` 1 ' ratio of @xmath78 in the outcome bit - strings corresponds to the state with @xmath79 . in table [ tab : tmm3 ] , we have intentionally grouped the history bit - strings into rows according to the label @xmath64 in fig . [ fig : figure2 ] . we immediately notice that for every possible state in the b - a - r model , the relative frequency of each outcome is a property of the _ group _ of histories having the same label @xmath64 rather than the individual history bit - string , i.e. all histories in a group have the same relative fraction of a given outcome . this observation is important in understanding the dynamics in the history space for different states in that it is no longer necessary to consider each of the @xmath9 history bit - strings in the history space . instead , it is sufficient to consider the four groups of histories ( for @xmath34 ) as shown in fig . [ fig : figure2](a ) . analysis of results for higher values of @xmath5 show the same feature . for the state characterized by @xmath66 , the outcome bit - string is persistently ` @xmath4 ' and the path in the history space is repeatedly 111@xmath801 . therefore , the path is restricted to the history labelled by @xmath70 and simply corresponds to an infinite number of loops around the history node 111 . since there is no ` @xmath3 ' in the outcome bit - string , we will also refer to this state as @xmath81 state . the system is effectively frozen into one node in the history space . in this case , there are effectively only two kinds of strategies , which differ by their predictions for the history 111 . the difference in predictions for the other ( @xmath82 ) history bit - strings become irrelevant . obviously , the ranking in the performance of the two effective groups of strategies is such that the group of strategies that suggest an action ` 1 ' for the history ` 111 ' , outperforms the group that suggests an action ` 0 ' . for a uniform initial distribution of strategies , there are @xmath83 agents taking the action 0 and @xmath84 agents taking the action ` 1 ' , since half of the strategies predict 0 and half of them predict 1 . to sustain a winning outcome of ` 1 ' , the criterion is that the resource level @xmath2 should be higher than the number of agents taking the action ` 1 ' . therefore , we have for the state with @xmath85 that @xmath86 and @xmath87 these results are in agreement with the results obtained by numerical simulations . for @xmath30 , @xmath88 for @xmath58 . note that equations and are valid for _ any _ values of @xmath5 . table [ tab : tmm3 ] shows that the states with @xmath89 , @xmath36 , @xmath35 have very similar features in terms of the bit - string statistics . they differ only in the frequency of giving an outcome of @xmath4 following the history of @xmath68 . note that the @xmath90 and @xmath91 histories do not occur . the results imply that as the system evolves , the path in history space for these states is restricted to the two groups of histories labelled by @xmath67 and @xmath71 . the statistics show that the outcome bit - strings for the states with @xmath92 , @xmath36 and @xmath35 exhibit only one @xmath3-bit in a period of @xmath93 , @xmath94 and @xmath95 bits , respectively . we refer to these states collectively as @xmath96 states , since the portion of allowed history space is bounded by the @xmath71 histories . graphically , the path in history space consists of a few self - loops at the node 111 , i.e. from 111 to 111 , then passing through the @xmath71 group of histories once and back to 111 , as shown in fig . [ fig : figure2](b ) . the states with @xmath42 , @xmath38 , @xmath97 involve the other groups of histories and exhibit complicated looping among the histories . we refer to them collectively as higher ( i.e. @xmath98 ) states . we now proceed to derive an expression for the observed value of @xmath16 for the @xmath96 states . recall that each strategy consists of a prediction or action for all of the @xmath5-bit histories . an important idea is that for states corresponding to paths restricted to a certain portion of the history space , only that part of a strategy corresponding to the histories in question is being used in making decisions . strategies that only differ in their predictions for the history bit - strings which _ do not _ occur ( i.e. the avoided histories ) are now effectively identical . in the context of the crowd - anticrowd theory @xcite , two previously uncorrelated strategies could now be correlated when viewed within this restricted history subspace . as @xmath2 decreases , the system is allowed to explore a larger portion of the history space . the ranking in the strategies performance becomes more complicated . for the @xmath96 states , the paths in history space involve the ( @xmath99 ) histories labelled by @xmath1000 and 1 ( see fig . [ fig : figure2](b ) ) . for @xmath34 , only four out of a total of @xmath101 entries in a strategy corresponding to the histories ` @xmath68 ' , ` @xmath102 ' , ` @xmath103 ' , and ` @xmath104 ' now matter . for general @xmath5 , there are @xmath5 histories with one ` 0'-bit : hence a total of ( @xmath99 ) entries in each strategy now matter . for later discussions , it is useful to first classify all strategies into two groups according to their prediction for the history ` 11 1 ' ( @xmath70 history ) . these strategies can further be classified according to their @xmath5 predictions for the @xmath71 histories . a strategy having @xmath105 bits predicting 1 and @xmath106 bits predicting 0 for the @xmath5 histories belonging to @xmath71 , can be labelled as ( @xmath107 ; @xmath105 , @xmath106 ) where @xmath108 is the prediction for the @xmath70 history . the first three columns in table [ tab : strategies10 ] show this classification of strategies for general values of @xmath5 describing the @xmath96 states . for @xmath109 , the outcomes are no longer persistently ` @xmath4 ' , and the system explores both the @xmath70 and @xmath71 groups of histories . assume that for @xmath2 just below @xmath15 , the outcome must be ` @xmath4 ' for the @xmath71 histories , i.e. the system only visits the @xmath71 histories once in a cycle . consider the @xmath34 case , for example . as time evolves , the system exhibits periodic visits to the histories . in each period , each history in @xmath110 occurs once and the history in @xmath111 occurs @xmath112 times . among these @xmath112 occurrences of the @xmath70 history , the outcomes are ` @xmath4 ' for @xmath113 timesteps and ` @xmath3 ' for one timestep ( fig . [ fig : figure2](b ) ) , since the system must go from ` @xmath68 ' to ` @xmath103 ' after @xmath113 loops in order to sustain the path . it turns out that paths in the history space for the @xmath89 , @xmath36 , and @xmath35 states ( @xmath34 ) correspond to that shown in fig . [ fig : figure2](b ) with @xmath114 , @xmath65 , and @xmath44 loops at the @xmath70 history . since only @xmath99 different histories are involved , the performance of a strategy depends only on the predictions for this subset of histories . consider the path in fig . [ fig : figure2](b ) . the strategy labelled by @xmath115 predicts the correct outcome @xmath116 times for @xmath117 and @xmath118 times for @xmath119 , respectively , in going through the path once . in table [ tab : strategies10 ] , we list the performance of the strategies labelled by @xmath120 according to the number of successful predictions ( which reflects the vps ) in a closed path ( see fig . [ fig : figure2](b ) ) consisting of @xmath113 loops at @xmath70 history , i.e. a total of @xmath121 timesteps from @xmath122 to @xmath123 . it is important to note that there may be overlaps in strategies performance between the @xmath117 and @xmath119 groups of strategies for small values of @xmath113 , i.e. strategies with the label @xmath117 and @xmath119 may win the same number of timesteps in a cycle and hence belong to the same rank in performance of the strategies . for example for @xmath34 and @xmath114 , strategies labelled by @xmath124 and @xmath125 belong to the same rank in performance . the number of turns @xmath113 around the @xmath70 history which is consistent with the condition of @xmath126 , is restricted to the range @xmath127 . this criteria is related to the number @xmath128 of overlapping performances between the @xmath129 and @xmath130 groups of strategies ( see table [ tab : strategies10 ] ) . note that for @xmath131 , we have @xmath132 , i.e. strategies with @xmath117 do not have overlapping vps with strategies with @xmath119 . this implies that the vps of the strategies predicting ` @xmath4 ' for the history in @xmath133 , are always higher than those predicting ` @xmath3 ' . this further implies that agents will take action ` @xmath4 ' for the @xmath70 history if one of their strategies belongs to the @xmath117 category . for @xmath30 , there will then be @xmath15 agents taking the action ` @xmath4 ' . since @xmath134 , the outcome must be ` @xmath3 ' . therefore , the upper bound for the number of self - loops is @xmath99 , and thus @xmath135 . in other words , for paths with @xmath131 the system must have @xmath58 and the path in history space will be that of an infinite number of loops around @xmath70 history , i.e. the @xmath81 state . for @xmath136 , strategies in @xmath129 will perform worse than or equally well to those in @xmath130 for @xmath137 . as @xmath138 , strategies that predict more @xmath4 s should perform better . therefore , @xmath136 leads to inconsistency and thus @xmath139 . thus @xmath127 for the @xmath96 states , with the corresponding @xmath128 of overlapping groups of strategies being in the range @xmath140 . since each possible allowed value of @xmath113 or @xmath128 gives one @xmath96 state , there are altogether @xmath5 possible @xmath96 states for a given @xmath5 . the values of @xmath16 for the @xmath96 states can be readily found . for a given value of @xmath113 , the best performance among the strategies is to have @xmath141 correct predictions in a path consisting of ( @xmath142 ) timesteps with @xmath113 loops at the @xmath70 history . therefore , for the @xmath96 states @xmath143 for @xmath34 , we have @xmath144 and hence @xmath145 . there are three @xmath96 states with @xmath89 , @xmath36 , and @xmath35 , exactly as observed in the numerical simulations . we now derive the critical values of the resource level at which transitions occur from one value of @xmath16 to another for the @xmath96 states . note that the transition from the @xmath66 state to the @xmath146 state for @xmath34 and @xmath30 , occurs at @xmath147 as predicted in eq.(2 ) . the condition for transitions from @xmath69 to @xmath148 state is that the value of @xmath2 can no longer support @xmath149 loops at the @xmath133 history before giving an outcome of ` 0 ' for the history 111 . from table [ tab : strategies10 ] , we note that the performance of the @xmath117 group of strategies becomes increasingly better than the @xmath119 group as @xmath113 increases , and the number of agents taking the action ` 1 ' increases towards @xmath15 . therefore , the highest number of agents who take the action ` 1 ' _ and _ win will arise at the last turn among the @xmath113 loops where the history 111 is followed by an outcome ` 1 ' . similarly , the lowest number of winning agents will be at the turn when the history 111 is followed by an outcome ` 0 ' , i.e. breaking away from the @xmath113 loops at 111 . the number of agents choosing the action ` 1 ' given the history 111 , is related to the number @xmath128 ( and hence @xmath113 ) of overlapping performances among the @xmath129 and @xmath130 strategies . from table [ tab : strategies10 ] , the number of correct predictions @xmath150 of a strategy @xmath120 in going through a path with @xmath113 loops at the @xmath70 history , is given by @xmath151 for general values of @xmath5 , there are only @xmath152 histories which matter for the @xmath96 states . therefore there are @xmath153 _ effectively different _ strategies , each of which represents a group of @xmath154 strategies . the number of effectively different strategies predicting @xmath107 for the @xmath70 history and having @xmath105 predictions of ` 1 ' for the @xmath5 @xmath71 histories , is given by @xmath155 for both @xmath156 and @xmath3 , where @xmath157 is the binomial coefficient . the performance of the strategies can be ranked by a label @xmath158 , with @xmath159 and @xmath160 representing the best performing group of strategies . a general situation in which there are @xmath128 overlapping performances between the @xmath129 strategies and @xmath130 strategies , is shown in table [ tab : vprank10 ] . for the @xmath96 states , the best performing strategies belong to the @xmath117 group and the worst performing ones belong to the @xmath119 group , with @xmath128 overlapping rankings in between where the allowed range of @xmath128 is @xmath140 . the ranking @xmath158 of the strategies @xmath120 is related to @xmath105 by the simple relation @xmath161 for a given value of @xmath128 , there are a total of @xmath162 ranks . therefore , the label @xmath158 is restricted to the range @xmath163 with @xmath164 given by @xmath165 as the number of loops @xmath113 increases , @xmath128 decreases and the strategies spread more widely in terms of performance . it follows from eqs . and that the number of effectively different strategies @xmath166 in rank-@xmath158 is given by @xmath167 the fraction of rank-@xmath158 strategies among all the strategies is then given by @xmath168 on the critical turn that determines the minimum value of @xmath2 for sustaining a certain @xmath96 state , only the @xmath117 strategies win . using table [ tab : vprank10 ] together with eqs . and , the number of winning strategies @xmath169 is @xmath170 the corresponding fraction @xmath171 of winning strategies among all strategies of rank - r is given by @xmath172 each agent uses the strategy in his possession which has the best performance record , i.e. the one having the ranking with smaller @xmath158 , in order to make a decision . assuming a uniform initial distribution of any combination of @xmath25 strategies ( with repetitions allowed ) among the agents , the fraction of agents _ holding _ a rank-@xmath158 strategy as their best performing strategy @xmath173 is @xmath174^s -\bigg[\sum_{r'=r+1}^{r_{max}}\widetilde{c}(r')\bigg]^s.\end{aligned}\ ] ] with @xmath175 given by eq .. each @xmath96 state corresponds to a specific value of allowed @xmath113 and hence an allowed @xmath128 . for a given @xmath113 or @xmath128 , the resource level @xmath2 needed to accommodate all the agents that take the action ` 1 ' for the @xmath70 history , gives the criterion for the state : @xmath176 note that @xmath164 , @xmath177 , and @xmath173 are all @xmath128-dependent ( see eqs . , , and ) . equation gives the lower bounds of @xmath2 for each of the @xmath96 states . note that the lower bound for a state with a given @xmath128 is also the upper bound for the state with @xmath178 . for @xmath33 , @xmath30 , and @xmath179 , @xmath180 , and @xmath65 , gives the lower bounds of @xmath181 , @xmath182 , and @xmath183 for the states characterized by @xmath146 , @xmath36 , and @xmath37 , respectively . these values are in excellent agreement with those obtained by numerical simulations ( see fig . [ fig : figure1](a ) and table [ tab : otsm3s2 ] ) . we note that our approach of focusing on strategy - performance ranking patterns and the fraction of strategies in each rank , represents a generalization of a similar approach @xcite that has already been successfully applied to the mg to cases in which some of the strategies have runaway vps . we have studied numerically and analytically the effects of a varying resource level @xmath2 on the success rate of the agents in a competing population within the b - a - r model . we found that the system passes through different states , characterized either by the mean success rate @xmath27 or by the highest success rate in the population @xmath16 , as @xmath2 decreases from the high resource level limit . the number of states depends on details of the system such as the memory size @xmath5 and the number of strategies per agent @xmath25 . transitions between these states occur at specific values of the resource level . for small values of @xmath5 , it is possible to explain these states by following the evolution of the performance of the strategies and the decision making dynamics of the system . more generally , we found that different states correspond to different paths covering a subspace within the whole history space . in the high resource level regime , namely @xmath184 , @xmath66 . the corresponding path in the history space is one that loops around the history ` 111 ... ' indefinitely . just below the high resource level regime is a range of @xmath2 that gives @xmath5 states corresponding to the fractions @xmath185 . this result is in excellent agreement with that obtained by numerical simulations . for these @xmath96 states , i.e. the outcome series consists of one bit of 0 in a cycle of @xmath142 bits , the path in history space is restricted to those @xmath5-bit histories with at most one - bit of ` 0 ' and with @xmath113 loops around the ` 111 ... ' history . this identification of an active portion within the history space implies that only part of each strategy is being used . by considering the performance of the strategies within this restricted portion of the history space , the number of loops @xmath113 consistent with the @xmath96 states was found to be @xmath127 . the number of agents using a strategy that predicts the action ` 1 ' given the history ` 111 ... ' increases as the number of loops @xmath113 increases . thus a criterion on the resource level for sustaining a state of given @xmath113 can be derived . the results are again in excellent agreement with numerical results . after passing through the @xmath96 states , the system goes into states with more than one ` 0'-bit per cycle in the outcome series as @xmath2 is further reduced . these @xmath186 states correspond to paths that explore an increasingly larger portion of the history space . while our analysis can also be applied to these @xmath186 states , the dynamics and the results are too complicated to be included here . a resource level @xmath2 that deviates from @xmath13 acts like a _ driving force _ in the history space . in response to this driving force , the system effectively adjusts its dynamics to occupy an increasingly restricted portion of the history space as @xmath2 increases . for @xmath12 , the system explores the whole history space by passing through trails that are almost eulerian @xcite . the random initial distribution of strategies and the random initial history bit - string that started the system , provide the seed for the diffusive behavior which develops in the history space as the system evolves . in fact , results of numerical simulations for @xmath187 show that slightly increasing @xmath2 beyond @xmath13 has the effect of suppressing this random wandering through the history space , and instead locks the system into the eulerian trail . however opposing mechanisms can arise to counteract this driving force , thereby enhancing the diffusive behavior . for example , this can be achieved by allowing the agents the chance of using a strategy besides the best - performing one @xcite or by allowing some agents to opt out of the system occasionally @xcite . alternatively , the system can be biased through the initial strategy scores , or by introducing a specially prepared non - random initial allocation of strategies @xcite . it is this competition between the diffusive and driven behavior that gives the non - trivial global behavior in the b - a - r model and its variations . for this reason , the present b - a - r system provides a fascinating laboratory for studying correlated , non - markovian diffusion on a non - trivial network ( i.e. history space ) . this non - trivial diffusion is in turn strongly coupled to the non - random temporal patterns arising in the vp dynamics . we note that the present results could also be used to generalize the crowd - anticrowd theory in order to incorporate the effect of restricted history - space dynamics : this would then allow identification of an appropriate set of correlated , uncorrelated and anti - correlated strategies in order to implement the crowd - anticrowd theoretical expressions . as a side - product , our analysis serves to illustrate the sensitivity within multi - agent models of competing populations , to tunable parameters . by tuning an external parameter , which we take as the resource level in the present work , the system is driven through different paths in the history space which can be regarded as a ` phase space ' of the system . the feedback mechanism , which is built - in through the decision making process and the evaluation of the performance of the strategies , makes the system highly sensitive to the resource level in terms of which states the system decides to settle in or around . these features are quite generally found in a wide range of complex systems . the ideas in the analysis carried out in the present work , while specific to the b - a - r model used , are also applicable to other models of complex systems . in closing , we remark that besides obtaining analytically the highest success rate @xmath16 and the criteria on the resource level @xmath2 , our treatment can also be extended to obtain the mean success rate @xmath27 for the @xmath96 states . the analysis is more complicated than that for obtaining @xmath16 . the procedure is to follow the evolution of the performance of the groups of strategies in _ each timestep _ through a path in the history space . the number of agents taking a particular action , and hence the number of winning agents , can be found from a strategies performance table like the one shown in table [ tab : strategies10 ] . we have carried out the analysis for @xmath27 for @xmath40 , @xmath180 , and @xmath65 , and results are found to be in excellent agreement with numerical results . our analysis can also be readily extended to consider connected populations in which agents have established links to connected neighbors for collecting additional information @xcite . quite generally , the effect of the links is to modify the number of agents _ using _ a strategy in a particular rank . for connected populations , an agent may use a strategy that he does not hold but has access to through his links . for an agent who uses the best performing strategy among his own @xmath25 strategies and those of his connected neighbors , the success rate behaves in a similar fashion as a function of @xmath2 as that reported here , only that the critical values of resource level at which transitions occur are shifted @xcite . these results can be understood by incorporating the effects of the linkages into eq .. results for the b - a - r model in a connected population will be reported elsewhere @xcite . this work was supported in part by the research grants council of the hong kong sar government through grant no . we acknowledge useful discussions with ho - yin lee , keven k. p. chan , charley s. choe and sean gourley . .table showing the states characterized by @xmath16 for b - a - r model with @xmath33 agents , @xmath34 and @xmath30 , together with the ratio of number of occurrences of ` 1'-bits to ` 0'-bits in the outcome bit - string and the range of resource level in which the state occurs . the results are obtained from the numerical data as shown in fig . [ fig : figure1](a ) . the last column shows the period observed in the outcome bit - string . [ cols="^,^,^,^",options="header " , ] ( dark solid line ) as a function of the resource level @xmath2 for a system with @xmath33 agents , @xmath30 strategies per agent , with memory length ( a ) @xmath34 and ( b ) @xmath40 . for each value of @xmath5 , data for different values of @xmath2 are taken in a system with the same initial distribution of strategies among the agents . also shown are the range corresponding to one standard deviation about @xmath27 in the success rates among the @xmath0 agents ( dotted lines ) and the spread in the success rates given by the highest and the lowest success rates ( thin solid lines ) among the agents . ] . the nodes correspond to the @xmath9 possible histories . the transition between nodes are indicated by the arrows , together with the outcome needed for the transitions to occur . the histories can be grouped into columns labelled by a parameter @xmath64 which gives the number of ` 0 ' bits in the histories . ( b ) for the @xmath96 states at @xmath188 , the system follows a path restricted to histories in the @xmath70 and @xmath71 columns , with @xmath113 loops at the @xmath70 history . ]

two unforeseen developments in recent years have made continued photometry of particularly important : distinctive brightness variations in the near ir accompany its mysterious 5.5-year spectroscopic cycle @xcite and , possibly independent of the cycle @xcite , the star brightened at a surprising rate after 1997 @xcite . neither of these phenomena has been explained and the data since 1998 are seriously incomplete . meanwhile , the long - term brightening trend continues ; for general information about @xmath0 car with many references , see @xcite , and @xcite . photometry of this bright object is difficult for at least two reasons : + 1 . at visual wavelengths , normal ground - based observations represent mainly the surrounding `` homunculus '' ejecta - nebula , which appears much brighter than the central star and has structure at all radii from 0.2 to 8 arcseconds . so far the only available measurements of just the central star have been made with the hubble space telescope ( hst ) . although the homunculus is primarily a reflection nebula , its apparent brightness measured with respect to the central star has changed greatly . during 1998 - 99 , while ground - based observations showed about a 0.3-magnitude brightening of homunculus plus star , the star itself nearly tripled in apparent brightness @xcite ! this discrepancy presumably involves dust along our line of sight to the star , but , as mentioned above , it is only vaguely understood at this time . 2 . numerous strong emission lines produced in the stellar wind perturb the results for standard photometric systems . h@xmath1 and h@xmath2 emission , for example , have equivalent widths of about 800 and 180 respectively in hst spectra of @xmath0 car . therefore , broad - band @xmath3 , @xmath4 , @xmath5 , and @xmath6 magnitudes , and most medium - band systems as well , are poorly defined for this object . photometry around 5500 , e.g. broad - band @xmath7 , is relatively free of strong emission lines , but transformations from instrumental magnitudes to a standard system are imprecise because they involve the other filters ( @xcite , @xcite , @xcite , and references therein ) . in this paper we report photometry _ of the central star _ obtained with two hst instruments since 1998 . for reasons noted above , these observations are quite distinct from all ground - based data ; and the star s photometric behavior during 2003 proved to be especially interesting . first we employ acquisition images produced by the space telescope imaging spectrograph ( stis ) , because these are numerous and internally reliable ; @xmath0 car has been observed many times with this instrument since 1998 , especially during mid-2003 when the most recent `` spectroscopic event '' occurred . these broad - band red data can not be transformed to any standard photometric system , but they are quite reliable for showing relative fluctuations by the star ( strictly speaking , the stellar wind ) , uncontaminated by light from the surrounding homunculus . we also report surprising variations in the star s h@xmath1 emission brightness accompanied by measurements of the nearby continuum in the stis slit - spectroscopy data . we supplement the stis data with similar measurements of a few images obtained in 200203 with hst s advanced camera for surveys ( acs ) . these represent a different set of wavelength bands . altogether these stis and acs results probably represent most of the photometry of @xmath0 car that will ever be obtained with the hst . very few additional stis observations are expected now that the 2003 event has passed . earlier hst / fos , foc , wfc - pc , and wfpc2 data were not obtained often enough to be suitable for our purposes , and in most cases would be more difficult to compare photometrically , for technical reasons . we report the stis and acs data now , rather than waiting for two or three more data points in 2004 , because the behavior during 2003 was rather unexpected and therefore deserves to be noted promptly . in addition , we briefly present a convenient summary of @xmath0 car s photometric record in the aavso archives . these ` v ' magnitudes represent , of course , the entire homunculus plus star but they are valuable for comparisons with the hst data and also with ground - based photometry using different filter systems , reported by , e.g. , @xcite , @xcite , @xcite , @xcite , @xcite , and @xcite . in section 2 we present the photometry of the central star based on hst / stis acquisition images and the acs / hrc observations . this is followed by a section on the ground - based data from the aavso and then a general discussion about how these results relate to the models which have been presented to describe @xmath0 carinae . normal ground - based imaging and photometry can not resolve the central star and small structures in the homunculus nebula . in this study , we have used the hst / stis acquisition images which have a resolution of about 0.1 arcseconds plus the recent acs / hrc images which have a resolution of 0.05 arcseconds to separate the contribution of central star from the surrounding nebulously . each set of stis observations of @xmath0 carinae has included a pair of acquisition images . these images are 100x100 pixel sub - frames ( 5 arcseconds square ) centered on the middle row and column of the ccd @xcite . each pair of acquisition images includes an initial targeting image and a post - acquisition image . only the post - acquisition images were used to measure the brightness of the central star because the target is rarely well centered in the initial targeting images , which raises concerns about ghost images and internal reflections that may unpredictably affect them ( see section 2.1.2 ) . the majority of the stis acquisition images of @xmath0 carinae have been taken using the f25nd3 filter , which is a neutral density filter that covers the wavelengths from 2000 to 11000 . acquisition images using other filters are ignored since they lack significant temporal baseline or coverage and there is little hope of transforming between filter functions for an object as peculiar as @xmath0 carinae . the f25nd3 filter includes several prominent features in the spectrum of @xmath0 carinae ( figure 1 ) and changes in them contribute to any measured brightness variations . we processed the stis acquisition images using the iraf procedure stas.hst_calib.stis.basic2d with version 3.0 of stdas and version 2.13 of calstis . en lieu of a bias over - scan region , each stis acquisition image has the uniform value of 1510 counts subtracted from each pixel as an approximate bias level correction for the target acquisition process . a small uncorrected bias level remains in the acquisition images after this initial correction , so we compensate for this tiny effect by modeling it as a function of time and ccd housing temperature . this residual bias was measured in acquisition images of the standard star after the normal bias correction and dark current were subtracted . the exposures of agk + 81 266 are short ( 2.1 seconds ) and it is more than 60 degrees above the ecliptic , so zodiacal light and other sources make a negligible contribution to the background @xcite . we found that the residual bias level in the images of agk + 81 266 ( figure 2 ) is roughly linear with respect to time prior to the failure of the ccd side-1 controller on may 16 , 2001 @xcite . there may be higher order terms in this relation however they are not significant in this study since they effect the measurements of the central star by less than 0.001 magnitude . subsequent to the failure of the side-1 controller the stis ccd temperature regulator has been held at a constant voltage permitting the ccd temperature to fluctuate . as a result , after the side-1 failure the residual bias level of the observations depends on both time and ccd housing temperature . using the measurements of the residual bias level in the agk + 81 266 observations , the following relations were obtained for the bias level correction as a function of modified julian date ( mjd ) and ccd housing temperature ( t ) : + mjd @xmath8 52045.5 : @xmath9 mjd @xmath10 52045.5 : @xmath11 when the modeled bias level is subtracted from the agk + 81 266 data the background levels average to @xmath12 counts / pixel / sec ( figure 2 ) . this correction was applied to the acquisition images for @xmath0 carinae in place of the bias over - scan correction normally implemented in the iraf basic2d procedure . typically the bias correction amounts to less than 0.005 magnitude for the measurements reported below . the flux from the central star was measured in each image by summing detector counts within a small circular virtual aperture . in order to avoid pixelization effects , the counts were weighted by a function of the distance of a pixel from the aperture center ( @xmath13 ) which decreases smoothly to zero at the aperture radius @xmath5 : + @xmath14 @xmath15 the centering on the star is determined by a peak - up algorithm which locates the position where the sum inside the virtual aperture is maximized . for the stis acquisition images we used @xmath16 pixels which corresponds to a diameter of 0.3 arcseconds and includes slightly fewer than 30 pixels . we did not attempt to subtract a background from the measured aperture flux , but this has little effect on the results since the star provides thousands of counts / pixel / second , far greater than any other contribution within the measuring aperture due to scattered and zodiacal light which contribute only about 0.2 counts / pixel / second or less @xcite . there is no feasible way to account for or remove the contribution from diffuse circumstellar material within 0.1 arcseconds of the star . however , stis spectroscopic data indicate the level of contamination from this source is much smaller than the other uncertainties . ghost images exist in the stis optics but do not contribute significantly to our measurements . when a bright source is centered on the ccd , a ghost image with an peak flux of a few percent relative to the psf maximum appears approximately 0.3 arcseconds ( 6 pixels ) to the right ( increasing column number ) of the peak of the point spread function . the 0.3 arcsecond diameter aperture used to measure the central star excludes this ghost . however , any future work that may attempt to measure the flux from the nebula immediately surrounding the central star in these images must confront this problem . the reduction procedure was tested using 80 pairs of acquisition images of the flux standard agk + 81 266 ( figure 3 ) . the measured aperture photometry for this standard shows a constant level with an r.m.s scatter of @xmath17 magnitudes . it should be noted that agk + 81 266 was observed using the f28x50lp `` long - pass '' filter rather than the f25nd3 filter which was used for the @xmath0 carinae observations . at the same time , the gross spectral energy distribution of agk + 81 266 ( spectral class b5 ) and @xmath0 carinae are very dissimilar . this is not significant since agk + 81 266 is only used to demonstrate that this method accurately measures a constant brightness over the temporal baseline for an established flux standard . the brightness of @xmath0 carinae is _ not _ measured relative to agk + 81 266 . in figure 4 and table 2 we present our results for the stis acquisition images as instrumental magnitudes calibrated for long term sensitivity effects using the photometric constants calculated by the reduction pipeline @xcite and scaled relative to the value on 1999 february 20 ( 1999.14 ) . these represent an extremely broad - band wavelength sample ( figure 1 ) . for the purpose of comparison with familiar photometric systems , table 1 gives approximate johnson @xmath7 @xcite and cousins @xmath5 and @xmath6 @xcite magnitudes at the reference epoch ( 1999.14 ) . these were calculated using flux - calibrated stis spectroscopic data @xcite , a flux - calibrated spectrum of vega @xcite , and the appropriate response functions @xcite . for reasons noted in section 1 and uncertainties in the stis slit throughput for an object which is not exactly a point source , entries for @xmath0 carinae in table 1 probably have uncertainties of roughly @xmath18 magnitudes and could possibly be somewhat worse . it is also possible that the photometric color may trend mildly blue - ward as the object brightens @xcite . the homunculus is currently of 5th magnitude according to _ ground - based _ johnson @xmath7 photometry ( section 3 below ) , but the central star is fainter than @xmath7 = 7.0 as stated in table 1 . its apparent continuum is red , while stellar - wind emission lines contribute a significant fraction of its brightness . in the 1999.14 spectrum , h@xmath1 and other emission features contribute about 16% of the flux in @xmath7 ( 4% from h@xmath1 ) and 38% of the flux in @xmath5 ( 35% from h@xmath1 ) . meanwhile , there is about a 5% affect on the @xmath6 flux from the hydrogen paschen lines which are in emission . because these emission features are constantly changing , it is not possible to formulate a consistent set of transformations to convert the stis magnitudes to more familar johnson @xmath7 and cousins @xmath5 and @xmath6 magnitudes for @xmath0 carinae . however , the values presented for 1999.14 can serve as a rough zero point . figure 4 shows that the dramatic brightening reported by @xcite leveled off by 2000 , but may have resumed at least temporarily around the mid-2003 spectroscopic event . the dip in brightness near the time of the event ( figure 6 ) may be reminiscent of the near - infrared behavior reported by @xcite . this will be discussed further in section 4 . eta carinae s broad h@xmath1 emission , produced in the stellar wind at radii of several au , is tremendously strong ( equivalent width usually about 800 ) . this feature did not vary much relative to the continuum in stis data from 1998 to 2002 , but recently it has faded to a surprising degree as reported below . since h@xmath1 contributes appreciably to the stis acquisition images , we have investigated the effects on our photometry . davidson ( 2003 ) has measured the equivalent width of h@xmath1 and estimated the absolute level of the nearby continuum in short - exposure stis slit data . h@xmath1 was integrated from 6509 to 6650 , and the continuum flux was averaged between 6740 and 6800 ( an interval that is practically uncontaminated by emission lines ) . these continuum measurements have the advantage of being approximately calibrated in physical units , but on the other hand may be underestimated if the slit was not perfectly centered on the star . fortunately , the h@xmath1 equivalent width measurements are quite insensitive to slit position . the results are shown in figure 5 . we estimate that h@xmath1 contributes from 8% to 16% of the stis / f25nd3 counts , varying with time as listed in table 2 . open triangles in figure 4 represent the instrumental magnitudes that would have been observed if h@xmath1 were not present ; evidently the brightness minimum in early 2003 and the subsequent rapid brightening are not due to this emission line . the measured brightness fluctuations in the central star appear to be mainly ( though not entirely ) changes in the continuum brightness . in the six months preceding the 2003.5 event ( december 2002 to june 2003 ) , the equivalent width of h@xmath1 declined by a factor of nearly 2 , while its line profile evolved in an interesting and unexpected way which is beyond the scope of this paper . the other balmer lines behaved in a similar fashion . moreover , in july and september 2003 they did not return to the state observed by stis in 1998 at the corresponding point in @xmath0 car s 5.54-year cycle . a more in - depth account of these effects will be reported in a paper which is now in preparation @xcite . hst acs / hrs observations of @xmath0 car were obtained for the hst treasury project beginning in october 2002 . bias corrected , dark subtracted , and flat fielded acs / hrc images were obtained from space telescope science institute via the multi - mission archive ( mast ) . observations which had overexposed the central star were not used and no drizzle or dithering correction is applied . the acs / hrc images were taken in four filters which cover near uv and optical wavelengths ( figure 1 ) . each of these filters is influenced by different types of spectral features . the uv filters ( hrc / f220w & hrc / f250w ) are heavily influenced by the `` fe ii forest '' @xcite . the opacity of this `` forest '' is known to dramatically increase during an event @xcite . the hrc / f330w filter is sensitive to both the balmer continuum and a number of emission features while the hrc / f550 m filter , strategically placed between h@xmath1 and h@xmath2 , almost exclusively measures the continuum . the relative brightness of the central star in the acs / hrc images is measured with the same weighted 0.3 arcsecond diameter aperture used to measure the stis acquisition images . on the scale of the acs / hrc this corresponds to @xmath19 pixels and includes about 79 pixels inside the aperture . the results are given in table 3 and figure 6 . the magnitudes given are on the stmag system @xcite as calibrated by the photometric keywords calculated by the reduction pipeline @xcite . note that the hrc / f220w and hrc / f250w curves show a significant dip at the time of the 2003.5 event , due to the absorption as noted above . interestingly , the visual continuum measured by the hrc / f550 m filter does _ not _ show the decrease in brightness seen at longer wavelengths in the stis data . while space - based photometry is able to resolve the central star , ground based photometry measures the integrated brightness of both the star and the homunculus nebula . however , ground - based photometry is still valuable because it has a longer temporal baseline and shows many of the same brightness variations , albeit at different amplitudes . also as more publicly funded small telescopes are closed the data from the aavso and groups like la plata observatory @xcite will become increasingly useful for monitoring objects like @xmath0 car . one of the few sources for long term monitoring of @xmath0 carinae is the american association of variable star observers ( aavso ) @xcite . from 1968 through mid-2003 there are 6640 observations from 110 observers situated around the globe . these data are from individual observers who have determined the brightness of @xmath0 carinae by visually comparing it with the brightness of other stars . ten observers account for 69% of these observations while 36% of the observations are from the two most prolific observers : m. daniel overbeek @xcite and albert f. jones . on average , there are 60 observations per observer but the median is 9 observations per observer . we averaged the observations in 90 day bins with a median of 32 observations in each bin . the bottom panel of figure 7 shows binned averages from 1968 through mid 2003 with a typical standard deviation of 0.2 magnitudes , which is consistent with the accepted usual r.m.s . error of 0.1 0.2 magnitudes for observations of this type . the top panel of figure 7 shows the photoelectric johnson @xmath7 band photometry ( pepv ) from stan walker @xcite , photoelectric observations from r. winfield jones obtained through the aavso , and the average @xmath20 magnitude from ground based observations reported in figure 2 of @xcite . the data in both panels follow the same general trend and compare well with the data presented in figure 2 of @xcite . even the most primitive aavso data track the professional photometry well enough to show irregular behavior . the maximum noted by @xcite at 1982.11 is visible in all the data , though the errors in the binned aavso data make it harder to recognize . each dataset also shows an overall brightening trend in @xmath0 carinae over the past thirty years along with a notable jump in brightness in 1998 - 1999 @xcite . the binned aavso data for the last five years ( figure 8) also follows a trend similar to the brightness variations in the central star measured from the stis acquisition images ( figure 4 ) . of particular note , both datasets show the brightening of @xmath0 carinae peaked in 2000 , fell off a bit , and then peaked again in early 2002 before brightening through the 2003.5 event . the data presented here from the hst and ground - based observations confirm the dramatic brightening in both the nebula and central star of @xmath0 carinae which was previously reported by @xcite and @xcite along with a more gradual brightening spanning the past thirty years . all the available data also show that the apparent brightness of @xmath0 carinae fluctuates with no obvious period . we also see intervals ( on the order of months ) with short lived increases in apparent brightness similar to those reported by @xcite . the 2003.5 event , as observed in the brightness fluctuations of the central star at visual wavelengths , was characterized by a roughly six month rise , then a sharp dip in most filters , followed by a recovery to the highest flux observed for the central star in recent years , which may be the start of another significant brightening episode . again we emphasize that since 1997 _ the central star has brightened substantially faster than the homunculus _ ( compare figures 4 and 7 ) . the stis , acs , h@xmath1 , and aavso data all have notable trends which began about 200 days before the 2003.54 event . the visual band photometry from the stis , acs , and aavso began to brighten significantly in january 2003 ( figures 9 & 6 and also @xcite ) which coincided with a steep drop in h@xmath1 luminosity ( figure 5 ) and the beginning of a dip in brightness in the near - uv ( figure 6 ) . the start time for this pre - event activity is important because most models which invoke a binary companion to trigger the events place the secondary on a very eccentric orbit where it is in close proximity to the primary for only a very brief period of time during each cycle . @xcite showed that the @xmath21 , @xmath22 , @xmath23 and @xmath24 bands brightened roughly 500 days before the time of the events over the past two decades with an additional sharp rise in the @xmath21 , @xmath22 , and @xmath23 bands about 80 days prior to the 1998.0 event . the x - ray flux from @xmath0 car ( @xcite and m. corcoran s web page for data on the recent event ) also increased before the 1998.0 and 2003.54 events , preceding them by about 350 and 700 days respectively . it is not necessary or expected that all of the pre - event activity will temporally coincide since each of these data measure activity which occurs at different radii from the star and a shock front which pushes deeper as the event proceeds will not affect each process simultaneously . also , as noted below there is compelling evidence that the state and distribution of the material immediately surrounding the central star has evolved over time so while the activity prior to the event should maintain roughly the same sequential order , the exact start times relative to the event could potentially change from cycle to cycle . the acs / hrc data show that the brief dip in apparent brightness during the 2003.5 event is not achromatic . the acs data show that the depth of the dip is wavelength dependent ; the dip is deeper at shorter wavelengths and the brightness continually increased in the hrc / f550 m filter through the last data point . however , @xcite showed that the brief decrease or dip in luminosity occurred at all of the near - ir wavelengths during the 1998.0 event ( figure 9 , reproduced from @xcite ) and @xcite reported similar activity during the 2003.5 event . this leaves two possibilities which are not mutually exclusive : 1 . the decrease in brightness occurs later at redder wavelengths and/or 2 . the decrease in brightness is driven entirely by increased line opacities , decreased emission lines , and balmer and paschen continua formed in the wind , rather than a change in the star s continuum flux . @xcite show that during the 1998.0 event , the dip in the @xmath21 band occurred about 35 days later than the dip in the x - rays . they also note that the epoch of the minimum in the dip increases slightly with increasing effective wavelength ( figure 9 ) . @xcite reported that during the 2003.5 event the dip in the @xmath21 and @xmath24 bands began sometime between mjd 52809.5 ( 2003.467 ) and 52814.5 ( 2003.481 ) , which preceded the observed dip in johnson - cousins @xmath25 by about one to two weeks @xcite . ( note that wavelengths around 2 @xmath26 m observed by @xcite presumably represent free - free emission in the wind at radii substantially larger than the photosphere at visual wavelengths . ) given only the hst data , one might suspect that dip in the hrc / f550 m filter may have occurred between data points and the slight rise observed at mjd 52803.1 ( 2003.45 ) is the rise in luminosity which precedes the dip . however , the hrc/550 m observation at mjd 52840.1 ( 2003.55 ) shows a _ rise _ in flux roughly coincident with the minimum observed in @xmath25 by @xcite and @xcite . therefore , it is unlikely that we missed the dip in hrc/550 m due to sampling . likewise , strong p - cygni absorption is present in the hydrogen lines during the event @xcite . since the hydrogen emission lines contribute significantly to the total brightness of the central star , anything that changes their profiles will change the brightness measured in a filter which they influence . the hrc / f550 m filter covers none of the prominent emission lines and it shows no dip concurrent with the dip in the other filters . therefore , this may indicate that the bolometric luminosity of the central star is largely unaffected by the mechanism which causes the observed decrease in brightness . @xcite discuss several possible explanations for the variability in the brightness of the central star of @xmath0 carinae . since the star itself radiates near the eddington limit , there is little room for rapid or dramatic changes in bolometric luminosity . this is supported by the fact that the surrounding nebula , which acts as a giant calorimeter for the central star , varies in brightness on a significantly smaller scale . if the bolometric luminosity of the central star stays roughly constant , then significant variations in its brightness may be more appropriately interpreted as changes in the density , temperature , and spatial distribution of intervening circumstellar material . an expanding shell of ejecta from an event may be able to produce the observed light curves @xcite . any ejected shell should start hot with high opacities from ionized metals and then later shift these opacities to the red as the shell cools and dust forms . particularly at early stages , this shell may be optically thin and as such would not have a significant impact on the continuum brightness . the data show that central star of @xmath0 carinae continues to brighten . we are unsure how long this trend can continue but it appears to be driven by an increase in the apparent continuum brightness which may be caused by changes in the intervening circumstellar material along the line of sight , although how or why the circumstellar material may have been been altered is not completely understood . continued monitoring is critical since coming out of the 2003.5 event there are indications that the central star may undergo another period of sudden dramatic brightening in the near future . the brief decrease in the star s brightness during the 2003.5 event was more complicated than expected and changes preceding the event in the optical and near - uv started as early as january 2003 . the wavelength dependence of the dip indicates that it can not be caused by a simple eclipse or occultation . if the primary star is eclipsed by the presumed secondary we would expect to observe some decrease at all wavelengths including the hrc/550 m band which measures the continuum . on the other hand if an eclipse of the secondary is sufficient to cause a dip of approximately 0.5 mag in the near uv flux of the central star , then we would expect to find some spectroscopic evidence for the secondary in the hst / stis spectra during the 5.5 year cycle which is not the case . however , the data may support a shell ejection model for the spectroscopic events . additional data , including spectra gathered by the stis during the 2003.5 event , should further constrain the mechanism powering these periodic phenomena . this research was conducted as part of the hst treasury project on @xmath0 carinae via grant no . go-9420 from the space telescope science institute . we acknowledge kris davidson , roberta m. humphreys , and kazunori ishibashi for directly assisting with the data and measurements and t.r . gull for preparing most of the hst observing plans . other participants in the hst treasury project are d. john hillier , augusto damineli , michael corcoran , otmar stahl , kerstin weis , sveneric johansson , fred hamann , h. hartman , nolan walborn , and manuel bautista . we are also grateful to the aavso international database for the variable star observations contributed by observers worldwide . we individually thank janet a. mattei and elizabeth o. waagen of the aavso who helped us obtain the data and stan walker and r. winfield jones who provided photoelectric photometry data . we are very appreciative of m. feast , and p. whitelock for allowing us to reproduce their figures . and we thank matt gray and qian an for providing the @xmath0 carinae project with computer technical support . altamore , a. , baratta , g. b. , cassatella , a. , rossi , l. , & viotti , r. 1986 , new insights in astrophysics . eight years of uv astronomy with iue , 303 bingham , r. g. & cousins , a. w. j. 1974 , monthly notes of the astronomical society of south africa , 33 , 15 bohlin , r. c. , dickinson , m. e. , & calzetti , d. 2001 , , 122 , 2118 cassatella , a. , giangrande , a. , & viotti , r. 1979 , , 71 , l9 clampin , m. , hartig , g. , baum , s. , kraemer , s. kinney , e. kutina , r. , pitts , r. , & balzano , v./ 1996 , stis instrument science report 96 - 030 , ( baltimore , stsci ) colina , l. , bohlin , r. , & castelli , f. 1996 , instrument science report cal / scs-008 , ( baltimore , stsci ) cousins , a. w. j. 1976 , mem . soc . , 81 , 25 damineli , a. 1996 , , 460 , l49 davidson , k. & humphreys , r. m. 1997 , , 35 , 1 davidson , k. , humphreys , r. m. , ishibashi , k. , gull , t. r. , hamuy , m. , berdnikov , l. , & whitelock , p. 1999a , , 7146 , 1 davidson , k. et al . 1999b , , 118 , 1777 davidson , k. , ishibashi , k. , gull , t. r. , & humphreys , r. m. 1999b , asp conf . ser . 179 : eta carinae at the millennium , 227 davidson , k. 2000 , cosmic explosions : tenth astrophysics conference . aip conference proceedings , vol . 522 . college park , maryland , 11 - 13 oct 1999 . edited by stephen s. holt and william w. zhang . american institute of physics , 2000 . , p.421 - 430 , 421 davidson , k. et al . 2003 , personal communication davidson , k. et al . 2004 , , in preparation davis , m. , campbell , d. , sticka , r. , faful , b. leidecker , h. , kimbel , r. , & goudfrooij , p. 2001 , stis failure review board final report ( baltimore : stsci ) downes , r. , clampin , m. , shaw , r. , baum , s. , kinney , e. & mcgrath , m. 1997,stis instrument science report 97 - 03b , ( baltimore , stsci ) feast , m. , whitelock , p. , & marang , f. 2001 , , 322 , 741 fernandez lajus , e. , gamen , r. , schwartz , m. , salerno , n. , llinares , c. , farina , c. , amorn , r. , & niemela , v. 2003 , informational bulletin on variable stars , 5477 , 1 gull , t. r. , davidson , k. , & ishibashi , k. 2000 , cosmic explosions : tenth astrophysics conference . aip conference proceedings , vol . 522 . college park , maryland , 11 - 13 oct 1999 . edited by stephen s. holt and william w. zhang . american institute of physics , 2000 . , p.439 - 442 , 439 holtzman , j. a. , burrows , c. j. , casertano , s. , hester , j. j. , trauger , j. t. , watson , a. m. , & worthey , g. 1995 , , 107 , 1065 ishibashi , k. , davidson , m. f. , corcoran , k. , drake , s. a. , swank , j. h. , & petre , r. 1999 , asp conf . ser . 179 : eta carinae at the millennium , 266 johnson , h. l. & morgan , w. w. 1953 , , 117 , 313 johnson , h. l. & morgan , w. w. 1951 , , 114 , 522 mattei , janet & fraser , brian 2003 , jaavso , 31 , 65 mattei , janet & foster , g. 1998 , iappp communications , 72 , 53 pavlovsky , c. , et al . 2003 , acs instrument handbook , version 4.0 , ( baltimore : stsci ) proffitt , c. r. et al . 2002 , the 2002 hst calibration workshop : hubble after the installation of the acs and the nicmos cooling system , proceedings of a workshop held at the space telescope science institute , baltimore , maryland , october 17 and 18 , 2002 . edited by santiago arribas , anton koekemoer , and brad whitmore . baltimore , md : space telescope science institute , 2002 . , p.97 , 97 kim quijano , j. , et al . 2003 , stis instrument handbook , version 7.0 , ( baltimore : stsci ) smith , e. , stys , d. , walborn , n. , & bohlin , r. 2000 , instrument science report stis 2000 - 03 , ( baltimore : stsci ) smith , n. , davidson , k. , gull , t. r. , ishibashi , k. , & hillier , d. j. 2003 , , 586 , 432 sterken , c. , de groot , m. j. h. , & van genderen , a. m. 1996 , , 116 , 9 sterken , c. , freyhammer , l. , arentoft , t. , & van genderen , a. m. 1999 , , 346 , l33 sterken , c. , freyhammer , l. m. , arentoft , t. , & van genderen , a. m. 2001 , asp conf . ser . 233 : p cygni 2000 : 400 years of progress , 39 stys , d. j. & walborn , n. r. 2000 , instrument science report stis 2001 - 01r , ( baltimore , stsci ) van genderen , a. m. , de groot , m. j. h. , & the , p. s. 1994 , , 283 , 89 van genderen , a. m. et al . 1995 , , 304 , 415 van genderen , a. m. , sterken , c. , de groot , m. , & burki , g. 1999 , , 343 , 847 van genderen , a. m. , sterken , c. , & allen , w. h. 2003a , , 405 , 1057 van genderen , a. m. , sterken , c. , allen , w. h. , & liller , w. 2003b , , 412 , l25 viotti , r. , rossi , l. , cassatella , a. , altamore , a. , & baratta , g. b. 1989 , , 71 , 983 whitelock , p. a. , feast , m. w. , koen , c. , roberts , g. , & carter , b. s. 1994 , , 270 , 364 whitelock , p. a. , marang , f. , & crause , l. 2003 , , 8160 , 2 zanella , r. , wolf , b. , & stahl , o. 1984 , , 137 , 79 carinae . the bottom panel shows the total relative response of each ccd and filter combination used in this study on the same wavelength scale as the top panel . for plotting purposes the curves are not representative of relative responses between filters . stis filters are plotted with a dashed line and acs / hrc filters are plotted with a solid line . the dotted line represents the product of the stis ccd+f25nd3 response curve and the photon flux from the central star . ] lllrrrrrrr o4j802qxq&50814.008&1997.999&3.196e-12&13.2&2.35e-12&0.65&0.65&&0.79 + o4j801y4q&50891.414&1998.211&3.720e-12&12.7&3.07e-12&0.49&0.49&&0.64 + o55601hkq&51142.172&1998.898&5.374e-12&14.2&4.47e-12&0.09&0.09&&0.25 + o55602qsq&51230.449&1999.140&5.831e-12&13.7&5.00e-12&-0.00&-0.00&&0.16 + o5f102bkq&51616.484&2000.197&7.476e-12&13.3&&-0.27&-0.28&0.01&-0.13 + o5kz01coq&51623.785&2000.217&7.645e-12&&&-0.29 & & & + o5kz02gbq&51624.285&2000.218&7.580e-12&&&-0.28&-0.26&0.02 & + o5f101yoq&51626.762&2000.225&7.288e-12&&&-0.24 & & & + o5f103a1q&51826.039&2000.769&6.753e-12&&&-0.16&-0.16 & & + o62r01h0q&52016.734&2001.291&6.518e-12&11.9&5.60e-12&-0.12&-0.12&&0.02 + o6ex03a9q&52183.086&2001.747&7.573e-12&13.5&6.52e-12&-0.28&-0.28&0.00&-0.12 + o6en01c2q&52183.297&2001.747&7.509e-12&&&-0.27 & & & + o62r02fvq&52240.070&2001.903&7.470e-12&13.5&6.10e-12&-0.27&-0.27&&-0.11 + o6ex02tkq&52293.977&2002.051&7.241e-12&14.5&5.47e-12&-0.24&-0.21&0.02&-0.04 + o6ex01wyq&52294.207&2002.051&6.923e-12&&&-0.19 & & & + o6mo02eaq&52459.480&2002.504&6.933e-12&14.1&5.69e-12&-0.19&-0.20&0.01&-0.04 + o6mo01k6q&52459.781&2002.505&7.089e-12&&&-0.21 & & & + o8gm01a1q&52624.043&2002.955&6.840e-12&15.5&5.01e-12&-0.17&-0.17&&0.01 + o8gm12ukq&52682.859&2003.116&6.730e-12&14.9&5.43e-12&-0.16&-0.12&0.03&0.05 + o8gm11zpq&52683.242&2003.117&6.326e-12&&&-0.09 & & & + o8gm21t9q&52727.234&2003.238&6.904e-12&13.5&5.29e-12&-0.18&-0.18&&-0.03 + o8gm41eqq&52764.266&2003.339&7.462e-12&11.7&5.54e-12&-0.27&-0.27&&-0.13 + o8gm33qsq&52776.398&2003.372&7.662e-12&11.2&5.95e-12&-0.30&-0.29&0.00&-0.16 + o8gm32c6q&52778.469&2003.378&7.623e-12&&&-0.29 & & & + o8gm31fnq&52785.797&2003.398&7.617e-12&10.6&6.24e-12&-0.29&-0.29&0.01&-0.17 + o8gm51fuq&52791.035&2003.412&7.752e-12&&&-0.31 & & & + o8gm52kyq&52791.707&2003.414&7.558e-12&&&-0.28 & & & + o8gm63ohq&52812.031&2003.470&8.264e-12&9.0&6.98e-12&-0.38&-0.37&0.00&-0.27 + o8gm61pxq&52812.250&2003.471&8.212e-12&&&-0.37 & & & + o8gm62dwq&52813.703&2003.475&8.222e-12&&&-0.37 & & & + o8ma71jnq&52825.020&2003.506&7.675e-12&8.9&7.05e-12&-0.30&-0.32&0.03&-0.22 + o8ma72liq&52825.355&2003.506&8.043e-12&&&-0.35 & & & + o8ma81mtq&52849.594&2003.573&7.763e-12&8.3&7.05e-12&-0.31&-0.32&0.01&-0.22 + o8ma82b8q&52851.906&2003.579&7.850e-12&&&-0.32 & & & + o8ma91ubq&52903.355&2003.720&8.830e-12&&&-0.45&-0.48&0.03 & + o8ma92ctq&52904.289&2003.723&9.309e-12&&&-0.51 & & & + o8ma83fzq&52960.590&2003.877&9.969e-12&&&-0.58 & & & + lllrrrr + + j8gm1aa7q&52561.039&2002.782&1.385e-12&8.547&8.549&0.011 + j8gm1aa8q&52561.039&2002.782&1.402e-12&8.533 & & + j8gm1aalq&52561.047&2002.782&1.366e-12&8.562 & & + j8gm1aamq&52561.047&2002.782&1.375e-12&8.554 & & + j8gm1abmq&52561.098&2002.782&1.426e-12&8.515&8.535&0.015 + j8gm1abnq&52561.098&2002.782&1.404e-12&8.531 & & + j8gm1abyq&52561.105&2002.782&1.394e-12&8.539 & & + j8gm1abxq&52561.105&2002.782&1.374e-12&8.555 & & + j8gm2as1q&52682.547&2003.115&1.562e-12&8.416&8.417&0.007 + j8gm2as5q&52682.551&2003.115&1.574e-12&8.407 & & + j8gm2asaq&52682.555&2003.115&1.555e-12&8.421 & & + j8gm2asoq&52682.566&2003.115&1.549e-12&8.425 & & + j8ma3adhq&52803.055&2003.445&1.316e-12&8.601&8.571&0.018 + j8ma3adlq&52803.086&2003.445&1.372e-12&8.557 & & + j8ma3adqq&52803.090&2003.445&1.373e-12&8.556 & & + j8ma3ae3q&52803.152&2003.446&1.355e-12&8.570 & & + j8ma4aqfq&52840.145&2003.547&8.093e-13&9.130&9.129&0.003 + j8ma4aqj&52840.148&2003.547&8.069e-13&9.133 & & + j8ma4aqnq&52840.156&2003.547&8.109e-13&9.128 & & + j8ma4ar2q&52840.168&2003.547&8.137e-13&9.124 & & + j8ma6ayjq&52957.809&2003.869&9.216e-13&8.989&8.970&0.01 + j8ma6aynq&52957.844&2003.869&9.531e-13&8.952 & & + j8ma6az2q&52957.855&2003.869&9.385e-13&8.969 & & + + + + j8gm1aaaq&52561.039&2002.782&2.935e-12&7.731&7.726&0.011 + j8gm1aabq&52561.043&2002.782&2.995e-12&7.709 & & + j8gm1aaoq&52561.047&2002.782&2.910e-12&7.740 & & + j8gm1aapq&52561.051&2002.782&2.951e-12&7.725 & & + j8gm1abpq&52561.098&2002.782&2.985e-12&7.713&7.722&0.018 + j8gm1abqq&52561.098&2002.782&3.022e-12&7.699 & & + j8gm1ac0q&52561.105&2002.782&2.895e-12&7.746 & & + j8gm1ac1q&52561.105&2002.782&2.937e-12&7.730 & & + j8gm2as2q&52682.547&2003.115&3.181e-12&7.643&7.646&0.006 + j8gm2as6q&52682.551&2003.115&3.198e-12&7.638 & & + j8gm2ascq&52682.559&2003.115&3.174e-12&7.646 & & + j8gm2asqq&52682.570&2003.115&3.148e-12&7.655 & & + j8ma3adiq&52803.055&2003.445&3.058e-12&7.687&7.651&0.021 + j8ma3admq&52803.086&2003.445&3.194e-12&7.639 & & + j8ma3adsq&52803.094&2003.445&3.184e-12&7.642 & & + j8ma3ae5q&52803.156&2003.446&3.208e-12&7.634 & & + j8ma4aqgq&52840.148&2003.547&2.178e-12&8.055&8.063&0.008 + j8ma4aqkq&52840.152&2003.547&2.137e-12&8.076 & & + j8ma4aqpq&52840.156&2003.547&2.171e-12&8.058 & & + j8ma4ar4q&52840.168&2003.547&2.159e-12&8.065 & & + j8ma6aqgq&52957.805&2003.869&2.619e-12&7.855&7.840&0.015 + j8ma6aykq&52957.809&2003.869&2.628e-12&7.851 & & + j8ma6aypq&52957.848&2003.869&2.710e-12&7.817 & & + j8ma6az4q&52957.859&2003.869&2.664e-12&7.836 & & + + + + j8gm1aacq&52561.043&2002.782&3.306e-12&7.602&7.621&0.014 + j8gm1aadq&52561.043&2002.782&3.253e-12&7.619 & & + j8gm1aaqq&52561.051&2002.782&3.243e-12&7.623 & & + j8gm1aarq&52561.051&2002.782&3.189e-12&7.641 & & + j8gm1abrq&52561.098&2002.782&3.378e-12&7.578&7.608&0.021 + j8gm1absq&52561.102&2002.782&3.292e-12&7.606 & & + j8gm1ac2q&52561.105&2002.782&3.273e-12&7.612 & & + j8gm1ac3q&52561.109&2002.782&3.201e-12&7.637 & & + j8gm2as3q&52682.547&2003.115&3.327e-12&7.595&7.598&0.002 + j8gm2as7q&52682.551&2003.115&3.312e-12&7.600 & & + j8gm2asgq&52682.563&2003.115&3.313e-12&7.599 & & + j8ma3adjq&52803.055&2003.445&3.913e-12&7.419&7.379&0.023 + j8ma3adnq&52803.090&2003.445&4.095e-12&7.369 & & + j8ma3adwq&52803.098&2003.445&4.112e-12&7.365 & & + j8ma3ae9q&52803.160&2003.446&4.119e-12&7.363 & & + j8ma4aqhq&52840.148&2003.547&3.671e-12&7.488&7.494&0.003 + j8ma4aqlq&52840.152&2003.547&3.649e-12&7.495 & & + j8ma4aqtq&52840.160&2003.547&3.648e-12&7.495 & & + j8ma4ar8q&52840.176&2003.547&3.642e-12&7.497 & & + j8ma6aqhq&52957.805&2003.869&4.643e-12&7.233&7.222&0.008 + j8ma6aylq&52957.809&2003.869&4.695e-12&7.221 & & + j8ma6aytq&52957.852&2003.869&4.738e-12&7.211 & & + j8ma6az8q&52957.852&2003.869&4.678e-12&7.225 & & + + + + j8gm1aafq&52561.043&2002.782&3.700e-12&7.479&7.485&0.006 + j8gm1aatq&52561.051&2002.782&3.663e-12&7.491 & & + j8gm1abuq&52561.102&2002.782&3.755e-12&7.463&7.475&0.014 + j8gm1ac5q&52561.109&2002.782&3.649e-12&7.494 & & + j8gm1ac4q&52561.109&2002.782&3.745e-12&7.466 & & + j8gm2as4q&52682.551&2003.115&4.129e-12&7.360&7.359&0.004 + j8gm2as8q&52682.555&2003.115&4.162e-12&7.352 & & + j8gm2asjq&52682.566&2003.115&4.134e-12&7.359 & & + j8gm2asxq&52682.578&2003.115&4.118e-12&7.363 & & + j8ma3adkq&52803.059&2003.445&4.963e-12&7.161&7.126&0.020 + j8ma3adoq&52803.090&2003.445&5.156e-12&7.119 & & + j8ma3adzq&52803.102&2003.445&5.168e-12&7.117 & & + j8ma3aecq&52803.164&2003.446&5.206e-12&7.109 & & + j8ma4aqiq&52840.148&2003.547&5.259e-12&7.098&7.099&0.003 + j8ma4aqmq&52840.152&2003.547&5.250e-12&7.100 & & + j8ma4aqwq&52840.164&2003.547&5.234e-12&7.103 & & + j8ma4arbq&52840.180&2003.547&5.272e-12&7.095 & & + j8ma6ayiq&52957.805&2003.869&6.851e-12&6.811&6.807&0.004 + j8ma6aymq&52957.813&2003.869&6.907e-12&6.802 & & + j8ma6aywq&52957.855&2003.869&6.890e-12&6.804 & & + j8ma6azbq&52957.867&2003.869&6.849e-12&6.811 & & +

coupled spin 1/2 chains have attracted much attention lately due to the large number of their experimental realizations , as well as to the variety of theoretical techniques , both analytical and numerical , available to study the relevant models . spin 1/2 chains , owing to the jordan - wigner transformation@xcite , show properties remarkably similar to those of interacting one dimensional fermions with their low energy properties described by an effective luttinger liquid theory@xcite . recently it has been realized that these properties are drastically modified when the spin 1/2 chains are coupled together forming ladder systems@xcite . in this case , in a way very similar to haldane s spin - s problem @xcite , a gap is found to open for an even number of chains while the system remains gapless if the number of chains is odd . this phenomenon has been thoroughly investigated both analytically @xciteand numerically @xcite , and corresponding experimental systems were identified . typical examples of two - chain spin 1/2 ladders exhibiting a gap are @xmath1@xcite and @xmath2@xcite . on the other hand , an example of a gapless three - chain ladder is @xmath3@xcite . from the theoretical point of view , the difference between odd and even number of legs@xcite or odd and even spin@xcite has been understood qualitatively through a generalization of the lieb - schultz - mattis theorem . the theorem states that if the ground state is unique the system will necessarily be gapless . this is indeed the case for an odd number of coupled spin 1/2 chains . more recent theoretical work has emphasized the role of boundary conditions in the transverse direction in the formation of spin gaps . the results quoted above - the opening of a gap for an even number of coupled chains , gaplessness if the number is odd - are valid for open boundary conditions ( obc ) . it turns out that when periodic boundary conditions ( pbc ) are imposed on the transverse direction a gap opens for both cases of even and odd number of chains although the underlying reasons and the natures of the gaps are different . in the case of an even number of coupled chains , the reason for the gap is the formation of spin singlets along the transverse direction , similarly to the case of chains with open boundary conditions . when the number of coupled chains is odd a two fold degenerate dimerized ground state is obtained in the case of pbc@xcite - in contrast to its uniqueness in the obc - allowing for a gap in the spin excitations . the degeneracy of the ground state in the case of pbc can be understood as a consequence of the fact that pbc are frustrating for an odd number of legs . to date , no experimental system described by coupled antiferromagnetic spin 1/2 chains with periodic boundary conditions and an odd number of legs has been reported . when , further , leg - leg biquadratic interactions are included new states emerge . a system of two spin-1/2 spin chains enters a spontaneously dimerized phase with a gapped spectrum exhibiting non - haldane spin - liquid properties@xcite . the elementary excitations are neither spinons nor magnons , but pairs of propagating triplet or singlet solitons connecting two spontaneously dimerized ground states . subsequently more non - haldane spin - liquid models have been proposed @xcite . recently , these models of spin ladders with biquadratic exchange have been advocated@xcite as possible models for the formation of a spin gap in @xmath0 and @xmath4 . in this paper we study the low energy physics of the three - leg ladder with periodic boundary conditions ( see fig . [ fig : spin_tube ] ) . this model is called the spin - tube model in the following . by analyzing in detail its effective low energy hamiltonian ( leh ) , which consists of two _ non - equivalent _ coupled xxz chains of spins and chiral degrees of freedom , in the presence of a biquadratic exchange@xcite , we will show that the dimerized ground state of this model falls in the universality class of the non - haldane spin liquids . in the case of xxz chains with a biquadratic coupling and for @xmath5 , we find that the ising antiferromagnetic order can compete with the dimer order , and we will describe the resulting phase diagram . the organization of the paper is as follows . in sec.[model ] we introduce the spin tube model . we sketch the derivation of the strong interchain coupling effective hamiltonian and discuss its relation with two spin 1/2 chains coupled by a biquadratic interaction . we then discuss the symmetries and recall the results already known on the two chains with biquadratic interactions , in particular the prediction of a spin gap . in sec . [ bosonization ] we discuss the bosonization treatment of two non - equivalent xxz chains coupled by a biquadratic interaction . we show that this hamiltonian contains a term @xmath6 that is responsible for the formation of a spin - gap and singlet order , as well as terms @xmath7 that cause antiferromagnetic order . to analyze the competitions of these terms , we derive renormalization group equations . using these equations , we estimate in which part of the phase diagram we should expect a competition of antiferromagnetism and dimer order . we also estimate the equation of the phase boundary between the singlet and the antiferromagnetic order . in sec.[sec : dimerized ] we analyze in details the dimerized phase . the dimerized ground state is two - fold degenerate and is formed of alternating singlets of spins and pseudospins . the elementary excitations above the ground state carry a spin @xmath8 as well as a pseudospin @xmath8 . they can be seen as formed by the introduction of a spin and a pseudospin in the pattern of alternating singlets . similar excitations were obtained in ref . and it was shown that they lead to response functions very different from those of a haldane spin liquid . numerical evidence for such excitations was also obtained in ref . . in section [ sec : gap_closure ] , we analyze the effect of a magnetic field strong enough to close the gap of the dimer phase . when both chains are magnetic , the situation is similar to the one obtained with two coupled spin ladders @xcite . in the more physical situation , when only one of the two chains is magnetic , the ground state properties are those of a two component luttinger liquid . in the case of spin - orbital models , spin and orbital degrees of freedom decouple completely but the presence of orbital modes should affect the specific heat . in the case of the spin - tube , we show that in contrast to spin - orbital models , the spin - correlation functions are affected by the presence of auxiliary gapless modes . we also show that for non - equivalent chains ( a case that is realized in spin - orbital models ) the exponent of the spin correlations is _ non - universal _ at the transition . finally , in sec.[conclusions ] we will give some concluding remarks . we wish to study the three - chain ladder hamiltonian , @xmath9 where @xmath10 ( resp . @xmath11 ) is a chain ( resp . site ) index , @xmath12 is the coupling along the chain and @xmath13 the transverse coupling . we impose periodic boundary conditions along the rungs by identifying @xmath14 . we call the resulting model the spin - tube ( see fig . [ fig : spin_tube ] ) since it can be realized by placing 3 spin 1/2 chains forming an equilateral triangle . in order to investigate the low - energy physics of the spin - tube , we consider the limit of strong interchain coupling ( @xmath15 ) . this is the appropriate starting point that yields a good effective description of the properties of the spin tube in the whole range of @xmath16 . ( starting from the opposite limit @xmath17 , and treating @xmath18 as a perturbation , one finds@xcite that it gives rise to relevant terms in the hamiltonian . as a result , the initial @xmath18 grows until it is of order of @xmath12 at which point the weak coupling bosonization scheme is no more valid . ) to derive the effective low - energy hamiltonian , let us first consider the case @xmath19 . the system is a collection of independent rungs , each described by the following hamiltonian @xmath20 the ground state of each rung is four - fold degenerate , composed of two doublets of spin 1/2 excitations , corresponding to the left and right chirality ( -/+ ) , with energy @xmath21@xcite , and the excited states form a spin 3/2 quadruplet with energy @xmath22 . turning on @xmath23 allows the rungs to exchange spins . in the limit @xmath15 , the quadruplet of s=3/2 excitations can be neglected , and only the degenerate low energy subspace of spin 1/2 states needs to be taken into account . in this subspace the hamiltonian transforms into an effective hamiltonian with a biquadratic coupling between the spin and chirality degrees of freedom@xcite , @xmath24 where @xmath25 is the total spin , and @xmath26 are operators exchanging left and right chiralities . the original spin operators can be expressed in terms of the effective spin @xmath25 and the chirality @xmath27 in the following way : @xmath28 the spin tube model has already been investigated numerically@xcite using dmrg in the case of coupled xxx chains . the system was shown to exhibit a spin gap @xmath29 , with exponentially decaying correlation functions @xmath30 and @xmath31 . this behavior was shown to be related to the formation of a dimer order . a qualitative discussion of the origin of this dimer order can be found in ref . @xcite . in ref . the dispersion relation of excitations having @xmath32 and @xmath33 as a function of momentum was obtained numerically for a system of 12 sites showing only gapped excitations . in ref . , a generalization of the effective hamiltonian eq . ( [ heff ] ) to coupled anisotropic spin chains has been derived ( see eq . ( [ eq : heff_anisotropic ] ) . numerical diagonalizations were performed . a gap to @xmath34 excitations was obtained for@xmath35 with results in good agreement with those of ref . for @xmath36 . it was also shown that the ground state was degenerate in agreement with the dimer order picture of ref . . the dispersion relation of excitations having @xmath37 and total spin @xmath38 or @xmath39 was obtained@xcite . it was shown to be the bottom of a two particle continuum . the fundamental particle , the spinon , was conjectured to have @xmath40 . the spinon dispersion relation was obtained numerically by considering a system with an odd number of sites@xcite , showing that the spinons were massive for all momentum . the hamiltonian ( [ heff ] ) is part of the class of the hamiltonians consisting of two non - equivalent coupled xxz chains in the presence of a biquadratic exchange , @xmath41 where h_0 & = & _ i _ = 1,2 j _ ( s_,i^xs_,i+1^x+s_,i^ys_,i+1^y ) + j_^z s_,i^zs_,i+1^z [ h0 ] + h_b&= & _ i ( s_1,i^xs_1,i+1^x+s_1,i^ys_1,i+1^y+ _ 1 s_1,i^zs_1,i+1^z ) ( s_2,i^xs_2,i+1^x+s_2,1^ys_2,i+1^y+ _ 2 s_2,i^zs_2,i+1^z).[hbiq ] in the case of the spin - tube , @xmath42 corresponds to spin @xmath43 , and @xmath44 is associated with the chiral degrees of freedom @xmath26 . another way of writing a class of hamiltonian generalizing the spin tube model ( [ heff ] ) is : @xmath45 the effective hamiltonian for the spin tube is obtained for , @xmath46 for a tube made of xxz chains , one has , instead@xcite : @xmath47 the parameters @xmath48 and @xmath49 in eq . ( [ ham ] ) measure the xxz anisotropy for spin and chirality , respectively . when both of them are equal to 1 , the hamiltonian eq . ( [ ham ] ) is @xmath50 symmetric . for @xmath51 , @xmath52 , @xmath53 the two chains are equivalent , and can be parameterized as : @xmath54 hamiltonians of the type eq . ( [ ham ] ) can be mapped onto hamiltonians of the type eq . ( [ eq : basic_hamiltonian ] ) . the correspondence is given by : @xmath55 with the identification @xmath56 , @xmath57 . since writing the hamiltonian in the form ( [ eq : basic_hamiltonian ] ) is less restrictive than in the form ( [ ham ] ) ( i.e. the former includes the case @xmath58 ) , we focus on the hamiltonian ( [ eq : basic_hamiltonian ] ) . hamiltonians of the type eq . ( [ eq : basic_hamiltonian ] ) are also encountered in a different context than the spin - tube model . in particular , a hamiltonian of the type ( [ eq : basic_hamiltonian ] ) has been proposed by mostovoy and khomskii as a model for the spin gap formation in the @xmath0 @xcite and @xmath4 @xcite compounds . in that case , the @xmath59 spins correspond to the real spin of the system whereas the @xmath60 spins are pseudospins associated with _ orbital _ degrees of freedom . these spin - orbital models can be derived from a multiband hubbard model . the derivation is reviewed for instance in ref . . let us discuss first the case @xmath61 , @xmath62 . the hamiltonian describes two coupled heisenberg chains with a biquadratic coupling preserving the @xmath63 symmetry . actually , the full symmetry group is larger than su(2 ) . one has : @xmath64=0 \hspace{0.5 cm } [ h,\vec{s}_{2,\text{tot.}}]=0\ ] ] and the full symmetry group is therefore @xmath50 rather than the su(2 ) symmetry that follows from @xmath65=0 $ ] . as a result the spectrum consists of @xmath66 multiplets@xcite . for @xmath67 , the hamiltonian ( [ eq : basic_hamiltonian ] ) has been shown to have an even larger @xmath68 symmetry and to reduce to an integrable @xmath68 spin chain @xcite . the spectrum of the su(4 ) spin chain has been obtained by the bethe ansatz @xcite , and the correlations functions have been obtained by non - abelian bosonization techniques@xcite , identifying the low energy effective theory describing the spin chain as the @xmath69 wznw model . the integrable @xmath68 spin - chain has also been intensively studied numerically using quantum monte carlo ( qmc ) @xcite or density matrix renormalization group ( dmrg ) and lanczos exact diagonalization ( ed)@xcite in the context of the spin - orbital models . the ground state energy , and excited state energy were obtained in good agreement@xcite with analytical calculation using the bethe ansatz@xcite . the numerical calculation of the correlation functions @xcite reproduces results of the the continuum field theoretical treatment @xcite . at @xmath70 perturbations are generated which lower the @xmath68 symmetry to @xmath71 and render the hamiltonian non - integrable . these perturbations have been recently studied @xcite by describing the chain away from the integrable point as a perturbed @xmath69 wzw model . it was found that for @xmath72 , a gapless phase is obtained while for @xmath73 , a gap is formed . a different field theoretical treatment in the limit @xmath74 also leads to the appearance of a gapped phase . for @xmath75 , the ground state wavefunction could be obtained exactly in matrix product form@xcite , of singlet states along the legs of the ladder . this picture is in good agreement with the predictions of the field theoretical treatment . the dependence of the gap on the coupling for the range @xmath76 was obtained by dmrg calculations@xcite . it was found that for @xmath77 , the gap increases proportionally to @xmath78 in agreement with the weak coupling bosonization treatment , and vanishes for @xmath79 , which is the su(4 ) symmetric point as predicted . the dmrg calculations of ref . showed however a power law gap opening . such power law gap opening can only be explained by the presence of a relevant operator in the continuum description . however , no such operator was obtained in the bosonization treatment@xcite . moreover , if a relevant operator was present in the continuum theory , the absence of a gap at @xmath80 would be the result of the coefficient of this operator vanishing precisely at @xmath80 . but then , in contrast to what is observed in numerical calculations , a gap would also obtain for @xmath81 . a solution to this puzzle has been suggested recently@xcite . in ref . , the gap has been calculated by assuming that the first excited state was in the subspace @xmath82 . this assumption was shown to be incorrect in the gapped phase in which the first excited state lies in the subspace @xmath83 . when corrected@xcite , a slow increase with @xmath84 is found , compatible with an exponential gap opening . the correlation functions @xcite do not show incommensuration , in agreement with the field theoretic approach @xcite . taking @xmath85 and @xmath62 in ( [ eq : basic_hamiltonian ] ) preserves the @xmath71 symmetry . this case has been investigated numerically@xcite with @xmath86 . it was shown that a gapless gapped phase transition obtained as @xmath87 was increased . this work was followed by analytical investigation based on the perturbed @xmath69 wzw continuum theory@xcite . the analytical investigations established the existence for a given @xmath78 of an extended gapless region in the plane @xmath88 that contains the line @xmath89 previously discussed . in the present work , we consider the general case of @xmath90 and @xmath91 . this case includes in particular the spin tube model . we will focus on the regime @xmath92 and apply methods similar to those of ref . . in this section , we derive the phase diagram of two xxz spin chains weakly coupled by a biquadratic exchange . we first recall the bosonization of a spin chain in sec . [ sec : single_chain ] . this section can be skipped by readers familiar with bosonization . then , we discuss the bosonization of the two coupled chains system in sec . [ sec : biquadratic_coupling ] . this allows us to derive renormalization group ( rg ) equations for the coupling constants of the problem . finally , in sec . [ phasediagram ] , we discuss the phase diagram deduced from the analysis of rg equations . in this section , we recall briefly the derivation of the bosonized hamiltonian and spin operators that describe an isolated xxz chain . we follow the well known abelian bosonization procedure for spins @xcite . the xxz spin chain is described by the hamiltonian : @xmath93 the xxz spin chain hamiltonian is first transformed into an interacting fermionic system on the lattice by expressing the spin operators @xmath94 , @xmath95 , @xmath96 in terms of fermion operators @xmath97 , @xmath98 , using the jordan - wigner transformation@xcite : @xmath99 this transformation turns the xxz hamiltonian into a model of spinless fermions with nearest neighbor interaction described by the hamiltonian , @xmath100 for @xmath101 , the hamiltonian ( [ spinlessf ] ) describes non - interacting fermions and is easily diagonalized . to proceed , we restrict our attention to the low - energy sector of the theory , captured by the continuum theory . introduce the left ( right ) chiral fermion fields @xmath102 containing momenta close to the fermi points @xmath103 , ( @xmath104 with @xmath98 the lattice spacing ) , @xmath105 the continuum fermi fields are then reexpressed in terms of bosonic field@xcite , as follows : @xmath106 where the pair of conjugate fields , @xmath107 satisfy the following commutation relation : @xmath108=\imath \delta(x - x^\prime),\ ] ] and the field @xmath109 , dual to @xmath110 , is defined as : @xmath111 the spin operators eqs . ( [ jwp])([jwz ] ) , can be expressed in the continuum limit as : @xmath112 \mbox { , } \mbox { } s^z(x)=\frac{s^z_n}{a}=-\frac { \partial_x \phi } \pi + \frac{e^{\imath \pi \frac x a } } { \pi a } \cos 2 \phi,\end{aligned}\ ] ] introducing normal ordering with respect to the fermion vacuum , one has : @xmath113 where @xmath114 indicates normal ordering . the average @xmath115 is the ground state energy of the chain and can be found by the bethe ansatz@xcite . however , as long as one is only interested in the correlation functions of the single chain , one can simply drop this contribution and focus on the normal ordered terms that describe the excitations above the ground state . our task is thus to derive a bosonized expression of the normal ordered product in eq . ( [ eq : normal_order ] ) . some care is needed in order to obtain correct results@xcite , and one finds : @xmath116 + \frac { e^{\frac { \imath \pi x } a}}{\pi } \sin 2 \phi \nonumber \\ : s_i^zs_{i+1}^z+ s_i^z s_{i+1}^z : & = & \frac 2 { \pi^2 } ( \partial_x \phi)^2 + \frac 2 { ( \pi a)^2 } e^{\imath \frac{\pi x } a } \sin 2 \phi + \frac { 2 \cos 4\phi}{(2\pi a)^2 } + \text{irrelevant terms \ldots}\end{aligned}\ ] ] the oscillating terms in eq . ( [ eq : bosonized_quad ] ) are dropped from the hamiltonian after integration over @xmath117 . the hamiltonian @xmath118 , eq . ( [ spinlessf ] ) then becomes : @xmath119-\frac { 2\delta}{(2\pi a)^2 } \int dx \cos(4\phi),\ ] ] where for @xmath120 , u&=&aj(1+)^1/2 + k&=&(1+)^-1/2 + & = & j_za.[deltaequiv ] thus , the bosonized form of @xmath118 reduces to a sine - gordon hamiltonian , where the cosine terms come from the intrachain umklapp process@xcite . the renormalization group treatment shows@xcite that in the vicinity of the xy point , the cosine terms are irrelevant , so that asymptotic properties are described by a free scalar field with renormalized @xmath121 . since the xxz chain is integrable , it can be shown that the gapless spectrum extends to the whole region @xmath122 . moreover , it is possible to obtain an analytic expression for the renormalized @xmath121 from the exact solution@xcite . one finds : @xmath123 @xmath124 the isotropic point @xmath36 ( antiferromagnetic heisenberg model ) corresponds , in the bosonization description , to @xmath125 and @xmath126 . at this point @xmath127 is marginally irrelevant . for @xmath128 , the exact solution shows that a gap opens in the excitations of the system@xcite , and that an ising order of the spins along the z axis is obtained . this result can also be found from a bosonization procedure valid in the vicinity of the isotropic point . we now proceed to derive , using the results reviewed in the previous subsection , the bosonized hamiltonian of two non - equivalent coupled xxz chains . to bosonize the hamiltonian @xmath129 , eq . ( [ h0 ] ) describing two decoupled xxz chain we introduce two pairs of dual fields ( one pair for each chain ) @xmath130 ( @xmath131 ) as defined in sec . [ sec : single_chain ] . the bosonized form of the hamiltonian @xmath129 is then : @xmath132-\frac{2 \delta_\alpha}{(2\pi a)^2}\int dx \cos 4\phi_\alpha \right\},\end{aligned}\ ] ] with the fields @xmath133 and @xmath134 having _ a prori _ different velocities and luttinger couplings , @xmath135 the bosonization formulas for the spins ( [ spin - operators ] ) are unchanged except for the obvious introduction of a chain index . in order to have the full bosonized hamiltonian , we now have to derive the bosonized form of the biquadratic exchange ( [ hbiq ] ) . the first step is to normal order using eq . ( [ eq : normal_order ] ) . this step is important since it leads to non trivial contributions to the quadratic part of the hamiltonian . we introduce the terms : @xmath136 where @xmath131 that measure the strength of these corrections . having expressed the hamiltonian in a normal ordered form , we can apply eq . ( [ eq : bosonized_quad ] ) to obtain the bosonized expression . the oscillating terms of eq . ( [ eq : bosonized_quad ] ) give rise to an interchain coupling term , the strength of which is given by : @xmath137 collecting everything , one obtains the following bosonized biquadratic exchange : @xmath138 \nonumber \\ & + & a_2 \int \frac{dx}{\pi } \left[(\pi \pi_1)^2 + ( 1 + 2\delta_1/\pi ) ( \partial_x \phi_1)^2 \right ] + \text{irrelevant terms \ldots},\end{aligned}\ ] ] the terms in @xmath139 form a mean - field like interchain interaction . for @xmath140 , these terms merely produce a renormalization of the velocities and luttinger liquid exponents with respect to the decoupled chains . however , in the case where @xmath141 or @xmath142 , these terms become crucial since they give a finite velocity to the @xmath133 ( resp . @xmath134 excitations ) excitations . such a case is realized in the spin - tube problem where the exchange constant of the pseudospins @xmath27 is zero . we see that the mean - field like contribution of the spin fluctuations contributes in that system to provide an exchange constant and thus a finite velocity to the pseudospin excitations . of course , in such case , the bosonization procedure is not really justified since interactions are of the order of magnitude of the bandwidth of the spin or pseudospin excitations . however , it is usual in quasi - one dimensional systems to have a continuity between the weak and the strong coupling regime . moreover , a mean field theoretical treatment in the xy limit ( @xmath143 ) leads to similar results to the bosonization treatment . details can be found in app . [ app : mean - field ] . we will therefore assume that although not fully justified in the spin tube case , bosonization nevertheless leads to qualitatively correct results concerning the phases of the system and the overall behavior of correlations in these various phases . a quantitative treatment ( in particular of the phase boundaries ) requires numerical simulations that are beyond the scope of this paper . the present treatment shows us that in weak coupling the two chains are described by a bosonized hamiltonian : @xmath144}\nonumber \\ & + & \frac { 2g}{(2 \pi a)^2}\int dx \sin 2\phi_1 \sin 2\phi_2 \end{aligned}\ ] ] we shall analyze the model perturbatively , with results valid up to a given value of @xmath78 : it is known that in the strong coupling regime of the two chains with biquadratic exchange there is a special value of the interchain coupling at which the model has @xmath68 symmetry . at this special point , the model develops a gapless phase described by a @xmath69 wznw model . such an effect is non - perturbative : the resulting critical point has a conformal anomaly @xmath145 whereas the original unperturbed model has @xmath146 . this implies by zamolodchikov s @xmath147theorem@xcite that the transition to the @xmath69 can not be predicted by a rg calculation that always leads to a decrease of @xmath147 . beyond this special value of @xmath78 , the weak coupling theory would lead to incorrect predictions and an alternative approach such as the one of lecheminant and azaria is needed . on the other hand , it is expected to give qualitatively correct predictions when the coupling is smaller than the critical value . in the remainder of the paper we shall thus work in the weak coupling regime where the weak coupling theory is valid . in the following section , we will discuss the rg treatment of the weak coupling model . in the following analysis , we will neglect the velocity difference between chains 1 and 2 since usually velocity differences do not play an important role in the derivation of the phase diagram by rg techniques . however , for the sake of completeness , we have given in app . [ app : momentum_shell_rg ] the rg equations for non - equal velocities derived using a momentum shell integration technique@xcite . when velocity differences are neglected , the renormalization group equations for @xmath148 can be easily derived from the hamiltonian ( [ non - equivalent - chains - ni])([hbiqbos ] ) using operator product expansions ( ope ) @xcite . the renormalization group equations for @xmath149 neglecting velocity differences are : @xmath150 while the renormalization group equations for @xmath151 are : @xmath152 we now proceed to deduce the weak coupling phase diagram of the model ( [ eq : scaling_rge ] ) indicates that @xmath153 is a relevant variable when @xmath154 , while the variables @xmath155 become relevant when respectively , @xmath156 , @xmath157 . there are therefore four cases to distinguish . in the first case : @xmath158 , there are no relevant operators and the system is in a gapless state . in the second case : @xmath159 , @xmath160 , there is a single relevant operator ; for @xmath161 ( or equivalently @xmath162 ) , there are two , while for @xmath163 there are three relevant operators . these four different regions are shown in fig . [ fig : phase_diagram ] . in the presence of relevant operators , rg equations cease to be valid as soon as the dimensionless coupling become @xmath164 , and we have to determine the nature of the strong coupling fixed points in order to predict the phase diagram . we see that @xmath151 are driven to zero by the flow of the rg when there is a relevant operator so that the fields become classical . as a result , @xmath165 are locked at average values @xmath166 that minimize the ground state energy and a gap opens in the excitations of these fields@xcite . when these fields are locked , it is also known that the exponentials of the dual field have exponential decay@xcite . let us consider first the case with @xmath153 the only relevant operator . then , the minimization of the ground state energy requires @xmath167 , i.e. @xmath168 . it is then clear that @xmath169 as well as @xmath170 , which are the staggered component of @xmath96 and @xmath94 , will display an exponential decay . it can also be shown that the uniform component of the spins also present an exponential decay . as a result we have a spin - liquid phase that presents only short range order in all its spin correlations . as we shall see , this spin - liquid phase is a dimerized phase whose properties are discussed extensively in sec . [ sec : dimerized ] . in the case @xmath171 ( decoupled chains ) the analysis is even simpler . then the two chains remain decoupled , and we are left with the analysis of the usual sine - gordon model . it is then known that when @xmath172 is relevant , the field @xmath173 is locked . the analysis of the resulting strong coupling fixed point can be found for instance in ref . . it is found that the strong coupling fixed point is associated with the ising antiferromagnetic phase of the xxz chain at @xmath128 in which the staggered component of @xmath96 has a non zero expectation value in the ground state . when @xmath153 and at least one of the @xmath172 are relevant , there are two possible candidates for the ground state . considering eq . ( [ eq : scaling_rge ] ) , we see that the effect of the biquadratic interchain interaction is to reduce the effective @xmath174 . physically , this means that the tendency to form singlets competes with the tendency to form an ising antiferromagnet . two scenarios are possible . one is that there is a well defined phase boundary between a pure spin - liquid state and a pure ising antiferromagnet state . the second scenario is that there is a crossover between the two pure states as @xmath175 is varied . in such case , increasing @xmath78 would lead to a gradual disappearance of antiferromagnetic order leaving a purely singlet state as @xmath176 . since in both phases it is the same field that orders , there is _ a priori _ no reason to exclude a mixed spin - liquid antiferromagnet order . thus , the first scenario appears extremely unlikely . a numerical investigation of the crossover could be very interesting as a toy model of a crossover from spin - liquid to antiferromagnetism . it is interesting to remark that if @xmath177 , and @xmath178 , the equations ( [ eq : scaling_rge ] ) can be integrated analytically . two phases are obtained , separated by a line @xmath179 . in the first one , @xmath180 and @xmath181 which corresponds to a ground state with singlet order . in the second one , @xmath182 and @xmath183 , which corresponds to antiferromagnetic order . the rg equations cease to be valid for @xmath184 or @xmath185 . if when this scale is reached @xmath153 and @xmath186 are of the same order of magnitude , there is a possibility of obtaining a mixing of antiferromagnetism and dimer order . note that even on the line @xmath187 , there is a finite correlation length . this is a further evidence for a progressive crossover from dominant antiferromagnetic order to dominant dimer order . it is also possible to give a purely classical treatment for @xmath177 by simply minimizing the ground state energy with respect to @xmath188 and @xmath189 . in the case @xmath190 , one finds that there are three different regimes . for @xmath191 one obtains @xmath192 corresponding to a purely dimerized phase . for @xmath193 , one obtains @xmath194 and @xmath195 . this corresponds to persistence of dimerization in the chain with the smallest tendency to antiferromagnetic order , whereas the chain with the strongest tendency to antiferromagnetism is found in state with mixed dimer and antiferromagnetic order . the antiferromagnetic order parameter in that chain , @xmath196 , then assumes the value @xmath197 . finally in the region @xmath198 , both chains display antiferromagnetism with @xmath199 . these results are summarized on figure [ fig : classical_phase_diag ] . for @xmath200 , the region with mixed antiferromagnetic and dimer order shrinks to a single point . it would be interesting to see how quantum fluctuations affect the present picture and in particular determine whether sharp transitions are preserved or if they evolve into crossovers . in this section , we try to estimate the position of the crossover between the spin - liquid and the antiferromagnet . this can be done roughly by comparing the correlation lengths in the antiferromagnet and in the spin - liquid phase . using the rg equations , for small @xmath201 , we can neglect the renormalization of @xmath151 . this leads to : @xmath202 when any of these quantities become of the order of the energy cutoff @xmath203 , the rg equations cease to be valid and the phase that is obtained is determined by minimizing the ground state energy . for @xmath153 , the strong coupling is obtained at a length scale : @xmath204 this is the correlation length of spin fluctuations in the spin - liquid phase . for @xmath205,@xmath206 the strong coupling is obtained respectively at length scale : @xmath207 it is clear that the shortest length corresponds to the first operator to attain strong coupling . therefore , the phase that is obtained is the one with the shortest correlation length . for @xmath177 , this is in agreement up to a constant with the criterion derived from the rg equations ( [ eq : scaling_rge])([eq : second_order_rge ] ) . the comparison of correlation length allows to draw a rough phase boundary between the antiferromagnet and the dimerized phase that could also be obtained by numerically integrating the rg equations ( [ eq : scaling_rge])([eq : second_order_rge ] ) starting from weak coupling and any @xmath151 . the equation of the phase boundary is in the case of equivalent chains @xmath178 , @xmath208 : @xmath209 let us note that in the isotropic spin - tube case , the operators causing antiferromagnetic order are ( marginally ) irrelevant so that there is only singlet order . however , in the case of an anisotropic spin tube ( 3 coupled xxz spin chain with @xmath210 ) , such competition becomes possible . completely decoupled chains exhibit for @xmath210 an ising antiferromagnetic phase . introducing a strong enough biquadratic interchain coupling favors on the other hand a spin liquid phase . the competition of the two should produce a crossover from the ising antiferromagnet to the spin liquid of the type discussed in the preceding section . in this section , we discuss the properties of the spin liquid phase . in the case of equivalent chains , further progress can be made by using symmetric and antisymmetric modes allowing in particular a refermionization of the problem and the calculation of some correlation functions@xcite . this will be discussed in sec . [ sec : equivalent_chains ] . in the general case with inequivalent chains , such decoupling is no longer possible . however , it is still possible to present a simple semiclassical picture of the nature of excitations above the ground state . this will be the subject of section [ sec : non_equivalent_chains ] . we will focus on region @xmath211 , @xmath159 , in which the only relevant operator is the biquadratic exchange , @xmath212 . the ground state shows long - range order of the fields @xmath133 and @xmath134 . the expectation values of the ordered fields are : @xmath213 and as a result , @xmath214 . using eq . ( [ eq : bosonized_quad ] ) , this implies that a dimerized order develops both in spin variables and pseudospin variables , i.e. @xmath215 ( @xmath131 ) . in parallel with that , we have @xmath216 , so that by eq . ( [ spin - operators ] ) @xmath217 , and the correlation functions @xmath218 as @xmath219 . it is also well known that when the fields @xmath220 are ordered , the correlation functions of the disorder operators @xmath221 decay exponentially at large distances . using again the bosonization formulas for the spins , eq . ( [ spin - operators ] ) , this implies an exponential decay of all correlation functions : @xmath222 and @xmath223 where @xmath224 . therefore , the dimerized phase appears as a spin liquid state formed of singlets of spins on both chain 1 and chain 2 . such a conclusion was reached previously in a numerical investigation of the spin tube @xcite and by considering the equivalent isotropic chains at the solvable point@xcite @xmath225 where the ground state wavefunction can be obtained exactly in matrix product form . our results show that such mechanism of spin liquid ground state formation does not require su(2 ) symmetry . this mechanism is somewhat reminiscent of the spin - peierls transition@xcite , the pseudospins playing here the role of the phonons . this is the essence of the mostovoy - khomskii model@xcite for the `` spin - peierls '' transition at @xmath226k in @xmath0 . in the case of the spin - tube the same picture of the ground state obtains , with the ground state of the spin - tube formed by _ singlet of spins _ on even bonds and _ singlet of chiralities _ on odd bonds or by _ singlet of spins _ on odd bonds and _ singlet of chiralities _ on even bonds ( see fig . [ fig : ground_state ] ) . for the moment , we have only been able to discuss the nature of the ground state . however , it is also important to discuss the nature of the excitations as well as the various correlation functions . in order to do that , it is worth to restrict first to the simple case of two equivalent chains , in which the physical picture is the clearest . when the two chains are equivalent , we introduce the fields@xcite : @xmath227 and their conjugate fields : @xmath228 so that the total hamiltonian can be completely decoupled into the symmetric and antisymmetric parts , @xmath229 @xmath230 - \frac{g}{(2\pi a)^2 } \int dx \cos \sqrt{8}\phi_s\ ] ] @xmath231 + \frac{g}{(2\pi a)^2 } \int dx \cos \sqrt{8}\phi_a,\ ] ] where the magnetic field couples only to @xmath232 , and only the most relevant operators have been taken into account . the elementary excitations can be discussed in terms of solitons of two decoupled sine - gordon models . the solitons of the hamiltonian @xmath233 carry a total spin @xmath234 whereas those of hamiltonian @xmath235 carry a spin 0 as they do not couple to the magnetic field . these solitons are represented in fig . [ fig : solitons ] . it is also convenient to use the canonical transformation @xmath236 , @xmath237 followed by a refermionization . introducing the fermion operators : @xmath238 where @xmath239 , one can finally rewrite @xmath240 in the form : @xmath241 where @xmath242 and @xmath239 . the couplings are given by : @xmath243 at the isotropic point ( @xmath244 ) , one has @xmath245 , so that @xmath246 is a free fermion hamiltonian @xcite . similarly to the spin ladder , where @xmath247=0 $ ] , the excitation spectrum can be split into a singlet and a triplet with spin @xmath248 . however , in contrast to the spin ladder case , the triplet and the singlet here have the same mass . this is the signature of a larger symmetry group , @xmath249 . the correlation functions can be obtained from mapping the free fermion hamiltonian onto two non - critical ising model @xcite exhibiting the @xmath250 symmetry . remarkably , although the system has a spin gap , these correlation functions are very different from those of a spin-1 chain @xcite or a spin ladder @xcite . in the chain with biquadratic exchange , the response functions do not show any particle - like delta function peak in their imaginary part , but only a two particle continuum@xcite even in the vicinity of @xmath251 . this is to be contrasted with the spin ladder @xcite which shows a delta function peak associated with a single particle excitation at @xmath251 . the two chains with biquadratic interactions thus form a `` non - haldane '' spin liquid . away from the isotropic point , the hamiltonian can still be refermionized but the fermions ( solitons ) have interactions which preclude a mapping on a non - critical ising model . however , in the anisotropic case , when the antiferromagnetic intrachain interaction is irrelevant one has @xmath252 . this implies that no bound states of solitons can be formed since there is no coherent propagation of two solitons@xcite , hence the absence of a single particle peak . a calculation of correlation functions is in principle possible using as form factors approach , but this is far beyond the scope of the present paper . since @xmath235 and @xmath233 remain decoupled , the excitations having @xmath234 do not interact with the excitations having @xmath253 . this can be understood as a consequence of the @xmath254 symmetry of the problem and the resulting separate conservation of @xmath255 and @xmath256 . in the case of two non - equivalent chains , one can not decouple the hamiltonian into two sine - gordon hamiltonians . therefore , determining the nature and the quantum numbers of elementary excitations is not as straightforward . the preceding canonical transformation leads to the following hamiltonian : @xmath257 + \frac{(u_1/k_1+u_2/k_2 ) } 2 [ ( \partial_x \phi_s)^2 + ( \partial_x \phi_a)^2]\right . \nonumber \\ & + & \left . \frac { u_1 k_1 -u_2 k_2}{2 } \pi^2 \pi_s \pi_a + \frac { u_1 /k_1 - u_2 /k_2 } 2 \partial_x \phi_s \partial_x \phi_a \right]+ \frac{2g}{(2\pi a)^2 } \int dx [ \cos \sqrt{8 } \phi_s -\cos \sqrt{8 } \phi_a ] \end{aligned}\ ] ] shifting @xmath258 , we see that the resulting hamiltonian has a @xmath259 invariance under @xmath260 . a rescaling @xmath261 , @xmath262 , and a refermionization brings the hamiltonian to the form : @xmath263 \nonumber \\ & + & g(\rho_{r , s}(x)\rho_{r , a}(x)+\rho_{l , s}(x)\rho_{l , a}(x))+ \tilde g_1 \sum_{r\ne r'}\rho_{r , r}(x)\rho_{l , r'}(x ) + \tilde g_2 \sum_r \rho_{r , r}(x)\rho_{l , r}(x)\end{aligned}\ ] ] as a result , there is now an interaction between the excitations of spin @xmath264 ( the @xmath98 fermions ) and the excitations of spin @xmath265 ( the @xmath266 fermions ) . we have : @xmath267\\ g&=&\pi \left [ u_1\left(k_1+\frac 1 { 4k_1}\right ) - u_2\left(k_2+\frac 1 { 4k_2}\right)\right ] \\ \tilde g_1&= & \pi\left [ u_1\left(\frac 1 { 4k_1}-k_1\right ) + u_2\left(k_2-\frac 1 { 4k_2}\right)\right ] \\ \tilde g_2&= & \pi\left [ u_1\left(\frac 1 { 4k_1}-k_1\right ) - u_2\left(k_2-\frac 1 { 4k_2}\right)\right ] \\ m & = & \frac{g}{4 \pi a } \end{aligned}\ ] ] the fermionic version of the model is a generalization of the massive thirring model . the fields carry spin and the interactions break spin rotation symmetry . it can be checked that the interaction of @xmath98 with @xmath266 fermions disappears only for @xmath268 , @xmath269 . not much can be said of the elementary excitations of hamiltonian ( [ eq : fermion_non_equivalent ] ) due to the absence of an exact solution . in order to gain some insight into the elementary excitations of the dimerized state in the case of non - equivalent chains , we resort to semi - classical approximations . clearly , we can search for a semi - classical minimum of ( [ eq : rotated_non_equivalent ] ) with either @xmath270 or @xmath271 . this corresponds to a single soliton in the system . in the general case , such excitations are associated with a single fermion , either @xmath98 or @xmath266 . in order to obtain a physical picture for this type of elementary excitations , let us calculate the average `` magnetization '' for the fields @xmath133 and @xmath134 when there is a soliton connecting the two degenerate dimerized ground - states . in the case @xmath270 , @xmath133 decreases from @xmath272 to @xmath273 while @xmath134 increases from @xmath273 to @xmath272 . we immediately get : @xmath274 another solution is obtained by reversing the sign of @xmath220 leading to @xmath275 . the case @xmath271 corresponds to @xmath276 . in each case , the elementary excitations are formed by breaking a singlet on neighboring sites on each chain . such objects then propagate coherently . in the case of the spin tube , this corresponds to having one unpaired spin associated with one unpaired chirality pseudospin . such an excitation has @xmath277 , @xmath278 . it is the spinon of ref . . thus the elementary excitations of the model can be easily visualized as an unpaired spin and an unpaired chirality forming a triplet or a singlet diagonal bond ( see fig . [ fig : singlet_triplet_ne ] ) . this is the soliton in the dimer order . this picture generalizes from isotropic equivalent spin chains @xcite to the case of inequivalent and anisotropic chains . an open question is whether in the case of inequivalent chains bound states of these elementary excitations can be obtained in contrast to the case of equivalent chains . a necessary condition is that there exists attractive interactions between @xmath98 and @xmath266 fermions . the study of such bound states could be of interest in relation with light scattering experiments on @xmath0 . besides semi - classical approximation , another approximate treatment is possible . in the case where @xmath268 , @xmath279 and @xmath280 , the hamiltonian ( [ eq : full_bosonized_hamiltonian ] ) is the double sine gordon model@xcite , or the bukhvostov - lipatov model@xcite . the model is known to be bethe ansatz solvable@xcite but its elementary excitations and @xmath25 matrix have only been obtained recently @xcite and shown to be identical to those of the double sine - gordon model on its integrable line . the spectrum of the model @xcite is consists of four massive particles carrying two quantum numbers @xmath281 , and the @xmath25 matrix is@xcite : @xmath282 where @xmath283 is the @xmath25 matrix of a sine gordon model having the parameter @xmath284 and @xmath285 . as a result , one of the sine gordon models is in the attractive regime and has a spectrum made of fermions and bound states of fermions whereas the second sine - gordon model is is the repulsive regime and has a spectrum containing only fermions . the mass of the fundamental fermion is @xmath286 . for @xmath287 , one recovers the equivalent chain and the mass of the bound states is then : @xmath288 since for @xmath289 , the operator @xmath290 is the most relevant , it is reasonable to expect that neither the marginal perturbations due to @xmath291 nor the less relevant perturbations @xmath292 change the gapped nature of the spectrum . in this regime , one should not observe any bound state , since the @xmath293 bound state only exists for @xmath294 . in terms of the original spin chain , the double sine - gordon regime should be accessible if one chooses to have one chain with @xmath210 and the other chain with @xmath128 in such way that @xmath280 and @xmath295 . however , the double sine - gordon regime would not be observed in the spin tube case , in which we should have @xmath296 . we expect that in the spin tube case , the system has more quantum fluctuations than in the double sine - gordon regime . therefore , no bound states of spinons will form . this heuristic argument agrees with the numerical calculations on the spin tube that show no bound state of spinons@xcite . in this section , we discuss the effect of the application of a magnetic field on the dimerized phase . in general , the application of a magnetic field to a gapped one dimensional spin liquid system results in the closure of the gap for a magnetic field of the order of magnitude of the gap . below the critical magnetic field the magnetization is zero . above the critical magnetic field the magnetization increases as @xmath297 , the system is in a single component luttinger liquid state and has incommensurate spin correlations@xcite . here , we have to consider two _ a priori _ different cases . in the first one and most academic , the two coupled chains carry real spins that couple to the magnetic field . in the second case , only one chain carry a spin that couple to the magnetic field , the other one carrying only a pseudospin degree of freedom that does not couple to a magnetic field . this last case corresponds to the spin - orbital models@xcite and to the spin tube model@xcite . the behavior of the spin - orbital model in a magnetic field has been discussed in refs . and . both references study models with @xmath50 symmetry in the absence of the magnetic field . it is shown that in the vicinity of the su(4 ) point , the spin - orbital model becomes equivalent to the @xmath298 gross neveu model describing the gapped modes plus a @xmath299 cft that describes the gapless magnetic modes . also , in ref . , the weak coupling case has been discussed . the case where both chains carry spin will be dealt with in sec . [ sec : both_w_spin ] and the case of a single chain carrying spin in sec . [ sec : one_w_spin ] . if both chains carry spin , the coupling to the magnetic field ( taken parallel to the z axis ) is : @xmath300 in that case , we can use the decoupling of sec . [ sec : equivalent_chains ] , eqs . ( [ hs ] ) and ( [ ha ] ) . the coupling to the magnetic field is : @xmath301 we consider the case where the two chains are equivalent . this problem has been discussed in the context of spin ladders@xcite . the gap in the antisymmetric modes @xmath220 is not affected by the presence of the magnetic field . on the other hand , for a sufficiently large magnetic field @xmath302 , the gap in the symmetric modes closes and the magnetization @xmath303 behaves as @xmath304 for @xmath302 . moreover , it can be shown that the low energy modes of @xmath232 are described by the following hamiltonian : @xmath305\ ] ] and the exponent @xmath306 takes the universal value 1/2 at @xmath307 . correlation functions in the incommensurate phase can be obtained in the isotropic case by a calculation similar to the spin - ladder case @xcite . this time , @xmath308 is finite so that @xmath309 has exponentially decaying correlations . as a result , one has : @xmath310 a simplified expression for the operator @xmath311 is : @xmath312 and the correlations of @xmath96 are : @xmath313 there are subleading power law corrections at @xmath314 . as shown by furusaki and zhang , these corrections are missed if one naively neglects the band curvature after refermionization @xcite . these corrections can also be obtained@xcite by using the haldane expansion of the spin operators and retaining the terms up to @xmath315 as we did here . in the case of non - equivalent chains , the problem gets more complicated . the magnetic field still couples only to @xmath232 . however , when the magnetic field @xmath316 exceeds the field @xmath317 needed to close the gap in the @xmath266 modes , the appearance of a non - zero @xmath318 creates an effective magnetic field that couples to @xmath319 . if this magnetic field is not strong enough to close the gap in the @xmath220 modes , the system remains in a one - component luttinger liquid for fields @xmath316 not much stronger than the critical field @xmath317 . for @xmath316 sufficiently large , the generated effective field can close the gap in the @xmath98 modes leading to a two component luttinger liquid . since no experimental ladder system with a biquadratic exchange much larger than the quadratic exchange much larger that the quadratic exchange and made of two non - equivalent chains is presently available , this two - step transition to a two component luttinger liquid is unlikely to be observed experimentally . this case includes the spin tube and the mostovoy - khomskii model and is therefore the most relevant physically . this case has been discussed for two coupled @xmath71 symmetric chains in ref . . in the case where only one chain , say chain @xmath320 to fix notations , carries spin , the interaction with the magnetic field is given by a term : @xmath321 in the case of equivalent chains , we can use the same decoupling of sec . [ sec : equivalent_chains ] , eqs . ( [ hs ] ) and ( [ ha ] ) as in the preceding section . however , there is an important difference . now , we have the coupling to the magnetic field in the form : @xmath322 as a result , now the magnetic field couples to both @xcite @xmath232 and @xmath220 . moreover , the strength of the couplings is exactly the same . we start with the discussion of equivalent chains . then , we discuss non equivalent chain . we will show that in both case , the closure of the gap leads to a two - component luttinger liquid behavior in contrast with the usual spin - liquid systems that lead to a single component luttinger liquid behavior@xcite . we will also give the expression of correlation functions in the luttinger liquid phase . as a result of the symmetry between @xmath220 and @xmath232 , the gap in the symmetric and the antisymmetric mode close simultaneously , leading to a two component luttinger liquid ground state under strong enough magnetic field . contrarily to the case where the magnetic field couples to both 1 and 2 spins , there is no intermediate single component luttinger liquid phase . we expect that the magnetization @xmath323 behaves as @xmath324 close to the threshold . it is also important to note that , the two sine - gordon model being equivalent , one has for the fixed point hamiltonian @xmath325 and @xmath326 . as a result , the fixed point hamiltonian is invariant under any rotation in the @xmath327 plane . we can thus write the fixed point hamiltonian as : @xmath328\ ] ] where @xmath329 and @xmath330 . moreover , in the case of @xmath50 symmetry , we have @xmath331 . it can the be shown easily , using the refermionization procedure that @xmath125 for any magnetic field . equation ( [ eq : fp_mf_equiv ] ) has important consequences for the correlation functions , which are of the form ( for equivalent chains ) : @xmath332 \nonumber \\ \langle ( s^z_{\alpha}(x , t)-m_\alpha ) ( s^z_{\alpha}(0,0)-m_\alpha ) \rangle = \cos(\pi x ( 1 - 2m_\alpha))(x^2-t^2)^{-k_\alpha}+ \text{constant } \mbox { } \frac{1}{4\pi^2 } \left ( \frac{x^2+t^2}{(x^2-t^2)^2 } \right ) , \nonumber \end{aligned}\ ] ] where the index @xmath333 indicates the spin in chain 1 or 2 , @xmath334 is the magnetization , with @xmath335 ( total magnetization ) and @xmath336 . from the above expressions we deduce the following : the correlation function parallel to the field , @xmath337 , has a staggered part shifted from the wave vector @xmath251 to @xmath338 , while the correlation function perpendicular to the field , @xmath339 , has an unshifted staggered mode and the uniform magnetization mode shifted to @xmath340 . the correlation functions for the spin of type 2 , instead , are completely unaffected by the presence of an external magnetic field . in this section , we consider the case where the two chains are not necessarily equivalent . in particular , this is the case that is realized in the spin - tube under a magnetic field . in this case , is convenient to use the rotation , @xmath341 to bring the hamiltonian to the form : @xmath342 + \frac{(u_1/k_1+u_2/k_2 ) } 2 [ ( \partial_x \phi_s)^2 + ( \partial_x \phi_a)^2]\right . \nonumber \\ & + & \left . \frac { u_1 k_1 -u_2 k_2}{2 } \pi^2 \pi_s \pi_a + \frac { u_1 /k_1 - u_2 /k_2 } 2 \partial_x \phi_s \partial_x \phi_a \right]+\nonumber \\ & + & \frac{2g}{(2\pi a)^2 } \int dx [ \cos \sqrt{8 } \phi_s -\cos \sqrt{8 } \phi_a ] -\frac h { \sqrt{2 } } \int dx \partial_x(\phi_s+\phi_a)\end{aligned}\ ] ] shifting @xmath343 , renders hamiltonian invariant under the interchange of @xmath232 and @xmath220 . as a consequence , the gaps in @xmath232 and @xmath220 are _ identical _ and have to close simultaneously . therefore , as in the case of identical chain , there is a transition from a gapped phase into a gapless two - component luttinger liquid phase . the most general hamiltonian for a two - component luttinger liquid is : @xmath344.\end{aligned}\ ] ] in the case of coupled xxz chain , the effective hamiltonian ( [ eq:2component_ll ] ) can be simplified by making use of symmetry . we know that the original hamiltonian , eq . ( [ eq : rotated_mf ] ) is invariant under the transformation : @xmath345 this symmetry must be preserved by the renormalized hamiltonian . therefore , one must have @xmath346 and @xmath347 . as a result , the effective hamiltonian ( [ eq:2component_ll ] ) is diagonalized by returning to the original variables @xmath348 . the effective hamiltonian is therefore the following : @xmath349\ ] ] the absence of coupling between @xmath133 and @xmath134 in this hamiltonian is somewhat surprising . this can however be understood by the fact that the hamiltonian even in the presence of a magnetic field is invariant by a rotation by @xmath350 @xmath351 . thus , @xmath352 for any @xmath316 , implying that @xmath353 terms can not appear in the two component luttinger liquid hamiltonian . in contrast to the spin ladder case@xcite , the exponents @xmath354 in the hamiltonian eq . ( [ eq : effective_gapless ] ) are _ non - universal_. this can be shown in the following way . let us consider the case where @xmath355 and @xmath356 are small compared to @xmath357 , namely where the couplings in the two chains are nearly identical . in that case , we can neglect in a first approximation the terms @xmath358 and @xmath359 in the hamiltonian [ eq : rotated_mf ] . we are thus left with two decoupled sine - gordon models under a magnetic field . these sine - gordon models undergo a commensurate - incommensurate transition at a critical magnetic field . it is well known@xcite that the exponent at the transition assumes a universal value that renders scaling dimension of the cosine term equal to one . therefore , in our case , the universal value of the exponent at the transition is : @xmath125 . as a result , close to the transition the effective hamiltonian is : @xmath360 + \int \frac{dx}{2\pi } \left [ \frac { u_1 k_1 -u_2 k_2}{2 } \pi^2 \pi_s \pi_a + \frac { u_1 /k_1 - u_2 /k_2 } 2 \partial_x \phi_s \partial_x \phi_a \right].\end{aligned}\ ] ] returning to @xmath133 and @xmath134 , one obtains an hamiltonian of the form eq . ( [ eq : effective_gapless ] ) with : @xmath361 indicating that except in the case of equivalent chains , one should not expect universal exponents at the transition . this should be the case in particular for the spin - tube @xcite . it may provide an experimental test for the spin - orbital model@xcite of @xmath0 since the exponent @xmath362 controls the temperature dependence of the nmr relaxation rate@xcite . however , since the transition temperature between the gapped and the gapless phase in @xmath0 is @xmath363 , the magnetic field needed to close the gap should be of order @xmath364 which could make the experiment impossible . in @xmath365 , with @xmath366 , the situation is even worse . the spin and pseudospin operators are , @xmath367 \nonumber \\ \tau^z(x)&=&-\frac{\partial_x \phi_2 } { \pi } + \frac{e^{\imath \frac{\pi x } a}}{\pi a } \sin 2\phi_2 , \mbox { } \mbox { } \;\;\;\;\;\;\;\;\;\;\;\;\;\ ; \tau^+(x)=\frac{e^{\imath \theta_2}}{\sqrt{\pi a } } \left [ e^{\imath { \pi x } a } + \sin 2\phi_2 \right],\end{aligned}\ ] ] where @xmath368 is the total magnetization . in the case of the spin - orbital model , the spin - spin correlation functions are therefore given by the usual formulas@xcite . the situation is however more interesting in the case of the spin tube . using the formula ( [ eq : spin_stot_chirality ] ) , one has : @xmath369,\ ] ] where @xmath370 , and @xmath10 is the chain index . explicitly , @xmath371 this correlation function , which enters in particular in the calculation of the nmr relaxation rate , contains power - law divergences at wave vector @xmath372 but also @xmath373 due to the fluctuations of chiralities . we have presented a field - theoretical analysis of the low - energy physics of the anisotropic spin - orbital model and the three - leg ladder with periodic boundary conditions ( the spin tube ) in the strong interchain coupling limit . we gave a derivation of the field theoretical model from the lattice hamiltonian and then analyzed the phase - diagram using renormalization group equations . the system is found to exhibit a gapless phase , a spin - liquid phase or an ising antiferromagnetic phase depending on the microscopic couplings . the spin liquid ground state is two - fold degenerate , formed either by singlets of spins on even bonds and singlets of orbital pseudospins ( chirality in the spin - tube case ) on odd bonds or the other way round . the antiferromagnetic phase competes with the spin - liquid and we have discussed this competition briefly . the spin liquid phase obtains in the case of the spin tube with @xmath63 symmetry , and we discussed the nature of excitations above this spin liquid ground state . these excitations have spin @xmath374 and pseudospin @xmath375 . they are formed by introducing a free spin as a defect in the spin singlet pattern as well as a pseudospin in the pseudospin singlet pattern . these excitations lead to a kind of non haldane spin liquid analogous to the one discussed by nersesyan and tsvelik@xcite . an interesting consequence is the absence of a magnon peak at @xmath376 in the spin - spin correlation functions of the spin tube . this behavior could be tested in numerical simulations . we have investigated the effect of an applied magnetic field @xmath316 on the dimerized phase . a strong enough magnetic field @xmath302 causes the closure of the gap and the disappearance of dimer order . the resulting gapless phase is a _ two _ component luttinger liquid in contrast to the one component luttinger liquid that is observed in spin ladders . the exponents appeared to be _ non - universal _ at the transition point , in contrast with the spin ladder case@xcite . it would be interesting to obtain numerically the luttinger liquid exponents for the spin tube or the anisotropic spin orbital model . for the spin tube , we have also shown that new soft modes appeared in the spin - spin correlation functions above @xmath317 by comparison with the soft modes of the single chain . this is the result of the presence of soft chirality modes . this may be tested numerically . e. o. acknowledges support from nsf under grant nsf - dmr 96 - 14999 . we thank p. lecheminant and h. saleur for discussions and useful comments on the manuscript . @xmath377 permanent address : cnrs - laboratoire de physique thorique de lecole normale suprieure 24 , rue lhomond 75231 paris cedex 05 france we start from the hamiltonian : @xmath378 after the usual jordan wigner fermionization , this gives : @xmath379 the mean field approximation is obtained by taking : ( recall , @xmath380 is a chain index ) . @xmath381 the mean field equation are then : @xmath382 \nonumber \\ t=-\frac{2}{\pi}(1+\lambda t)\sqrt{1 + \left(\frac{\delta}{1+\lambda t}\right)^2}\frac{\mathbf{e}\left(1-\left(\frac{\delta}{1+\lambda t}\right)^2\right)}{1-\left(\frac{\delta}{1+\lambda t}\right)^2}\end{aligned}\ ] ] where @xmath383 are the complete elliptic integrals of the first and second kind respectively . for small @xmath186 , the mean field equations ( [ eq : mean_field_symmetric ] ) reduce to : @xmath384 and one obtains : @xmath385 this result from the mean field theory agrees with the prediction of bosonization since bosonization would predict the same essential singularity in @xmath186 at small @xmath78 , the interchain coupling being marginal . in this section , we consider a problem with a hamiltonian of the form : @xmath386 after a jordan wigner transformation , the hamiltonian becomes : @xmath387 the mean field approximation is now : @xmath388 the mean field equations are now : @xmath389 \nonumber \\ t_2=-\frac{2}{\pi}(1+\alpha t_2)\sqrt{1 + \left(\frac{\delta_1}{t_1}\right)^2}\frac{\mathbf{e}\left(1-\left(\frac{\delta_1}{t_1}\right)^2\right)}{1-\left(\frac{\delta_1}{t_1}\right)^2 } \nonumber \\ \delta_2=-\frac{2\alpha \delta_1}{\pi}\sqrt{1 + \left(\frac{\delta_1 } { t_1}\right)^2}\left [ 2\mathbf{k}\left(1-\left(\frac{\delta_1}{t_1}\right)^2\right)-\frac{\mathbf{e}\left(1-\left(\frac{\delta_1}{t_1}\right)^2\right)}{1-\left(\frac{\delta_1}{t_1}\right)^2 } \right ] \\\end{aligned}\ ] ] for small @xmath333 , the mean field equations for @xmath390 and @xmath205 reduce to : @xmath391 one sees that a finite bandwidth @xmath392 is produced for the @xmath27 spin waves at least for small @xmath333 . the gap equations admit solutions at small @xmath186 . one expects at weak coupling an essential singularity in @xmath393 . therefore , @xmath394 should go to zero for @xmath395 . this implies that one can write : @xmath396 one can solve the mean field equations for @xmath205 and @xmath206 . one obtains : @xmath397 we obtain therefore self - consistently a gap much smaller than the smallest bandwidth . as a consequence , the correlation length is much larger than the lattice spacing , which justifies a continuum approximation . the bosonization treatment is therefore valid for @xmath395 and close to the xy limit . in this section , we derive rge for the quantum sine gordon model using the method of knops and den ouden @xcite . the quantum sine gordon model is defined by the following lattice hamiltonian : @xmath398 - \frac{2g}{(2\pi a)^2 } \cos 4\phi_i \right\}\end{aligned}\ ] ] the cutoff in this model is only on space and not on time . the euclidean action of the quantum sine gordon model is obtained as : @xmath399=\int_{-\infty}^{\infty } \frac{d\omega}{2\pi } \int_{-\frac \pi a}^{\frac \pi a } \frac{dk}{\pi } \frac{u}{2\pi k } \left(k^2+\frac{\omega^2}{u^2 } \right ) |\phi(k,\omega)|^2 -\frac{2g}{(2\pi a)^2 } \int dx d\tau \cos 4\phi\end{aligned}\ ] ] where @xmath400 . one can integrate from @xmath401 to @xmath402 using a cutoff function @xmath403 . this action breaks rotation invariance in the @xmath404 space in contrast with the action of the classical sine - gordon model @xcite . instead of the sharp cutoff , one can use any cutoff function @xmath405 such that @xmath406 and @xmath407 . an example is @xmath408 . where : @xmath410}\frac { d \theta}{\pi } \nonumber \\ f_2(\rho)=\int_0^{2\pi}\cos^2 \theta e^{16 k \left[v(\rho,\theta)+ \frac 1 2 \ln \rho\right]}\frac { d \theta}{\pi } \nonumber \\\end{aligned}\ ] ] and : @xmath411 the eqs . ( [ eq : rge_momentum_cutoff ] ) have to be contrasted with the usual rg equations for the sine gordon model with a cutoff isotropic in the @xmath414 space . in the latter case , the equations are such that @xmath415 , and only @xmath284 is flowing under renormalization . in the case we have considered , both @xmath416 and @xmath284 are flowing under rg transformation . however , for @xmath244 , the rg equations ( [ eq : rge_momentum_cutoff ] ) are considerably simplified . since @xmath417 and @xmath418 , one has the following simplified rg equations : @xmath419 these rg equations are identical to the ones obtained with a cutoff isotropic in @xmath414 space . we conclude that in the vicinity of the bkt transition point , the anisotropy between space and time does not matter . where : @xmath429 - 2(k_1-k_2)\ln \rho } \nonumber \\ f_8(\rho)=\int_0^{2\pi } \frac{d\theta}{2\pi } e^{4\left[k_1 v(\rho,\theta ) - k_2 v(\rho,\theta,\frac{u_2}{u_1})\right]-2(k_2-k_1)\ln \rho } \nonumber \\ f_9(\rho)=\int_0^\infty e^{8k_1 \left[v(\rho,\theta)+\frac 1 2 \ln \rho\right]}\end{aligned}\ ] ] a. furusaki and s .- c . zhang , dynamical spin correlations in the heisenberg ladder under a magnetic field and correlation functions in the so(5 ) ladder , to be published in phys . 60 , number 1 , 1999 , cond - mat/9807375 .

progress in single molecule techniques has enabled observations of single steps in many motor proteins , and accurate measurement of the distribution of dwell times , i.e. , the periods of little or no motion between steps . examples are the forward and backward steps of processive molecular motors like kinesin @xcite , myosin v @xcite , cytoplasmic dynein @xcite or rna polymerase @xcite , and stepwise rotations in atp synthase @xcite and the flagellar motor @xcite . observations of steps and dwell time distributions offer a route to gain insight into the microscopic mechanism and detailed motion of such systems , beyond what is available through knowledge of average turnover rates alone . dwell times are examples of first passage times , which have been extensively studied in the theory of random walks @xcite . an important modeling step is therefore to formulate a first passage problem that describes the experimental situation . close examination of step trajectories with high time resolution from several motor proteins @xcite reveals that the steps are very rapid events compared to typical dwell times . in a discrete - state description , it is therefore reasonable to identify a step with a single transition , and the direction of the step with the direction of that transition @xcite . the identification of steps with single transitions follows naturally from the assumption that each state has a well - defined average position , and is a basic assumption in this paper . motor proteins are cyclic enzymes , but contrary to earlier assumptions @xcite , the steps and waiting times obtained in many single - molecule experiments are not well described in terms of the cycle completions in enzyme kinetics . this was recently demonstrated for models where a single cycle accounts for both forward and backward steps @xcite . in this case , the average number of forward and backward steps differ from the average number of completed forward and backward cycles , and the average dwell times between steps differ from the average cycle completion times . the differences can be significant even in conditions where almost no backward steps occur . the basic reason for this difference is that a motor with one or more intermediate states per cycle is in a different state immediately after forward and backward steps . therefore , an experimental trajectory of forward and backward steps does not give explicit information about completed forward and backward cycles . the different states obtained just after forward and backward steps also implies that consecutive step directions are correlated @xcite . this prediction will be confirmed in a later section , where we analyze stepping data from @xcite for a flagellar motor , and find the clear step - step correlations shown in ( b ) . in contrast , consecutive cycle completions are statistically independent @xcite . the observation that steps are correlated has important implications for the interpretation of stepping experiments , and also motivates further theoretical study of stepping statistics that goes beyond the assumption @xcite of independent steps and dwell times . in this paper , we extend the theory of @xcite in two ways . first , we consider the distributions of conditional dwell times instead of mean values . second , we consider a larger class of models , including motor detachments , substeps , and multiple pathways . our main results are a distribution symmetry for conditional dwell times , and a simple relation between the dissipated free energy per cycle , @xmath0 , and certain conditional stepping probabilities . for a large number of models relevant to describe reversible motor proteins , we get [ mainresult ] @xmath1 here , @xmath2 and @xmath3 are the probability density functions for the conditional dwell times between two consecutive forward and backward steps respectively , @xmath4 are the corresponding integrated probability functions , @xmath5 is the probability that a forward step is followed by another forward step , and @xmath6 that a backward step is followed by another backward step . the different types of dwell times are illustrated in , and simply means that the conditional dwell times @xmath7 and @xmath8 are random variables with equal distributions . however , does not say anything about the distributions of dwell times between steps of different directions , which in general have different distributions . could be position of a linear motor , net rotation of a rotary motor , or the net number of a substrate or product in an enzymatic reaction , and changes stepwise in steps of size @xmath9 . the staircase line is an idealized running average . ] equation holds also at finite average velocity . for example , a forward - moving motor will take mostly forward steps , but might eventually produce two consecutive backward steps . the dwell times between such @xmath10 and @xmath11 step pairs have equal distributions , although the probability of observing two consecutive backward steps might be very small for a motor with strong forward bias . in practice , the need to observe a significant number of @xmath10 and @xmath11 events to test or apply can be an experimental challenge . for example , atp driven motors like kinesin or myosin v typically have @xmath12 _ in vivo _ @xcite . this means that @xmath11 events are very rare ( see ) , unless an external load is applied @xcite . to extract qualitative information out of an experimental test of dwell time symmetry , one needs a characterization of the class of models which satisfy . we try to formulate a general characterization of this class . this means that some of the models within this characterization will not be realistic descriptions of biological systems . on the other hand , a large number of candidate models can be excluded if dwell time symmetry is not observed in some system . we find two sufficient assumptions for to hold , which we call _ strong coupling _ and the _ bottleneck property_. within the discrete state modeling framework that we use , these assumptions are easy to formulate , but can also be disposed of @xcite . the bottleneck property is an assumption about the model topology . it means that both forward and backward steps correspond to transitions to or from a single state in the mechanochemical cycle , which we call the bottleneck state . this ensures that the state of the system is uniquely determined after each observed step , independent of previous step directions . the bottle - neck property alone results in a particularly simple form for the step - step correlations , given in , which can be tested experimentally . the assumption of strong coupling , defined mathematically in , is related to microscopic reversibility . physically , it means that there is no futile free energy dissipation in the mechanochemical cycle . in addition to tight coupling in the usual sense @xcite , i.e. , a one - to - one correspondence between fuel consumption and forward steps , it includes the assumption that backward steps is tightly coupled to synthesis of fuel . as a result , motors with strong coupling are effectively one - dimensional systems , in the sense that only one reaction coordinate , e.g. , position , is needed to describe their operation . strong coupling might not necessarily hold for all motors . for example , the forward steps of myosin v are tightly coupled to atp hydrolysis , while backward steps independent of atp binding was recently reported @xcite . another example is kinesin , for which atp binding during steps in both directions has been reported @xcite . on the other hand , strong coupling might be possible in some rotary motors , at least under certain conditions @xcite . several recent works @xcite use hidden markov models to estimate kinetic parameters directly from experimental trajectories , and thus in principle utilize all information in the data and bypass the difficulties associated with step detection . however , this approach does not replace the need for a theoretical understanding of the capabilities and limitations of the underlying stochastic models . our results should provide useful guidance when applying these techniques . equation corresponds to a similar result in cycle kinetics @xcite . using the symbol `` @xmath13 '' to denote cycle time properties , one has @xmath14 where @xmath15 are the probability density functions for waiting times before forward ( @xmath16 ) or backward ( @xmath17 ) cycle completions , @xmath18 are the relative frequencies of completed forward and backward cycles respectively , and @xmath19 are one - way cycle fluxes @xcite . equation is based on similar assumptions as @xcite . however , the waiting times in do not describe the experimental dwell times observed in stepping experiments on motor proteins @xcite . distinguishing between cycle completion times and dwell times results in significant differences when modeling and interpreting stepping data . one example of such a difference is the possibility to calculate @xmath0 from observations of forward and backward steps . in this case , using the cycle completion result in on stepping data can give large systematic errors in the estimated @xmath0 , which can be avoided if is used instead . a related example concerns how to interpret the ratio of observed forward and backward steps , which has been measured for kinesin over a range of forces and atp concentrations @xcite . the force dependence of this ratio can be described as roughly proportional , where @xmath20 nm is the step length , @xmath21 is the applied load , and @xmath22 is a numerical factor significantly smaller than unity . the dissipated free energy per cycle depends on applied load as @xmath23 this means that predicts @xmath24 if steps are assumed to correspond to completed cycles . in contrast , gives no general reason to even expect an exponential behavior of the ratio of forward and backward steps @xcite , and suggests that @xmath25 is a more relevant quantity to study . as mentioned above , kinesin might not satisfy strong coupling , in which case one must look at more complicated models . in any case , making the distinction between steps and completed cycles is clearly important in order to interpret the experiments correctly . the rest of the paper is organized as follows . in the next section , we sketch a derivation of our results for the models studied in refs . @xcite . we also discuss step - step correlations . after that , we generalize our results to a broader class of models , and discuss detachments and multiple pathways in the mechanochemical cycle . we also generalize to the case of several detectable substeps . to illustrate our results , we then analyze stepping data for the flagellar motor , before the concluding discussion . detailed derivations are given in the appendices . states per cycle , @xmath26_l,[2]_l,\ldots,[n-1]_l$ ] . the model can be viewed as a biased random walk on the periodic lattice of states . numbers in brackets represent different states , and indices different cycles . state @xmath27_l$ ] denotes the same structural state as @xmath28_l\equiv [ k]_{l+1}$ ] , but with position or angular orientation @xmath29 ( for linear and rotary motors respectively ) shifted by the step length @xmath9 . arrows indicate the allowed transitions between neighbor states with transition rates @xmath30 , @xmath31 . fat solid arrows denote the major conformational changes that produce observable steps like those in . the other transitions are indicated by dashed arrows . in a motor protein , these transitions could for example be binding or release of ligands , or small conformational changes that are hidden in the experimental noise . ] in this section , we derive and some related results in the simplest case , i.e. , a sequential model . consider a model with a single cycle consisting of a sequence of @xmath32 states ( see ) , where we defined @xmath27_l$ ] to denote state @xmath33 in cycle @xmath34 , associated with position @xmath35 . this is a basic model in enzyme kinetics @xcite and has been used to describe motor proteins like kinesin @xcite , @xcite and f@xmath36-atpase @xcite . by construction , it satisfies both strong coupling and the bottleneck property . we denote the forward ( backward ) transition rates from state @xmath37 to adjacent states by @xmath30 ( @xmath31 ) as indicated in . the transition rates are positive and periodic , i.e. , @xmath38 , @xmath39 , and possibly functions of external loads and concentrations of various species in the surrounding solution . for the purpose of this discussion , the rates are assumed to be arbitrary positive constants , some of which can be tuned experimentally . we assume that transitions between different cycles , @xmath40_l\rightleftharpoons[0]_{l+1}$ ] , produces observable forward ( @xmath16 ) or backward ( @xmath17 ) steps . if the model describes an enzyme , the observable step could be the release or uptake of a product or substrate molecule . counting steps is different from counting cycle completions , which can be illustrated with the following thought experiment . consider the sequential model in with @xmath41 states per cycle , and a trajectory where the motor goes through the states @xmath26_0\to[1]_{-1}\to[0]_0\to[1]_0\to[0]_1 $ ] . this completes one forward cycle from @xmath26_0 $ ] to @xmath26_1 $ ] , but produces three steps : one backward ( @xmath26\to[1]_{-1}$ ] ) , followed by two forward ( @xmath26_{-1}\to[0]_0 $ ] , and @xmath42_0\to[0]_1 $ ] ) . due to such events , neither the number of steps nor the dwell times between steps are accurately described in terms of cycle completions . a different treatment is needed . the time evolution of the system is a random walk on the periodic one - dimensional lattice of states , where the average velocity may be nonzero . to simplify the notation , we now use @xmath37 to denote a state in any cycle , with the convention that @xmath37 and @xmath43 are equivalent states in different positions . the probability @xmath44 to be in state @xmath37 at time @xmath45 evolves according to a master equation @xcite , in this case @xmath46 state @xmath37 has free energy @xmath47 , and according to detailed balance , the free energy difference between two adjacent states are related to the transition rates through @xmath48 . from the periodicity of the rates we find @xmath49 the direction of the average drift is positive ( to the right in fig.[seqpath ] ) if the dissipated free energy per cycle , @xmath0 , is negative , and zero if @xmath50 @xcite . tracking the position produces a series of forward ( @xmath16 ) and backward ( @xmath17 ) steps separated by random dwell times , as sketched in . our aim is to describe the statistics of such a trajectory , i.e. , the fluctuations of the dwell times and the number of forward and backward steps , which can then be compared with experimental data . following ref . @xcite , we introduce the pairwise splitting probabilities @xmath5 , @xmath51 , @xmath52 , and @xmath6 , where @xmath5 and @xmath51 is the probability that a forward step is followed by a forward step or a backward step , respectively , and similar for @xmath53 . the splitting probabilities satisfy @xmath54 similarly , we introduce random variables @xmath7 , @xmath55 , @xmath56 , and @xmath8 for the conditional dwell times , where @xmath7 is the dwell time between two consecutive forward steps , and so on , as illustrated in . once the proper first passage problem to describe the dwell times has been formulated , explicit expressions for the splitting probabilities and mean conditional dwell times can be computed for arbitrary @xmath32 @xcite , using standard methods @xcite . we now describe the first passage problem for dwell times in sequential models , which we then use to derive . the dwell time symmetry is due to a one - to - one mapping between each @xmath11 event , i.e. , two consecutive backward steps and the dwell time between them , and a corresponding @xmath10 event . this is the key observation for the generalization of to more complex models in later sections . we model the splitting probabilities and probability distribution functions for the dwell times as a first escape problem from the interval of states @xmath57 . one way to approach this problem is the approach with absorbing boundaries @xcite , which means solving a reduced master equation for the states @xmath57 , with absorbing boundaries at both ends @xcite . from the assumption that steps are produced by the transition @xmath40_l\rightleftharpoons[0]_{l+1}$ ] , it follows that immediately after a @xmath16 step , the system is in the state @xmath58 ( @xmath59 ) , and just after a @xmath17 step , it is in the state @xmath60 ( @xmath59 ) . we write the reduced master equation in matrix form , , where the matrix @xmath61 has elements @xmath62 and @xmath63 . the element @xmath64 is the transition rate from state @xmath37 to state @xmath65 . the probability functions are given by the outgoing probability current , @xmath66 where @xmath67 indicates an initial condition @xmath68 , given by the direction the of the previous step ( see ) . normalization of the distributions requires that @xmath69 to derive , we show that the taylor series of @xmath70 and @xmath71 are identical up to a factor @xmath72 , from which follows . the actual calculation is given in appendix a , and we now discuss some of its consequences . first , it is interesting to note that the periodicity of the model is not necessary for the dwell time symmetry , only to get the simple relation between @xmath5 , @xmath6 and @xmath0 in . second , the @xmath73 probability distribution has a simple closed form for ( periodic ) sequential models . as we show in appendix b , @xmath74 where the @xmath75 are the eigenvalues of the matrix @xmath61 this is the distribution of a sum @xmath32 of independent exponential random variables with mean values @xmath76 . explicit expressions for all four conditional dwell time distributions for the case @xmath77 are given in ref . as expected from the calculations in appendix b , there is no simple relation between @xmath78 and @xmath79 . equation is only valid if the eigenvalues @xmath75 are distinct . this is the generic situation , since there are no symmetries or other reasons to expect degeneracies . however , is valid also in the degenerate case . finally , it is worth noting that the sequential models in this section can also be obtained as a discretization of overdamped one - dimensional diffusion @xcite in an arbitrary potential @xmath80 between two points a and b ( see ) . in this case , our results mean that the waiting time between last touch at a and first touch at b , and the reverse waiting time between last touch at b and first touch at a , have equal distributions . this pair of first passage problems was previously studied by @xcite . the fact that the motor is in a different state immediately after steps in different directions means that the step directions might be correlated . in this section , we formulate this observation mathematically , and point out some experimentally relevant consequences . let @xmath81 and @xmath82 be the probabilities that step @xmath33 in a trajectory is a forward or backward step respectively . in the simplest case of no substeps or detachments , the definitions of the pairwise splitting probabilities give @xmath83 equation describes the sequence of forward and backward steps as a markov chain @xcite , and the normalization constraints in leave two independent parameters in the @xmath84 transition matrix . the average frequencies @xmath85 of forward and backward steps have been measured for several motor proteins @xcite . in our model , those frequencies are given by the stationary distribution of @xcite , @xmath86 in contrast to the ratio @xmath87 of the number of forward and backward cycles in , the ratio @xmath88 of forward and backward steps is in general not equal to @xmath89 . looking back at , we see no general reason to expect the ratio of forward and backward steps to depend exponentially on the applied load . several experiments have divided the dwell times according to the direction of the following step @xcite . our model gives the forward and backward dwell time probability density functions as @xmath90 since @xmath78 and @xmath79 are in general not related in a simple way , one should not expect equal dwell times before forward and backward steps , in contrast to the case of forward and backward cycle completion times @xcite . equation implies that step directions are correlated , and gives the step - step correlation function @xmath91 where @xmath92 indicates the direction of step @xmath93 , and @xmath94 is an eigenvalue of the transition matrix in @xcite . the other eigenvalue is one . note that @xmath95 since @xmath96 . in addition , it is reasonable to expect @xmath97 , since a forward step following a backward step can be accomplished by a single transition , while the whole cycle must be passed before a forward step is followed by another forward step . inserting @xmath97 in , and using the normalization in , we get @xmath98 apparently , we should expect negatively correlated steps , @xmath99 . uncorrelated steps ( @xmath100 ) occur in systems with @xmath101 , i.e. , simple random walks , as well as some special cases of the extended models in the next section . an upper bound for the number @xmath102 of @xmath11 events is useful to estimate the regime where our results can be applied in practice . a rough estimate is given by @xmath103 hence , the number of steps @xmath104 required to get adequate statistics grows at least exponentially with decreasing @xmath0 . beyond the simplest descriptions in terms of a sequential model for the dominating pathway ( if there is one ) , more complicated situations with multiple pathways are possible @xcite . a general ( coarse - grained ) description of a motor protein would keep track of both the position and the consumption of fuel molecules , which requires the model to include an effective chemical coordinate in addition to the position of the motor @xcite . are the dwell time symmetry properties of sequential models valid for more general cases as well ? as we will see next , the answer is yes for a large class of models , in which the motion is still effectively one - dimensional . with the extended models , we look for general properties that imply dwell time symmetry , even if not all models with these properties are biologically relevant . we find that the strong coupling and bottleneck assumptions mentioned in the introduction are sufficient . this means that if dwell time symmetry is observed not to hold for a particular system , then one can conclude that the strong coupling or bottleneck properties are absent . to model such a system , one must go beyond the extended models presented below , for example along the lines discussed in refs . the special situation without strong coupling but with the bottleneck property is illustrated below for an example model ( fig . [ n2example2d ] ) . the extension of our results to detachments and observable substeps in later subsections is directly motivated by experimental observations of such events @xcite . in this section , we describe a large class of extended models which display dwell time symmetry in a little more detail , and indicate how the dwell time symmetry is derived . we use chemical kinetics , but write a master equation with arbitrary transition rates @xmath105 from state @xmath37 to @xmath65 , @xmath106 non - zero `` diagonal '' rates @xmath107 are allowed , for example to describe irreversible detachments of motors from their tracks @xcite , or other events that can be filtered out experimentally . we also need to specify the condition of strong coupling . one way is to demand that the transition rates around any closed loop @xmath108 of transitions satisfy @xmath109 this property is automatically fulfilled for sequential models , and makes it possible to define a free energy @xmath47 for each state @xmath37 , up to an additive constant . from detailed balance , we have @xmath110 for the free energy difference along the transition @xmath111 . hence , says that the sum of free energy differences along any closed loop is zero , which means that free energy dissipation that produces no net motion , e.g. , futile atp hydrolysis , is ruled out . ( note that the mechanochemical cycle , say , from a state @xmath112 to @xmath113 , is not a closed loop in our description . ) we also assume that no transition is irreversible , i.e. , , although the transition rates may be arbitrarily small . to reflect the cyclic operation of motor proteins , the transition rates are periodic with period @xmath32 , and there is a well defined free energy per period , @xmath114 an important step in the derivation of is to conclude that the state of the motor immediately after a step is independent of earlier steps . in a sequential model , this occurs because there is only one step - producing transition per cycle . this condition can be relaxed somewhat . it is sufficient to assume the bottleneck property mentioned in the introduction , i.e. , that all transitions corresponding to a step are either to or from a single state , the bottleneck state , as illustrated in . as the term indicates , the system must visit the bottleneck state each time it goes through the cycle . whether the bottleneck assumption holds must be determined for each system separately . for example , deviations from the step - step correlations predicted in means that the assumption is not valid . although the models considered in this section are more general than the sequential ones , several properties of the sequential models are retained . these include strong coupling , periodicity , and that steps ( roughly ) correspond to one transition in the enzymatic cycle . they therefore describe an effectively one - dimensional motion , where the motion along the spatial and chemical reaction coordinates @xcite are tightly coupled to each other for motion in both directions . hence , backward and forward motion proceeds in opposite directions along the same reaction paths , just as for the sequential models . states per cycle . all states inside the gray area belong to the same cycle . the bottleneck states are @xmath115 . black arrows denote transitions coupled to a mechanical step of length @xmath9 , and dashed arrows indicate transitions within a cycle , which are undetectable . it is not possible to go from one cycle to the next without either leaving or arriving at a bottleneck state . in general , transitions between any states within a cycle are allowed , and bottleneck states might have a transition to any state in the neighbor cycle . ] dwell time symmetry for the extended models with detachments can be derived using the same methods as for sequential models , and further details on that derivation are given in appendix c. just as in the sequential case , periodicity is only necessary to establish , while the dwell time symmetry , , follows also without assuming periodicity . examples of the non - periodic case are the subcycles in , which shows a model with two observable substeps per cycle . each step in the example satisfies the bottleneck property , but both bottleneck states are in the same subcycle . deriving dwell time symmetry for the different subcycles proceeds as in appendix c , with slightly different initial conditions . as mentioned above , detachments of a motor from its track are observed in several systems @xcite . we model detachments by introducing death rates @xcite as extra terms in the diagonal elements of the master equation and @xmath61 matrix of eqs . and . the presence of detachments affect the long time behavior of the model @xcite , as well as dwell time distributions and pairwise splitting probabilities . conditional detachment probabilities also have to be added to eqs . and . however , in the derivations of dwell time symmetry in appendix c , the transition rates always enter as ratios of forward and reverse rates , e.g. , @xmath116 in , so that diagonal elements ( @xmath117 ) always cancel . hence , remains valid even in the presence of detachments . and subcycle 1 has @xmath119 states , and subcycle 2 has @xmath120 states . both steps have the bottleneck property , although the bottleneck states @xmath58 and @xmath121 are both in subcycle 1 . if strong coupling is satisfied , both subcycles display dwell time symmetry . ] for some motors , the full forward and backward steps are divided into observable substeps @xcite . in this section , we show how the relation between splitting probabilities in can be generalized , if strong coupling holds and all substeps satisfy the bottleneck property . to start with , consider the system in , which produces substeps @xmath122 and @xmath123 during each cycle , and assume that each sub - cycle displays dwell time symmetry . from now on , we denote a forward step of length @xmath122 with @xmath124 , and similarly for the other substeps . keeping track of only the @xmath118-steps , we can analyze the system as in previous sections , using four splitting probabilities @xmath125 and dwell times @xmath126 , which satisfy . on the other hand , if we keep track of and discriminate between all substeps , we could instead measure eight splitting probabilities : @xmath127 , @xmath128 , @xmath129 , @xmath130 , and so on . summing over all paths between two consecutive @xmath131-steps , we get @xmath132 and similarly @xmath133 this means that the relation between free energy and splitting probabilities for a system with only one step per cycle generalizes to @xmath134 note that we summed over all paths between two @xmath131 steps without explicit reference to all paths that start with a @xmath131 step and end with something else . this means that is valid also for a system with detachments , or with several parallel pathways . for systems which go through substeps @xmath122 , @xmath118 @xmath135 in each cycle , we can use the same argument to relate the splitting probabilities of the full analysis ( all substeps included ) to one where the first substep ( @xmath122 ) is ignored : @xmath136 iterating this transformation to ignore substeps @xmath137 as well , we find the following relation for a cycle with @xmath138 visible substeps : @xmath139 in a complicated system with parallel pathways where a cycle can be completed using different sequences of substeps , holds for every such sequence separately . states per cycle , with two parallel pathways that produce a step in the @xmath29 direction . solid arrows are step - producing transitions , while dashed arrows indicate hidden transitions that do not produce an observable step . shaded boxes are bottleneck states . ] in this section , we consider a small but non - trivial model with 2 states per cycle , to present an explicit example that illustrates the results of the previous sections . the model has two parallel pathways as sketched in , and the steady state velocity and effective dispersion are known exactly @xcite . the bottleneck property is satisfied , and strong coupling is equivalent to @xmath140 computing the dwell time distributions and splitting probabilities ( see appendix d for details ) , we get @xmath141 the eigenvalues @xmath142 are given in appendix d , and @xmath143 . for this model , is obviously equivalent to the dwell time symmetry of . for the other two pairs of steps , we get @xmath144 we see that @xmath78 and @xmath79 differ both from each other and from the @xmath10 and @xmath11 distributions , independent of whether holds or not . the step directions are anti - correlated , since @xmath145 and @xmath93 respectively . shaded boxes are bottleneck states . projecting out the chemical coordinate as explained in the text , this model can be brought to the same form as in . in general , the resulting effective one - dimensional model does not satisfy the strong coupling assumption of . ] the example illustrates the logic of our results . strong coupling and the bottleneck property together are sufficient conditions for the dwell time symmetry . the bottleneck property alone is sufficient for the markov - chain description of step directions in , which remains valid even if is not satisfied . a possible reason for why the model in might not satisfy is illustrated in this is again an @xmath41 state model , but with an independent `` chemical '' reaction coordinate ( the superscript @xmath93 on the states ) , e.g , the number of hydrolyzed atp molecules @xcite . for example , the path @xmath26_l^m\leftrightharpoons [ 1]_l^m\leftrightharpoons [ 0]_{l+1}^{m+1}$ ] could be a step driven atp hydrolysis , while the reaction @xmath26_l^m\leftrightharpoons [ 0]_{l+1}^m$ ] could describe an atp independent step . this model can be transformed to the 1d model in through the projection @xmath146_l=\sum_m [ j]_l^m$ ] , which leaves the rates unaffected . however , the open - ended reaction path @xmath26_l^m\leftrightharpoons [ 1]_l^m\leftrightharpoons [ 0]_{l+1}^{m+1}\leftrightharpoons [ 0]_l^{m+1}$ ] is transformed into the closed loop @xmath26_l\leftrightharpoons[1]_l\leftrightharpoons [ 0]_{l+1}\leftrightharpoons [ 0]_l$ ] . since strong coupling places no constraints on the rates in the original model in , there is no reason to assume that holds for the projected model . in the effective 1d description , this loop then becomes a slip loop , which dissipates free energy without producing net motion . hence , the simple step - step correlations of is retained , but not the dwell time symmetry . in systems where it is not possible to measure the `` chemical position '' with single molecule precision , the resulting violation of dwell time symmetry is a useful test for such projected slip loops , if the bottle - neck property is satisfied . this property can in turn be ruled out , for example if the step - step correlations are more complicated than predicted in . the most interesting candidates for such tests seem to be the rotary motors driven by ion fluxes , e.g. , the f@xmath147 part of atp synthase , or the bacterial flagellar motor , since it is difficult to measure the ion flow with single molecule precision . we are not aware of any experiments on atp synthase under conditions that produces both forward and backward steps . data from a flagellar motor will be analyzed in the next section . @xmath148 in this section , we apply our theoretical results to analyze stepping data from a recent experiment with a chimaeric flagellar motor @xcite . as we will see , the data is consistent with the predicted dwell time symmetry and step - step correlations within the experimental uncertainty . we also estimate the free energy per cycle , and show that the estimate based on cycle completions can have a significant systematic error , compared to . the flagellar motor propels many swimming bacteria , by driving the rotation of flagellar filaments . each filament is driven at its base by a transmembrane rotary motor , powered by ion flux ( na@xmath149 in this case @xcite ) down an electrochemical gradient across the cell membrane . the motor is about 45 nm in diameter , and is believed to contain 13 torque - generating units @xcite . in this experiment , only one unit was active , and the flagellum is expected to switch between 26 distinct orientations per turn , corresponding to a step length of @xmath150 @xcite . we analyzed one trajectory with close to zero net velocity , and one with finite velocity . aided by chung - kennedy filtering @xcite to enhance the steps , and the step - finding algorithm described in ref . @xcite , the raw data was converted to stair - case stepping data , as shown in ( a ) . upon close inspection , the trajectory seems to be divided into intervals where the step lengths are consistent with each other , but out of phase with steps outside . this apparent drift might be due to dynamical exchange of the stator units @xcite that anchor the motor to the cell . we identified intervals of consistent stepping by inspection ( see ( a ) ) , and excluded the drift from our analysis by treating inconsistent steps as detachments , i.e. , each interval of consistent stepping was treated as an independent run . the result is summarized in tab . [ stepdir ] , and we now proceed to compare step - step correlations and dwell time distributions from trajectory a with the theoretical predictions . trajectory b contained too few steps to make such comparisons meaningful . . stepping is consistent with the theoretical step length in intervals interrupted by drift events ( arrows ) , and adjacent intervals are out of phase with each other . ( b ) step - step autocorrelation function as defined in . the measured correlation function ( squares ) is consistent with the theoretical prediction ( circles ) using conditional splitting probabilities from tab . [ stepdir ] . symbols are the same size as the estimated standard deviations . ( c ) distribution functions @xmath151 for the four different dwell times . ] the step - step autocorrelation function for trajectory a is shown in ( b ) . the prediction @xmath152 of is valid for trajectories without detachments . in order to compare theory and experiments , the splitting probabilities in tab . [ stepdir ] must therefore be renormalized by a factor @xmath153 to account for the fact that intervals of consistent stepping ( by definition ) contain no drift events . this gives the theoretical prediction @xmath154 a least squares fit to the experimental correlations gives @xmath155 , in agreement with , but inconsistent with uncorrelated step directions . having confirmed that the step - step correlations are consistent with , we go on to test the dwell time symmetry . the empirical distribution functions @xmath151 are plotted in ( c ) for trajectory a. error bars are standard deviations from bootstrap estimates @xcite . using the kolmogorov - smirnov test @xcite , we conclude , with 95% confidence , that @xmath56 have different distribution than @xmath7 and @xmath8 , and that @xmath55 have different distribution than @xmath8 . however , the test could not detect significant differences between the other three pairs ( @xmath73 , @xmath156 , and @xmath157 ) . this is again consistent with the theoretical predictions , and also shows that the statistics is good enough to detect differences between distributions that are not equal . it is also interesting to note from ( c ) that @xmath7 and @xmath8 are longer than @xmath55 and @xmath56 on average . this is reasonable , since @xmath157 events in principle only requires a single transition , while the system must go through a complete cycle to complete a @xmath10 or @xmath11 event . the ionic motive force was not measured independently @xcite , so it is not possible to compare the free - energy estimate of with an experimental value . however , we can compare our result with that based on , which was used in ref . the results are summarized in tab.[dgestimates ] . as expected , the free energy per cycle is close to zero in trajectory a , which has almost no net velocity , while trajectory b clearly has a finite free energy per cycle to drive the rotation . also note the differences in the estimates based on and , which show that the cycle - completion estimate can give rise to significant systematic errors in estimated free energies if applied to stepping data . the statistical properties of steps in reversible molecular motors are not the same as that of cycle completions in enzyme kinetics . to interpret stepping trajectories correctly , this difference must be taken into account . in this paper , we have extended the theory for dwell times in sequential models @xcite , and derived a symmetry relation for the conditional dwell time distributions for a large class of experimentally relevant models , including parallel pathways , visible substeps , and detachments . in contrast to the statistics of cycle completions @xcite , the dwell time distributions and splitting probabilities of steps depend on the step directions of both steps before and after the dwell period . the dwell time symmetry is a consequence of strong coupling , i.e. , tight coupling for both forward and backward steps , and a bottleneck property of the underlying kinetic scheme . this means that the motion of the system is essentially a one - dimensional random walk in the ( two - dimensional ) space of spatial and chemical reaction coordinates . this makes it possible to use our results to infer information about the coupling and efficiency of a system from kinetic data , without detailed assumptions about the underlying kinetic scheme . in this respect , it is similar to the relation @xmath159 between the randomness parameter @xmath160 and the number of steps @xmath32 in the mechanochemical cycle of molecular motors @xcite . an alternative to the discrete formalism used here , is to include continuous spatial degrees of freedom , as is common for example when modeling ratchet - type motor mechanisms @xcite . since such models can be discretized @xcite , it is in principle possible to inquire about strong coupling and the bottleneck property in continuous models as well . an example of this is diffusion in one dimension ( see fig . [ contlim ] ) , which strictly satisfies strong coupling and the bottleneck property . an interesting problem for further research is to quantify how well the dwell time symmetry is preserved in system with small deviations from these assumptions . we analyzed stepping data from single motor subunits in a flagellar motor @xcite . the data seems to be consistent with the predicted dwell time symmetry and step - step correlations , although only one trajectory with almost zero velocity had enough steps to make such a comparison meaningful . the form of the step - step correlations is consistent with an underlying kinetic scheme that satisfies the bottleneck property . so , does the dwell time symmetry in fig . [ flagellapc ] indicate that the flagellar motor subunits are strongly coupled to the driving ion flow ? another possibility is that the steps are equilibrium fluctuations , as indicated by the low @xmath0 estimated from the stepping probabilities . in equilibrium , there is no free energy change associated with ion transport , which means that dwell time symmetry could be obtained also for a system with loose coupling . to say anything about the coupling in the flagellar motor , the dwell time symmetry must be tested in a regime with finite ( and constant ) velocity and ionic driving force . this is in principle a question of observing more steps in such conditions , but a systematic way to identify and separate steps from drift events would also be useful . the step - step correlations predicted in might be present also in systems that violates the dwell time symmetry . deviations from this form would say something about the topology of the kinetic pathways in such systems , and it would probably be useful to apply our analysis also to kinesin and myosin v. an accurate theoretical model is often crucial in order to correctly interpret experiments on such complex systems as motor proteins . many previous theoretical works on steps and dwell times in molecular motors @xcite derive or assume descriptions where consecutive steps or cycles are statistically independent of each other . as we have demonstrated , this does not apply to the flagellar motor of ref . we have also presented a large class of simple models where easily accessible quantities like the dwell times and step directions are correlated . we expect our results to be of practical use in both data analysis and design of experiments . in particular , further experimental efforts are motivated in order to detect correlations and collect significant statistics for both forward and backward steps . we now give a detailed derivation of for a sequential model . the initial condition is the state immediately after a step , and can be written @xmath161 to derive , we compute @xmath162 and @xmath163 using and the initial conditions in . this gives @xmath164 where the summation over @xmath165 goes from @xmath58 to @xmath60 . the same calculation for @xmath166 yields @xmath167 the products of matrix elements in eqs.([seqffdiff]-[seqbbdiff ] ) correspond to @xmath168-step paths between states @xmath58 and @xmath60 , plus the extra escape step . for sequential models , the shortest such path is @xmath32 steps , so the first @xmath60 derivatives are zero . for @xmath169 , we note that there is a one - to - one correspondence between the nonzero terms in and . for each term including a path from @xmath58 to @xmath60 in , there is a corresponding term for the reverse path from @xmath60 to @xmath58 in . the ratio of two corresponding non - zero terms is @xmath170 going back to for the elements of @xmath61 , and using @xmath171 by periodicity , we see that @xmath172 for all pairs of corresponding terms . moreover , @xmath173 , as there is no transition which produces two steps at once . therefore , @xmath174 for all @xmath175 . the underlying reason is that the sum of free energy changes is the same along all possible paths going forward one cycle from @xmath58 . since @xmath176 and @xmath166 are integrals of the solutions of the reduced master equation , which is a finite system of ordinary differential equations with constant coefficients , they are smooth functions which have taylor series . hence , together with the normalization in implies . a similar correspondence between paths was used in ref . @xcite to derive . the periodicity is necessary to get the simple relationship between @xmath5 , @xmath6 and @xmath0 in , but not to establish the dwell time symmetry in . without the periodicity , is replaced by @xmath177 to derive explicit expressions for @xmath178 and @xmath179 for sequential models , we first note that the occupation probabilities @xmath44 in are the solution of a system of linear ordinary differential equations with constant coefficients , namely the elements of @xmath61 in . if the eigenvalues @xmath180 of @xmath61 are non - degenerate , the solutions have the form @xmath181 . note that the eigenvalues have negative real parts , to guarantee that the system eventually leaves the interval . we make the ansatz @xmath182 where the term 1 ensures proper normalization . we need @xmath32 equations to determine the coefficients @xmath183 . these are @xmath184 for @xmath185 . the case @xmath186 was argued just before , and @xmath187 follows from the argument between eqs . and . the resulting system of equations for @xmath183 is of the vandermonde type , @xmath188 solving with cramer s rule and setting @xmath189 we arrive at . the distributions of @xmath55 and @xmath55 can be computed with the same ansatz , but the derivatives @xmath190 and @xmath190 are non - zero for @xmath191 , so the results are more complicated and there is in general no simple relation between @xmath78 and @xmath79 . in this appendix , we derive for a periodic extended model like the one in , generalizing the first passage problem as we go along . the derivation proceeds much as for the sequential model , by solving a reduced master equation for the states @xmath57 , @xmath192 with absorbing boundaries at @xmath193 and @xmath32 , and allowing @xmath194 to model detachments . note that the sum in have at most @xmath195 terms , due to the bottleneck property . after a forward step , the system is in state @xmath58 , while it could in principle be anywhere in the cycle after a backward step . the initial conditions , describing the distribution of states just after a @xmath67 step , are therefore @xmath196 since there are several escape transitions for each step , for the dwell time probability functions is generalized to @xmath197 we now proceed to compute derivatives of @xmath198 and @xmath199 , and note that by eqs . ( [ micrev]-[gendg ] ) , the free energy difference @xmath200 is independent of path : @xmath201 @xmath202 from the periodicity of @xmath105 expressed in , we get @xmath203 , which proves and thereby . the derivation for the case when backward steps ( instead of forward steps ) end in the bottleneck state is analogous . note that our definition of the bottleneck property means that there are no transitions that produce two steps at the same time . this rules out discontinuities at @xmath204 in the distribution functions @xmath151 , and justifies our use of taylor expansions . in this appendix , we solve the example model in fig . [ n2example1d ] in some detail . in addition to deriving eqs . ( [ ffbbexample]-[n2xgamma ] ) , this also illustrates a method that can easily be generalized to larger systems and implemented on computer algebra systems . for @xmath205 states per cycle , the eigenvalues @xmath75 must be calculated numerically . with this exception , the method gives explicit expressions for the dwell time distributions in the real time domain , that can be directly compared to experimental dwell time histograms . the first escape problem for the model in is governed by a reduced master equation as described in , with the matrix @xmath206 the eigenvalues of @xmath61 are @xmath207 and satisfy the relations @xmath208 the initial conditions after forward and backward steps are given by , and the conditional dwell time distributions are given by : @xmath209 using the ansatz in together with eqs.([aic]-[apdfdef ] ) to compute @xmath210 and @xmath211 , we get the following systems of linear equations : @xmath212 solving this and using @xmath213 , we eventually arrive at the probability density functions given in eqs.([ffbbexample]-[fbbfexample ] ) . finally , we need expressions for the splitting probabilities , which we derive using the adjoint equations @xcite . let @xmath214 , with @xmath215 , be the probability that a system starting in state @xmath37 will next produce a @xmath67 step . starting with @xmath216 , we note that from state @xmath58 , a backward step can be accomplished either by the next transition being a backward step , which has probability @xmath217 , or by jumping to state @xmath218 and from there eventually get a backward step . the last possibility has total probability @xmath219 , so that @xmath220 next , we multiply with @xmath221 and rearrange the terms . applying the same procedure for the other @xmath214 , we finally get @xmath222 to get the pairwise splitting probabilities in eqs.([ffbbexample]-[n2xgamma ] ) , we first solve for the @xmath214 , and then weight them according to the initial conditions of : @xmath223 we are grateful to richard berry and yoshiyuki sowa for helpful comments about the stepping data and flagellar motors in general , to michael e. fisher , denis tsygankov and r. dean astumian for stimulating discussions , and to hong qian for sharing ref.@xcite with us prior to publication . financial support from the royal institute of technology , the wallenberg foundation ( m.l . ) and the swedish research council , grant 2003 - 5001 , ( m.w . ) is gratefully acknowledged .

surface plasmon polaritons ( spp ) are electromagnetic bound modes responsible for the transport of light at the interface separating a metal from a dielectric . their ability to confine light at an air - dielectric interface offers the prospect of developing a new technology consisting of photonic nano - devices@xcite . active research is currently focusing on the possibility of achieving control over the propagation of spps by means of optical elements that would couple or decouple light to them@xcite . in order to conceive optical elements ( lenses , mirrors , beam - splitters ) able to manipulate spps propagation , we need to learn more about the interaction of surface plasmons with a sub - wavelength modification of the underling dielectric metal interface . indeed the interaction of spps with surface sub - wavelength defects on a metal surface is of great interest from a theoretical standpoint.@xcite + in this article we shall study scattering of spps by a _ shallow _ surface defect . we will consider both indentations of the metal surface ( grooves ) and protrusions on it ( ridges ) . we shall only deal with bi - dimensional defects , which are deemed infinite in one dimension parallel to the interface ( the @xmath0-direction ) . different aspects of this problem have been studied before with a variety of numerical techniques @xcite . here we present a systematic comparison between the different scattering coefficients and provide both analytical expressions and qualitative explanations . + it must be noted that in a previous work we presented such a comparison@xcite but within an approximate numerical scheme . within that framework it was found that ridges and grooves exhibited the same scattering , whenever they are shallow enough . here we will revise that result , which turns out to be valid only for long ( elongated ) defects . the mistaken outcome of ref . [ ] for short defects may be traced back to the breakdown of the assumption of small curvature in the defect geometry that was made there . in this paper we solve the maxwell equations through a discretization method , which does not assume the previous approximation and whose accuracy depends only on the discretization mesh . we found that , as in the previous work , long asymmetric ridges or grooves with the width much larger than the depth , do scatter very similarly . however square shallow defects manifest a different scattering efficiency , differing in the relative radiative loss and radiation pattern . the lack of distinction between these two cases did not emerge in the previous approximate treatment . on the whole the problem needs to be revisited so as to : @xmath1 substantiate why the approximate result does work in the case of elongated defects , @xmath2 point out what is the correct result in the case of shallow and short symmetric defects , and @xmath3 explain qualitatively how the scattering properties of short and shallow symmetric defects are gradually transformed into the scattering properties of elongated defects , as the aspect ratio of the defect increases . + this paper is organized as follows . in sec . [ s1 ] we state the basic assumptions on the scattering system as well as the solution method . in sec . [ coeff ] we rearrange the asymptotic expansions of the far - field to produce the scattering coefficients . namely we express the far - field and the related poynting vector in terms of the field inside the defect . still in this section we look at an approximation for the scattering coefficients of shallow ridges . in sec . [ ray ] we explain that , in general , we can not quantitatively represent a scatterer ( however small ) by one mesh . we explain how we associate a small symmetric ridge or groove to a point dipole . in sec . [ num ] we look at exact numerical results for the scattering of shallow defects of various horizontal lengths . we analyze these results and , in the case of square defects , we associate a ridge to a vertical dipole and a groove to a horizontal dipole . in sec . [ hv ] we produce an analytical model that explains the radiation pattern of the surface plasmons scattered by small square ridges and grooves . in sec . [ needlesection ] we look at the solutions for the case of shallow and long defects and we present a clear - cut interpretation to support the results of the previous treatment@xcite . finally in sec . [ transition ] we explain qualitatively that the aspect ratio of the defect determines the orientation of the field induced in a shallow defect . the considered defects are infinite in the @xmath0-dimension and shallow with depth @xmath4 , where @xmath5 is the free space wavelength . the defects are going to be illuminated by a monochromatic surface plasmon at normal incidence @xmath6 , associated to an impinging energy flux @xmath7 , defined and derived in appendix [ sp ] . therefore , only radiation into p - polarized(tm ) waves needs to be considered . after we drop , out of symmetry , the @xmath0-dependence on the whole problem the field is expressed as : @xmath8 . the wavevector in vacuum is : @xmath9 , where @xmath10 . the material making the slab shall be lossless silver@xcite , that is : @xmath11 . absorption is neglected as we consider non - resonant defects with widths much smaller than the spp propagation length . + as represented in fig . [ scheme ] , we shall be expressing the source orientation in a cartesian basis @xmath12 , and the scattered fields in a right - handed orthogonal polar basis : @xmath13 finally a question of notation : throughout we shall refer to a bi - dimensional point - source simply as a dipole , but it is meant that their emission as all of the fields are cyclical in the @xmath0-direction . as represented in fig . [ scheme ] , each object lying in the vacuum semi - space shall be labeled by the superscript @xmath14 while any object lying in the metal shall be labeled by the superscript @xmath15 . in particular , scattering quantities related to ridges have the superscript @xmath14 while the ones related to grooves have the superscript @xmath15 . the field within the cross - sectional area of the ridge is labeled @xmath16 and the one within that of the groove is labeled @xmath17 . a groove is a one dimensional defect located in the metal and labeled by the index @xmath18 ] the green tensor approach is a standard method to solve electromagnetic scattering problems @xcite . our first task in this section , is to arrive at an explicit expression for the scattered electric far - field . this is attained by propagating the field induced by a dipole density @xmath19 ( where @xmath20 ) inside the area of a ridge , to a point @xmath21 very far from the source . for a groove we have the same relation between polarization and field ( except for a change of sign ) @xmath22 . to propagate the field from any of the two , we use the standard formula @xcite : @xmath23 + where @xmath24 is the green tensor for the air - metal background . the green tensor propagates the emission of a point source at @xmath25 to the distant point of the @xmath21 . one of the advantages of the green tensor technique is that once the fields inside the defects @xmath26 ( and thus @xmath27 ) are computed numerically the asymptotic expansions of scattered fields become analytic . this takes us to our second task , which is making a direct connection between the orientation of the induced polarization inside the defects and the far - field radiation pattern , and in so doing define the scattering coefficients . + first of all , finding the scattered electric far - field @xmath28 requires the asymptotic expansions of the green tensor . the derivation is sketched in the appendix [ gt ] . in what follows we give some simplifying rearrangements that will let us focus directly on the angular radiation pattern of surface defects . the asymptotic green s tensor in the radiative zone for either a ridge or a groove can be written in a compact form as : @xmath29 & \times&e^{-i k_z^\nu z'}\;\g_{\infty}^{(\nu)}(\alpha,\r').\end{aligned}\ ] ] in such form we can factor the asymptotic scalar green function out of the dyadic part of the green tensor . from eq.([es ] ) , the direction of @xmath30 results from superposition of @xmath31 , the emission from all induced point polarization elements , or dipole density elements . yet the direction of each contribution @xmath32 must be independent of @xmath33 . in other words since electromagnetic waves are transverse waves in vacuum , far from their source , the field emitted by a dipole @xmath31 must be proportional to @xmath34 . in fact using the standard asymptotic expansions ( see the appendix [ gt ] ) we can write : @xmath35\;\u_\alpha .\end{aligned}\ ] ] where for a ridge : @xmath36 and for a groove : @xmath37 the vectors @xmath38 are p - waves defined in vacuum , while @xmath39 are defined in the metal . a reminder of their expressions at normal incidence , in terms of the angle @xmath40 of fig . [ scheme ] , is reported in the appendix [ pmodes ] , along with the expression for the fresnel reflection and transmission coefficients : @xmath41 , @xmath42 . + we are now in a position to write the expressions for the radiative fields . plugging eq.([asyp ] ) and eq.([inf ] ) into eq.([es ] ) we can separate the electric far field dependence into its radial and angular parts as : @xmath43 here the angular amplitude can be written as : @xmath44 where @xmath45 is the scattering coefficient into radiative - modes : @xmath46 in the last expression the scattered field in the far zone consists of a cylindrical wave , transverse to the direction of propagation @xmath47 , and with a net angular amplitude determined by the integral over the source region @xmath48 . the latter is actually the important bit in the formula as its squared module determines the radiation pattern . as seen from eq.([angularfield ] ) this angular amplitude results from the superposition of each scattering element taken with its own amplitude , phase and optical path in analogy to how an antenna array determines its effective radiation pattern . the radiation is given by the intensity or poynting vector in the far field . accordingly the differential angular scattering cross - section is : @xmath49 finally , the net radiative loss @xmath50 is defined as the integrated angular radiation : @xmath51 whenever the height of the defect is small enough , typically much smaller than the wavelength of the incident light , we can make the approximation @xmath52 . that allows for some simplification for the angular amplitude of a scattering element _ above the surface_. consider : @xmath53 hence , for shallow defects the green tensor dependence of eq.([asyp ] ) is entirely given by the exponential factors @xmath54 , for both a source in the vacuum semi - space and a source in the metal semi - space . indeed this turns out to be a major simplification for the relative amplitude of the scattering elements in the air semi - space , which we shall perform in detail section[hv ] . + before that we need to highlight the relation between the green tensor of a defect on the metal slab and in the metal slab , under this approximation . such relation emerges from the boundary conditions for the green s tensor at the interface , which are : @xmath55 \cdot\u_x=0&\nonumber \\\end{aligned}\ ] ] @xmath56\cdot\u_z=0.&\nonumber\\\end{aligned}\ ] ] notice that , in the unperturbed system , space is translationally invariant in the horizontal direction @xmath57 and this is reflected in is the @xmath57-component of the vector in eq.([vector ] ) . because of eq.([asyp ] ) and eq.([inf ] ) , we can turn eq.([gsurfacex ] ) into : @xmath58 the presence of surface charges at the interface implies , from eq.([gsurfacez ] ) , that the @xmath59-components of the vector @xmath60 on either sides of the interface have the relation : @xmath61 let us derive the scattering coefficient into surface plasmon modes . note that , in this one dimensional problem , scattering will be into both the forward surface plasmon @xmath62 , propagating in the positive @xmath57 direction and the backwards plasmon @xmath63 propagating in the negative @xmath64 direction , as defined in the appendix [ sp ] . the emission by a point dipole or a point polarization element must result into a plasmon final state : @xmath65 , as shown in the derivation sketched in the appendix [ gt ] . the asymptotic greens tensor for a source _ upon _ ( @xmath14 ) or _ in _ ( @xmath15 ) the metal is : @xmath66&{\bf e } _ { spp\pm}({\bf r}).\end{aligned}\ ] ] notice that @xmath67 complies with eq.([gsurfacex ] ) and eq.([gsurfacez ] ) . consequently the field of the scattered plasmons are : @xmath68 \scap_{p\pm}^{(\nu)}&= & \int_{a^{(\nu ) } } d\r'\;{\bf e } ^{*}_{spp\pm}(\r')\cdot { \bf p } ( \r')\end{aligned}\ ] ] furthermore the magnetic field related to the field scattered into spps is : @xmath69 where @xmath70 is the magnetic field of a spp , as proved in the appendix [ sp ] . + now , if the the source @xmath71 is produced by an incident surface plasmon field ( as is our case ) , we can define the scattering cross - section of into spps as : @xmath72 finally , we can define the total scattering cross - section , which in the lossless case is equivalent to the extinction cross - section : @xmath73 next we are going to develop solutions to point sources in a metal plane background . however one question may be raised : how do we associate the field induced by a surface plasmon inside a ridge or a groove to a point dipole ? the answer is the argument of this section . + when the field inside a defect is obtained by mesh discretization we assume that the field inside a single mesh is uniform , and deviations from the field at its center are deemed negligible . yet , in general , the field in a defect , can not be represented by the field at its center alone . let us explain a little bit further this point . + for simplicity let us consider a defect in a homogenous medium with dielectric constant @xmath74 , but the argument is the same in other backgrounds . as usual@xcite , the field at every mesh is found by solving self - consistently a system of @xmath75 coupled equations : @xmath76where @xmath77 and @xmath78 and @xmath79 is the field at the mesh center . @xmath80 is a term related to the depolarization of light and comes about from the quasi - static contribution of the green tensor . @xmath81 is a correction term to the green tensor in the region of the scatterer useful to improve the accuracy of the calculation , when the inhomogeneity is discretized @xcite . + in practice , the number of mesh points @xmath75 is increased until the calculation converges to the required precision . then scale variations @xmath82 of @xmath83 are properly represented in the solution . in the rayleigh limit , for a defect of area @xmath84 so small that @xmath85 , the scatterer behaves like a point source or a point dipole and the background field ( in this case the illumination ) can be considered uniform over @xmath84 : @xmath86 . exceptionally , for a circular defect in a homogenous medium with dielectric constant @xmath74 , the net field at any point @xmath87 converges to : @xmath88 in vacuum , and a circular defect of the same area , illuminated by a plane wave . the dashed line represents extinction coefficient calculated with the mie theory for the circle . the dielectric constant in the defect is @xmath89 at the wavelength of @xmath90 . the inset represents the geometry of a discretized circle when inscribed in a square represented by 30x30 mesh points . ] this is because for the field inside an infinitesimal ( very sub - wavelength ) circular shape is actually uniform and thus scattering by such circular defects can be described by one mesh . in fact the extinction coefficient@xcite can be derived from the field at the center alone : @xmath91=\\ \label{xtn2}&= & a \im\left [ \delta\varepsilon \e^{*}_b\cdot\e \right]\end{aligned}\ ] ] to prove this numerically we have calculated @xmath92 for a cylinder represented by a single mesh , as in eq.([mie ] ) , and illuminated by a plane wave . first of all we have checked that the one - mesh cross - section of eq.([xtn2 ] ) , coincides with the mie theory result . secondly , we have subdiscretized the cylinder into square meshes as rendered in the inset of fig . [ cylinder3 ] . as also rendered in the figure , applying eq.([ls ] ) we found that , as the number of meshes grows , the scattering cross - section calculated by the collection of meshes eq.([xtn1 ] ) converges to the initial value of one single mesh of eq.([xtn2 ] ) . however the field inside of a square scatterer can never be uniform if it is to satisfy real boundary conditions even in a homogenous medium or vacuum . thus , it can not be faithfully described by one mesh . this is illustrated in fig . [ cylinder3 ] , which renders the extinction coefficient for a square defect of the same area as the circle . as it turns out , the converged value is @xmath93 larger than that obtained by the one - mesh approximation . remarkably this error is not reduced with the defect size : we obtained the same error for squares with side @xmath94 or @xmath95 . this is just for reference in the optical range , since we found that the error actually depends on type of defect and on the dielectric constant . + however , even if the field is not uniform , a small defect in the rayleigh limit can be represented by a point source at the center of the mesh , with its field equal to the average field over the mesh @xmath96 . + indeed if the variation of @xmath97 is negligible over the area of the defect we have : @xmath98\end{aligned}\ ] ] so the object behaves as a point - dipole @xmath99 . + the previous results were for a homogeneous background , but they also hold for the inhomogeneous one considered in this paper . we find that , for a defect above the surface in the optical range , the relative error is about @xmath100 , while it can reach @xmath101 for a defects below the surface . + with very small non - elongated ridges and grooves , such that @xmath102 , the equivalent point dipoles are attained by averaging the fields over the area of the defects as follows : @xmath103 accordingly if we set @xmath104 eq.([angularfield ] ) and eq.([ep1 ] ) for small non - elongated defects become : @xmath105\scap_{p\pm}^{(\nu)}(\alpha)&= & \left[{\bf e } _ { p\pm}^{(\nu)}(0)\right]^ * \cdot\p^{(\nu)}\end{aligned}\ ] ] as an illustration consider a square ridge and a groove of side @xmath106 . we have calculated the scattering into radiative modes and spps without associating the defect to a point dipole but rather using eq.([sigmaout ] ) and eq.([spmas ] ) . in this case the major task is computing the green s tensor for the plane metal surface required to attain the exact field within the surface defect . this can be achieved following the prescriptions of ref . + similar numerical results for the case of shallow grooves were found in ref . [ ] using a different computational technique . + the out of plane radiation pattern of a surface plasmon scattered by such defects is given in fig . [ radpati ] . + calculations show that , for symmetric defects , the net radiative loss is greater for a groove than for a ridge . this is so because , while both the scattering into spps and the radiation close to the surface ( at @xmath107 ) are similar , their radiation patterns greatly differ normal to the surface ( @xmath108 ) , where the groove radiation is maximum while the ridge radiation goes to zero . angular radiative cross section ( eq.([sigmaout ] ) ) and surface plasmon cross section ( eq.([spmas ] ) represented by the almost horizontal lines at @xmath109 and @xmath110 ) , for square defects with 10 nm side , illuminated by a spp on silver at @xmath111 . the scale is linear but the units are arbitrary . each concentric line indicates an equal increment of the cross - sections , from the minimum at the the center to the maximum at the outermost . ] ) ) and surface plasmon cross section ( eq.([spmas ] ) represented by the almost horizontal lines at @xmath109 and @xmath112 ) , for rectangular defects with @xmath113 height and 50 nm width , illuminated by a spp on silver at @xmath114 . the scale is linear but the units are arbitrary . each concentric line indicates an equal increment of the cross - sections , from the minimum at the the center to the maximum at the outermost . ] the ridge radiation pattern is distributed into two lobes on either sides of @xmath108 but the groove radiation pattern forms a single lobe . this is one of our main result and shall be analyzed in detail in the next section . the result is not in agreement with those obtained in the approximate treatment ref . we associate the discrepancy to the breakdown of the condition that the curvature of a short and shallow defect does not vary rapidly , used in that work . + notice the fraction of energy scattered into spps , i.e @xmath115 of eq.([spmas ] ) , is large . the values of @xmath115 are represented by the horizontal lines of fig . [ radpati ] ( the concentric lines indicate their amplitude in a linear scale and in arbitrary units , say , for instance from @xmath116 at the center to @xmath117 at the outermost ) . for both ridges and grooves @xmath118 and @xmath119 are roughly equal . however in the case of ridges , @xmath120 is greater than the maximum value of the scattering cross - section into radiative modes @xmath121 of eq.([sigmaout ] ) , by a factor slightly greater than @xmath122 . for grooves , @xmath115 is greater than the maximum value of @xmath123 by a factor slightly smaller than @xmath122 . + let us now keep the defects height at @xmath124 and enlarge the width @xmath125 . fig . [ rect1 ] renders the radiation pattern for a rectangular defect of width @xmath126 nm ( @xmath124 ) . the emergence of directivity in the out of plane radiation , is part of a transitional behavior , in which the radiation patterns tend to align and , simultaneously , one of the lobes is shrunk while the other is blown up in the ridge radiation . notice that the scattered energy into spps exhibits the same directivity , going mainly in reflection . eventually , if we keep enlarging the defects until they are considerably asymmetric the radiation patterns for both ridges and grooves tend to be single overlapping lobes ( see fig . [ rect2 ] ) . noticeably , the scattering into spps is greatly reduced . such similarity is explainable in the approximate framework presented in ref . [ ] which turns out to be quite acceptable in this limit of large enough defects , as we shall substantiate in sec . [ needlesection ] . in sec . [ transition ] we shall account qualitatively for the transition observed in fig . [ rect1 ] , explaining why the radiation pattern changes when the defects are enlarged . ) ) and surface plasmon cross section ( eq.([spmas ] ) represented by the almost horizontal lines at @xmath109 and @xmath112 ) , for rectangular defects with @xmath113 height and @xmath127 width , illuminated by a spp on silver at @xmath90 . the scale is linear but the units are arbitrary . each concentric line indicates an equal increment of the cross - sections , from the minimum at the the center to the maximum at the outermost . [ rect2 ] ] the equivalence between non - elongated subwavelength defects and point dipoles gives us a chance to investigate in depth the individual radiation pattern of a single scattering element . [ mode ] shows the averaged the field inside the 10 nm ridges and grooves , as prescribed in eq.([sqraverage ] ) and eq.([sqraverage2 ] ) . the field induced in a groove is mainly longitudinal while the field inside the ridge is mainly transversal . this is due to both the illumination and the polarizabilty of the scatterers . when defects are almost symmetric their polarizibilities @xmath128 are nearly isotropic and so the induced field and the incident field are virtually parallel . hence the field induced in a ridge and a groove are nearly parallel to the incident surface plasmon @xmath129 , which is mainly perpendicular to the plane in the vacuum semi - space and is mainly parallel to the plane in the metal semi - space . therefore , in the rayleigh limit , a ridge scatters spps into radiative modes like a vertical dipole on the plane , while the groove scatters them into radiative modes like a horizontal dipole on a plane . the results for grooves is in agreement with ref . + interestingly , we also have found numerically in fig . [ mode ] that : @xmath130 especially at short wavelengths . we have devised a virtual source , that can condense the orientation of the equivalent dipole representing a non - elongated symmetric ridges and grooves . this virtual dipole is defined as : @xmath131 . the fields inside a groove and a ridge , are respectively , represented as:@xmath132 | \overline{\e}^{r}|&\simeq & |\delta\varepsilon\,\overline{e}^r_{z}{\bf q}(90 ^ 0)|\end{aligned}\ ] ] at least as long as eq.([exz ] ) holds . + in reality we can see what happens by means of eq.([mie ] ) . despite the fact that this equation is only exact for a circle in a homogenous background ( as explained ) we can use it to show _ qualitatively _ the relation between the field inside the groove and the ridge , when their shapes are symmetric . if we approximate the polarizability of a ridge for that of a circle in vacuum ( whose polarizability is calculated through eq.([mie ] ) ) , so @xmath133 . if we also approximate the groove polarizability by that of a hole in a homogenous metal medium , we have : @xmath134 . hence the field induced inside each object is : @xmath135 since these polarizabilities also have the property : @xmath136 ( the polarizability of a hole in a material is @xmath137 times larger than the polarizability of a particle of the same material and the same shape ) then @xmath138 . ) , eq.([sqraverage2 ] ) ) for a square groove and a square ridge of 10 nm side in silver , as a function of the wavelength . the scale is logarithmic with arbitrary units . ] + the symmetry of the polarizations @xmath128 and the property @xmath139 are strictly true for circular defects in homogeneous media . our numerical calculations of fig . [ mode ] shows that , even though the field inside a ridge and a groove are quantitatively different from those of circular defects in homogenous media , the assumption that their mutual relation is preserved is in very good agreement with the exact result . because of the symmetry of the square shape , the _ averaged field _ inside the square is very nearly parallel to the incident field . as a corollary of the properties of the fields in a ridge and a groove @xmath140 notice that these define @xmath141 through eq.([spmas ] ) . once @xmath50 from eq.([net ] ) and @xmath142 are determined the value of the transmission of the surface plasmon is a constrained variable : @xmath143 , at least for the lossless case@xcite . since @xmath50 is greater for grooves than for ridges , the groove transmission is smaller . the first part of the expression eq.([sigmaout ] ) is a pre - factor @xmath144 whereas the second part is the _ the radiation pattern _ of a point dipole : @xmath145 a groove emits like a horizontal dipole . the angular amplitude of the field radiated by a horizontal unit dipole @xmath146 , placed close to the interface @xmath147 , is @xmath148 , and it does not matter on which side of the interface it is placed . @xmath148 can be derived using the relations in the appendix [ key ] and the explicit result is : @xmath149 and the radiation pattern is @xmath150 . notice @xmath148 presents a mirror symmetry about the angle @xmath151 , the normal to to the plane . furthermore since @xmath148 never changes sign between @xmath116 and 180@xmath152 ( nor goes to zero ) , the field of a horizontal dipole has one single symmetric lobe , where the field always has the same sign . of a horizontal point dipole at an air - metal interface . the radiation patterns varies as the metal dielectric constant is varied . the scale is logarithmic with arbitrary units . ] the field intensity @xmath150 of such lobe is rendered in fig . [ ux ] for different dielectric constants . this radiation pattern of a groove shown in fig . [ ux ] , is in agreement with the one represented by ref . [ ] , obtained with a different numerical method . notice that for @xmath153 : @xmath154that is , when @xmath137 increases this radiation pattern tends to become simultaneously isotropic and vanishing . in fact a horizontal dipole does not radiate on a perfect conductor@xcite . on a small digression it is interesting to notice an apparent contradiction between treatments such as ref . [ ] , which considered that a defect in a perfect metal were equivalent to a magnetic dipole , while another work@xcite explains a defect in a real metal corresponds to an electric dipole . actually we have just reconciled the two results . we know that a horizontal dipole on a plane tends to emit isotropically for large @xmath137 . this means that on a first order expansion in @xmath155 , the radiation pattern of a horizontal dipole on a plane and that of a magnetic dipole in vacuum , are identical . + for finite @xmath137 the field @xmath148 of a horizontal dipole within a real metal would not be thoroughly screened , and while the pattern remains symmetric , its isotropy is disrupted parallel to the surface ( i.e. @xmath156 ) to accommodate the emergence of the surface plasmons density of states . + for an individual vertical dipole @xmath157 , which represents a ridge , the angular amplitude of the field is ( see appendix [ key ] ) : @xmath158 the field from a vertical dipole also goes to zero at @xmath156 for a finite @xmath159 , but since dipoles only radiate transversally , the field has a third zero at @xmath160 . the field is antisymmetric with respect to the normal of the plane , while the intensity @xmath161 is symmetric , and is made up of the two lobes separated by a zero at 90@xmath152 , see fig . [ symmetric ] . yet it is important to keep in mind that the field of one lobe is in anti - phase with the field of the other . from the virtual dipole @xmath162 at @xmath163 . the scale is linear with arbitrary units . ] + unlike a horizontal dipole , the vertical dipole radiative field does not vanish for @xmath153 in fact : @xmath164 the total radiation from a vertical dipole has a larger weight than the radiation by a horizontal one , by a factor of @xmath165 . this can be seen , in fact , from eq.([gammax ] ) if we assume @xmath159 is large , we get the following relation : @xmath166 in fig . [ symmetric ] we represent radiation pattern of @xmath167 for the horizontal and vertical orientations respectively , @xmath168 , @xmath169 , which corresponds to our analytic analog of the emission pattern of square ridges and grooves respectively . while we will consider an intermediate orientation in the next section , we want to remark here that , due to eq.([uguale ] ) , the radiation by both the horizontal moment @xmath170 and a vertical moment @xmath171 vanish parallel to the plane at @xmath156 in a similar manner , as illustrated in fig . [ radpati ] + at the same time the far - field emissions of ridges and grooves become increasingly different as we approach the direction normal to the plane . for shallow and long defects @xmath172 and @xmath173 we define the following height - averaged polarization densities and fields : @xmath174 where the last equation defines @xmath175 . likewise for a groove we can define @xmath176 and @xmath177 through the following equation : @xmath178 notice for @xmath153 we can make the approximation @xmath179 . + the benefit of using @xmath180 is that the scattered - field coefficients for these defects in the far zone , @xmath45 and @xmath181 , are those emitted by a chain of point - dipoles on the surface over the segment @xmath125 , and set at @xmath182 and @xmath183 for ridges and grooves , respectively . + the scattered field angular amplitude @xmath184 from eq.([angularfield ] ) and eq.([short ] ) is obtained as : @xmath185 the averaged fields component inside of a ridge and a groove , @xmath186 , the quantity @xmath187 and the rest of the components @xmath186 , @xmath188 , @xmath189 , for rectangular defects of @xmath190 and @xmath191 . the system is illuminated by a spp in lossless silver at @xmath192 nm . the scale is linear with arbitrary units . ] this holds for the scattering into surface plasmon modes as well since we have : @xmath193^*\cdot \int_0^w dx'\;e^{\mp i k_{px}x'}\ ; { \bf \widetilde{p}}^{(\nu)}(x').\end{aligned}\ ] ] when we illuminate a shallow and long defect , with a spp , an equivalent linear density of dipole sources @xmath194 stems from how the induced fields are distorted inside the scatterer , namely by its polarizability . when the defect is larger in the horizontal direction than in the vertical one , ridges and grooves were found to give the same scattering by an approximated rayleigh expansion@xcite . we have an alternative first principles argument to justify the rayleigh expansion result , which is based entirely on the assumption that these defects are needle shaped . the field induced in these defects tends to be that induced in a needle - shaped protrusion placed horizontally on the surface @xmath182 in the case of a ridge . for a groove we have a horizontal needle - shaped cavity at @xmath183 . in such idealistic simplification it is clear - cut to deduce the fields inside the defects from the boundary conditions . namely the parallel component of the incident field is always continuous and equal , as in eq.([ispp1])and eq.([ispp2 ] ) : @xmath195 which preserves the continuity of eq.([gsurfacex ] ) . however , we are generating fields which , normal to the surface , make up for the discontinuity perpendicular to the metal surface of eq.([gsurfacez ] ) . in fact , for a horizontal needle - like ridge , the boundary conditions imposed by the continuity of the displacement vector are : @xmath196 while for a needle - like slit : @xmath197 ultimately : @xmath198 which , matched with eq.([gsurfacex ] ) and eq.([gsurfacez ] ) , yields : @xmath199 and thus the property of producing the same scattering coefficients , previously found in ref . of course this is just an approximation , but it explains why elongated defects have similar scattering properties . in real life the plasmon scattering by protrusions and indentations is similar because , far from the edges , a shallow but elongated defect behaves as an infinitely elongated one , as confirmed by numerical calculations . as an example we report in fig . [ fieldlong ] a numerical calculation of the fields averaged over the height for defects of @xmath190 and @xmath124 . this shows that eq.([needle ] ) and eq.([needle2 ] ) are quite accurate at the center of the defect , and deviate from the needle model prediction due to fringe effects at the edges . + it is worth mentioning that this equivalence is valid in the rayleigh limit when the defect size is much smaller than the wavelength , and may be altered at resonant wavelengths . everything we just said for symmetric surface defects was based on the fact that their aspect ratio equals one . as the defect width is increased , the aspect ratio becomes larger and this leads , progressively , to an asymmetric polarizability tensor . the first effect is that the field induced is gradually less and less parallel to the incident field . therefore a ridge would develop a non - negligible horizontal electric field component , thus ceasing to be equivalent to a vertical dipole . likewise the groove , which in the symmetric case behaves as a horizontal dipole , gradually starts having a non - negligible vertical component as its shape is elongated . the process goes on until we recover the case of a needle shaped defect of section [ needlesection ] . the fields inside a defect having intermediate width , as in fig . [ rect1 ] , are intermediate between those for the needle case and the square symmetric case . therefore in these cases defects emit qualitatively like _ oblique dipoles _ , with the orthogonal components out of phase . + in order to understand better the radiation pattern by ridges and grooves we decompose the oblique dipole in its horizontal and vertical components . + first of all , we focus on the mechanisms involved radiation pattern for a ridge @xmath14 . from eq.([p ] ) a dipole with arbitrary orientation emits close to the surface , with a field angular amplitude : @xmath200 where @xmath201 and equals : @xmath202@xmath203 shows that the contribution to the radiative field coming from the vertical and horizontal dipole on a metal plane have a phase difference of @xmath160 . this was already evident from eq.([uguale ] ) , when @xmath204 . such phase difference arises from the impedance of a metal plane@xcite @xmath205 . + the radiation pattern for a dipole with arbitrary orientation and lying above the metal , is written in our formalism as : @xmath206 . for a point dipole : @xmath207 , lying on top of a metal surface . the scale is linear with arbitrary units . ] the net angular amplitude for an oblique dipole is resolved into the superposition of the angular envelope of the horizontal dipole ( shown in fig . [ ux ] ) , with the other radiation factor @xmath208 . this last factor contains both the orientation and phase of the field . to envisage how these combine we may develop @xmath208 into three terms . these consist in the individual emission from the horizontal and vertical dipole plus an interference term : @xmath209\cos\alpha & \end{aligned}\ ] ] in the presence of the plane metal background , we have that horizontal and vertical dipoles behave as individual sources but their interaction presents an intrinsic added phase difference of @xmath160 , which is due to the different interaction of a horizontal and a vertical dipole with the plane . as a result , when in phase they do not interfere , and their radiation pattern is always symmetric regardless of the orientation of the dipole . this is the case for @xmath210 where , as in fig . [ symmetric ] , the radiation pattern is the sum of the angular intensity of a vertical and a horizontal dipole , so that at @xmath160 there is a minimum due to the vanishing of the vertical dipole contribution , and yet never goes to zero because of the horizontal dipole contribution . nevertheless , when the dipole components are not in phase , we can get asymmetric radiation patterns and additional zeros ( to those at @xmath211 and @xmath212 ) , because the interaction term can be negative . in such case the interaction of the horizontal radiative field ( with only one lobe ) with the vertical radiative ( with two lobes of different sign ) is responsible for an asymmetric radiation pattern and exhibits directionality . for a point dipole : @xmath213 , lying on top of a metal surface . the scale is linear with arbitrary units . ] this is illustrated in fig . [ dipoles2 ] for a dipole emission whose main contribution comes from the vertical dipole . in fig . [ dipoles3 ] we show the radiation pattern for a dipole whose main contribution comes from the horizontal dipole radiation . + for the case of a grooves ( @xmath15 ) , the radiative angular field amplitude is , from eq.([piz ] ) : @xmath214 \delta^{(2)}(\alpha)&=&p_{x}^{(2)}+ i\ ; \frac { \cos\alpha}{\sqrt{\cos^2\alpha+|\varepsilon|}}\;p_{z}^{(2)}\end{aligned}\ ] ] where remember we have also added the approximation : @xmath215 for @xmath153 . + remarkably , as opposed to the the dipole emission over the surface , in the net emission from a dipole under the surface the horizontal dipole contribution has a greater weight than the vertical dipole contribution . apart from this , all the arguments used for a dipole over the surface apply . + the interaction between the vertical and horizontal components of the field induced in the field generates the directional patterns of fig . [ rect1 ] . for a ridge with length slightly larger than its height the directional radiation is dominated by its vertical component . [ dipoles2 ] exemplifies the effect of the interference of a dominant vertical component with a smaller but non - negligible horizontal component . for even larger aspect ratios the contribution from the other component may be comparable . + likewise when a groove has a small aspect ratio it is predominantly a horizontal source interfering with a smaller vertical source . the result is in an interference pattern that looks like the one rendered in fig . [ dipoles3 ] . yet again this can be modified by increasing the aspect ratio . this transition is in good agreement with fig . 11 of ref . [ ] where , using a different numerical method , the radiation pattern of a groove was computed for different aspect ratios . our analysis of the surface plasmon scattering by square shallow defects into radiative modes and plasmon modes , reveals that a groove scatters more of the incident energy than a ridge does . the reflection by a symmetric ridge and a groove is similar and so is the radiative emission close to the horizontal direction . indeed their scattering essentially differs in the vertical direction , where a groove scatterers while a ridge does not . when defects start to become longer in width we saw the polarizability gets more asymmetric . correspondingly , since both components of the incident plasmon are out of phase , defects are equivalent to interfering horizontal and vertical dipoles on a plane , which interfere constructively in some direction , thus producing directionality in the radiation pattern . finally when ridges and grooves are shallow and long they tend to produce the same scattering as , apart for fringe effects , their polarizability exactly counterbalances the discontinuity of the incident surface plasmon field at the air - metal interface . the authors acknowledge financial support from the spanish ministry of science and innovation under grants no.ap2005-5185 , mat2008 - 06609-c02 and csd2007-046-nanolight.es . the incident illumination is the field of a surface plasmon wave mode propagating in the positive x direction @xmath216 or negative x direction @xmath217 is : @xmath218 @xmath219 where @xmath220 , @xmath221 and @xmath222 . this can , alternatively , be written as : @xmath223 for @xmath224 ; and @xmath225 for @xmath226 . + the magnetic field associated is continuous at the interface and equal to : @xmath227 now consider a lossless metal , characterized by a real and negative dielectric constant @xmath137 and consider a plasmon moving in the forward direction , ( the subscript + will be omitted ) . the incident poyinting vector of the plasmon in the air side is : @xmath228 while in the metal is @xmath229 where @xmath230 . the total poynting vector energy flux associated to a plasmon mode in a lossless metal is : @xmath231 we shall repeat , out of completeness , the explicit expression for p - waves , particularly in the far field when @xmath232 . in this case these modes are expressed in terms of the direct space polar angle @xmath40 by noticing that @xmath233 and @xmath234 in the air semi - space and @xmath235 in the metal . hence @xmath236 for reference , we give here the fresnel coefficients for an air metal interface . in the present treatment we only deal with the reflection coefficient for a p - wave propagating from air to metal , and this is : @xmath237 where notice that , for the sake of tidiness , we omit the superscript throughout . + as to the transmission coefficients the one for a wave ( 2,1 ) propagating from the metal to air is @xmath238 , while the one for a p - wave transmitted from the air medium to the metal is @xmath239 . @xmath240 notice that the transmission coefficients are related as follows : @xmath241 the following expressions for the reflection and transmission coefficients are essential to derive eq.([gammax ] ) and eq.([gammaz ] ) : @xmath242 @xmath243 1-r_p(\alpha)=\frac{2 \varepsilon\sin\alpha } { \sqrt{\varepsilon-\cos^2\alpha}+\varepsilon \sin\alpha}\end{aligned}\ ] ] the asymptotic expressions for the green tensor for @xmath244d scatterers are found in references@xcite . we have already presented the derivation scheme for bi - dimensional defects in appendix b of ref . [ ] , for a groove . as explained therein the surface plasmon green tensor and the far - field green tensor are obtained from its angular spectrum . from the relevant sommerfeld integral the surface plasmon contribution is obtained by applying the residue theorem and the far - field green tensor instead is obtained by applying the method of the steepest descent . + for the case of the ridge we use the total green tensor of the background in the vacuum semi - space . this can be written as the sum of the direct green tensor ( the free space green tensor ) and the indirect green tensor ( which gives the contribution due to the reflections at the metal plane interface ) . hence @xmath245 where the spectral representation for the direct green tensor is : @xmath246 while for the indirect green tensor : @xmath247 applying the residue theorem and the steepest descent method to @xmath248 we end up with eq.([asyp ] ) and eq.([gp1 ] ) for ( @xmath14 ) . + for the groove case we need to expand the green tensor connecting a point in the metal to a point in air . this is just : @xmath249 applying the residue theorem and the steepest descent method to @xmath250 we end up with eq.([asyp ] ) and eq.([gp1 ] ) for ( @xmath15 ) . notice that the form of @xmath251 given in sec . [ coeff ] , is obtained by recognizing @xmath252 . + one more subtlety , that might be confusing , is how we pass from the transmission coefficient @xmath253 in the integral to the transmission coefficient @xmath254 in the asymptotic form @xmath255 . this comes about because when we apply the method of the steepest descent to the integral we get eq.([asyp ] ) with : @xmath256 where , in the last equation , we have used the identity eq.([a6 ] ) .

collective phenomena in systems far from thermal equilibrium have been of considerable interest in recent years @xcite . unlike systems in thermal equilibrium where the gibbs picture provides a theoretical framework within which such phenomena can be studied , here no such framework exists and one has to resort to studies of specific models in order to gain some understanding of the phenomena involved . one class of such models is driven diffusive systems ( dds ) @xcite . driven by an external field these systems do not generically obey detailed balance so that the steady state has non - vanishing currents . theoretical studies of dds have revealed basic differences between systems in thermal equilibrium and systems far from thermal equilibrium . for example , it is well known that one dimensional ( @xmath0 ) systems in thermal equilibrium with short - range interactions do not exhibit phenomena such as phase transitions , spontaneous symmetry breaking ( ssb ) and phase separation ( except in the limit of zero temperature or in the context of long - range interactions ) @xcite . in contrast , some examples of noisy @xmath0 dds with local dynamics have been found to exhibit such phenomena . one example of a noisy system which exhibits ssb in @xmath0 is the asymmetric exclusion model of two types of charge studied in @xcite . in this model , two types of charge are biased to move in opposite directions on a @xmath0 lattice with open ends . the charges interact via a hard - core interaction , and are injected at one end of the lattice and ejected at the other end . this model is symmetric under the combined operations of charge conjugation and parity ( pc symmetry ) . however , this symmetry is broken in the steady state , where the currents of the two charges are not equal . the reason for symmetry breaking in this model lies to some extent in the open boundaries . other examples of models in which there is ssb in @xmath0 have also been found in the context of cellular automata @xcite and surface growth @xcite . in the latter , ssb was due to the fact that one of the rates for a local dynamical move in the models is zero . once this zero rate changes to a non - zero rate ssb disappears . a closely related problem to spontaneous symmetry breaking , is that of _ phase separation _ in @xmath0 noisy systems . this has been observed in driven diffusive models with inhomogeneities , such as defect sites @xcite or particles @xcite . in these models it has been found that macroscopic regions of high densities are formed near the defect , much like a high density of cars behind a slow car in a traffic jam @xcite . here the phase separation is triggered by the defects . it is of interest to study whether phase separation can occur in @xmath0 noisy homogeneous systems such as on a ring geometry with no defects , where _ all _ possible local transition rates which are consistent with the symmetry and conservation laws of the model are non vanishing . recently , lahiri and ramaswamy have introduced a lattice model in the context of sedimenting colloidal crystals , where phase separation is found to take place without any inhomogeneities @xcite . in this model , there are two rings coupled to each other and particles on each ring undergo an asymmetric exclusion process . the hopping rate between sites @xmath1 and @xmath2 on each ring depends on the occupation at the @xmath1th site on the other ring . however , this model is studied mainly using monte carlo simulations and no analytical results are available so far . in a recent letter @xcite we introduced a simple three - species driven diffusive model exhibiting phase separation and spontaneous breaking of the translational symmetry on a ring . in the model nearest - neighbor particles exchange with given rates and the numbers of each species are conserved under the dynamics . the rates of all local dynamical moves that obey the conservation laws are non zero . an argument indicating that generically the system phase separates , thus breaking the translational invariance , was given for the case when none of the species of particle has zero density . in the special case of equal number of particles of each type , it was shown that the local dynamics obeys detailed balance with respect to a long - range asymmetric ( chiral ) hamiltonian . in this special case , using the hamiltonian , we have found the steady state of the model exactly and have been able to prove the existence of phase separation analytically . the existence of a hamiltonian for this special case is of interest in the light of speculation that non - equilibrium systems exhibiting generic long - range correlations might be described by effective hamiltonians containing long - range interactions @xcite . here we explicitly demonstrate that for the special case where the three densities are equal the model is exactly described by a long - range asymmetric hamiltonian . the model not only has long - range correlations but has generic long - range _ order_. the mechanism found in this study suggests that systems with dynamical rules defined completely locally and _ a priori _ without respect to any hamiltonian , may have a steady state where the configuration space is sampled according to a measure that is intrinsically global . the hamiltonian also allows us to identify the analog of a temperature in the microscopic dynamics as related to the drive of the system ; for zero drive , that is symmetric diffusion of the particles , the effective temperature is infinite and phase separation is lost . we note that a related but distinct three - species model has recently been introduced by arndt _ et al_. this model also exhibits phase separation@xcite . in the present work we analyze in detail the @xmath3 species model which was introduced in @xcite and then generalize it to larger @xmath4 . we provide the complete proof of phase separation which follows from the exact calculation of the partition sum in the thermodynamic limit . we also provide numerical evidence of phase separation in the general case where the densities of the three particles are not equal . in order to study the coarsening process monte - carlo simulations are performed . however , simulation of the microscopic model is hampered by slow dynamics which makes it difficult to access the scaling regime . the system becomes trapped in metastable states comprising several domains of each type of particle . the number of metastable states is exponentially large in the system size . the lifetimes of the metastable states increase exponentially with the average domain size as the fully phase separated state is approached . thus the model provides an example of slow dynamics in a system without any quenched disorder @xcite . to ameliorate the difficulty of numerically studying such slow dynamics we employ a toy model wherein it is the domains that are updated rather than the individual particles this allows the long - time scaling behavior of the domain size to be investigated and to confirm a logarithmic growth of the average domain size with time . the toy model also affords a mean - field solution for the long - time dynamical behavior , that again confirms the scaling behavior . returning to the case of equal numbers of particles of different species it is of interest to investigate the steady - state behavior in finite - size systems . we have found it convenient to do this by employing a matrix product technique previously used to solve the steady state of asymmetric exclusion processes @xcite . however , in the case of three species the simplest form of this technique @xcite is applicable only to a limited class of systems @xcite . for the present model we generalize the matrix product to a product of rank 6 tensors and write the steady state by taking an appropriate contraction . the partition sum and steady - state correlation functions can be conveniently computed numerically using this tensor product ansatz . the paper is organized as follows : in section ii we define the model introduced in @xcite and we present an argument which indicates that the system should phase separate as long as none of the species of particles has zero density . in section iii we study the special case where the model satisfies detailed balance and explicitly write down the steady - state weight for the three - species model . the existence of phase separation in the model for any non - infinite temperature is proved analytically by calculating some bounds on the two - point correlation functions . section iv contains numerical evidence for phase separation in the general case where the densities of the three species of particles are not equal . the toy model , which facilitates efficient monte carlo simulations , is used to study the dynamics of phase separation . a mean - field analysis of this toy model is presented , the details being left to appendix a. in section v we present results for finite systems obtained via the tensor product ansatz . using these results we study finite - size scaling in the system . in section vi we address phase separation in systems with more than three species of particle and a proof of detailed balance for special cases is given in appendix b. we conclude in section vii and discuss some open questions . we start by defining a three - species model which exhibits phase separation in @xmath0 . consider a one - dimensional , ring - like ( periodic ) lattice of length @xmath5 where each site is occupied by one of the three types of particles , @xmath6 , @xmath7 , or @xmath8 . the model evolves under a random sequential update procedure which is defined as follows : at each time step two neighboring sites are chosen randomly and the particles at these sites are exchanged according to the following rates @xmath9 the particles thus diffuse asymmetrically around the ring . the dynamics conserves the number of particles , @xmath10 and @xmath11 of the three species . the @xmath12 case is special . here the diffusion is symmetric and every local exchange of particles takes place with the same rate as the reverse move . the system thus obeys detailed balance reaching a steady state in which all microscopic configurations ( compatible with the number of particles @xmath10 and @xmath11 ) are equally probable . this state is homogeneous , and no phase separation takes place . we now present a simple argument suggesting that for @xmath13 the steady state of the system is not homogeneous in the thermodynamic limit . for simplicity the case @xmath14 is examined . as a result of the bias in the exchange rates an @xmath6 particle prefers to move to the left inside a @xmath7 domain and to the right inside a @xmath8 domain . similarly the motion of @xmath7 and @xmath8 particles in foreign domains is biased . consider the dynamics starting from a random initial configuration . the configuration is composed of a random sequence of domains of @xmath6 , @xmath7 , and @xmath8 particles . due to the bias a local configuration in which an @xmath6 domain is placed to the right of a @xmath7 domain is unstable and the two domains exchange places on a relatively short time scale which is linear in the domain size . similarly , @xmath15 and @xmath16 domains are unstable too . on the other hand @xmath17 , @xmath18 and @xmath19 configurations are stable and long lived . thus after a relatively short time the system reaches a state of the type @xmath20 in which @xmath21 and @xmath8 domains are located to the right of @xmath22 and @xmath7 domains , respectively . the evolution of this state takes place via a slow diffusion process in which , for example , the time scale for an @xmath6 particle to cross an adjacent @xmath7 domain is @xmath23 , where @xmath24 is the size of the @xmath7 domain . the system therefore coarsens and the average domain size increases with time as @xmath25 @xcite . eventually the system phase separates into three domains of the three species of the form @xmath26 . in a finite system the phase separated state may further evolve and become disordered due to fluctuations . however , the time scale for this to happen grows exponentially with the system size . for example it would take a time of order of @xmath27 for the @xmath6 domain in the totally phase separated state to break up into smaller domains . hence in the thermodynamic limit , this time scale diverges and the phase separated state remains stable provided the density of each species is non - zero . note that there are always small fluctuations about a totally phase separated state . however , these fluctuations affect the densities only near the domain boundaries . they result in a finite width for the domain walls . the fact that any phase separated state is stable for a time exponentially long in the system size amounts to a breaking of the translational symmetry . since the exchange rates are asymmetric , the system generically supports a particle current in the steady state . to see this , consider the @xmath6 domain in the phase separated state . an @xmath6 particle near the @xmath28 boundary can traverse the entire @xmath7 domain to the right with an effective rate proportional to @xmath29 . once it crosses the @xmath7 domain it will move through the @xmath8 domain with rate @xmath30 . similarly an @xmath6 particle near the @xmath31 boundary can traverse the entire @xmath8 domain to the left with a rate proportional to @xmath32 . once the domain is crossed it moves through the @xmath7 domain with rate @xmath30 . hence the net @xmath6 particle current is of the order of @xmath33 . since this current is exponentially small in system size , it vanishes in the thermodynamic limit . for the case of @xmath34 , this argument suggests that the current is strictly zero for any @xmath5 . in sections iii and v we study this case in detail . the arguments presented above suggesting phase separation for @xmath35 may be easily extended to @xmath36 . in this case , however , the phase separated state is @xmath37 rather than @xmath38 . this may be seen by noting that the dynamical rules are invariant under the transformation @xmath39 together with @xmath40 . in this section we show that the dynamics ( [ eq : dynamics ] ) , for the special case @xmath34 , satisfies detailed balance . the corresponding hamiltonian , which determines the steady - state distribution , is found to have long - range asymmetric interactions . using this hamiltonian , we calculate analytically the partition sum and bounds on the correlation functions in the thermodynamic limit . these are then used to prove the existence of phase separation in the model . later , in section [ matrixans ] we study finite systems for this case and the approach to the thermodynamic limit . the general argument presented in the previous section suggests that for the special case @xmath34 , the steady state carries no current for any system size . we demonstrate this explicitly by showing that the _ local dynamics _ of the model satisfies detailed balance with respect to a _ long - range asymmetric _ hamiltonian @xmath41 . we define the occupation variables @xmath42 and @xmath43 as follows : @xmath44 the variables @xmath45 and @xmath43 are defined similarly . clearly the relation @xmath46 is satisfied . a microscopic configuration is thus described by a set @xmath47 . using these variables , we will show that the hamiltonian @xmath41 and the steady - state distribution @xmath48 corresponding to the dynamics ( [ eq : dynamics ] ) for the case @xmath49 are given by @xmath50 \ ; , \ ] ] @xmath51 here @xmath52 is the partition sum given by @xmath53 , where the sum is over all configurations in which @xmath34 . note that the hamiltonian @xmath41 does not determine the dynamics of the system , it just governs the steady - state distribution as given in ( [ eq : hamilton ] ) and ( [ eq : weight ] ) . ( [ eq : weight ] ) suggests that @xmath54 serves as a temperature variable with @xmath55 . thus , @xmath56 is the infinite - temperature limit . the hamiltonian ( [ eq : hamilton ] ) is written in a form which is not manifestly translationally invariant . however , careful examination reveals that when the relation @xmath34 is taken into account , the hamiltonian as given by ( [ eq : hamilton ] ) is indeed translationally invariant ( see appendix b ) . therefore site @xmath57 may be chosen arbitrarily . an expression for @xmath41 which is manifestly translationally invariant will be derived at the end of this section . note that @xmath58 since the lhs yields the number of @xmath19 ( and @xmath15 ) pairs in the system . using this relation the hamiltonian may also be written in a form where the cyclic symmetry is more apparent : @xmath59-(n/3)^2 \ ; .\ ] ] the proof of eqs . ( [ eq : hamilton],[eq : weight ] ) is straightforward . this is done by considering a nearest - neighbor particle exchange and verifying that detailed balance is satisfied with respect to ( [ eq : weight ] ) . we start by considering nearest - neighbor sites in the interior of the lattice , namely pairs other than @xmath60 . for example consider the exchange @xmath61 taking place at two adjacent sites @xmath62 and @xmath63 , where @xmath64 . this exchange results in the contribution of one more @xmath65 term in @xmath66 and hence the energy of the resulting configuration is higher by @xmath57 . it is easy to see using eq . ( [ eq : weight ] ) that @xmath67 , as required by detailed balance . similar relations are easily derived for exchange of @xmath18 and @xmath19 pairs . now consider an exchange taking place between sites @xmath5 and @xmath57 , say @xmath68 . according to ( [ eq : hamilton ] ) this exchange costs an energy of @xmath69 . therefore the exchange satisfies the detailed balance condition @xmath70 only when @xmath71 . similarly , by considering the exchanges @xmath72 and @xmath73 , one deduces that the detailed balance condition is satisfied for any exchange at sites @xmath5 and @xmath57 as long as @xmath34 . in appendix b we consider the most general nearest - neighbor exchange rates for @xmath4 species and arbitrary densities and derive conditions ( [ condition ] ) for exchange rates which satisfy detailed balance . to write @xmath41 in a manifestly translationally invariant form we define @xmath74 as the hamiltonian in which site @xmath75 is the origin . namely , @xmath76 \ ; , \ ] ] where the summation over @xmath1 and @xmath77 is modulo @xmath5 . summing ( [ hi0 ] ) over all @xmath75 and dividing by @xmath5 , one obtains , @xmath78 where in the summation the value of the site index @xmath79 is modulo @xmath5 . in the hamiltonian ( [ eq : h_tinvar ] ) the interaction is linear in the distance between the particles , and thus is long - ranged . the distance is measured in a preferred direction from site @xmath1 to site @xmath80 . moreover it is asymmetric in the sense that @xmath66 is not invariant under the parity operation . it is also related to chiral hamiltonians @xcite . before proceeding further to evaluate the partition sum and some correlation functions associated with the hamiltonian ( [ eq : hamilton ] ) let us make a few observations . the ground state of the hamiltonian is given by the fully separated state @xmath81 and its translationally related states . the degeneracy of the ground state is thus @xmath5 and its energy is zero . a simple way of evaluating the energy of an arbitrary configuration is obtained by noting that nearest - neighbor ( nn ) exchanges @xmath82 and @xmath83 cost one unit of energy each while the reverse exchanges result in an energy gain of one unit . the energy of an arbitrary configuration may thus be evaluated by starting with the ground state and performing nn exchanges until the configuration is reached , keeping track of the energy changes at each step of the way . the highest energy is @xmath84 and it corresponds to the totally phase separated configuration @xmath85 and its @xmath5 translations . in considering the excited states of the hamiltonian ( [ eq : hamilton ] ) we note that the model exhibits a set of metastable states which correspond to local minima of the energy : any exchange of nn particles results in an increase of the energy . in these states no @xmath86 and @xmath15 nn pairs exist ; only @xmath87 and @xmath19 nn pairs may be found in addition to @xmath88 , @xmath89 , and @xmath90 . any metastable state is thus composed of a sequence of domains in which @xmath21 and @xmath8 domains follow @xmath22 and @xmath7 domains , respectively . therefore each metastable state has an equal number of domains , @xmath91 , of each type with @xmath92 . the @xmath93 case corresponds to the ground state while @xmath94 corresponds to the @xmath95 state , composed of a total of @xmath5 domains each of length @xmath57 . ( the total number of domains in an @xmath91-state is @xmath96 . ) for calculating the free energy and some correlation functions corresponding to the hamiltonian ( [ eq : hamilton ] ) we find it useful to first derive some bounds for the number @xmath97 of @xmath91-states and their energies . in the following such bounds are presented . they are then used , in the next section , to evaluate the free energy and correlation functions of the model . to obtain a bound for @xmath97 we note that the number of ways of dividing @xmath98 @xmath6 particles into @xmath91 domains is @xmath99 . the number of ways of combining @xmath91 divisions of each of the three types of particles is clearly @xmath100 ^ 3 $ ] . there are at most @xmath5 ways of placing this string of domains on a lattice to obtain a metastable state ( the number of ways need not be equal to @xmath5 since the string may possess some translational symmetry ) . one therefore has @xmath101 ^ 3 \le{\cal n}(s ) \le n \left[{n/3 - 1}\choose { s-1}\right]^3\ ; . \label{nmss}\ ] ] thus , the total number of metastable states is exponential in @xmath5 . we now consider the energy of the metastable states . it is easy to convince oneself that among all @xmath91-states , none has energy lower than the following configuration , @xmath102 where the @xmath103 rightmost domains are of size @xmath57 and the three leftmost domains are of size @xmath104 each . the energy of this state , @xmath105 satisfies the following recursion relation @xmath106 with @xmath107 . to see this one notes that the @xmath91-state may be created from the @xmath108-state by first moving a @xmath7 particle from the leftmost @xmath7 domain across @xmath109 @xmath8 particles to the right and then moving an @xmath6 particle from the leftmost @xmath6 domain to the right across the adjacent @xmath7 and @xmath8 domains . the energy cost of these moves is @xmath109 , yielding ( [ rr ] ) . the recursion relation ( [ rr ] ) , together with @xmath110 , is then readily solved to give @xmath111 the energy of all metastable @xmath91-states is larger or equal to @xmath105 as given by eq.([emss ] ) . in the following section we use the bounds ( [ nmss ] ) and ( [ emss ] ) to calculate the partition sum and some correlation functions corresponding to the hamiltonian ( [ eq : hamilton ] ) . in this section we prove that , in the large @xmath5 limit and for all @xmath35 , the partition sum is given by @xmath112 ^ 3 \ ; , \label{psum}\ ] ] where @xmath113 the partition sum for @xmath36 may be obtained by replacing @xmath54 by @xmath114 in ( [ psum ] ) . note that the partition sum is linear and not exponential in @xmath5 , meaning that the free energy is not extensive . this is a result of the long - range interaction in the hamiltonian and the fact that the energy excitations are localized near the domain boundaries , as will be shown in the following . for @xmath54 close to @xmath57 , @xmath115 has an essential singularity , @xmath116}\;.\ ] ] this suggests that extensivity of the free energy could be restored in the double limit @xmath117 and @xmath118 with @xmath119 finite . this scaling behavior is only suggestive since expression ( [ psum ] ) may not be valid in this limit . for @xmath12 all configurations with @xmath34 are equally probable , so that the partition sum is given by @xmath120 which goes like @xmath121 for large @xmath5 . it is instructive to first present the proof of ( [ psum ] ) for small values of @xmath54 . this proof will then serve as the basis for the proof for any @xmath35 . we start by noting that in calculating the partition sum ( [ psum ] ) , configurations with energy larger then @xmath122 ( @xmath123 ) may be neglected in the thermodynamic limit . for simplicity we first demonstrate this for @xmath124 , although later we show it for any @xmath35 . the contribution to the partition sum from these energy states , @xmath125 , is given by @xmath126 where @xmath127 is the number of configurations of energy @xmath128 . clearly , the number of possible configurations in the system is bounded crudely from above by @xmath121 . this bound implies @xmath129 thus , for @xmath124 , the contribution to the partition sum arising from energies larger than @xmath122 is exponentially small in @xmath5 and may be neglected in the thermodynamic limit . the calculation of @xmath52 is thus reduced to calculating a truncated partition sum in which only energies up to @xmath122 are summed over . to proceed we consider @xmath130 and take into account configurations with energy less than @xmath131 . this simplifies the calculations considerably since all configurations with energy @xmath132 may be decomposed into @xmath5 disjoint sets of states , each corresponding to a unique ground state ( see previous section for a discussion of ground states ) . we label the sets by @xmath133 , the position of the rightmost @xmath6 particle in the @xmath6 domain of the ground state belonging to the set ( see fig . [ ground ] for an example of an @xmath134 ground state ) . each state in a specific set can be obtained from the corresponding ground state by exchanging nearest neighbors so that the energy always increases along the intermediate states . note that this is correct only if excitations of energy less than @xmath131 are considered . this is because not all higher energy states can be reached by uphill steps from a ground state . thus , using translational invariance , the partition sum can be written as @xmath135 where @xmath136 is the truncated partition sum of one of the @xmath5 sets of configurations . we proceed to calculate @xmath136 . this is done by considering all the possible energy excitations with energy less than @xmath131 above one ground state . consider a specific domain boundary , say @xmath17 . excitations of energy @xmath128 at this boundary can be created by moving one or more @xmath6 particles into the @xmath7 domain ( this is equivalent to moving @xmath7 particles into the @xmath6 domain ) . an @xmath6 particle moving into the @xmath7 domain is considered as a walker . the excitation energy increases linearly with the distance the walker has moved . thus , in this picture an excitation of energy @xmath128 is created by @xmath137 walkers , traveling a total distance @xmath128 . the number of excitations of energy @xmath128 is then given by the number of ways , @xmath138 , of partitioning an integer @xmath128 into the sum of a sequence of non - increasing positive integers . taking into account excitations at all three boundaries , an excitation of energy @xmath128 in the system is created by three independent excitations of energy @xmath139 and @xmath140 at the different domain boundaries such that @xmath141 . the number of excitations of this form is just given by @xmath142 . thus , @xmath136 is given by @xmath143 taking the thermodynamic limit we obtain @xmath144 using a well known result from number theory , attributed to euler , for the generating function of @xmath138 @xcite , @xmath145 and using ( [ thelim ] ) and ( [ exp ] ) , eq.([psum ] ) is obtained . so far we have proved that for @xmath146 , eq.([psum ] ) is exact in the thermodynamic limit . we now extend these results for any @xmath35 . first we have to show that the states ignored in the previous calculation for @xmath146 may be ignored for all @xmath35 . to do this we calculate upper and lower bounds on @xmath52 and show they converge for large enough @xmath5 . for this we have to consider the entire energy spectrum of the hamiltonian . any configuration of the system which is neither a ground state nor a metastable state can be obtained from at least one ground state ( @xmath93 ) or a metastable state ( @xmath147 ) as follows : starting from this @xmath91-state exchange nearest neighbors such that the energy always increases along the path until the configuration is reached . in what follows it is demonstrated that none of the configurations which can be obtained from @xmath91-states , with @xmath147 , by the above procedure of particle exchange , contributes to the partition sum in the thermodynamic limit . an upper bound on the partition sum may be calculated as follows : using the same steps of derivation used for computing @xmath136 , it is straightforward to show that the contribution to the partition sum from an @xmath91-state and associated configurations is _ at most _ @xmath148^{-3s}$ ] . the prefactor @xmath149 arises from the minimum energy ( [ emss ] ) of this metastable state . therefore by considering the contributions from all the @xmath91-states and using ( [ nmeta ] ) the following bound is found @xmath150 ^ 3 + \sum_{s=2}^{n/3}n { n/3 - 1\choose s-1 } q^{(s-1)n/3 -s(s-1)/2 } [ ( q_{\infty})]^{-3s}\;. \label{zupperb1}\ ] ] the second term on the rhs represents the contribution from excitations around the metastable states . replacing @xmath151 by an upper bound @xmath152 one can sum the binomial series . the resulting expression is exponentially small in @xmath5 for any @xmath35 . a lower bound on @xmath52 can be calculated by neglecting configurations with energy greater than @xmath131 as follows , @xmath153 ^ 3 - n \sum_{m = n/3 - 1}^{\infty}q^m\sum_{m_i=0}^mp(m_1)p(m_2)p(m_3)\delta_{m_1+m_2+m_3,m } \\ & > & n/[(q)_{\infty}]^3 - n \sum_{m = n/3 - 1}^{\infty}q^m ( mp(m))^3 \ ; . \label{lowbound}\end{aligned}\ ] ] the asymptotic behavior of @xmath138 @xcite is given by @xmath154 thus , for large @xmath5 the lower bound ( [ lowbound ] ) converges to ( [ psum ] ) as does the upper bound ( [ zupperb1 ] ) . whether or not a system has long - range order in the steady state can be found by studying the decay of two - point density correlation functions . for example the probability of finding an @xmath6 particle at site @xmath1 and a @xmath7 particle at site @xmath77 is , @xmath155 where the summation is over all configurations @xmath156 in which @xmath34 . due to symmetry many of the correlation functions will be the same , for example @xmath157 . a sufficient condition for the existence of phase separation is @xmath158 since @xmath159 we wish to show that @xmath160 . in fact we will show below that for any given @xmath161 and for sufficiently large @xmath5 , @xmath162 this result not only demonstrates that there is phase separation , but also that each of the domains is pure . namely the probability of finding a particle a large distance inside a domain of particles of another type is vanishingly small in the thermodynamic limit . to prove eq . ( [ corrf3 ] ) , we use the relation @xmath163 , and show that the correlation function @xmath164 is of @xmath165 and @xmath166 is of @xmath167 . here we show only the proof for @xmath164 , since the proof of @xmath166 is similar . we also restrict ourselves to @xmath168 , which is sufficient for proving eq . ( [ corrf3 ] ) . we have already seen that the contribution to the partition sum from the metastable states and excitations above them are exponentially small in the system size and hence may be neglected . therefore , for calculating the correlation function it is sufficient to consider the @xmath5 ground states and excitations above them which may be reached by moves which only increase the energy . as we have seen , these states form @xmath5 disjoint sets of states , each associated with one of the ground states . using this we now show that @xmath169 . for this purpose we use a restricted partition sum @xmath170 , which is defined as the partition sum @xmath136 calculated with the constraint that one of the walkers , say of type @xmath6 , has traveled at least distance @xmath91 . it is given as @xmath171 by @xmath172 here @xmath173 is the number of partitions of integer @xmath128 with the constraint that in all the partitions the integer @xmath91 occurs at least once . noting that @xmath174 it is easy to show that @xmath175 where @xmath176 . we now proceed to derive a bound for @xmath177 . recall that @xmath24 is the position of the rightmost @xmath6 particle in the @xmath6 domain in the ground state labeled @xmath24 . if we define @xmath178 as the correlation function calculated within the set of states labeled @xmath24 , we can write @xmath179 up to exponentially small corrections in the system size . for convenience we break the summation over @xmath24 into @xmath180 sums according to the values of @xmath181 and @xmath182 in the ground state . these @xmath180 parts correspond to ( i ) ground states where @xmath183 ; ( ii ) ground states where @xmath184 ; ( iii ) ground states where @xmath185 and ( iv ) ground states where @xmath186 . we now consider each of them in detail and give an upper bound for @xmath187 in each case . \(i ) _ ground states where @xmath183 _ : in this case the site @xmath57 is inside the @xmath6 domain and site @xmath161 is inside the @xmath7 domain . since we consider only @xmath168 , these states correspond to the @xmath24 ground states with @xmath188 . using the fact that @xmath189 one finds @xmath190 ( ii ) _ ground states where @xmath184 _ : in principle site @xmath161 might be either inside the @xmath6 domain or inside the @xmath8 domain . however , since site @xmath57 is in the @xmath6 domain and we consider only @xmath191 , site @xmath161 must be in the @xmath6 domain . the ground states @xmath24 for which this takes place satisfy @xmath192 . clearly , only the excited states where @xmath193 contribute to @xmath187 . in such excited states one of the @xmath7 walkers travels at least a distance @xmath194 into the @xmath6 domain ( see fig . [ a10bk1 ] ) . for this case we can give the upper bound @xmath195 . from eq . ( [ psums1 ] ) , @xmath196 . hence , @xmath197 \(iii ) _ ground states where @xmath185 _ : again , since @xmath168 site @xmath57 has to be inside the @xmath7 domain . the values of @xmath24 satisfying this condition are in the range @xmath198 . in this case only excited states in which one of the @xmath6 walkers travels at least a distance @xmath199 into the @xmath7 domain will contribute to @xmath200 . hence we can use the upper bound @xmath201 , in this case . therefore , @xmath202 \(iv ) _ ground states where @xmath186 _ : there are three possibilities here . @xmath203 site @xmath57 is inside the @xmath8 domain and site @xmath161 is inside the @xmath6 domain ( @xmath204 ) , @xmath205 both the sites @xmath57 and @xmath161 are inside the @xmath8 domain ( @xmath206 ) , @xmath207 site @xmath57 inside the @xmath7 domain and site @xmath161 inside the @xmath8 domain ( @xmath208 ) . since all these are consistent with @xmath209 , all these cases can occur . it can be shown that the minimal energy needed to create an excited state where @xmath210 and @xmath193 is @xmath211 for the case @xmath203 , @xmath212 for the case @xmath205 and @xmath213 for the case @xmath207 . the resulting expression for the bound is @xmath214 the summations on the rhs of eqs . ( [ ubound2]-[ubound4 ] ) can be carried out explicitly . to leading order , the summations gives @xmath215 for each of eqs . ( [ ubound2 ] ) and ( [ ubound3 ] ) . the summation on the rhs of eq . ( [ ubound4 ] ) vanishes exponentially in the thermodynamic limit . using eqs . ( [ a1br]-[ubound3 ] ) , we get the following expression for the upper bound on @xmath216 @xmath217 \;.\ ] ] therefore @xmath218 . similarly one can show that @xmath219 . thus for all @xmath35 , @xmath220 , proving the existence of a complete phase separation . we have demonstrated that in the thermodynamic limit the system is phase separated when @xmath34 . the general arguments given in section ii indicate that when the global densities of the three species are non - vanishing and @xmath221 , the system phase separates , even when the three densities are not equal . the argument suggests that the typical time , @xmath222 , in which the system leaves a specific phase separated configuration increases exponentially with the system size . thus , a phase separated state is stable in the thermodynamic limit . in the following we use monte carlo simulations to support these arguments . the time , @xmath222 , can be measured using the auto - correlation function defined as , @xmath223 where @xmath224 and @xmath225 are the values of the occupation variables @xmath42 and @xmath43 at time @xmath226 , and @xmath227 denotes an average over histories of evolution . clearly , @xmath228 while @xmath229 , the value of the autocorrelation between two independent configurations . thus , @xmath222 may be defined as the decay time of @xmath230 to @xmath231 when at @xmath232 the system is totally phase separated . we have measured the time scale @xmath222 using monte carlo simulations for different system sizes for @xmath34 and for @xmath233 for several @xmath54 values . an example of such measurements for @xmath234 , @xmath235 and @xmath236 is presented in fig . [ fig_corrt ] . in the plot @xmath222 is plotted versus system size for several values of @xmath54 . it agrees with the exponential growth of @xmath222 with the system size suggested by the simple argument of section ii . the same behavior seems to occur for all @xmath237 and different choices of @xmath238 , @xmath239 and @xmath240 . therefore we conclude that the monte carlo simulations support the claim that for any @xmath241 the system will phase separate into three domains in the thermodynamic limit , even when the number of particles of each species is not equal . in the thermodynamic limit the translational symmetry is spontaneously broken in this state . due to the slow dynamics , which reflects escape from metastable states , monte carlo simulations could be performed only for relatively small system size ( @xmath242 ) . in order to study the coarsening process for larger systems we employ , in the following , a toy model which mimics the dynamics of the model ( [ eq : dynamics ] ) . the toy model can be conveniently simulated for systems larger by about two orders of magnitude . we now construct a simple toy model which captures the essential physics of the coarsening process in the model at large times and enables us to simulate systems much larger than those accessible by monte carlo simulation . using the toy model we examine another characteristic scale of the system . namely , the average domain size @xmath243 as a function of the time @xmath226 . the results support the simple argument leading to a domain growth law @xmath244 . a mean - field version of the toy model is then solved analytically . we consider a system at time @xmath226 such that the average domain size , @xmath243 , is much larger than the domain wall width . at these time scales , the domain walls can be taken as sharp and we may consider only events which modify the size of domains . this means that the dynamics of the system can be approximated by considering only the movement of particles between neighboring domains of the same species . using this we represent a configuration by a sequence of domains of the form @xmath245 , where the @xmath1th domain of , say @xmath6 , particles is represented by @xmath246 , as shown in fig . [ fig_toy1 ] . the exchange of particles between domains , say @xmath247 and @xmath248 , takes place at a rate dictated by the size of the domains @xmath249 and @xmath250 which separate them . since intermediate configurations of the form @xmath251 rearrange on short time scales compared with the evolution between metastable states , only metastable configurations are considered in the toy model . events in which a domain splits into two are ignored . using these ideas we define the dynamics of the toy model as follows : at each time step two neighboring domains of the same species of particle are chosen randomly , say @xmath246 and @xmath252 . let @xmath253 , @xmath254 and @xmath255 denote the lengths of the domains @xmath256 and @xmath257 respectively . the length of the domain chosen is then modified by carrying out one of the following processes : @xmath258{ll } \left . \begin{array}[]{ll } 1 ) & a_i \rightarrow a_i - 1 \\ & a_{i+1}\rightarrow a_{i+1 } + 1 \\ \end{array } \right\ } & { \rm with~rate } \quad q^{b_i } \\ & \\ \left . \begin{array}[]{ll } 2 ) & a_i \rightarrow a_i + 1 \\ & a_{i+1}\rightarrow a_{i+1 } - 1 \\ \end{array } \right\ } & { \rm with~rate } \quad q^{c_i } \\ \end{array}\ ] ] where , as before , @xmath14 is considered . if @xmath253 becomes zero , then delete the domain @xmath246 from the list of domains , and merge @xmath259 and @xmath257 with @xmath260 and @xmath261 , respectively . then for @xmath262 , shift the indices of the domains from @xmath77 to @xmath263 , so that @xmath264 becomes @xmath265 . the rules for updating @xmath266 and @xmath267 domains can be obtained from ( [ toy_rule ] ) using cyclic permutations and a slight change of indices . note that the toy model is only relevant to the description of the coarsening dynamics . this is because here , once the system is left with three domains , it remains in that state . to simulate the toy model efficiently , an algorithm suitable for rare event dynamics must be used due to the small rate of events @xcite . we use an algorithm which is performed by repeating the following steps : 1 . list all possible events @xmath268 and assign to them rates @xmath269 according to the rules of the model . 2 . choose an event @xmath128 with probability @xmath270 where @xmath271 . advance time by @xmath272 , where @xmath273 . the algorithm would be equivalent to a usual monte carlo simulation , where @xmath57 time step is equivalent to one monte carlo sweep , if in step @xmath274 , @xmath275 would be drawn from a poisson distribution @xmath276 $ ] . however , here we make an approximation by using the deterministic choice @xmath277 . we have simulated the dynamics for lattices of size up to @xmath278 . for simplicity we consider the case @xmath34 . an example of a typical behavior of the average domain size as a function of @xmath226 is shown in fig . [ fig_domsize ] . one can see that after an initial transient growth time the data fits very well with a @xmath279 behavior . simulations for different @xmath54 values indicate that , @xmath280 with @xmath281 . the toy model enables one to verify the scaling behavior ( [ loga ] ) and estimate the constant @xmath282 . this would be very difficult to do by simulation of the full model ( [ eq : dynamics ] ) . here we present the solution of a mean - field version of the toy model based on ideas presented by rutenberg and bray @xcite and derrida _ et al._@xcite in the study of the ordering dynamics in a one - dimensional scalar model . to construct the mean - field model we notice that since all steps in the toy model which involve exchange of particles between domains occur at a rate exponentially small in the size of domains , one can consider a model where dynamics occurs only in the vicinity of the smallest domains . the mean field approximation assumes that different domains are uncorrelated and does not distinguish between domains of different species . the second assumption relaxes the conservation of particles of each species . thus , in contrast to the systems studied by @xcite and @xcite we do not expect the mean field to become exact in the scaling limit . we define the mean - field model as follows : 1 . pick one of the smallest domains @xmath283 . 2 . pick @xmath284 domains @xmath285 and @xmath286 randomly and treat them as the neighbors of @xmath283 . 3 . pick @xmath274 more domains randomly say , @xmath287 and @xmath288 . 4 . eliminate @xmath289 and @xmath290 from the system and change the length of the domains @xmath287 and @xmath288 by @xmath291 steps @xmath57 to @xmath180 are performed simultaneously for all of the smallest domains in the system . here @xmath292 is the length of domain @xmath1 , and @xmath293 is the length of the smallest domain . steps @xmath57 to @xmath274 choose the smallest domain @xmath283 and its nearest neighbors ( see fig . [ fig_toy2 ] ) . step @xmath180 uses the fact that the dynamics occur in the model only in the vicinity of the smallest domain and eliminates the three domains @xmath283 , @xmath294 and @xmath295 joining them appropriately with the other domains @xmath296 , @xmath297 and @xmath298 . note that this mean - field dynamics does not take into account the time taken for these events to happen . this will be done later when we derive the growth law of domains . to solve the mean - field model we follow the method used in @xcite . let @xmath299 be the number of domains of size @xmath24 , irrespective of the type of particle it consists of . let @xmath300 be the length of the smallest domain and @xmath301 be the total number of domains at time @xmath226 . we will assume that @xmath299 has the following scaling form in the large - time limit . @xmath302 a solution of the model given in appendix [ meanfield ] yields @xmath303 where @xmath304 is given by @xmath305 the growth law for @xmath306 can be derived following @xcite : after the elimination of the smallest domain , @xmath307 increases by @xmath57 . this happens at a rate @xmath308 , namely , the inverse time required by a typical domain to cross a distance @xmath309 ( thus , causing the annihilation of @xmath283 ) . using @xmath310 we write @xmath311 from this one can obtain the scaling form of the average domain size , @xmath312 note that in this eq . @xmath304 does not depend , according to the mean - field solution , on either @xmath226 or @xmath54 . one can see from ( [ sform ] ) that for large @xmath24 , @xmath313 which was confirmed by the simulations of the toy model where @xmath314 ( see ( [ loga ] ) ) . a numerical evaluation of ( [ avexpp ] ) yields @xmath315 as compared with @xmath316 obtained from the toy model simulations . in section iii the partition function , @xmath52 , and correlation function , @xmath317 for finite @xmath161 , have been calculated in the thermodynamic limit . it is also of interest to obtain results for finite systems for the study of finite - size effects and the approach to the thermodynamic limit . recently a matrix ansatz method has been introduced to study one - dimensional non - equilibrium systems @xcite . it has been shown that in certain three - species models the steady - state weight and correlation functions can be represented as a product of matrices @xcite . in the ansatz a specific matrix is associated with each type of particle . then the unnormalized probability of a certain configuration is obtained from a matrix product . the matrices corresponding to the different species of particles satisfy an algebra derived from the dynamics of the model . a scalar , _ i.e. _ the weight of a configuration or some correlation function , is usually obtained by performing a trace over the product of matrices , or by multiplying both sides of the product of matrices by vectors . generalizing this method to replace matrices by tensors @xcite , we have been able to obtain recursion relations for the partition function and correlation function for finite systems for the special case @xmath34 . the recursion relations are then used to obtain the partition function and correlation function @xmath318 for any @xmath161 in small systems . the results are used to study the scaling of the correlation function near the critical point @xmath12 ( infinite temperature ) , where the typical domain wall width diverges . it is convenient to consider the unnormalized weights , @xmath319 , defined through @xmath320 where @xmath321 is the probability of being in configuration @xmath322 . the partition sum @xmath52 is given by @xmath323 where the sum is over all configurations with @xmath34 . we generalize the matrix ansatz and construct the steady - state weight , @xmath324 , from a product of tensors each corresponding to a particle located in a specific place on the lattice . the contraction of the tensors yields a tensor which is then contracted with ` left ' and ` right ' tensors to generate a scalar . the three tensors which represent the different type of particle are defined as rank @xmath325 tensors through the following tensor products of square matrices @xmath326 here @xmath327 is a unit matrix . the matrices @xmath328 and @xmath329 will be chosen in what follows to satisfy a commutation relation which will be dictated by the detailed balance condition . to define @xmath324 we introduce the following notation : the contraction of two rank @xmath325 tensors @xmath330 and @xmath331 , where @xmath332 and @xmath333 are square matrices , according to the rule @xmath334 is denoted by @xmath335 . the contraction of a rank 6 tensor @xmath336 with a ` left ' rank @xmath274 tensor @xmath337 , where @xmath338 are transposed vectors , and a ` right ' rank @xmath274 tensor , @xmath339 , where @xmath340 are vectors , defined through @xmath341 is denoted by @xmath342 . using these definitions we write the steady - state weight of the system as @xmath343 \vert { \cal v } \rangle \ ; , \ ] ] where @xmath344 , @xmath45 and @xmath43 are the occupation variables defined in ( [ ocup ] ) . the expression states that a tensor @xmath345 is present at place @xmath1 in the tensor product if site @xmath1 is occupied by an @xmath6 particle , a tensor @xmath346 is present if site @xmath1 is occupied by a @xmath7 particle , and a tensor @xmath347 is present if site @xmath1 is occupied by a @xmath8 particle . the action of the tensor product on @xmath348 and @xmath349 ( to be specified later ) produces the scalar @xmath324 . it is straightforward to show using detailed balance that a necessary condition for eq . ( [ eq : ansatz ] ) to be the steady - state weight is that the following commutation relations are satisfied between @xmath350 , @xmath351 , and @xmath352 : @xmath353 using ( [ eq : tensors ] ) one can verify that these commutation relations are satisfied provided that @xmath354 . this deformed commutator is of relevance in other stochastic systems @xcite . a representation of the matrices @xmath328 and @xmath329 which satisfies this commutation relation can be obtained as follows : let @xmath355 denote a basis set ( @xmath356 ) forming a vector space . in this basis we choose the matrices so that , @xmath357 while for @xmath358 , @xmath359 . an explicit form for @xmath329 and @xmath328 is given by the following @xmath360 square matrices @xmath361 to obtain @xmath324 , @xmath362 and @xmath349 have to be specified . this should obviously be done so that @xmath324 is non - zero if the ansatz is to give a non - trivial result . we consider a general tensor product which corresponds to some configuration . the product has @xmath98 tensors of each type @xmath363 and @xmath352 , which results in a tensor product of three matrix products . using ( [ eq : tensors ] ) it can be seen that each matrix product contains @xmath98 matrices of each type @xmath328 , @xmath329 and @xmath327 . since @xmath329 and @xmath327 are diagonal , while @xmath328 acts to its left as a lowering matrix ( see ( [ eq : matrix ] ) ) , choosing @xmath364 , and @xmath365 will give a non - zero @xmath366 . this makes clear that the minimal size choice for the vector - spaces is @xmath367 . under this choice it is easy to see that in the ground states one has @xmath368 which corresponds to a ground state energy @xmath84 . however , the choice of @xmath369 and @xmath370 is determined only up to some multiplicative factors . these factors may be used to shift the ground state energy of the system . for example choosing @xmath369 as before with @xmath371 will shift the ground state energy to @xmath372 . in the following the factors are taken to be @xmath57 . finally we would like to remark that usually when using the matrix ansatz for systems with periodic boundary conditions it is often convenient to use a trace of the matrix product @xcite . in this case this is not possible since the trace of our tensor product is always zero . in terms of the ansatz ( [ eq : ansatz ] ) the partition function @xmath52 is given by @xmath373 note that we are using the canonical ensemble since all tensor products with unequal number of particles do not contribute to @xmath52 this is easily seen since in these cases there are always more than @xmath98 @xmath328 matrices acting on one of the vectors @xmath374 . to obtain the partition function we derive a recursion relation for @xmath375 one can see that @xmath376 . rewriting @xmath377 as @xmath378 and using relations ( [ eq : tensors ] ) and ( [ eq : matrix ] ) the following recursion relation can be derived @xmath379 the boundary conditions for this recursion relation is given by the no particle partition function @xmath380 if @xmath381 and is zero otherwise . for small systems ( up to @xmath382 ) for which the recursion relation is tractable analytically on mathematica , we obtained the partition function @xmath383 as a polynomial in @xmath54 . as expected the first @xmath384 terms of the polynomial match the first @xmath384 terms of the expansion of ( [ psum ] ) up to a factor of @xmath385 due to the energy shift in the ground state . for larger @xmath5 we solve the recursion relation numerically . we note that ( [ eq : reccursion1 ] ) could have been derived directly from the definition of the partition function without recourse to the tensor ansatz . however , we believe the utility of the ansatz lies in the ease with which correlation functions can be manipulated and relations such as that of the next subsection derived . the correlation function @xmath317 is given in terms of the ansatz by @xmath386 using relations ( [ eq : tensors]),([eq : commutation ] ) and ( [ eq : matrix ] ) we obtain @xmath387 where we define the object @xmath388 through , @xmath389 a recursion relation for @xmath388 can be obtained , similarly to the recursion relation for the partition function ( [ eq : reccursion1 ] ) . using ( [ eq : tensors ] ) and ( [ eq : matrix ] ) gives @xmath390 the boundary conditions for the recursion relation are obtained by noting that @xmath391 , _ i.e. _ , @xmath392 . using the same methods one can obtain recursion relations for all other correlation functions . the recursion relations are solved numerically for finite systems . this is done by first solving numerically for @xmath393 , and then using the result as boundary conditions for the recursion relation ( [ eq : u recurr ] ) . owing to ( [ a1ar ] ) we are ultimately interested in @xmath394 . using these recursion relations we have calculated the correlation function @xmath395 , which is a measure of the phase separation in the system , for various system sizes . when @xmath396 is close to zero the system is phase separated . in the disordered case , @xmath12 , the value of the correlation function is @xmath397 , approaching @xmath398 in the thermodynamic limit . the results are shown in the inset of fig . [ order ] one can see that the system is phase separated for small values of @xmath54 while for @xmath54 close to @xmath57 the system is disordered . the range of @xmath54 values for which the system is phase separated increases as the system size increases . the natural scaling variable near the critical point @xmath12 is @xmath399 , the ratio between the domain size @xmath98 and the domain wall width @xmath400 . in fig.([order ] ) the correlation function is plotted as a function of the scaling variable . one can see that the data collapse improves as the system size increases . this scaling variable was also suggested by the form of the partition function ( [ psum ] ) in section iii c. in the following we discuss possible generalization of the model to @xmath401 species . to demonstrate how this might be done we first discuss the case @xmath402 . we then comment briefly on @xmath403 . we now define a @xmath180-species model and argue that it phase separates . consider a ring where each site is occupied by either an @xmath6 , @xmath7 , @xmath8 or @xmath404 particle . the model evolves according to the following procedure : at each time step two nearest neighbors are chosen randomly and exchanged according to the rates , @xmath405 as before the model conserves the number of particles of each species . note that several other generalizations of the model to four species are possible . however , for simplicity , we discuss only the model defined by ( [ eq : four ] ) with @xmath35 . we now argue that the system phase separates into a configuration of the form @xmath406 ( where each letter now indicates a domain ) as long as the densities of particles of each species are non - zero . note that in the model @xmath17 , @xmath18 , @xmath407 , @xmath408 , @xmath15 and @xmath409 boundaries are stable while reverse boundaries @xmath410 , @xmath16 , @xmath411 , @xmath412 , @xmath19 and @xmath413 are unstable . as in the case of the three species model the system , starting from a random initial condition , evolves on a short time scale ( _ i.e. _ , which is not determined by the size of the system ) into a metastable configuration where only stable domains are present . this configuration then slowly coarsens by slow diffusion of particles through neighboring domains . the system will finally reach the most stable state where the number of domains is minimal . one can easily check that this configuration is given by @xmath406 . note that the system may exhibit other metastable states . for example , a state composed of @xmath414 is also stable under the choice of rates ( [ eq : four ] ) . however , since this state is composed of more domains than the @xmath180-domain state , some of the domains are necessarily smaller . according to the argument presented in section ii the relaxation time of this sequence ( proportional to @xmath415 where @xmath128 is the typical domain size ) is much shorter than the relaxation time of the @xmath180-domain state . therefore the @xmath180-domain state is more stable so that the system will finally evolve into it . in considering @xmath403 models one finds that for some choices of transition rates several states with the minimum number of domains may become metastable . for example , for @xmath416 it is possible to choose transition rates for which both @xmath417 and @xmath418 are locally stable . the relative stability ( and thus the resulting phase separated state ) may be found by determining the relaxation time of these states using simple considerations such as those presented in section ii . as is the case of @xmath3 detailed balance is found to be satisfied for certain densities and transition rates for @xmath419 . the condition for this is derived in appendix [ dbappen ] . in this case the relative stability of metastable states could be determined by comparing free energies . in this paper a model of three species of particles diffusing on a ring previously introduced in @xcite has been studied . the model is governed by local dynamics in which all moves compatible with the conservation of the three densities are allowed . we argue that phase separation should occur as long as all densities are non - zero . in the special case of equal densities we find that the steady state generated by the local stochastic dynamics is exactly given by a long - range asymmetric hamiltonian . phase separation for this case is explicitly demonstrated . the model provides an explicit example for the mechanism leading to phase separation or breaking of ergodicity in systems with _ local _ stochastic dynamics . although we did not succeed in solving the steady state in the case of non - equal densities of particles , there is a strong evidence that phase separation still occurs . in order to investigate further the case of equal densities we employed a generalized matrix ansatz to calculate correlation function for finite - size systems . the novel structure of the ansatz may give some clue as to handle other @xmath0 models which have so far resisted solution . the dynamics of phase separation reduces to a coarsening problem where the typical domain size grows logarithmically in time . this results from the elimination of the domains at a rate exponentially small in their size . the slow dynamics poses a problem of how to access the scaling regime numerically . with direct numerical simulations only small systems can be studied ( see fig . however , by employing a toy model in which domains rather than individual sites are updated one can simulate much larger systems and probe the scaling regime ( see fig . 5 ) . such ideas of updating domains have been used before in the study of coarsening @xcite . with the aid of the toy model it should be possible to study other aspects of the scaling regime associated with the slow dynamics and escape from metastable states . generalizations are possible to models with @xmath419 species . we have discussed some possibilities and have shown that phase separation may take place , although the structure of the set of metastable states is more complicated . as was the case for @xmath3 , conditions for detailed balance with respect to a long - range hamiltonian may be determined . the problem of phase separation and coarsening is of interest also in the broader context of phase transitions in one dimensional systems . here the existence of conserved quantities results in certain local transition rates being zero . it would be interesting to generalize this study to models in which no conserved quantity exists and all local rates are non - vanishing . also , another open problem is to calculate the steady state of the present model in the case of non - equal densities . * acknowledgments * : mre is a royal society university research fellow and thanks the weizmann institute and the einstein center for warm hospitality during several visits . the support of the minerva foundation , munich germany , the israeli science foundation , and the israel ministry of science is gratefully acknowledged . computations were performed on the sp2 at the inter - university high performance computer center , tel aviv . we thank s. wiseman for helpful advice on programming and m. j. e. richardson for careful reading of the manuscript . to solve the mean - field model we follow the method used in @xcite . let @xmath299 be the number of domains of size @xmath24 , irrespective of the type of particle it consists of . let @xmath300 be the length of the smallest domain and @xmath301 be the total number of domains at time @xmath226 . we will assume that @xmath299 has the following scaling form in the large - time limit . @xmath420 after the elimination of the smallest domain as given by ( [ toy2 ] ) , @xmath421 and @xmath4 change according to @xmath422 using the scaling form ( [ nlform ] ) we have , @xmath423\ ] ] where @xmath424 and we have expanded in @xmath425 . now substituting these in ( [ modif ] ) , it is straightforward to show that @xmath426 using the laplace transform , @xmath427~f(x)\ ] ] one can show that @xmath428 satisfies the differential equation , @xmath429[2 \phi(p ) + e^{-p}]\ ] ] since @xmath430 , by expanding ( [ phip ] ) to order @xmath431 one obtains @xmath432 . the solution of ( [ phip ] ) with boundary conditions @xmath433 and @xmath434 for @xmath435 is @xmath436 where @xmath437/t = -\log(x ) - \gamma - \sum_{n=1}^{\infty } ( -x)^n/(n!\ n)$ ] and @xmath438 is the euler constant . inverse laplace transform of ( [ solu ] ) gives the domain size distribution . from ( [ solu ] ) and using the expansion @xmath439 , where the average is with respect to @xmath440 , we get @xmath441 we now define the most general @xmath4 species model , where @xmath442 . let @xmath443 denote a variable at site @xmath1 of a ring of size @xmath5 , which takes values @xmath444 . @xmath445 means that site @xmath1 is occupied by a particle of type @xmath128 . the system evolves by a random sequential , nearest - neighbor exchange dynamics , with the following rates : consider a particle exchange between sites @xmath62 and @xmath63 , where @xmath451 and @xmath452 ( _ i.e. _ in the bulk , note that site @xmath57 is chosen arbitrarily ) . expanding the product in ( [ eq : sweight ] ) , it is easy to verify that @xmath453 since this hold for any @xmath454 , and is irrespective of the number of particles of each species , the steady - state weight ( [ eq : sweight ] ) satisfies detailed balance for all nearest - neighbor exchanges in the bulk . if the weights ( [ eq : sweight ] ) are translationally invariant then detailed balance will also hold for exchanges between sites @xmath57 and @xmath5 . thus , to complete the proof of detailed balance it is sufficient to demand that ( [ eq : sweight ] ) is translationally invariant . to do this we relabel sites @xmath455 . the weight then becomes @xmath456 where @xmath457 is identical to @xmath458 . rewriting this equation by relabeling the indices we obtain , @xmath459 \ \prod_{k=1}^{n-1}\frac{q(x_k , x_n)}{q(x_n , x_k ) } \ ; .\ ] ] comparing ( [ eq : sweight3 ] ) with ( [ eq : sweight ] ) and noting for example that , @xmath460^{n_l } \ ; , \ ] ] one can see that ( [ eq : sweight ] ) is translational invariant if @xmath461^{n_{l } } = 1 \ ; , \ ] ] for every @xmath462 . thus , detailed balance holds if ( [ condition ] ) is satisfied . we note that for given densities @xmath463 the manifold of solutions for the rates is of @xmath464 dimensions .

recent progress in the physics of bose - einstein condensation ( bec ) of atomic gases ( see refs . @xcite and references therein ) stimulated a further development of theoretical methods used previously for the description of superfluidity , superconductivity , and collective effects in heavy atoms and nuclei ( see refs.@xcite ) . these methods treat condensates as mean fields , while interactions between particles , according to the bogolubov theory , lead to formation of fluctuations ( see refs.@xcite ) . in the case of a positive scattering length , the condensate ( mean field ) is stable , while fluctuations can be treated by using a bogolubov transformation to quasiparticle states . recent experiments @xcite drew attention to the physics of condensates with a negative scattering length . in this case , a free condensate ( with a homogeneous mean field ) is unstable , collapsing within a finite time @xcite , while a trapped condensate is stable at the low occupations of the experiment @xcite . in the jila experiments on rb @xcite , the effect of feshbach resonance ( see ref . @xcite ) was used for changing the sign of the scattering length . this effect allows the formation of a condensate with a positive scattering length , followed by a study of the dynamics of collapse after applying a sudden change of sign of the scattering length . a formal extension of the bogolubov transformation to the case of negative scattering lengths would lead to complex energies of the quasiparticles , but actually such a transformation does not exist in this case , as was proven by bogolubov ( see ref . nevertheless , the development of fluctuations can substantially affect the dynamics of an unstable condensate ( see the theoretical analysis @xcite of the jila experiments @xcite ) . the present paper describes the dynamics of fluctuations by using the parametric approximation employed in ref . @xcite for a model problem of a two - mode atom - molecule condensate , and in ref.@xcite for a description of quantum effects on curve crossing . this method is similar to the parametric approximation used in quantum optics ( see ref . @xcite ) , and in the description of interaction of electromagnetic and matter waves in refs . @xcite . it is also closely related to the linearization approximation in quantum soliton theory ( see ref . @xcite and references therein ) and to the theory of quantum integrals of motion ( see refs . @xcite and references therein ) . the fluctuations are treated here by a linearization of the exact quantum equations of motion for the field operators . in the case of a stable bec , the resulting linear equations describe the motion of quantum oscillators , corresponding to the bogolubov quasiparticles . however , in the case of unstable condensates , these equations describe inverted oscillators , leading to a rapid growth of the fluctuations . this growth remains super - exponential as long as the fluctuations do not reach a macroscopic occupation . this dynamics offers a more precise picture compared to the exponential growth obtained from a liapunov analysis of the classical mean field . initially , the dynamics can be described as a quantum process of spontaneous decay of the condensate into the fluctuation vacuum ( see refs.@xcite ) , accompanied by the formation of entangled pairs ( see also refs . @xcite ) . as in the case of fluctuations in a stable condensate ( see ref . @xcite ) , the pairs are formed in two - mode squeezed states . the formally calculated energy of a bogolubov quasiparticle becomes complex also when the energy of the excitation is lower than the energy of a particle in the condensate state . this case can take place if the condensate itself occupies an excited translational state , as realized recently in the nist experiments on four - wave mixing of atomic condensate waves @xcite . the loss of condensate atoms in such a situation was studied in ref . @xcite , using the method of complex scattering lengths , without taking into account second - quantized properties of the atomic scattering . the results of the present theory , which take into account these properties , are in agreement with ref . @xcite at short collision times [ see eq.([ndotstat ] ) below ] . however , at long collision times the effects of bose enhancement lead to a substantial increase of the losses [ see eq . ( [ ndotenh ] ) below ] . notice that the necessity of accounting for the quantum effects was mentioned in ref . @xcite itself . these effects should appear under the conditions of the nist experiments @xcite . the present method can describe only qualitatively this case of large losses , comparable in magnitude to the initial occupation . this case may be treated by the incorporation of quantum effects into the slowly - varying - envelope approximation developed in ref . @xcite . quantum effects on the loss incurred by collisions of two condensate wavepackets were studied in ref . @xcite , using an approach similar to the present one . however , the results of the present work disagree with those of ref . the losses quoted in ref . @xcite , which have been evaluated only for short collision times , differ also from the results of ref.@xcite . these disagreements are apparently due to some unjustified neglections in ref . @xcite ( see discussion in sec . [ gauss ] ) . as in the case of fluctuations in a ground - state condensate , the scattered particles are formed here in entangled ( squeezed ) states . therefore the processes considered here may open up possibilities for the formation of an entangled gas remained after the decay of a condensate , or after the collision of wavepackets . the entangled gas can have many application ( see ref . @xcite ) . this paper is organized as follows . section [ ground ] considers the dynamics of instabilities in a ground - state condensate with a negative scattering length . the equations of motion for the condensate and the fluctuations are derived in sec . [ condsep ] by using a projection onto a coherent state . the equations for the fluctuations are solved in sec . [ stabunstmod ] . in the dependence on the excitation energy , the fluctuations are divided into two classes , of stable and unstable modes , which are respectively associated with regular and inverted oscillators in sec.[reginvosc ] . the dynamics and statistical properties of the instabilities are considered in sec . [ instdyn ] , and sec.[conddepl ] describes the dynamics of condensate depletion . the collision of two matter waves is studied in sec.[twocond ] . general equations for the condensate and the fluctuations are derived in sec . [ tcsep ] . section [ loss ] considers elastic losses in collisions of two matter waves . the collisions of rectangular and gaussian wavepackets are treated in detail in secs . [ rect ] and [ gauss ] , respectively . a system of units in which planck s constant is @xmath0 is used below . consider a homogeneous gas of bosons of mass @xmath1 , in a finite quantization box of volume @xmath2 , described by the following hamiltonian in the momentum representation ( see refs . @xcite ) , @xmath3 here @xmath4 is the annihilation operator for bosons in the momentum @xmath5-mode , @xmath6 is the kinetic energy , and @xmath7 describes the elastic scattering . in the model case of a zero - range contact potential , @xmath8 is a constant , and can be expressed in terms of the elastic scattering length @xmath9 as @xmath10 ( a modification of the contact - potential model was recently proposed in ref . @xcite which allows to avoid ultraviolet divergences ) . the hamiltonian ( [ h ] ) leads to the following equations of motion for the annihilation operators : @xmath11 following a general method @xcite let us separate the condensate part by a projection onto a coherent state @xmath12 defined so that @xmath13 describes the ground - state condensate occupying the momentum state @xmath14 . the annihilation operator can be written as @xmath15 where @xmath16 , the operators @xmath17 represent the non - condensate part ( fluctuations ) having the zero expectation value , @xmath18 and @xmath19 is the conserved total number of bosons . the exponential factor in eq . ( [ psiphi ] ) shifts the zero energy to the energy of non - excited condensate . substitution of eq . ( [ psiphi ] ) into eq . ( [ pside ] ) , combined with the property ( [ csiev ] ) , produces equations of motion for @xmath20 and @xmath21 . assuming a small occupation of the fluctuations , let us keep in these equations only terms of lowest non - vanishing order in @xmath22 and @xmath23 , obtaining the following equations : @xmath24 where @xmath25 and @xmath26 is the condensate density . this approximate method , that leads to linear operator equations for the fluctuations , is similar to the parametric approximation used in quantum optics ( see ref . it has been applied to the study of a molecular bec dissociation in refs . @xcite ) , and is valid as long as the depletion of the condensate is negligibly small . an applicability criterion for this method is presented in sec . [ conddepl ] below [ see eq . ( [ valran ] ) ] . the present approximation is gapless [ see eq . ( [ lambda ] ) below ] and , therefore , non - conserving ( see ref . @xcite ) , although the average number of the atoms is conserved [ see eq . ( [ nbos ] ) above ] . equation ( [ csip ] ) combined with the corresponding equation for @xmath27 provide a set of two linear operator equations ( similar to the set considered in refs . this set can be solved exactly , using common methods reducing it to an eigenproblem , by assuming the time - independence of @xmath28 and @xmath29 . this assumption is corroborated by eq . ( [ phi ] ) , which shows that the variation of @xmath30 is of the second order in @xmath21 and therefore can be neglected . indeed , the validity of the parametric approximation requires a small variation of the condensate density , and the dominant part of the condensate phase is accumulated in the exponential factor in eq . ( [ psiphi ] ) . as in the case considered in ref . @xcite , the form of the solution depends on the ratio of the detuning @xmath31 to the coupling strength @xmath29 . all momentum modes observing @xmath32 are unstable , and the corresponding solutions attain the aperiodic form @xmath33 where @xmath34 the condition ( [ inst ] ) can be obeyed only when @xmath35 , i. e. , with a negative scattering length , and for excitations with an energy @xmath36 . otherwise , whenever @xmath37 , the modes are stable and the solutions are oscillatory functions of the form @xmath38 in a case of intervening mode ( @xmath39 ) the solution is a linear function of time @xmath40 this case can take place for the spurious ( goldstowne ) mode with a vanishing excitation energy ( @xmath41 , see refs.@xcite ) , or at @xmath42 . the stable modes can be treated by the bogolubov transformation . it is therefore instructive to establish a connection between the present and the bogolubov methods . the equations of motion ( [ csip ] ) can be derived from the following hamiltonian constituent involving two counter - propagating modes , @xmath43 the transformation @xmath44 introduces a new pair of field operators , @xmath45 and @xmath46 , whose equations of motion are decoupled by separating the hamiltonian into @xmath47 , where @xmath48 ( hereafter in this subsection the arguments @xmath5 and @xmath49 are dropped out , for simplicity s sake . ) a system described by the hamiltonian ( [ hj ] ) was solved in ref . @xcite , and the solutions have a form similar to eq . ( [ solap ] ) , or eq . ( [ solp ] ) , according to the ratio of @xmath50 to @xmath51 . let us introduce canonically conjugate operators for the coordinates @xmath52 and the momenta @xmath53 , such that @xmath54 , writing out the hamiltonian ( [ hj ] ) as @xmath55 if @xmath56 , this hamiltonian describes a harmonic oscillator and can be transformed into the standard form representing free bogolubov quasiparticles [ the total transformation , incorporating ( [ csi12 ] ) , being indeed the bogolubov transformation ] . the motion of the harmonic oscillator is periodic , in agreement with the periodic behavior of the solution ( [ solp ] ) . a different situation takes place if the condition ( [ inst ] ) holds . in this case one of the coefficients ( preceding @xmath57 or @xmath58 ) is negative , and the hamiltonian ( [ hpq ] ) represents an inverted oscillator . it can not be transformed to a form representing free quasiparticles , and therefore a bogolubov transformation can not exist , in agreement with ref . the infinite aperiodic motion of the inverted oscillator is exhibited in eq . ( [ solap ] ) and in the instability of the mode . in a case of intervening mode ( @xmath39 ) eq . ( [ lambda ] ) gives @xmath59 , one of the coefficients ( preceding @xmath58 or @xmath57 ) vanishes , and the hamiltonian ( [ hpq ] ) represents a motion without restoring forces or an oscillator with infinitely large mass , respectively ( see ref . such systems demonstrate linear time - dependence of @xmath53 or @xmath52 , respectively , in agreement with the solution ( [ soli ] ) . the bogolubov transformation does not exist in this case as well . in the case of the goldstowne mode ( @xmath60 ) the hamiltonian ( [ hp ] ) has already the form ( [ hj ] ) , and the transformation ( [ csi12 ] ) is not necessary . this mode brings only one term @xmath61 into the total hamiltonian , in agreement with ref . @xcite ( up to the exchange of the coordinate and the momentum ) . consider the dynamics of fluctuations in an unstable bec , starting from the state of a pure condensate @xmath62 this problem is not only of a speculative interest ; it actually describes the recent jila experiment @xcite , in which the elastic scattering length was varied very fast from the zero value ( of the ideal condensate ) to a negative value ( of an unstable condensate ) . given the initial conditions ( [ purecond ] ) , eq . ( [ solap ] ) leads to the following dynamics for the occupation of the unstable modes , @xmath63 and for the two - particle correlations , @xmath64 it is not surprising that this dynamics is similar to the dynamics of spontaneous dissociation of a molecular condensate , studied in ref . @xcite , as the non - condensate modes can be described in terms of the fields @xmath65 [ see eq . ( [ csi12 ] ) ] by the hamiltonian ( [ hj ] ) , which is equivalent to the one considered in ref . @xcite . by analogy to the results of ref . @xcite , the non - condensate atoms are formed in squeezed states involving atom pairs of opposite momenta . these states are similar to the two - mode squeezed states of fluctuations in a stable bec studied in ref . @xcite . the states are perfectly squeezed and have zero variance in the relative number of particles with momenta @xmath5 and @xmath66 , being optimally entangled in the entropic sense . the uncertainty of the generalized coordinates @xmath67 where the operators @xmath68 ( with @xmath69 ) are defined by eq . ( [ csi12 ] ) , attains the maximal and minimal values @xmath70 at the two orthogonal values of the respective phase angles @xmath71 , @xmath72 a common method for studying bec stability is the liapunov analysis ( see e. g. ref . this method produces equations for small perturbations @xmath73 to the mean field . these equations are equivalent to eq . ( [ csip ] ) , in which the operators @xmath22 are replaced by @xmath74-number functions @xmath73 . in the case studied here one obtains for the liapunov exponents the expression @xmath75 given by eq.([lambda ] ) . this type of analysis attributes an exponential gain to the small perturbation , @xmath76 , when the condition ( [ inst ] ) is obeyed . this result is in qualitative agreement with the quantum analysis used here , which attributes an even faster ( super - exponential ) gain to the fluctuations [ see eq . ( [ den ] ) ] , as @xmath77 the dynamics of the fluctuations has been studied in ref.@xcite . it is shown there that the quantum ( squeezed ) state always grows faster than a classical ( coherent ) state . therefore the liapunov analysis concerning classical fields underestimates the fluctuation growth , becoming correct only when @xmath78 , where the fluctuations are large enough to be considered as classical ones . a formal application of the bogolubov transformation at @xmath79 ( although incorrect , as had been observed by bogolubov himself ; see ref.@xcite ) produces quasiparticles with imaginary energies @xmath80 , which correspond to exponential growth . however , the transformed field operators do not obey the proper bose commutation rules . the homogeneous boson gas with a negative scattering length considered above always has unstable modes due to the continuous nature of the excitation spectrum . the situation changes in the case of a trapped gas having a discrete excitation spectrum . following ref . @xcite , consider @xmath81 as a function of the condensate density @xmath82 for a fixed value of @xmath83 chosen so that @xmath84 , where @xmath85 is the energy of an excited state of the trapped condensate . at low condensate density @xmath86 is imaginary , corresponding to the well - known stability of a trapped condensate with a negative scattering length at low occupations ( see ref.@xcite ) . at @xmath87 the value of @xmath88 becomes real , and therefore describes an instability of the condensate ( leading to its collapse ) . although the present approach does not take into account the inhomogeneity of the condensate and the variation of its size , the dependence of @xmath89 on @xmath82 [ see eq . ( [ lambda ] ) ] is in qualitative agreement with the one obtained from the numerical calculations of ref . @xcite for the monopole mode . let us consider now the dynamics of condensate depletion . in the limit @xmath90 , eqs . ( [ den ] ) , ( [ cor ] ) , and ( [ phi ] ) lead to the following equation for the condensate density , @xmath91 where @xmath92 ) and stable ( @xmath93 ) modes , respectively , @xmath94 is the boundary between the stable and unstable modes , and the contact potential ( [ ua ] ) is used hereafter in this section . the statistical weight of intervening mode with @xmath39 is negligible compared to the continuums of the stable and unstable modes . unlike a case of stable condensate , where the unrestricted behavior of the goldstowne mode requires a special treatment ( see ref . @xcite ) , in the present case the effect of linearly - growing intervening modes is negligible compared to the effect of exponentially - growing unstable modes . substitution of @xmath95 allows us to express @xmath96 in terms of the weber functions @xmath97 ( see ref . @xcite ) : @xmath98 where @xmath99 , and @xmath100 is the nonlinear interaction time ( see ref . approximate expressions for @xmath96 , holding at small and large values of @xmath49 , have the following respective forms , @xmath101 the contribution of the stable modes can be evaluated by substituting @xmath102 , as a result of which @xmath103 its limiting values can be approximately expressed as @xmath104 finally , the rate of condensate depletion can be expressed as @xmath105 @xmath106 at the two distinguishable loss regimes . at the regime of non - stationarity @xmath107 the depletion is mostly due to an occupation of the stable modes , which carry much more statistical weight than the unstable modes [ @xmath108 , see eqs.([iinsta ] ) and ( [ istaba ] ) ] . the statistical weight is infinitely large , leading to a singularity in the depletion rate ( [ ndot ] ) , at @xmath109 . this singularity is absent in eq . ( [ deltan ] ) , and can be eliminated by using of an interatomic interaction with a finite range in place of the contact potential ( [ ua ] ) . at @xmath110 the occupation of the modes with @xmath111 begins to decrease due to their oscillating behavior , leading to decrease of the depletion rate . at the regime of bose - enhancement @xmath112 the depletion is dominated by the exponentially increasing occupation of the unstable modes [ @xmath113 , see eqs.([iinsta ] ) and ( [ istaba ] ) ] . the approximate expressions ( [ deltan ] ) are compared in fig . [ fg ] to results of numerical integration of eq . ( [ ndot ] ) . one should be reminded that the parametric approximation , on which all the analysis above rests , is applicable only as long as @xmath114 . in the case of one excited mode @xcite , this approximation is in good agreement with numerical results while @xmath115 . in the case of a dilute condensate ( @xmath116 , see ref . @xcite ) , the validity range @xmath117 includes a region @xmath118 where the exponential bose - enhancement factor can reach a rather large value , of the order of @xmath119 ) predicts small depletion at times @xmath120 ( with @xmath121 , given the dilution parameter @xmath122 is small ) . therefore , the development of instabilities in bec , including collapse , starts with a delay time @xmath123 after turning on the interatomic interactions , although the collapse itself is a nonlinear phenomenon that can not be treated by the parametric approximation . under the conditions of the @xmath124rb experiment @xcite ( where @xmath125@xmath126 and @xmath127 nm ) , the expected delay is @xmath128ms ( cf . [ fg ] ) , in agreement with the experimentally observed delay time for the collapse . in this case the parametric approximation is valid as long as @xmath129 , allowing for values of the exponential bose - enhancement factor in eq . ( [ deltane ] ) exceeding 100 . it should be noted that the condensate dynamics can be sensitive to initial correlations and the spatial inhomogeneity neglected here . the approach developed in sec . [ ground ] above can also be applied to a collision of two free counterpropagating bec waves with momenta @xmath130 ( see ref . @xcite ) . in this case the annihilation operator should be represented as @xmath131 where @xmath132 here @xmath133 describe the two condensate fields with momenta @xmath130 , respectively , the coherent states @xmath134 are defined by @xmath135 and the conserved total number of atoms is @xmath136 substitution of eq . ( [ psi2p ] ) into eq.([pside ] ) gives the following equations of motion for the fluctuations , @xmath137 where @xmath138 and @xmath139 are the densities of the condensate waves . the term @xmath140 \hat{\xi } \left ( { \bf p}-2{\bf p}_{0},t\right ) \nonumber \\ & & + \phi ^{2}_{+}\left ( t\right ) u\left ( |{\bf p}-{\bf p}_{0}|\right ) \hat{\xi } ^{\dag } \left ( -{\bf p}+2{\bf p}_{0},t\right ) \ } \nonumber \\ \times&&\exp\{2i[s_+(t)-s_-(t)]\ } \nonumber \\ + \{&&\phi ^{*}_{+}\left ( t\right ) \phi _ { -}\left ( t\right ) \left [ u\left ( 2{\bf p}_{0}\right ) + u\left ( |{\bf p}+{\bf p}_{0}|\right ) \right ] \hat{\xi } \left ( { \bf p}+2{\bf p}_{0},t\right ) \nonumber \\ & & + \phi ^{2}_{-}\left ( t\right ) u\left ( |{\bf p}+{\bf p}_{0}|\right ) \hat{\xi } ^{\dag } \left ( -{\bf p}-2{\bf p}_{0},t\right ) \ } \nonumber \\ \times&&\exp\{2i[s_-(t)-s_+(t)]\}\end{aligned}\ ] ] describes couplings of fluctuations with momenta different by @xmath141 . whenever @xmath142 the disparity of the energies of such fluctuations far exceeds the coupling strengths , and the term @xmath143 can be neglected in eq . ( [ csip2 ] ) . this allows us also to neglect correlations of such fluctuations in the equations of motion for the condensate fields , which then attain the form @xmath144 equation ( [ csip2 ] ) with @xmath143 neglected has solutions of the form eqs.([solap ] ) , ( [ solp ] ) , or ( [ soli ] ) , according to the ratio of @xmath145 to @xmath146 , in which @xmath51 , @xmath50 , and @xmath75 are respectively replaced by @xmath147 , @xmath146 , and @xmath148 the condition of instability @xmath149 can be obeyed now even for positive values of the scattering length ( and hence of @xmath8 ) , as @xmath146 contains a negative term @xmath150 [ see eq . ( [ d2 ] ) ] . in the case of a contact potential ( [ ua ] ) an instability takes place when @xmath151 or when @xmath152 where @xmath153 whenever condition ( [ highp ] ) is obeyed , the momenta @xmath154 are given approximately by @xmath155 the instability interval can include the region @xmath156 . it may appear that this fact can lead to the absurd conclusion that a ground - state condensate ( @xmath157 ) with a positive scattering length is unstable . however , the analysis above is valid only when the condition ( [ highp ] ) is obeyed . otherwise , the terms neglected in eqs.([csip2 ] ) and ( [ phi2 ] ) have to be taken into account , leading to stability of the condensate at @xmath157 . as in the case of a ground - state condensate , the scattered atoms with momenta @xmath5 and @xmath66 are correlated , forming optimally entangled and optimally squeezed states [ see discussion following eq.([cor ] ) ] . consider now the elastic - scattering loss of atoms from the two colliding counterpropagating waves bounded only by the normalization box . substitution of eqs . ( [ den ] ) and ( [ cor ] ) ( with the replacement of @xmath50 , @xmath51 , and @xmath75 by @xmath146 , @xmath147 , and @xmath158 , respectively ) into eq . ( [ phi2 ] ) leads to the following equations for the two condensate densities , @xmath159 where latexmath:[\[\begin{aligned } & & i_{\text{unst}}={1\over m p_0}\int\limits^{p { } _ { + } } _ { p { } _ { -}}p^{2}{\sinh\left ( 2\lambda _ { 2}\left ( p\right ) t\right ) \over \lambda _ { 2}\left ( p\right ) } d p \nonumber \\ & & i_{-}={1\over m p_0}\int\limits^{p { } _ { -}}_{0}p^{2}{\sin\left ( 2 p\right ) |}d p \label{iunstpm } \\ & & i_{+}={1\over m p_0}\int\limits^{\infty } _ { p { } _ { + } } p^{2}{\sin\left ( 2|\lambda _ { 2}\left ( p\right ) |t\right ) \over the contributions of the unstable ( @xmath96 ) and stable ( @xmath161 ) modes , respectively . the substitution of @xmath162 allows us to express @xmath96 in terms of the modified struve function @xmath163 ( see ref . @xcite ) , as @xmath164 where @xmath99 , and here @xmath165 . approximate expressions for @xmath96 at small and large values of @xmath166 have the respective forms @xmath167 given the condition @xmath168 , the substitution of @xmath169 produces the estimate @xmath170 the integral @xmath171 can be estimated under the same condition by retaining only the first term in the taylor series for @xmath172 . given the opposite condition , @xmath173 , the contributions of the stable modes @xmath174 can be evaluated by substituting @xmath175 resulting in @xmath176 where @xmath177 and @xmath178 ( with @xmath179 ) . the total contribution of the stable modes , given by @xmath180 where @xmath181 is the modified bessel function ( see ref.@xcite ) , can be approximately expressed as @xmath182 combining eqs . ( [ npm ] ) , ( [ iinst2 ] ) , and ( [ ipm ] ) , one obtains the total rate of the condensate loss @xmath183 unlike the case of ground - state condensate , characterized by a single time - scale @xmath184 , an additional time - scale , @xmath185 , appears in the present case , leading to appearance of a third regime of the condensate loss . at short times , @xmath107 , the loss is dominated by the occupation of the stable modes due to their higher statistical weight . at the regime of non - stationarity @xmath168 , the wavefunction of the scattered particles does not have enough time to become stationary , and the estimate ( [ imnonst ] ) gives the loss rate @xmath186 the singularity at @xmath109 has the same nature as in the case of the ground - state condensate with a negative scattering length [ see discussion following eq.([deltan ] ) ] . at the regime of stationarity , @xmath187 , the loss rate becomes time independent , and given by @xmath188 in full agreement with theory @xcite . at the regime of bose - enhancement , @xmath112 , the loss is principally determined by the occupation of unstable modes which increase exponentially due to the effect of bose enhancement . the loss rate can then be approximately expressed as @xmath189 the approximate expressions ( [ npmdotn ] ) , ( [ ndotstat ] ) , and ( [ ndotenh ] ) are compared in fig . [ fc ] to results of numerical integration in eqs . ( [ iunstpm ] ) . the contribution of stable modes decreases at long times as @xmath49 increases [ see eq.([ipmlt ] ) ] . in order to explain the decrease , let us consider the asymptote of @xmath161 using , in the spirit of scattering theory @xcite , the substitution @xmath190 in the absence of many - body effects ( @xmath191 ) , one obtains @xmath192 , expressing the regular dispersion law for free particles . also , @xmath193 , @xmath194 , and hence @xmath195 , which also hold for small @xmath49 [ see eq . ( [ ipmst ] ) ] . therefore the scattering process can be described by a stationary scattering theory at any time @xmath196 ( see ref.@xcite ) . contrarily , in the case of a finite @xmath8 , many - body effects modify the dispersion law , so that @xmath197 , and hence @xmath198 . the discussion above is related to the scattering of two condensate waves bounded only by a normalization box . a more realistic yet simple model is that of a half - collision of rectangular wavepackets . consider a homogeneous condensate occupying a region of length @xmath199 along the @xmath200 axis ( leaving the results independent of size and shape in the @xmath201-plane ) . at the moment @xmath109 a bragg scattering pulse is applied , creating a wavepacket of density @xmath202 and momentum @xmath141 directed along the @xmath200 axis , while the rest of the atoms , with density @xmath203 , remains in the @xmath14 wavepacket . losses take place only at the overlapping parts of the two wavepackets . the overlapping time @xmath204 varies from 0 to @xmath205 , between the leading and trailing edges of the moving wavepacket , where @xmath206 is the total duration of the collision of the two wavepackets . the average density loss can be expressed as @xmath207 where @xmath208 is defined by eq.([npmdota ] ) . at the regime of non - stationarity , @xmath209 , or @xmath210 , the loss rate is given by eq . ( [ npmdotn ] ) , and @xmath211 at the regime of stationarity , @xmath212 , the loss rate is determined by eq . ( [ ndotstat ] ) , and the loss has a form @xmath213 the same expression for the loss can also be obtained by using the method of complex scattering length @xcite . at the regime of bose - enhancement , @xmath214 , the loss rate eq . ( [ ndotenh ] ) gives the average density loss @xmath215 the approximate expressions ( [ dnpmn ] ) , ( [ dnpms ] ) , and ( [ dnpmbe ] ) are compared in fig . [ fc ] to results of numerical integration in eqs . ( [ iunstpm ] ) and ( [ dnpma ] ) . the parametric approximation and , hence all the analysis above , are applicable only as long as @xmath216 . using eq . ( [ dnpmbe ] ) , this condition can be rewritten in the form of a limit on the duration of collision consistent with the approximation : @xmath217 for example , in the nist na experiment @xcite ( where @xmath218@xmath126 , @xmath219 nm , @xmath220 m , and @xmath221@xmath222 ; see ref . @xcite ) , the parametric approximation is applicable while @xmath223 , which allows for values of the exponential bose - enhancement factor in eq . ( [ dnpmbe ] ) of more than a 100 . this validity range substantially exceeds the range of applicability of the complex scattering - length method , @xmath224 . however , in the four - wave - mixing nist experiment @xcite , @xmath225 , which requires taking into account a considerable depletion of the condensates during the collision . this task may be performed by combining the present approach with that of the slowly - varying - envelope approximation @xcite . the consideration of wavepackets of a more realistic shape requires the solution of eq . ( [ csip2 ] ) for time - dependent @xmath146 and @xmath147 . this solution can be obtained by using a perturbation theory , which is applicable as long as @xmath107 , and does not describe the effects of bose enhancement . the evaluation is made simpler in the coordinate representation , in which case the linearized equations of motion for the field operators of the fluctuations @xmath226 have the form @xmath227 where @xmath228 is a unit - normalized wavefunction describing particles in the colliding wavepackets , @xmath229 is the total number of particles in the wavepackets , and @xmath230 is the interatomic interaction potential . the first - order perturbative solution of eq . ( [ csir ] ) can be expressed in the form @xmath231 here the zero - order solution @xmath232 is expressed in terms of free ( plane ) waves @xmath233 and the annihilation operators @xmath22 for the fluctuation in the momentum representation . as the following equations turn out to be independent of the first term in eq . ( [ xi1 ] ) , including the function @xmath234 , it is not necessary to specify that function . the second term includes the function @xmath235 considering scattering into the vacuum of fluctuations , such that @xmath236 , and therefore @xmath237 one can express the number of the lost atoms in the form @xmath238 where @xmath239 here @xmath240 and @xmath241 is the heavyside step function . as one can see @xmath242 is a time - dependent green function for two free particles . therefore @xmath243 is nothing else but the second - order term for the diagonal matrix element of the transition amplitude , obtained by time - dependent perturbation theory for a collision of two particles , each in the state @xmath244 . if the size of the wavepacket obeys @xmath245 , the result can be reduced to a form given by the stationary scattering theory ( see ref . @xcite ) , which is proportional to the imaginary part of the elastic scattering length , in agreement with ref . @xcite . consider two colliding gaussian wavepackets described by the wavefunction @xmath246 where @xmath247 and @xmath248 is the number of particles in each wavepacket . substituting eq . ( [ phig ] ) into eq . ( [ deltan ] ) one can obtain @xmath249 where @xmath250 is the collision duration . equation ( [ deltanst ] ) coincide with the expression which can be obtained by using of the complex scattering length method @xcite and neglecting the depletion of the wavepackets . the recent preprint @xcite , posted during preparation of the present paper , treats the loss of atoms from colliding bec wavepackets by using an approach similar to the one presented here , with solutions of the form of eq . ( [ solap ] ) published previously in ref . @xcite . however , unjustified neglections of substantial terms in the hamiltonian lead to an expression for @xmath147 which is smaller by a factor of four than the one given here by eq . ( [ g2 ] ) . this neglection should therefore result in an underestimation of the condensate loss . the above - mentioned work , ref . @xcite , considers collision of gaussian wavepackets to first - order perturbation theory only , which do not describe the effects of bose enhancement . finite analytical expressions of the kind presented here were not derived there , and the results of numerical calculations disagree with results of the complex scattering length method @xcite even after the correcting for the missing factor four in @xmath147 are taken into account . in contrast , the analytical expression eq . ( [ deltanst ] ) of the present paper , based on the same approximations , is in an agreement with the complex scattering length method . fluctuations in bec are treated by using a parametric approximation for the exact quantum equations of motion for the field operators . depending on the excitation energy , the fluctuations are divided into stable modes , which can be represented as bogolubov quasiparticles ( field oscillators ) , and unstable modes , which can be represented as inverted oscillators and demonstrate a rapid growth . initially the dynamics of unstable modes is a quantum process of spontaneous decay into the vacuum of the fluctuations with a super - exponential rate . this approach is applied to two physical problems . the first one involves the dynamics of a bec after a fast change of the scattering length from a non - negative to a negative one . in this case , the experimentally observed @xcite delay of the condensate collapse is described . the second problem involves the elastic scattering losses on collision of two bec wavepackets . at short collision times [ see eq.([ndotstat ] ) ] the losses obtained here are in agreement with those of the method of complex scattering length @xcite , with specific reference to the collisions of non - confined waves , of rectangular and of gaussian wavepackets . at longer collision times [ see eq . ( [ ndotenh ] ) ] quantum effects lead to bose - enhancement of the elastic scattering losses . the atoms lost from the condensates in both cases form correlated pairs in optimally entangled and optimally squeezed states . the author is very grateful to p. s. julienne and k. burnett for helpful discussions . special appreciation is expressed to a. ben - reuven for invaluable discussions and the help in preparation of the present paper . this work was partially supported by the national science foundation through a grant for the institute for theoretical atomic and molecular physics at harvard university and the smithsonian astrophysical observatory . parkins and d.f . walls , phys . rep . * 303 * , 1 ( 1998 ) . f. dalfovo , s. giorgini , l.p . pitaevskii , and s. stringari , rev . phys . * 71 * , 463 ( 1999 ) . e. m. lifshitz and l. p. pitaevskii , _ statistical physics , pt . 2 _ ( pergamon , oxford , 1980 ) . a. l. fetter and j. d. walecka , _ quantum theory of many - body systems _ ( mcgraw , san francisco , 1971 ) . n. n. bogolubov and n. n. bogolubov , jr . _ an introduction to quantum statistical mechanics _ ( gordon and breach , lausanne , 1994 ) . j .- blaizot and g. ripka , _ quantum theory of finite systems _ ( mit press , cambridge , ma , 1986 ) . n. p. proukakis and k. burnett , j. res . nist * 101 * , 475 ( 1996 ) . n. p. proukakis , k. burnett , and h. t. c. stoof , phys . rev a * 57 * , 1230 ( 1998 ) . a. griffin , phys . b * 53 * , 9341 ( 1996 ) . r. walser , j. williams , j. cooper , and m. holland , phys . a * 59 * , 3878 ( 1999 ) . sackett , j.m . gerton , m. welling and r.g . hulet , phys . lett . * 82 * , 876 ( 1999 ) . j. m. gerton , d. strekalov , i. prodan , and r. g. hulet , nature ( london ) * 408 * , 692 , ( 2000 ) . j. l. roberts _ et al . * 86 * , 4211 ( 2001 ) . e. a. donley _ et al . _ , nature ( london ) * 412 * , 295 ( 2001 ) . e. timmermans , p. tommasini , m . hussein , and a. kerman , phys . * 315 * , 199 ( 1999 ) . duine and h. t. c. stoof , phys . lett . * 86 * , 2204 ( 2001 ) . a. vardi , v. a. yurovsky , and j. r. anglin , cond - mat/0105439 ; phys . rev a ( in press ) . v. a. yurovsky , a. ben - reuven , and p. s. julienne , cond - mat/0108372 ; phys . ( submitted ) . m. o. scully and m. s. zubairy , _ quantum optics _ ( university press , cambridge , 1997 ) . m. g. moore and p. meystre , phys . a * 59 * , r1754 ( 1999 ) ; c. k. law and n. p. bigelow , phys . rev a * 58 * , 4791 ( 1998 ) . h. a. haus and c. x. yu , j. opt . b * 17 * , 618 ( 2000 ) . v. v. dodonov and v. i. manko , in _ invariants and evolution of nonstationary quantum systems _ , proceedings of lebedev physics institute vol . 183 , edited by m. a. markov ( nova science publishers , commack , ny , 1989 ) , p. 3 ; o. v. manko , phys . a * 228 * , 29 ( 1997 ) . a. vardi and j. r. anglin , phys . lett . * 86 * , 568 ( 2001 ) . j. r. anglin and a. vardi , phys . rev a * 64 * , 013605 ( 2001 ) . u. v. poulsen and k. molmer , phys . rev a * 63 * , 023604 ( 2001 ) . d. c. roberts , t. gasenzer , and k. burnett , quant - ph/0106060 . et al . _ , nature ( london ) * 398 * , 218 ( 1999 ) . y. b. band , m. trippenbach , j. p. burke , and p. s. julienne , phys . rev . lett . * 84 * , 5462 ( 2000 ) . m. trippenbach , y. b. band , and p. s. julienne , phys . a * 62 * , 023608 ( 2000 ) . r. bach , m. trippenbach , and k. rzazewski , cond - mat/0107117 . m. olshanii and l. pricopenko , cond - mat 0101275 . m. lewenstein and l. you , phys . lett . * 77 * 3489 , ( 1996 ) . r. j. dodd _ et al _ , phys . a * 54 * , 661 ( 1996 ) . h. bateman and a. erdely , _ higher transcendental functions , vol . 2 _ ( mc graw - hill , new york , 1953 ) . m. l. goldberger and k. w. watson , _ collision theory _ ( wiley , new york , 1964 ) .

comprehension of the phenomenon of _ decoherence _ in open quantum systems has always attracted much attention , in particular as a prerequisite to understand the transition from quantum to classical behavior . the dissipative two - state or _ spin - boson _ model has been thoroughly studied in wide regions of the parameter space with diverse methods and techniques since the 80 s @xcite . in the last decade , the subject of decoherence has experienced renaissance following the growing interest in the field of quantum state manipulation and quantum computation @xcite . any noise source sensitively leads to a narrowing of the quantum coherence domain . this entails severe limitations for coupled qubits to perform logic quantum operations . for this reason , extensive understanding of the decoherence mechanisms is indispensable . in this work , we focus upon a model which is a generalization of the single spin - boson model to the case of two spins which mutually interact via an ising - type coupling and are coupled to independent environments made up by bosons . the first analysis of this model relying on the influence functional method was given by dub and stamp @xcite . they obtained results for the dynamics in analytic form in restricted regions of the parameter space by omitting certain classes of path contributions and bath correlations . several other previous studies on the same or related models relied on the master equation and/or perturbative redfield approach @xcite . besides the weak - coupling assumption , often the secular approximation @xcite is made , which breaks down however when the spectrum becomes degenerate . the model allows , for instance , to study decoherence and relaxation of two coupled qubits @xcite , or the influence of a bistable impurity on the qubit dynamics @xcite . the latter may significantly degrade coherence in josephson phase qubits @xcite other possible application is study of coherence effects in coupled molecular magnets @xcite . in earlier works , the model has been analyzed in the pure dephasing regime both by the feynman - vernon method @xcite and the lindblad approach @xcite . here we extend the work in ref . @xcite beyond the pure dephasing regime and include the full dynamics of the qubit . in particular , we are interested in the competition between decoherence and relaxation to the equilibrium state . here we focus on the white - noise regime . we shall derive the exact solution for the reduced density matrix without restriction on the parameters of the model and analyze it in the coherent and incoherent domains and in the crossover regions in between . the model and relevant quantities of the reduced dynamics are introduced in section 2 . section 3 deals with the exact formal solution for the reduced dynamics . in section 4 , the path sum is carried out in the white - noise domain without any further approximation , and analytic expressions for the relevant expectation values in laplace space are presented . after an overview of the qualitative features of the dynamics in section 5 , we present in section 6 explicit expressions for decoherence and relaxation in the various parameter regimes ranging from low temperture and/or weak coupling to high temperature and/or strong coupling . finally , we study in section 7 the influence of a nonlinear spin - boson environment on the second spin in the various limits . we demonstrate that it behaves in the weak - coupling limit as a bosonic ( linear ) bath with a resonant spectral structure . we consider two two - state systems which are coupled to each other via an ising - type coupling and to independent bosonic environments . in pseudospin representation , we choose the generalized spin - boson hamiltonian ( we use units where @xmath0 ) @xmath1 in the basis formed by the localized eigen states @xmath2 and @xmath3 of @xmath4 and @xmath5 , respectively , @xmath6 and @xmath7 represent the tunneling couplings between the localized states , and the coupling term @xmath8 acts as a mutual bias energy of strength @xmath9 . the collective bath modes @xmath10 $ ] ( @xmath11 ) represent fluctuating bias forces . the hamiltonian is very rich in content and may model diverse physical situations . it may describe two coupled qubits or a qubit @xmath12 in contact with a complex environment formed by a bistable dissipative impurity @xmath13 . other possible realizations are coupled molecular magnets of which the low - energy states can be viewed as a spin @xcite . for the model ( [ ham1 ] ) , all effects of the environments are captured by the power spectrum of the collective bath modes @xmath14 with the spectral density of the coupling @xcite @xmath15 here , the second form represents the ohmic case with a high - frequency cut - off @xmath16 . alternatively , one may choose that the two spins are coupled to a common bath @xcite . here we study the effects of independent environments . this case is realistic in most physical systems of actual interest . the density matrix of a single spin has four matrix elements , the two populations that we shall label as @xmath17 and @xmath18 , and the two coherences with labels @xmath19 and @xmath20 . the two - spin density matrix has 16 matrix elements @xmath21 . we choose for convenience that the first ( second ) index refers to the states @xmath22 ( @xmath23 ) of the @xmath12spin ( @xmath13spin ) . the matrix elements can be expressed in terms of expectation values of 15 operators , @xmath24 , @xmath25 and @xmath26 ( @xmath27 and @xmath28 ) . the 4 pure populations may then be written as @xmath29/4 \ ; , \\[3 mm ] \rho_{\rm 1,3}^{}(t ) & = & [ \ , 1 \,+\,\langle\sigma_z^{}\rangle_t \,-\,\langle\tau_z^{}\rangle_t \,-\,\langle\sigma_z^{}\tau_z^{}\rangle_t\,]/4 \ ; , \\[3 mm ] \rho_{\rm 3,1}^{}(t ) & = & [ \ , 1 \,-\,\langle\sigma_z^{}\rangle_t \,+\,\langle\tau_z^{}\rangle_t \,-\,\langle\sigma_z^{}\tau_z^{}\rangle_t \,]/4\ ; , \\[3 mm ] \rho_{\rm 3,3}^{}(t ) & = & [ \ , 1 \,-\,\langle\sigma_z^{}\rangle_t \,-\,\langle\tau_z^{}\rangle_t \,+\,\langle\sigma_z^{}\tau_z^{}\rangle_t \,]/4\;. \end{array}\ ] ] corresponding expressions hold for the 4 pure coherences and the 8 hybrid states . for instance , we have @xmath30/4 \ ; .\ ] ] here we are predominantly interested in the populations . throughout we will choose that the reduced system starts out from the initial state @xmath31 while the heat reservoirs are in thermal equilibrium at temperature @xmath32 . in the absence of the environment , the hamiltonian @xmath33 can be easily transformed into diagonal form @xmath34 the eigenfrequencies are @xmath35 @xmath36 and they obey the vieta relations @xmath37 \omega_-^2+\omega_+^2 & = & 2 ( \delta_1 ^ 2+\delta_2 ^ 2+v^2 _ { } ) \ ; , \quad\ ; \omega_+^2\omega_-^2 \;=\ ; ( \delta_1 ^ 2+\delta_2 ^ 2+v^2_{})^2_{}- 4\delta_1 ^ 2\delta_2 ^ 2 \ ; . \end{array}\ ] ] the liouville equations @xmath38 $ ] ( @xmath39 ) , where the set @xmath40 represents the above 15 operators , yield 15 coupled equations . these are conveniently solved in laplace space . for instance , we get @xmath41 \langle \sigma_z\tau_z ( \lambda ) \rangle & = & { \displaystyle \frac{(v^2 + \lambda^2 ) ( v^2 + \delta_1 ^ 2 + \delta_2 ^ 2 + \lambda^2)}{\lambda\ , ( \lambda^2+\omega_{+}^{2})(\lambda^2+\omega_{-}^{2 } ) } } \ ; . \end{array}\ ] ] here we are interested in the evolution of the two - spin system without restricting ourselves to weak damping . therefore , we refrain from employing the perturbative redfield approach . rather we calculate the reduced dynamics with use of the feynman - vernon influence functional method . we show that the solution is available in analytic form in the white noise limit for general parameters @xmath42 and @xmath32 . within the feynman - vernon method , the exact formal expression for the rdm of the two - spin system is the quadruple path integral @xmath43\ , { \mathcal a}^ * [ \sigma']\ , { \mathcal a}[\tau]\ , { \mathcal a}^ * [ \tau ' ] \,{\cal b}[\sigma,\sigma';\tau,\tau ' ] \,\mathcal{f}[\sigma,\sigma';\tau,\tau']\ ] ] with appropriately chosen boundary values for the spin paths . here , each of the paths @xmath44 starts out from the localized state @xmath2 at time zero . they end up at time @xmath45 in the states @xmath46 , respectively , where @xmath47 . the functional @xmath48 $ ] is the amplitude for the free spin @xmath12 to follow the path @xmath49 , the functional @xmath50 $ ] represents the coupling of the two spins ( see below ) , and the functional @xmath51 $ ] introduces the environmental influences . for uncorrelated baths , we have @xmath52\,=\,\mathcal{f}_1[\xi_1^{},\eta_1^{}]\,\,\mathcal{f}_2[\xi_2^{},\eta_2^{}]$ ] , where @xmath53 = \int_0^t \!\!\!dt ' \int_0^{t'}\!\!\ ! dt '' \left [ \dot{\xi}_\zeta ( t')q_\zeta'(t'-t'')\dot{\xi}_\zeta(t '' ) \,+\,{\rm i}\,\dot{\xi}_\zeta ( t')q_\zeta''(t'-t'')\dot{\eta}_\zeta(t '' ) \right ] \;.\ ] ] here we have introduced symmetric and antisymmetric spin paths , @xmath54\ ; , \hspace{1 cm } \eta_1^{}(t')\;=\ ; \frac{1}{2}[\sigma(t')+\sigma'(t ' ) ] } \;,\\[3 mm ] \xi_2^{}(t ' ) & = & { \displaystyle \frac{1}{2}[\tau(t')-\tau'(t')]\ ; , \hspace{1 cm } \eta_2(^{}t')\;=\;\frac{1}{2}[\tau(t')+\tau'(t ' ) ] } \ ; . \end{array}\ ] ] the correlator @xmath55 is the second integral of the force autocorrelation function @xmath56 ( see eq . ( [ powerspec ] ) ) . in the ohmic scaling limit , we have @xmath57 + i \pi k_\zeta \ , { \rm sgn}(t)\ ; , \qquad \zeta=1,\,2\ ; .\ ] ] here , @xmath58 is the usual dimensionless ohmic coupling strength for the spin @xmath59 , and @xmath60 is the inverse temperature . to handle the quadruple path integral ( [ rdm ] ) , we follow the procedure for the single spin - boson problem @xcite and write it as an integral over two paths , one for each spin . each such path visits the diagonal `` sojourn '' states and the off - diagonal `` blip '' states of the respective spin . a path which starts and ends in a sojourn state must contain an even number of transitions with amplitude @xmath61 for each flip of spin @xmath59 . the flips occur at times @xmath62 for spin 1 and at times @xmath63 for spin 2 . upon labeling the sojourn and blip states with charges @xmath64 ( for spin 1 ) and @xmath65 ( for spin 2 ) , each with values @xmath66 , the paths with @xmath67 and @xmath68 transitions , respectively , may be written as @xmath69\ ; , \\ \xi^{(n_1^{})}_1(t ' ) & = & \sum_{j=1}^{n_1^ { } } \xi_{1,j}^ { } \left[\theta(t'-t_{2j-1})-\theta(t'-t_{2j } ) \right]\ ; , \\ \eta^{(n_2^{})}_2(t ' ) & = & \sum_{j=0}^{n_2^ { } } \eta_{2,j}^ { } \left[\theta(t'-s_{2j}^{})-\theta(t'-s_{2j+1}^ { } ) \right]\ ; , \\ \xi^{(n_2^{})}_2(t ' ) & = & \sum_{j=1}^{n_2^ { } } \xi_{2,j}^ { } \left[\theta(t'-s_{2j-1}^{})-\theta(t'-s_{2j}^ { } ) \right]\ ; . \end{aligned}\ ] ] upon introducing the notation @xmath70 , we may write the bath correlations between the blip pair @xmath71 of spin @xmath59 in the compact form @xmath72 with this , the influence functional for the paths ( [ sojblip ] ) reads @xmath73\ , \exp\big[-\sum_{j=2}^{n_\zeta^{}}\sum_{k=1}^{j-1}\xi_{\zeta j}^{}\xi_{\zeta , k}^{}\lambda_{\zeta;j , k}^{}\big ] } \\[4 mm ] & & \times\ ; { \displaystyle \,\exp\big[{\rm i } \,\pi k_\zeta^{}\sum_{k=0}^{n_\zeta^{}-1 } \xi_{\zeta , k+1}^{}\eta_{\zeta , k}^ { } \big ] } \ ; . \end{array}\ ] ] here the first and second term represent the intrablip and interblip correlations , respectively . the phase term is specific to the ohmic scaling limit and represents correlations of the sojourns with their subsequent blips . the sum over all paths now means ( i ) to sum over all possible intermediate sojourn and blip states of the two spins the paths with a given number of transitions can visit , ( ii ) to integrate over the ( for each spin ) time - ordered jumps of these paths , and ( iii ) to sum over the possible number of transitions the two spins can take , @xmath74 & \times & { \displaystyle \sum_{n_2^{}=0}^\infty \sum _ { \{\eta_{2,j}^{}=\pm 1\ } } \sum _ { \{\xi_{2,j}^{}=\pm 1\ } } \int_{t_{\rm i}^{}}^{t_{\rm f}^ { } } \!\ ! { \rm d}s_{2n_2^{}}^ { } \int_{t_{\rm i}^{}}^{s_{2n_2^ { } } } \!\ ! { \rm d}s_{2n_2^{}-1}^ { } \cdots \int_{t_{\rm i}^{}}^{s_2^ { } } \!\!{\rm d}s_1^{}\;\cdots } \ ; . \end{array}\ ] ] for coherences , the number of transitions in the respective spin path is odd . next , we rewrite the double path sum ( [ pathsum ] ) with time - ordering for each spin in terms of a single path over the 16 possible states with time ordering of all the flip times @xmath75 . in this representation , the system lingers for some period in a particular state of the double - spin rdm and then it flips with amplitude @xmath76 or @xmath77 to another state . for the longitudinal spin - spin coupling @xmath78 in ( [ ham1 ] ) , the coupling factor @xmath79 in eq . ( [ rdm ] ) is unity when the system dwells on one of the four pure soujourn or one of the four pure blip states , and it is sensitive to the coupling when it stays , say for a period @xmath80 , in one of the eight hybrid states . in detail , we have @xmath81 & = & { \rm e}^{+i v u_j^{}}_{}\ ; , \quad\quad\mbox{for}\quad\quad j \;\in\ ; ( 1,2),\,(2,1),\,(4,3),\,(3,4)\ ; , \\[3 mm ] { \cal b}[\,j\ , ] & = & { \rm e}^{-i v u_j^{}}_{}\ ; , \quad\quad\mbox{for}\quad\quad j \;\in\ ; ( 1,4),\,(4,1),\,(2,3),\,(3,2)\ ; . \end{array}\ ] ] combination of the above expressions yields the exact formal solution for the dynamics of the rdm of the two - spin system in the ohmic scaling limit . evidently , because of the nonconvolutive form of the bath correlations in the influence functional ( [ infl ] ) , the path sum can not be performed in analytic form . alternatively , one may recast the exact formal series expression for the populations in the form of generalized master equations in which the kernels , by definition , are the irreducible components of path segments with diagonal initial and final states @xcite . in the general case , the kernels are given by an infinite series in @xmath82 and @xmath83 , with the time integrals in each summand being again in nonconvolutive form . additional difficulties in performing the path sum ( [ pathsum ] ) arises from the spin - spin coupling ( [ intfac ] ) . in the white - noise limit of eq . ( [ powerspec ] ) , @xmath84 , where @xmath85 is a scaled thermal energy , the bath correlation function takes the form accounts for deviation of the actual high frequency behavior of @xmath86 from the exponential cut - off form in eq . ( [ specdens ] ) @xcite . ] @xmath87 \,+\ , \vartheta_\zeta^ { } |t| \,+\ , { \rm i}\ , \pi k_\zeta \ , { \rm sgn}(t ) \;.\ ] ] this expression emerges directly from eq . ( [ ohmcorr ] ) in the high temperature or long - time limit @xmath88 . the first term in eq . ( [ wncorr ] ) leads to an adiabatic ( franck - condon - type ) renormalization factor made up by modes in the frequency range @xmath89 . it is natural to assimilate this term , together with the phase term , into an effective temperature - dependent tunneling matrix element , @xmath90 where @xmath91 is the standard renormalized tunneling matrix element . all dynamical effects of the environmental coupling are captured by the second term @xmath92 in eq . ( [ wncorr ] ) . from this we see that the weight of the thermal energy relative to the systems energies @xmath93 and @xmath9 is assessed by the scaled thermal energy @xmath94 . based on experience from the single spin - boson system we should expect that the form ( [ wncorr ] ) is a remarkably good approximation in the parameter range @xcite @xmath95 and this is corroborated indeed by our study . for a single unbiased spin with ohmic damping @xmath96 , the coherent - incoherent `` phase''-transition is at temperature @xmath97 with @xmath98 @xcite . therefore we should expect , that , for @xmath99 , the white noise form ( [ wncorr ] ) is valid not only in the incoherent regime but also in a sizeable domain of the coherent regime . this shall be confirmed subsequently . for the form ( [ wncorr ] ) of @xmath100 , the interblip correlations cancel out exactly in eq . ( [ blipcor ] ) , @xmath101 . as a result , each term of the infinite series for the rdm becomes a convolution . this makes the path sum accomplishable . we shall now exemplify , by taking @xmath102 and @xmath103 as examples , that the path sum for the laplace transform of the rdm can be carried out exactly in analytic form . this is achieved by first calculating the kernels and then summing up the respective geometrical series of these objects . by definition , the kernels represent irreducible path segments which interpolate between pure sojourn states . irreducibility means that these segments can not be separated into uncorrelated pieces without removing bath correlations . the analysis gives that every contribution to the kernel @xmath106 of @xmath107 displays initially and finally a transition of spin @xmath12 with any number of even hops of spin @xmath13 at intermediate times , as shown in the diagrams of fig . there are no other contributions . for all other irreducible diagrams , one may think of , e.g. , where either the first or the last flip , or both of them , are flips of spin @xmath13 , the respective contributions from the two different final states of spin @xmath13 cancel each other . sketch of the irreducible kernels ( with amputated legs ) @xmath108 ( a ) , @xmath109 ( b ) , and @xmath110 ( c ) . the solid and dashed curve represent the correlations of bath 1 and of bath 2 , respectively . ] it is convenient to write @xmath111 , where @xmath112 is of order @xmath113 . the sum over all spin states of order @xmath114 yields the expressions @xmath115 @xmath116 & & \!\!\!\!\times \ ; { \displaystyle \,\frac{1}{2}\,[\ , \cos(\pi k_1^ { } + v\tau_1^ { } + v \tau_3^ { } ) - \cos(\pi k_1^ { } + v\tau_1^ { } - v \tau_3^{})\ , ] \cos(\pi\,k_2^ { } ) } \ ; , \end{array}\ ] ] @xmath117 here we have taken into account that , for the white - noise form ( [ wncorr ] ) , a correlation of bath @xmath59 stretching over an interval between neighboring hops effectively leads to a shift of the laplace variable in the respective time integral , @xmath118 . it is convenient to split the kernels into the contributions which are even and odd in the coupling @xmath9 , @xmath119 . the resulting expressions may be written as @xmath120 & \times & { \displaystyle \!\!\!\left\ { 1 \;+\ ; \frac{v^2_{}}{(\lambda + \vartheta_1^{})(\lambda + \vartheta_1^{}+\vartheta_2^{})}\ , \frac{\bar\delta_{2}^2 } { ( \lambda + \vartheta_1^{})^2_{}+ v^2 _ { } + \frac{\lambda+\vartheta_1^{}}{\lambda + \vartheta_1^ { } + \vartheta_2^{}}\bar\delta_2^{2 } } \right\ } } \ ; , \end{array}\ ] ] @xmath121 & \times & { \displaystyle \!\!\!\left\ { 1 \;-\ ; \frac{\lambda+\vartheta_1^{}}{\lambda + \vartheta_1^{}+\vartheta_2^{}}\ , \frac{\bar\delta_{2}^2 } { ( \lambda + \vartheta_1^{})^2_{}+ v^2 _ { } + \frac{\lambda+\vartheta_1^{}}{\lambda + \vartheta_1^ { } + \vartheta_2^{}}\bar\delta_2^{2 } } \right\ } } \ ; . \end{array}\ ] ] paths which visit a pure sojourn state at intermediate times yield reducible contributions . taking into account all possibilities of such visits yields a geometrical series in the kernel @xmath122 , while @xmath123 occurs only once as initial irreducible contribution . thus we get for @xmath107 the concise form @xmath124 algebraic manipulation gives @xmath107 in the form of a simple fraction with denominator and numerator in the form of polynomials of degree four . we get @xmath125 & + & \!\!\!(v^2_{}\vartheta_1^ { } + \bar\delta_2 ^ 2\vartheta_1^ { } + \vartheta_1 ^ 3 + v^2_{}\vartheta_2^ { } + \vartheta_1 ^ 2\vartheta_2^{})\lambda + v\bar\delta_1^{}\tan(\pi k_1^ { } ) ( \vartheta_1^{}+\vartheta_2^ { } + \lambda ) \ ; , \end{array}\ ] ] @xmath126 & + & ( v^2\ , \vartheta_1^ { } + 2\ , \bar\delta_{1}^2 \,\vartheta_1^ { } + \bar\delta_{2}^2 \ , \vartheta_1^ { } + \vartheta_1 ^ 3 + v^2 _ { } \,\vartheta_2^ { } + \bar\delta_{1}^2\ , \vartheta_2^ { } + \vartheta_1 ^ 2\ , \vartheta_2^{})\,\lambda \\[3 mm ] & + & \bar\delta_{1}^2 \,\vartheta_1 ^ 2 + \bar\delta_{1}^2\ , \vartheta_1^{}\ , \vartheta_2^ { } + \bar\omega_{}^2 \,\bar\delta_{}^2 \ ; . \end{array}\ ] ] here we have introduced the eigenfrequencies @xmath127 , @xmath128 of the undamped coupled two - spin system . the bar denotes adiabatic renormalization in eq . ( [ def_eigen1 ] ) according to eq . ( [ adia ] ) . the pole at @xmath129 in eq . ( [ expecsigma ] ) yields the equilibrium value @xmath130 hence @xmath131 is negligibly small for @xmath132 , while , as @xmath133 , it takes the proper ( white noise ) equilibrium value @xmath134 of the single biased spin boson system . the dynamical poles @xmath135 ( @xmath136 ) are given by a quartic equation with real coefficients . upon collecting the various pole contributions , we get in the time domain @xmath137 evidently , the expectation @xmath138 takes similar form , @xmath139 here , the polynomials @xmath140 and @xmath141 follow from the expressions ( [ sz1 ] ) and ( [ sz2 ] ) by interchange of the indices 1 and 2 . sketch of the irreducible kernels @xmath142 ( left ) and @xmath143 ( right ) . the intervals are dressed by self - energy contributions of spin @xmath12 ( black circle ) and spin @xmath13 ( black square ) , as sketched in fig . [ f4 ] . ] self - energy terms due to spin @xmath12 ( circle ) and spin @xmath13 ( square ) . ] consider first contributions to the kernel of @xmath145 in which the first and last flip are made by spin @xmath12 , as sketched in fig . [ f2 ] . in the intervals of the bare diagrams , either spin @xmath12 or spin @xmath13 or both stay offdiagonal . every interval in which spin @xmath59 dwells in a sojourn state is dressed by selfenergy contributions schematically given for spin @xmath12 ( circle ) and spin @xmath13 ( square ) in fig . diagram [ f2](a ) yields @xmath146 \,\alpha_1^{}(\lambda ) \ ; , \end{aligned}\ ] ] where the functions @xmath147 @xmath148 are given by @xmath149 the nested diagram in fig . [ f2](b ) produces the additional factor @xmath150 , @xmath151 higher - order nested diagrams of the type sketched in fig . [ f2 ] class into a geometrical series . all these terms are readily summed up to the contribution @xmath152 similarly , we find that diagrams ( a ) and ( b ) in fig . [ f3 ] yield the expressions @xmath153 \,\alpha_1^{}(\lambda)\alpha_2^{}(\lambda ) \ ; , \ ] ] @xmath154 with all higher - order nested diagrams of this type added , one finds again a geometrical series in @xmath155 , @xmath156 clearly , we must add those contributions resulting from the terms @xmath157 and @xmath158 by interchange of the spins @xmath12 and @xmath13 . the analysis shows that there are no other contributions . again , we split the kernel @xmath159 into the parts which are even and odd in the coupling @xmath9 . we readily get for @xmath160 $ ] the forms @xmath161 @xmath162 these expression represent the entity of irreducible path segments . next , we observe that the sum of two - spin paths with any number of interim visits of pure sojourn states yields a geometrical series of these objects . in the part which is odd in @xmath9 , the first irreducible path section is again described by the kernel @xmath163 . thus we get @xmath164 the second form is a simple fraction with polynomials @xmath165 and @xmath166 of sixth order , @xmath167 \\[2 mm ] & & + \ ; v \tan ( \pi k_1^ { } ) \ , \bar\delta_1 ^ 2\,(\vartheta_2^{}+\lambda)\,[\,\bar\delta_1 ^ 2\ , -\ , \bar\delta_2 ^ 2\ , + \ , ( \vartheta_2+\lambda)(\vartheta_1^{}+\vartheta_2^{}+\lambda ) \ , ] \\[2 mm ] & & + \;v \tan ( \pi k_2^ { } ) \ , \bar\delta_2 ^ 2\,(\vartheta_1^{}+\lambda)\,[\,\bar\delta_2 ^ 2 \,-\ , \bar\delta_1 ^ 2 \,+\ , ( \vartheta_1+\lambda ) ( \vartheta_1^{}+\vartheta_2^{}+\lambda ) \ , ] \\[2 mm ] & & + \ ; \lambda \ , v^4_{}(\vartheta_1^{}+\vartheta_2^{}+\lambda ) \,+\,\lambda v^2_{}[\,\bar\delta_2 ^ 2\,(\vartheta_1^{}+\lambda ) + \bar\delta_1 ^ 2\,(\vartheta_2^{}+\lambda)\ , ] \\[2 mm ] & & + \ ; \lambda \,v^2 _ { } ( \vartheta_1^{}+\vartheta_2^{}+\lambda ) [ \,\vartheta_1 ^ 2+\vartheta_2 ^ 2 + 2\lambda ( \vartheta_1^{}+\vartheta_2^{}+\lambda)\,]\\[2 mm ] & & + \ ; \lambda \,(\vartheta_1^{}+\lambda)(\vartheta_2^{}+\lambda ) [ \,\bar\delta_1 ^ 2\,(\vartheta_1+\lambda ) + \bar\delta_2 ^ 2\,(\vartheta_2+\lambda)\ , ] \\[2 mm ] & & + \ ; \lambda\,(\vartheta_1^{}+\lambda)^2_{}(\vartheta_2^{}+\lambda)^2_{}(\vartheta_1^{}+\vartheta_2^{}+\lambda ) \ ; , \end{array}\ ] ] @xmath168 & & + \;(\vartheta_1 + \vartheta_2)(2\bar\omega_+^2 + 2 \bar\omega_-^2 + \vartheta_1 ^ 2 + 6\vartheta_1\vartheta_2 + \vartheta_2 ^ 2)\,\lambda^3 _ { } \\[2 mm ] & & + \;[\,\bar\omega_+^2\,\bar\omega_-^2 + \vartheta_1 \vartheta_2(2 \vartheta_1 + \vartheta_2)(\vartheta_1 + 2\vartheta_2 ) + v^2 ( 3\vartheta_1 ^ 2 + 4\vartheta_1\vartheta_2 + 3\vartheta_2 ^ 2 ) \\[2 mm ] & & + \ ; \bar\delta_1 ^ 2(2\vartheta_1 + \vartheta_2)(\vartheta_1 + 3\vartheta_2 ) + \bar\delta_2 ^ 2(2\vartheta_2 + \vartheta_1)(\vartheta_2 + 3\vartheta_1 ) ] \,\lambda_{}^2\\[2 mm ] & & + \;(\vartheta_1 + \vartheta_2)[\,\bar\omega_+^2\ , \bar\omega_-^2 + \vartheta_1 ^ 2\vartheta_2 ^ 2 + ( v^2 + \bar\delta_2 ^ 2)\vartheta_1 ^ 2 + ( v^2_{}+\bar\delta_1 ^ 2)\vartheta_2 ^ 2 \\[2 mm ] & & + \ ; 3\vartheta_1^{}\vartheta_2^{}(\bar\delta_1 ^ 2+\bar\delta_2 ^ 2 ) \,]\,\lambda \;+ \ ; v^2(\vartheta_1 + \vartheta_2)(\bar\delta_1 ^ 2\vartheta_1 + \bar\delta_2 ^ 2 \vartheta_2)\\[2 mm ] & & + \ ; \vartheta_1\vartheta_2 [ \ , ( \bar\delta_1 ^ 2 - \bar\delta_2 ^ 2)^2 + ( \bar\delta_2 ^ 2\vartheta_1 + \bar\delta_1 ^ 2\vartheta_2 ) ( \vartheta_1^{}+\vartheta_2^ { } ) \ , ] \ ; . \end{array}\ ] ] sketch of the irreducible kernels @xmath169 ( left ) and @xmath170 ( right ) . again , the intervals are dressed as sketched in fig . [ f4 ] . ] the odd powers of the pole function can be removed with a shift . putting @xmath171 \equiv \tilde d(x)$ ] , we obtain a polynomial with even powers , @xmath172\,x^4 _ { } \\[3 mm ] & & + \;[\,\bar\omega_+^2\bar\omega_-^2\ , + \,(\bar\delta_1 ^ 2+\bar\delta_2 ^ 2 - 2v^2_{})\vartheta_1^ { } \vartheta_2^ { } -\bar\delta_1 ^ 2\vartheta_1 ^ 2-\bar\delta_2 ^ 2\vartheta_2 ^ 2 \\[3 mm ] & & \quad\ ; + \;\frac{1}{16 } ( \vartheta_1^{}-\vartheta_2^{})^2_{}(3\vartheta_1 ^ 2 + 2\vartheta_1^{}\vartheta_2^{}+3\vartheta_2 ^ 2 ) \,]\,x^2 _ { } \\[3 mm ] & & -\frac{1}{4 } [ \,v^2_{}(\vartheta_1^{}+\vartheta_2^ { } ) \,+\ , ( \bar\delta_1 ^ 2-\bar\delta_2 ^ 2)(\vartheta_2^{}-\vartheta_1^ { } ) \,+\,\frac{1}{4}(\vartheta_1 ^ 2-\vartheta_2 ^ 2)(\vartheta_1^{}-\vartheta_2^ { } ) \,]^2 _ { } \;. \\[3 mm ] \end{array}\ ] ] thus we have in the time domain @xmath173 where the @xmath174 are the zeros of @xmath175 , and the equilibrium value is @xmath176 the expressions ( [ expecsigma ] ) ( [ expectau ] ) and ( [ expecsigtau ] ) ( [ sigtauz3 ] ) are the main results of this work . they represent the exact analytical solutions for @xmath107 , @xmath177 , and @xmath145 , in the white - noise limit for general coupling @xmath9 and general effective reservoir couplings @xmath178 and @xmath179 . except for use of the form ( [ wncorr ] ) , no other approximation has been made . we remark that for all other initial and final states of the rdm we would find the same pole functions ( [ sz2 ] ) and ( [ sigtauz1 ] ) . only the numerator function would be different . the behaviors of the four dynamical poles of @xmath180 and the six dynamical poles of @xmath181 , and the respective amplitudes are quite multifarious . in this section , we sketch the characteristics for the symmetric system , @xmath182 and @xmath183 . in the coupling range @xmath184 , there are three crossover temperatures , denoted by @xmath185 , and @xmath186 ( see figs . [ fig : sz_crosstemp ] and [ fig : sz_re_im ] ) . in the regime @xmath187 the dynamics is coherent and described by a superposition of two damped oscillations . for @xmath188 , the oscillations have different frequency and the same damping rate , and the amplitudes are comparable in magnitude . on the other hand , in the range @xmath189 , they have the same frequency , but different decrement and the amplitude belonging to the larger decrement is negligibly small . in the temperature regime @xmath190 , the dynamics is incoherent . in the regime @xmath191 , the four poles are real , and the two smallest rates have largest amplitudes and dominate the relaxation process . in the so - called kondo regime @xmath192 , the dominant pole is real and approaches @xmath193 , and its residuum goes to @xmath194 , as temperature is increased . the other real pole takes the value @xmath195 , while its residuum drops to zero . there is also a damped oscillation of which the frequency and rate approach asymptotically @xmath9 and @xmath196 , but the amplitude becomes negligibly small . the phenomenon that , in the kondo regime for @xmath197 , incoherent relaxation slows down with increasing temperature , is already well - known in the single spin - boson problem @xcite . [ fig : sz_fourier_time ] shows the transition from coherent to incoherent dynamics as @xmath196 is raised . one can also see that at high @xmath196 the effective damping decreases with increasing @xmath196 . for @xmath198 , there is only one crossover . it separates the regime with two complex conjugate poles from the regime with one pair of complex conjugate poles and two real poles . above @xmath199 , the relaxation is governed by the kondo pole . for a symmetric system , two poles are cancelled . the reduced pole equation reads @xmath200 the characteristic behavior of the poles as function of the scaled temperature @xmath196 is shown in fig . [ fig : sztz_re_im ] . at low @xmath196 , all poles contribute to the dynamics . in the kondo regime @xmath201 , the leading pole behaves as @xmath202 , and the amplitudes of the other contributions are negligibly small . : crossover temperatures as functions of @xmath9 , @xmath203.,width=6,height=4 ] : real ( a ) and imaginary ( b ) part of @xmath135 , @xmath204 , @xmath203.,title="fig:",width=6,height=4 ] : real ( a ) and imaginary ( b ) part of @xmath135 , @xmath204 , @xmath203.,title="fig:",width=6,height=4 ] : plots of @xmath205 in the fourier regime ( a ) and time regime ( b ) , @xmath204 , @xmath203 , @xmath206 . red ( small - dashed ) @xmath207 , green ( dashed ) @xmath208 , blue ( medium - dashed ) @xmath209 , grey ( long - dashed ) @xmath210.,title="fig:",width=6,height=4 ] : plots of @xmath205 in the fourier regime ( a ) and time regime ( b ) , @xmath204 , @xmath203 , @xmath206 . red ( small - dashed ) @xmath207 , green ( dashed ) @xmath208 , blue ( medium - dashed ) @xmath209 , grey ( long - dashed ) @xmath210.,title="fig:",width=6,height=4 ] : real and imaginary part of @xmath135 , @xmath204 , @xmath203.,title="fig:",width=6,height=4 ] : real and imaginary part of @xmath135 , @xmath204 , @xmath203.,title="fig:",width=6,height=4 ] below the first crossover temperature , @xmath211 , the real parts of all poles are varying linearly with @xmath32 . in detail , the poles and amplitudes are as follows . there is a superposition of two damped oscillations with frequencies @xmath212 \lambda_{3,4}^ { } & = & { \displaystyle \pm\,\mathrm{i}\,\bar\omega\,-\,\frac{2\,\bar\omega^2 _ { } - \bar\delta^2 _ { } - \bar\delta_1 ^ 2}{2\,(\bar\omega^2 - \bar\delta^2)}\,\vartheta_1 \,-\ , \frac{\bar\delta_2 ^ 2-\bar\delta^2_{}}{2\,(\bar\omega^2_{}- \bar\delta^2_{})}\,\vartheta_2^ { } \,+\,{\cal o } ( \vartheta^2 _ { } ) \ ; , } \end{array}\ ] ] and amplitudes @xmath213 there is a superposition of two damped oscillations and two contributions describing incoherent relaxation towards the equilibrium value @xmath214 , @xmath215 \lambda_5^ { } & = & { \displaystyle -\,\frac{\bar\omega^2 _ { } - \bar\delta _ 1 ^ 2}{\bar\omega^2 _ { } - \bar\delta^2 _ { } } \vartheta_1^ { } \,-\ , \frac{\bar\omega^2 _ { } -\bar\delta_2 ^ 2}{\bar\omega^2 _ { } - \bar\delta^2 _ { } } \vartheta_2^ { } \ ; , \quad \lambda_6^ { } \;=\ ; -\,\frac{\bar\delta_1 ^ 2 - \bar\delta^2_{}}{\bar\omega^2 _ { } - \bar\delta^2 _ { } } \vartheta_1^ { } \,-\ , \frac{\bar\delta_2 ^ 2 - \bar\delta^2_{}}{\bar\omega^2 _ { } - \bar\delta^2 _ { } } \vartheta_2^ { } \ ; , } \end{array}\ ] ] where terms of order @xmath216 are disregarded . in leading order , the amplitudes are @xmath217 when the coupling @xmath9 and/or the scaled temperatures @xmath218 are large compared to the other frequencies , the amplitudes of three ( five ) pole contributions to @xmath102 ( @xmath103 ) are negligibly small , and among the dynamical poles only the real pole with smallest modulus is relevant . hence the two spins essentially behave as a single spin which relaxes incoherently to the equilibrium state according to @xmath219 \langle\sigma_z^{}\tau_z^{}\rangle_t & = & { \displaystyle \big(\ , 1\,-\,\langle\sigma_z^{}\tau_z^{}\rangle_{\rm eq}\,\big)\ , \rme^{-\gamma_{\sigma\tau}^{}t } _ { } \ , + \ , \langle\sigma_z^{}\tau_z^{}\rangle_{\rm eq}^ { } } \ ; . \end{array}\ ] ] the relaxation rate is found from @xmath220 with the form ( [ sz2 ] ) as @xmath221 this reduces in the parameter regime @xmath222 to @xmath223 in this regime , @xmath180 is independent of the coupling @xmath9 and hence independent of the dynamics of the @xmath13-spin . the temperature dependence @xmath224 distinguishes the so - called kondo regime , in which , for @xmath225 , the relaxation dynamics slows down as temperature is increased . on the other hand , when @xmath226 , the two spins are locked together @xcite , and the effective tunneling matrix element is @xmath227 , as follows from ( [ def_eigen1 ] ) with ( [ def_eigen2 ] ) . we then get from eq . ( [ relaxrate ] ) @xmath228 \ ; .\ ] ] this yields the limiting expressions @xmath229 the former is the relaxation rate of the biased single spin - boson system at low @xmath178 . the latter describes kondo - like joint relaxation of the locked spins . in the incoherent regime , the relaxation rate @xmath230 of the effective single spin receives rate contributions from both the @xmath12- and the @xmath13-spin as if these were independent biased spins in contact with their own heat reservoir . in the large - coupling limit , @xmath231 , the relaxation rate is found as @xmath232 the individual contributions are single - spin rates in the large - bias regime . in the high temperature limit @xmath233 , on the other hand , both rate contributions are kondo - like , @xmath234 consider next the regime @xmath235 , in which spin @xmath12 behaves kondo - like , as in eq . ( [ kondo1 ] ) . hence the dynamics of the @xmath12-spin is slow compared to that of the @xmath13-spin . thus we should expect that @xmath103 approaches the dynamics of the biased single spin - boson case as @xmath178 is increased . taking into account terms of linear order in @xmath236 in the pole equation , the expression ( [ expecsigtau ] ) with ( [ sigtauz2 ] ) and ( [ sigtauz1 ] ) assumes the form @xmath237 indeed , in the limit @xmath238 , this form reduces just to the analytic expression for @xmath177 of the biased single spin - boson system in the white - noise limit @xcite . let us now view spin @xmath13 with reservoir 2 as an environment for spin @xmath12 . this complex environment is in general non - gaussian and non - markovian @xcite . recently , the same model has been studied numerically using a markovian master equation approach @xcite . to proceed , we first note that in the absence of bath 1 , @xmath239 , the pole equation @xmath240 is still of fourth order . there is no reduction in the general case . in fig . [ fig : sbe_re_im ] we show plots of the four poles as functions of @xmath179 for a particular set of parameters . with spin - boson environment : real ( a ) and imaginary ( b ) part of @xmath135 . at low @xmath179 , @xmath180 is a superposition of two damped oscillations . at high @xmath179 , there is one damped oscillation and one relevant relaxation contribution . the parameters are @xmath241 , @xmath242 , @xmath243.,title="fig:",width=6,height=4 ] with spin - boson environment : real ( a ) and imaginary ( b ) part of @xmath135 . at low @xmath179 , @xmath180 is a superposition of two damped oscillations . at high @xmath179 , there is one damped oscillation and one relevant relaxation contribution . the parameters are @xmath241 , @xmath242 , @xmath243.,title="fig:",width=6,height=4 ] simplification occurs , however , when @xmath179 is very large compared to the other frequencies . in this regime , the kernel ( [ kplus ] ) reduces to the form @xmath244 where @xmath245 is the relaxation rate of spin @xmath13 in the kondo regime . with this high - temperature expression for the kernel , the quantity @xmath107 is found to read @xmath246 this expression describes the dynamics of spin @xmath12 coupled to a spin - boson environment , where the latter is in the kondo regime . to leading order in @xmath247 , the poles of the expression ( [ szsb ] ) are ( see fig . [ fig : sbe_re_im ] at large @xmath179 ) @xmath248 and the amplitudes read @xmath249 the expressions ( [ szsb ] ) ( [ szsb2 ] ) may now be compared with the corresponding ones of a fictive single biased spin - boson system with parameters @xmath82 and @xmath9 in the white noise limit at scaled temperature @xmath250 . the part that is symmetric in the bias reads @xmath251 we see that with the identification @xmath252 the expressions ( [ szsb ] ) and ( [ szsb3 ] ) are quite similar . observe , however , that the damping rate of the oscillation is somewhat different because of the term @xmath253 in eq . ( [ szsb3 ] ) instead of @xmath254 in eq . ( [ szsb ] ) @xcite . most importantly and interestingly in the correspondence , temperature maps on the inverse of it . we should expect that , in the weak - coupling limit , the spin - boson environment is gaussian and can be represented by a resonant power spectrum of a bath of bosons . the gaussian approximation of the spin - boson environment is found by matching the power spectrum of the coupling of the @xmath12-spin to the spin - boson environment @xcite ( with normalization as in eq . ( [ powerspec ] ) ) @xmath255\,\big\rangle\ , \rme^{{\rm i}\,\omega t}_{},\ ] ] with that of a harmonic oscillator bath . in the white - noise limit of an unbiased spin , the symmetrized equilibrium correlation function @xmath256 coincides with the expectation @xmath257 . thus we obtain @xmath258 the resulting power spectrum is that of a structured bath of bosons with a resonance of width @xmath179 at frequency @xmath259 , @xmath260 due to the coupling to the spin - boson environment , the spin @xmath12 performs damped oscillation , @xmath261 . upon calculating the decoherence rate in order @xmath262 , i.e. , the so - called one - boson - exchange contribution of the effective boson bath , we obtain @xmath263 } \ ; .\ ] ] the analysis is completed by observing that this form emerges also upon calculating directly @xmath264 from the pole equation @xmath220 with the form ( [ sz2 ] ) . we have studied the dynamics of a spin or qubit @xmath12 coupled to another spin , which could be , for instance , another qubit , or a bistable impurity , or a measuring device . we have solved the dynamics exactly for white - noise reservoir couplings , and we have studied the rich behaviors of the dynamics in diverse limits ranging from weak coupling and/or low temperatures to strong coupling and/or high temperature . we have also analyzed the effects of a spin - boson environment on the spin dynamics in the gaussian and non - gaussian domains . this paper has not attempted to perform applications to already available experiments , instead we have tried to make some general points on complementary regimes and on the crossovers in between . one possible simple generalization beyond the white noise limit would be to replace the white - noise bath correlations in time intervals in which the laplace variable is irrelevant by the full quantum noise correlation , for instance @xmath265 the advantage would be two - fold : ( i ) the noise integral is known in analytic form for the ohmic correlation function ( [ ohmcorr ] ) @xcite , and ( ii ) with this substitution the algebraic form of pole equation and residua would be left unchanged . one further generalization of the hamiltonian ( [ ham1 ] ) is an applied bias acting on one or both of the spins , e.g. , of the form @xmath266 and @xmath267 . this important extra ingredient could be taken into account exactly in the white noise regime , and would lead to additional shifts of the laplace variable in all blip states of the @xmath12- and @xmath13-spin . then one would end up with expressions of the form ( [ expecsigma ] ) and ( [ expecsigtau ] ) with polynomials ramped up by bias terms . this extension will be discussed elsewhere . finally , extension of the analysis of the dynamics to the regime @xmath268 requires to revert to the original expression ( [ powerspec ] ) and compute its effect perturbatively in the one - boson - exchange approximation . this can be done either with the self - energy method presented in ref . @xcite or with the redfield approach . one then find , e.g. , that the actual equilibrium state of @xmath214 for @xmath269 is @xmath270 \ ; .\ ] ] this reduces for @xmath271 to the previous form ( [ sztzeq ] ) found in the white - noise regime . the corresponding extension of the analysis and of the results given in subsection [ subseclowtemp ] to the domain @xmath272 will be reported elsewhere . the authors wish to thank g. falci and e. paladino for valuable discussions . financial support by the dfg through sfb / tr 21 is gratefully acknowledged .

although much can be learned by studying stellar nurseries and the fascinating process of stellar birth , we have much yet to learn in the field of stellar geriatrics . stars that do not proceed to explosive ends , the low- and intermediate - mass stars , undergo a period of mass loss , often extreme , in which 50% or more of the star s initial mass is transferred back to the ism . the rates of this mass loss vary widely from @xmath0 per year to as much as @xmath1 per year . such prodigious mass loss and the large number of low- and intermediate mass stars results in the fact that most of the interstellar medium perhaps as much as 80 to 90 percent has been cycled through a star and ejected via this process . flash - in - the - pan supernovae do have a significant impact , especially enriching the heavier metals in the ism , but the bulk of the material is provided by the aging process of common stars similar to the sun . understanding how the mass loss process proceeds and its implications on the chemical modification of the ism in our own galaxy has obvious implications for the study of more distant galaxies as well as being of interest itself . the mass loss process proceeds from the formation of dust in the upper atmospheres of evolved stars , a few stellar radii from the optical photosphere . this process has long been thought to be driven by the pulsations inherent in these kinds of stars , but it now appears likely that it is driven by dramatic temperature fluctuations caused by the formation of tio in the stellar atmosphere , which changes the physical conditions in the dust formation region @xcite . although the details of dust formation remain an unknown factor , we know roughly that when the temperature and density conditions are appropriate , nucleation can occur , leading to the formation of dust . this dust , exposed to the radiation field of the evolved star , absorbs outward momentum and begins to accelerate . gas not incorporated into dust grains is carried along with the dust through momentum coupling . as conditions allow , molecules can form from the gas that is carried along with the outward - moving dust . using existing centimeter and millimeter wave interferometers , studies of the molecules formed in these winds have been completed showing more or less spherically symmetric mass loss @xcite with some interesting results such as rotation @xcite and perfect spherical symmetry of an apparently single ejection event @xcite . certain molecules are capable of maser emission ( e.g. sio , h@xmath2o and oh ) . when such masers are found in the winds of evolved o - rich stars , they are powerful probes of the mass loss kinematics . some c - rich stars do exhibit hcn masers at high frequencies , but the o - dominated species are absent . however , they can provide only rough information about the physical conditions of the wind itself , provided by the physical conditions required for maser emission . using vlbi techniques , which provide resolutions as fine as 100 @xmath3arcseconds , the motions of masing gas can be tracked with high accuracy and the kinematics of the wind modelled . in practice , this has proven to be a challenging undertaking . the non - linear emission process and apparently complex distributions of the masers make modelling difficult . only rough models have yet been made placing the masers in ellipsoidal distributions undergoing a variety of kinematic motions . the maser observations do indicate more - or - less spherically or elliptically symmetric mass loss with acceleration occurring to the outermost regions of the wind where acceleration ceases due to decoupling of the gas from the dust . a mild controversy about the relative angular scales of the oh and h@xmath4o maser distributions ( e.g. the oh masers , although predicted to be at large radii , appear at about the same angular scale as the h@xmath4o masers ) is likely due to beaming effects . water masers are preferentially tangentially beamed as they reside in an accelerating portion of the wind while the oh masers are radially beamed as they reside in a constant velocity region of the wind @xcite . as difficult as the physics and geometries are , our current understanding is limited due to lack of adequate modelling in my opinion . other results indicate non - negligible rotation @xcite of the envelopes and the influence of magnetic fields on the shape of the shell @xcite . we have yet to understand the dust formation process in these objects . although infrared interferometric observations @xcite hint at a very clumpy and dynamic process , we have few tools available to probe this process in detail . we have only a very rough picture of the structure of the extended photosphere and wind of evolved stars . the role of magnetic fields in agb stars has not been explored in any detail , though they must impact the dust formation process , the wind itself and obviously provide information on the star itself . although detailed studies have been made using millimeter interferometers of the chemical structure of the nearest and largest evolved stars , much of the chemical structure in these objects remains a mystery . alma will help here but will miss the low - frequency line transitions . a detailed understanding of the structure of these objects awaits the square kilometer array . the ska , with its high resolution , sensitivity to a range of emission mechanisms and low - frequency observing capability will allow studies of evolved stars that have not been possible before and provide complementary observations to those provided by alma and other instruments . i discuss the anticipated observations ska can provide in the sections below . it has been shown @xcite that certain agb stars , the mira variables , undergo temperature changes of 30% and luminosity changes of a factor of two during their visual fluctuation period ( @xmath5 ) . as shown in @xcite , these changes should result in stellar radius fluctuations of about 40% . such dramatic fluctuations would lead to both dramatic shock waves that propagate from the star into its extended photosphere and also measurable changes in light curves at radio , infrared and optical wavelengths . the visual light curve fluctuates dramatically ( extreme cases show fluctuations of 8 magnitudes ) while the infrared light curves rarely fluctuate by more than a magnitude and the radio light curves fluctuate only by a few percent at most . observed light curves do not match those predicted by the radial fluctuations implied by the temperature and luminosity fluctuations @xcite . figure [ reidmira ] shows the model from @xcite that ( to first order ) reproduces the radio , infrared and visual light curves . the model predicts that tio is formed in the upper atmosphere as the star approaches minimum light and this additional opacity source can greatly decrease the observed light at visual wavelengths while having less impact at infrared wavelengths and almost no impact at radio wavelengths . this new discovery shows that much remains to be learned about evolved stars . after all , mira variables are one of the oldest astronomical phenomena studied and only now has an adequate first - order model been developed to explain their fluctuations . the ska will allow further testing of these models at far greater sensitivity . observations of the flux from the photospheres of agb stars are exceedingly difficult . typical fluxes are on the order of 200 microjy and require special calibration techniques with current interferometers . the ska , with a sensitivity of about 0.1 microjy at 20 ghz will provide the most accurate agb star light curves across all wavelengths . such measurements will allow improved modelling of the opacity source fluctuations in the star . the discovery of particularly large extrasolar planets orbiting close to their host star opens up the possibility for the observation of eclipses using the ska , as has been observed in the star hd 209458 . however , the eclipse type that is potentially observable would be an active radio - emitting planet similar to jupiter being eclipsed by its host star rather than the more typical eclipse . as pointed out in @xcite , jupiter - like planets will produce detectable radio emission out to distances of 10 pc . the passage of a planet of this type behind its host agb star would be detectable , since the emission from the planet would be of order 10 microjy , compared to the photospheric flux of 200 microjy . the diameter of the radio photosphere of a typical agb star is of order 5 au . at 1kpc , such a source would have a maximum angular diameter of 6 milliarcseconds . at 22 ghz and with 1000 km baselines , the ska will have a resolution of roughly 3 milliarcseconds . this resolution corresponds to linear resolution of 3 au at 1 kpc . for agb star diameters of 3 - 5 au , they can be moderately resolved with ska . thus , for only the nearest agb stars will any degree of imaging be possible . the number of agb stars closer than 1kpc is limited . without a substantial increase in the highest frequency observed by the ska or the maximum baselines , imaging of only the nearest agb stars will be possible . that said , some very interesting imaging projects can be undertaken for large agb stars not further than 1 kpc from the earth . for example , the well - known and nearby ( 150 pc ) carbon star irc+10216 has a photospheric size of 35 milliarcseconds and an extended envelope diameter of nearly 1. with a resolution element of 3 mas , the surface of the star would be imaged well and the overall envelope , especially in spectral lines ( see below ) would be highly resolved . the imaging design goal of 0.1 arcsecond resolution at 1.4 ghz over a 1 degree field ( and scaled with frequency ) is sufficient to provide high spatial dynamic range imaging at high sensitivity for objects of this type . the science the ska will allow is the direct imaging of the dust formation process and connection with stellar pulsation for the nearest and largest agb stars . maser emission from gas in the outflowing winds of agb stars is a common phenomenon in o - rich agb stars . masers are regions of gas in the stellar wind that have sufficient velocity coherence to amplify background photons via amplified emission of radiation . such amplification is possible due to a population inversion of the molecular species in question and a fortuitous alignment of molecular rotational , vibrational or ro - vibrational energy levels . several species are found . sio masers are located close to the star ( within a few stellar radii and below the dust formation zone ) . masers are located at intermediate distances from a few tens of stellar radii to a few hundred . the remote oh masers are located up to several thousand stellar radii from the host star . in addition to knowing the location of the various maser species , we have a good understanding of the overall shell structure around these stars @xcite . figure [ shell ] shows graphically our current understanding of the circumstellar region around an agb star . the star itself is from between 1 and 5 au in size . above this surface is a chromospheric region followed by a molecular photosphere ending between 1 and 2 au above the optical photosphere . the radio photosphere ( about 0.5 to 1au in thickness ) is located near the sio maser formation region . beyond this zone , wind acceleration begins as dust forms in a region from roughly 5 au to 10 au ( depending on the properties of the star and pulsation phase ) . the h@xmath4o and oh masers begin to appear at radii of 15 au or more and the oh masers are found further out from the water maser shell . the exact sizes of the various regions , their exact locations and how they interact remain rough measurements . , width=283 ] masers are imaged using vlbi techniques and typically have resolved sizes of a milliarcsecond or so , but observations with merlin show that a weak diffuse emission can also be present @xcite . depending on the upper frequency cutoff for the ska , the oh ( 1.6 ghz ) , methanol ( 6.7 ghz ) and water masers ( 22 ghz ) could be observable . however , it is not the detection of maser emission with the ska that is of greatest interest ( though the sensitivity of the instrument would allow detection of extragalactic masers to a much greater distance than currently available ) . it is the sensitivity to both the stellar photospheric emission and the dust continuum in the wind combined with the vlbi observations that will be of prime interest . vlbi imaging techniques are sensitive only to very high brightness temperatures and the smallest angular sizes ( 1 - 5 milliarcseconds ) and therefore only the maser spots themselves and not the environment in which they are located can be imaged . with the angular resolution of the ska at 1.4 ghz ( 0.1 `` ) , the thermal emission across a typical oh maser distribution 1''-2 " in diameter could be mapped with sufficient resolution and sensitivity to allow alignment of the vlbi maser observations with the overall dust distribution and star itself . combined with infrared interferometric observations , which are now beginning to show the details of the dust distribution at high angular resolutions ( see figure [ monnierfigure ] ) @xcite ( @xmath6 milliarseconds , but over limited fields of view ) , the ska will play a critical role in providing information on the largest scales . outstanding problems to be addressed include the details of dust formation , such as whether the process proceeds uniformly as a function of pulsation cycle or at particular times , the degree of clumpiness of the dust formation and the exact physical conditions that lead to dust formation . the transition of agb stars from roughly spherically symmetric mass - losing objects to the asymmetric planetary nebulae has yet to be understood completely and the combination of the kinematic information provided by vlbi maser observations and the dust distribution will hopefully shed new light on this area . in the frequency range of the ska are 634 molecular line transitions , many of which have not been well studied , only detected , and some of which have still not been identified @xcite . for convenience , these transitions are provided in table [ lovastable ] . many of these species are expected to be present in the winds of evolved stars . as pointed out by zijlstra ( 2003 ) , both the dust and gas created in the stellar winds of agb stars survive in the ism . the dust , as indicated by reddening and particles found in meteorites or as micrometeoroid particulates in our own upper atmosphere @xcite , survives in interstellar space . the presence of the diffuse interstellar bands are the main piece of evidence for the existence of rather complex molecules in interstellar space . as yet , we do not know if the molecules were formed within an envelope of an agb star and survived ejection into interstellar space or if they were incorporated onto dust grains and later evaporated from their surfaces upon exposure to interstellar uv radiation . it is likely that the true situation will be a mixture of both of these cases . a number of molecules with transitions in the frequency range of the ska are of particular interest for astrobiology . these include the building block molecule for simple sugars such as ribose and deoxyribose , furan ( @xmath7 ; e.g. with a transition at 10.6 ghz ) @xcite . the same authors detected c - c@xmath4h@xmath8o , one of the few cyclic molecules in space . they note that the presence of these molecular species in cold dark clouds suggests that rather complex organic molecules may have been present in the solar system before the planets formed , a first step toward explaining the origin of life on the early earth . magnetic fields of agb stars are now thought to be fairly strong from observations of sio masers @xcite . depending on the exact models used , the field strengths seem to be between 5 - 10 g at radii of 3 au or so . such strong magnetic fields will have obvious impacts on both the molecular gas ( zeeman splitting for a number of species such as ccs and so ) and possibly produce circularly polarized radio emission from the star itself ( analogous to the emission observed from the sun ) . although requiring careful instrumental polarization characterization , observations of these effects will provide confirmation of the magnetic field strength implied by the sio maser polarization observations and further constraints on agb stars themselves . potential movies of regions of magnetic field enhancements on the surfaces of nearby agb stars could be tracked with time , testing maser observations of rotation . 9 j. h. bieging and rieu , n.q . , astronomical journal 112 , ( 1996 ) , 706 . d. a. boboltz & k. b. marvel , astrophysical journal 545 , ( 2000 ) , 149 . dickens , w. m. irvine , a. nummelin , h. mollendal , s. saito , s. thorwith , a. hjalmarsoni , m. ohishi , spectrochimica acta part a : molecular and biomolecular spectroscopy 57 , ( 2001 ) , 643 . a. j. kemball & p.j . diamond , astrophysical journal , 481l ( 1997 ) , 111 . lovas , nist recommended rest frequencies for observed interstellar molecular microwave transitions : 2002 revision , ( 2002 ) . s. messenger , l.p . keller , f.j . stadermann , r.m . walker and e. zinner , science 300 , ( 2003 ) , 105 . monnier , r. millan - gabet , p.g . tuthill , w.a . traub , n.p . carleton , v. coud du foresto , w. c. danchi , m.g . lacasse , s. morel , g. perrin , i.l . porro , f.p . schloerb , c.h . townes , astrophysical journal 605 , ( 2004 ) , 436 . monnier , p.g . tuthill , b. lopez , p. cruzalebes , w. c. danchi , c.a . haniff , astrophysical journal 512 , ( 1999 ) , 351 . k. murakawa , j.a . yates , a.m.s . richards , r.j . cohen monthly notices of the royal astronomical society 344 , ( 2003 ) , 1 . h. olofsson , p. bergman , r. lucas , k. eriksson , b. gustafsson and j.h . bieging , astronomy and astrophysics 330 , ( 1998 ) , l1 . e. petit and s. b. nicholson , astrophysical journal 78 , ( 1933 ) , 320 . reid and k.m . menten , astrophysical journal 456 , ( 1997 ) 327 . reid and j.e . goldston , astrophysical journal 568 , ( 2002 ) 931 . reid in _ cosmic masers : from protostars to blackholes _ , astronomical society of the pacific , 2002 . richards , j. a. yates and r. j. cohen , monthly notices of the royal astronomical society 306 , ( 1999 ) 954 . richards , j. a. yates and r. j. cohen , monthly notices of the royal astronomical society 299 , ( 1998 ) 319 . rieu and j.h . bieging , astrophysical journal 359 , ( 1990 ) , 131 . taylor and r. braun , _ science with the square kilometer array _ , 1999 . zijlstra , the scientific promise of the ska : ska workshop oxford , 2003 , kramer & rawlings , eds . web reference http://www.skatelescope.org/documents/workshop_oxford2002.pdf lll|lll + frequency ( mhz ) & formula & quantum numbers & frequency ( mhz ) & formula & quantum numbers + frequency ( mhz ) & formula & quantum numbers & frequency ( mhz ) & formula & quantum numbers + + 701.679 & ch & 23/2 j=3/2 f=2 - 2 & 19418.796 & c - c@xmath9hd & 1(1,0)-1(0,1 ) f=1 - 0 + 704.175 & ch & 23/2 j=3/2 f=2 + -1- & 19426.679 & ch@xmath4chcn & 2(1,1)-1(1,0 ) f=2 - 1 + 722.303 & ch & 23/2 j=3/2 f=1 + -2- & 19427.851 & ch@xmath4chcn & 2(1,1)-1(1,0 ) f=3 - 2 + 724.791 & ch & 23/2 j=3/2 f=1 - 1 & 19429.098 & ch@xmath4chcn & 2(1,1)-1(1,0 ) f=1 - 0 + 834.285 & ch@xmath9oh & 1(1,0)-1(1,1 ) a-+ & 19430.85 & unidentified & + 1065.076 & ch@xmath9cho & 1(1,0 ) - 1(1,1 ) a-+ & 19609.78 & unidentified & + 1371.722 & ch@xmath4chcn & 2(1,1)-2(1,2 ) f=1 - 1 & 19682.50 & unidentified & + 1371.797 & ch@xmath4chcn & 2(1,1)-2(1,2 ) f=3 - 3 & 19692.50 & unidentified & + 1371.934 & ch@xmath4chcn & 2(1,1)-2(1,2 ) f=2 - 2 & 19755.111 & hc@xmath10n & 34 - 33 + 1538.108 & nh@xmath4cho & 1(1,0)-1(1,1 ) f=1 - 1 & 19757.538 & nh@xmath9 & 6(3)-6(3 ) + 1538.676 & nh@xmath4cho & 1(1,0)-1(1,1 ) f=1 - 2 & 19771.50 & unidentified & + 1539.264 & nh@xmath4cho & 1(1,0)-1(1,1 ) f=2 - 1 & 19780.800 & cccn & 2 - 1 j=5/2 - 3/2 f=5/2 - 3/2 + 1539.527 & nh@xmath4cho & 1(1,0)-1(1,1 ) f=1 - 0 & 19780.826 & cccn & 2 - 1 j=5/2 - 3/2 f=3/2 - 1/2 + 1539.832 & nh@xmath4cho & 1(1,0)-1(1,1 ) f=2 - 2 & 19781.094 & cccn & 2 - 1 j=5/2 - 3/2 f=7/2 - 5/2 + 1540.998 & nh@xmath4cho & 1(1,0)-1(1,1 ) f=0 - 1 & 19799.951 & cccn & 2 - 1 j=5/2 - 3/2 f=3/2 - 1/2 + 1570.805 & nh@xmath11cho & 1(1,0)-1(1,1 ) f=2 - 2 & 19800.121 & cccn & 2 - 1 j=5/2 - 3/2 f=5/2 - 3/2 + 1584.274 & @xmath12oh & 23/2 j=3/2 f=1 - 2 & 19838.346 & nh@xmath9 & 5(1)-5(1 ) + 1610.247 & ch@xmath9ocho & 1(1,0)-1(1,1 ) a & 19871.344 & hccnc & 2 - 1 + 1610.900 & ch@xmath9ocho & 1(1,0)-1(1,1 ) e & 19967.396 & ch@xmath9oh & 2(1,1)-3(0,3 ) e + 1612.2310 & oh & 23/2 j=3/2 f=1 - 2 & 19974.50 & unidentified & + 1624.518 & @xmath13oh & 23/2 j=3/2 f , f1=7/2,4 - 7/2,4 & 20064.21 & unidentified & + 1626.161 & @xmath13oh & 23/2 j=3/2 f , f1=9/2,4 - 9/2,4 & 20109.547 & ch@xmath4cn & 1 - 0 3/2 - 1/2 5/2 - 3/2 5/2 - 5/2 + 1637.564 & @xmath12oh & 23/2 j=3/2 f=1 - 1 & 20115.77 & ch@xmath4cn & 1 - 0 1/2 - 1/2 3/2 - 3/2 5/2 - 5/2 + 1638.805 & hcooh & 1(1,0)-1(1,1 ) & 20117.43 & ch@xmath4cn & 1 - 0 3/2 - 1/2 5/2 - 3/2 3/2 - 1/2 + 1639.503 & @xmath12oh & 23/2 j-3/2 f=2 - 2 & 20118.014 & ch@xmath4cn & 1 - 0 3/2 - 1/2 5/2 - 3/2 5/2 - 3/2 + 1665.4018 & oh & 23/2 j=3/2 f=1 - 1 & 20118.16 & ch@xmath4cn & 1 - 0 3/2 - 1/2 1/2 - 1/2 3/2 - 3/2 + 1667.3590 & oh & 23/2 j=3/2 f=2 - 2 & 20119.606 & ch@xmath4cn & 1 - 0 3/2 - 1/2 5/3 - 3/2 7/2 - 5/2 + 1692.795 & @xmath12oh & 23/2 j=3/2 f=2 - 1 & 20121.61 & ch@xmath4cn & 1 - 0 3/2 - 1/2 3/2 - 3/2 3/2 - 3/2 + 1720.5300 & oh & 23/2 j=3/2 f=2 - 1 & 20123.96 & ch@xmath4cn & 1 - 0 3/2 - 1/2 1/2 - 1/2 3/2 - 3/2 + 2661.61 & hc@xmath14n & 1 - 0 f=1 - 1 & 20124.22 & ch@xmath4cn & 1 - 0 1/2 - 1/2 3/2 - 1/2 3/2 - 1/2 + 2662.87 & hc@xmath14n & 1 - 0 f=2 - 1 & 20124.22 & ch@xmath4cn & 1 - 0 3/2 - 1/2 3/2 - 3/2 1/2 - 1/2 + 2664.76 & hc@xmath14n & 1 - 0 f=0 - 1 & 20124.45 & ch@xmath4cn & 1 - 0 3/2 - 1/2 3/2 - 1/2 3/2 - 3/2 + 3139.404 & h@xmath4cs & 2(1,1)-2(1,2 ) & 20124.49 & ch@xmath4cn & 1 - 0 1/2 - 1/2 3/2 - 3/2 5/2 - 3/2 + 3195.162 & ch@xmath9cho & 2(1,1 ) - 2(1,2 ) a-+ & 20126.031 & ch@xmath4cn & 1 - 0 3/2 - 1/2 3/2 - 3/2 3/2 - 1/2 + 3263.794 & ch & 21/2 j=1/2 f=0 - 1 & 20128.770 & ch@xmath4cn & 1 - 0 1/2 - 1/2 3/2 - 1/2 3/2 - 3/2 + 3335.481 & ch & 21/2 j=1/2 f=1 - 1 & 20139.76 & ch@xmath4cn & 1 - 0 1/2 - 1/2 1/2 - 3/2 3/2 - 5/2 + 3349.193 & ch & 21/2 j=1/2 f=1 - 0 & 20168.48 & unidentified & + 4388.7786 & h@xmath4c@xmath12o & 1(1,0)-1(1,1 ) f=1 - 0 & 20171.089 & ch@xmath9oh & 11(1,11)-10(2,8 ) a+ + 4388.7960 & h@xmath4c@xmath12o & 1(1,0)-1(1,1 ) f=0 - 1 & 20203.31 & unidentified & + 4388.7963 & h@xmath4c@xmath12o & 1(1,0)-1(1,1 ) f=2 - 2 & 20209.209 & ch@xmath15co & 1(0,1)-0(0,0 ) + 4388.8011 & h@xmath4c@xmath12o & 1(1,0)-1(1,1 ) f=2 - 1 & 20281.00 & unidentified & + 4388.8035 & h@xmath4c@xmath12o & 1(1,0)-1(1,1 ) f=1 - 2 & 20303.946 & hc@xmath16n & 18 - 17 + 4388.8084 & h@xmath4c@xmath12o & 1(1,0)-1(1,1 ) f=1 - 1 & 20336.135 & hc@xmath10n & 35 - 34 + 4592.9563 & h@xmath11co & 1(1,0)-1(1,1)1/2,1/2 - 1/2,3/2 & 20357.226 & ch3c@xmath8h & 5(1)-4(1 ) + 4592.9738 & h@xmath11co & 1(1,0)-1(1,1)1/2,1/2 - 3/2,3/2 & 20357.423 & ch3c@xmath8h & 5(0)-4(0 ) + 4592.9759 & h@xmath11co & 1(1,0)-1(1,1)3/2,1/2 - 1/2,3/2 & 20371.45 & nh@xmath9 & 5(2)-5(2 ) + 4592.9857 & h@xmath11co & 1(1,0)-1(1,1)3/2,1/2 - 5/2,3/2 & 20460.01 & hdo & 3(2,1)-4(1,4 ) + 4592.9934 & h@xmath11co & 1(1,0)-1(1,1)3/2,1/2 - 3/2,3/2 & 20501.5 & unidentified & + 4593.0494 & h@xmath11co & 1(1,0)-1(1,1)1/2,1/2 - 1/2,1/2 & 20533.235 & unidentified & + 4593.0690 & h@xmath11co & 1(1,0)-1(1,1)3/2,1/2 - 1/2,1/2 & 20533.289 & c@xmath17h & 23/2 17.5 - 16.5 + 4593.0800 & h@xmath11co & 1(1,0)-1(1,1)1/2,1/2 - 3/2,1/2 & 20723.5 & unidentified & + 4593.0812 & h@xmath11co & 1(1,0)-1(1,1)1/2,3/2 - 1/2,3/2 & 20728.67 & unidentified & + 4593.0864 & h@xmath11co & 1(1,0)-1(1,1)3/2,3/2 - 1/2,3/2 & 20735.452 & nh@xmath9 & 9(7)-9(7 ) + 4593.0865 & h@xmath11co & 1(1,0)-1(1,1)5/2,3/2 - 5/2,3/2 & 20765.80 & unidentified & + 4593.0942 & h@xmath11co & 1(1,0)-1(1,1)5/2,3/2 - 3/2,3/2 & 20790.00 & unidentified & + 4593.0961 & h@xmath11co & 1(1,0)-1(1,1)3/2,3/2 - 5/2,3/2 & 20792.563 & h@xmath4ccc & 1(0,1)-0(0,0 ) + 4593.0985 & h@xmath11co & 1(1,0)-1(1,1)1/2,3/2 - 3/2,3/2 & 20792.872 & c@xmath18h & 23/2 j=15/2 - 13/2 f=8 - 7 e + 4593.0994 & h@xmath11co & 1(1,0)-1(1,1)3/2,1/2 - 3/2,1/2 & 20792.945 & c@xmath18h & 23/2 j=15/2 - 13/2 f=7 - 6 e + 4593.1039 & h@xmath11co & 1(1,0)-1(1,1)3/2,3/2 - 3/2,3/2 & 20794.444 & c@xmath18h & 23/2 j=15/2 - 13/2 f=8 - 7 f + 4593.1741 & h@xmath11co & 1(1,0)-1(1,1)1/2,3/2 - 1/2,1/2 & 20794.512 & c@xmath18h & 23/2 j=15/2 - 13/2 f=7 - 6 f + 4593.1795 & h@xmath11co & 1(1,0)-1(1,1)3/2,3/2 - 1/2,1/2 & 20804.830 & nh@xmath9 & 7(5)-7(5 ) + 4593.2003 & h@xmath11co & 1(1,0)-1(1,1)5/2,3/2 - 3/2,1/2 & 20838.20 & unidentified & + 4593.2046 & h@xmath11co & 1(1,0)-1(1,1)1/2,3/2 - 3/2,1/2 & 20847.50 & unidentified & + 4593.2099 & h@xmath11co & 1(1,0)-1(1,1)3/2,3/2 - 3/2,1/2 & 20852.527 & nh@xmath9 & 10(8)-10(8 ) + 4617.121 & nh@xmath4cho & 2(1,1)-2(1,2 ) f=2 - 2 & 20878.00 & unidentified & + 4618.967 & nh@xmath4cho & 2(1,1)-2(1,2 ) f=3 - 3 & 20908.848 & ch@xmath9oh & 16(-4,13)-15(-5,10 ) e + 4619.993 & nh@xmath4cho & 2(1,1)-2(1,2 ) f=1 - 1 & 20917.157 & hc@xmath10n & 36 - 35 + 4660.242 & oh & 21/2 j=1/2 f=0 - 1 & 20970.658 & ch@xmath9oh & 10(1,10)-11(,9 ) a+ t=1 + 4750.656 & oh & 21/2 j=1/2 f=1 - 1 & 20994.617 & nh@xmath9 & 6(4)-6(4 ) + 4765.562 & oh & 21/2 j=1/2 f=1 - 0 & 20999.79 & unidentified & + 4829.6412 & h@xmath15co & 1(1,0)-1(1,1 ) f=1 - 0 & 21070.739 & nh@xmath9 & 11(9)-11(9 ) + 4829.6587 & h@xmath15co & 1(1,0)-1(1,1 ) f=0 - 1 & 21134.311 & nh@xmath9 & 4(1)-4(1 ) + 4829.6594 & h@xmath15co & 1(1,0)-1(1,1 ) f=2 - 2 & 21143.18 & unidentified & + 4829.6639 & h@xmath15co & 1(1,0)-1(1,1 ) f=2 - 1 & 21231.00 & unidentified & + 4829.6664 & h@xmath15co & 1(1,0)-1(1,1 ) f=1 - 2 & 21285.275 & nh@xmath9 & 5(3)-5(3 ) + 4829.6710 & h@xmath15co & 1(1,0)-1(1,1 ) f=1 - 1 & 21301.261 & hc@xmath14n & 8 - 7 + 4916.312 & hcooh & 2(1,1)-2(1,2 ) & 21322.50 & unidentified & + 5005.3208 & ch@xmath9oh & 3(1,2)-3(1,3 ) a-+ & 21431.932 & hc@xmath16n & 19 - 18 + 5289.015 & ch@xmath4nh & 1(1,0)-1(1,1 ) f=0 - 1 & 21447.8 & unidentified & + 5289.678 & ch@xmath4nh & 1(1,0)-1(1,1 ) f=1 - 0 & 21453.93 & unidentified & + 5289.813 & ch@xmath4nh & 1(1,0)-1(1,1 ) f=2 - 2 & 21470.4 & unidentified & + 5290.614 & ch@xmath4nh & 1(1,0)-1(1,1 ) f=2 - 1 & 21480.809 & c@xmath14h & 21/2 j=9/2 - 7/2 f=5 - 4 e + 5290.879 & ch@xmath4nh & 1(1,0)-1(1,1 ) f=1 - 2 & 21481.299 & c@xmath14h & 21/2 j=9/2 - 7/2 f=4 - 3 e + 5291.680 & ch@xmath4nh & 1(1,0)-1(1,1 ) f=1 - 1 & 21484.695 & c@xmath14h & 21/2 j=9/2 - 7/2 f=5 - 4 f + 5324.058 & hc@xmath14n & 2 - 1 f=2 - 2 & 21485.248 & c@xmath14h & 21/2 j=9/2 - 7/2 f=4 - 3 f + 5324.270 & hc@xmath14n & 2 - 1 f=1 - 0 & 21488.255 & h@xmath4cccccc & 8(1,8)-7(1,7 ) + 5325.330 & hc@xmath14n & 2 - 1 f=2 - 1 & 21498.182 & hc@xmath10n & 37 - 36 + 5325.421 & hc@xmath14n & 2 - 1 f=3 - 2 & 21546.94 & unidentified & + 5327.451 & hc@xmath14n & 2 - 1 f=1 - 1 & 21550.342 & ch@xmath9oh & 12(2,11)-11(1,11 ) a+ t=1 + 6016.746 & oh & 23/2 j=5/2 f=2 - 3 & 21569.5 & unidentified & + 6030.747 & oh & 23/2 j=5/2 f=2 - 2 & 21576.5 & unidentified & + 6035.092 & oh & 23/2 j-5/2 f=3 - 3 & 21582.6 & unidentified & + 6049.084 & oh & 23/2 j=5/2 f=3 - 2 & 21587.400 & c - c@xmath9h@xmath4 & 2(2,0)-2(1,1 ) + 6278.628 & h@xmath4cs & 3(1,2)-3(1,3 ) & 21592.1 & unidentified & + 6389.933 & ch@xmath9cho & 3(1,2 ) - 3(1,3 ) a-+ & 21595.8 & unidentified & + 6668.5192 & ch@xmath9oh & 5(1,6)-6(0,6 ) a++ & 21598.4 & unidentified & + 7761.747 & oh & 21/2 j=3/2 f=1 - 1 & 21606.30 & unidentified & + 7820.125 & oh & 21/2 j=3/2 f=2 - 2 & 21615.5 & unidentified & + 7895.989 & hc@xmath16n & 7 - 6 f=6 - 5 & 21703.3580 & nh@xmath9 & 4(2)-4(2 ) + 7896.010 & hc@xmath16n & 7 - 6 f=7 - 6 & 21715.8 & unidentified & + 7896.023 & hc@xmath16n & 7 - 6 f=8 - 7 & 21930.476 & cc34s & 2,1 - 1,0 + 7987.782 & hc@xmath14n & 3 - 2 f=2 - 1 & 21980.5453 & hnco & 1(0,1)-0(0,0 ) f=0 - 1 + 7987.994 & hc@xmath14n & 3 - 2 f=3 - 2 & 21981.4706 & hnco & 1(0,1)-0(0,0 ) f=2 - 1 + 7988.044 & hc@xmath14n & 3 - 2 f=4 - 3 & 21982.0854 & hnco & 1(0,1)-0(0,0 ) f=1 - 1 + 8135.870 & oh & 21/2 j=5/2 f=2 - 2 & 22079.204 & hc@xmath10n & 38 - 37 + 8189.587 & oh & 21/2 j=5/2 f=3 - 3 & 22235.044 & h@xmath4o & 6(1,6)-5(2,3 ) f=7 - 6 + 8775.088 & ch@xmath9nh@xmath4 & 2(0,2)-1(0,1 ) f=1 - 0 aa & 22235.077 & h@xmath4o & 6(1,6)-5(2,3 ) f=6 - 5 + 8777.442 & ch@xmath9nh@xmath4 & 2(0,2)-1(0,1 ) f=3 - 2 aa & 22235.120 & h@xmath4o & 6(1,6)-5(2,3 ) f=5 - 4 + 8778.200 & ch@xmath9nh@xmath4 & 2(0,2)-1(0,1 ) f=2 - 2 aa & 22235.253 & h@xmath4o & 6(1,6)-5(2,3 ) f=6 - 6 + 8778.260 & ch@xmath9nh@xmath4 & 2(0,2)-1(0,1 ) f=1 - 1 aa & 22235.298 & h@xmath4o & 6(1,6)-5(2,3 ) f=5 - 5 + 8779.496 & ch@xmath9nh@xmath4 & 2(0,2)-1(0,1 ) f=2 - 1 aa & 22258.173 & cco & 2,1 - 1,0 + 8815.814 & h@xmath19cccn & 1 - 0 f=1 - 1 & 22307.670 & hdo & 5(3,2)-5(3,3 ) + 8817.096 & h@xmath19cccn & 1 - 0 f=2 - 1 & 22344.030 & ccs & 2,1 - 1,0 + 8819.019 & h@xmath19cccn & 1 - 0 f=0 - 1 & 22471.180 & hcooh & 1(0,1)-0(0,0 ) + 9024.009 & hc@xmath16n & 8 - 7 & 22559.915 & hc@xmath16n & 20 - 19 + 9058.447 & hc@xmath19ccn & 1 - 0 f=1 - 1 & 22624.8892 & 15nh@xmath9 & 1(1)-1(1 ) f , f1=1.5,1 - 1.3,1 + 9059.318 & hcc@xmath19cn & 1 - 0 f=1 - 1 & 22624.9331 & 15nh@xmath9 & 1(1)-1(1 ) f , f1=1.5,1 - 0.8,1 + 9059.736 & hc@xmath19ccn & 1 - 0 f=2 - 1 & 22624.9410 & 15nh@xmath9 & 1(1)-1(1 ) f , f1=0.5,1 - 0.8,1 + 9060.6080 & hcc@xmath19cn & 1 - 0 f=2 - 1 & 22624.9469 & 15nh@xmath9 & 1(1)-1(1 ) f , f1=1.5,2 - 1.5,2 + 9097.0346 & hcccn & 1 - 0 f=1 - 1 & 22639.3 & unidentified & + 9098.3321 & hcccn & 1 - 0 f=2 - 1 & 22644.3 & unidentified & + 9100.2727 & hcccn & 1 - 0 f=0 - 1 & 22649.843 & 15nh@xmath9 & 2(2)-2(2 ) + 9118.823 & ch@xmath9och@xmath9 & 2(0,2)-1(1,1 ) aa & 22653.022 & nh@xmath9 & 5(4)-5(4 ) + 9119.671 & ch@xmath9och@xmath9 & 2(0,2)-1(1,1 ) ee & 22660.225 & hc@xmath10n & 39 - 38 + 9120.509 & ch@xmath9och@xmath9 & 2(0,2)-1(1,1 ) ae & 22678.6 & unidentified & + 9120.527 & ch@xmath9och@xmath9 & 2(0,2)-1(1,1 ) ea & 22688.312 & nh@xmath9 & 4(3)-4(3 ) + 9235.119 & nh@xmath4cho & 3(1,2)-3(1,3 ) f=3 - 3 & 22732.429 & nh@xmath9 & 6(5)-6(5 ) + 9237.034 & nh@xmath4cho & 3(1,2)-3(1,3 ) f=4 - 4 & 22789.421 & 15nh@xmath9 & 3(3)-3(3 ) + 9237.704 & nh@xmath4cho & 3(1,2)-3(1,3 ) f=2 - 2 & 22827.741 & ch@xmath9ocho & 2(1,2)-1(1,1 ) e + 9486.71 & unidentified & & 22828.134 & ch@xmath9ocho & 2(1,2)-1(1,1 ) a + 9493.061 & c@xmath8h & 3/2 - 1/2 f=1 - 0 & 22834.1851 & nh@xmath9 & 3(2)-3(2 ) + 9496.4 & unidentified & & 22878.949 & dc5n & 9 - 8 + 9497.616 & c@xmath8h & 3/2 - 1/2 f=2 - 1 & 22924.940 & nh@xmath9 & 7(6)-7(6 ) + 9508.005 & c@xmath8h & 3/2 - 1/2 f=1 - 1 & 23046.0158 & 15nh@xmath9 & 4(4)-4(4 ) + 9547.953 & c@xmath8h & 1/2 - 1/2 f=1 - 0 & 23098.8190 & nh@xmath9 & 2(1)-2(1 ) + 9551.717 & c@xmath8h & 1/2 - 1/2 f=0 - 1 & 23121.024 & ch@xmath9oh & 9(2,7)-10(1,10 ) a+ + 9562.904 & c@xmath8h & 1/2 - 1/2 f=1 - 1 & 23122.983 & cccs & 4 - 3 + 9703.508 & c@xmath18h & 23/2 j=3.5 - 2.5 f=4 - 3 e & 23142.2 & unidentified & + 9703.600 & c@xmath18h & 23/2 j=3.5 - 2.5 f=3 - 2 e & 23228.0 & unidentified & + 9703.835 & c@xmath18h & 23/2 j=3.5 - 2.5 f=4 - 3 f & 23232.238 & nh@xmath9 & 8(7)-8(7 ) + 9703.936 & c@xmath18h & 23/2 j=3.5 - 2.5 f=3 - 2 f & 23241.246 & hc@xmath10n & 40 - 39 + 9877.606 & hc@xmath10n & 17 - 16 & 23421.9823 & 15nh@xmath9 & 5(5)-5(5 ) + 9885.89 & cccn & 1 - 0 j=3/2 - 1/2 f=5/2 - 3/2 & 23444.778 & ch@xmath9oh & 10(1,9)-9(2,8 ) a- + 9936.202 & ch@xmath9oh & 9(-1,9)-8(-2,7 ) e & 23565.160 & c@xmath18h & 23/2 j=17/2 - 15/2 f=9 - 8 e + 9978.686 & ch@xmath9oh & 4(3,2)-5(2,3 ) e & 23565.226 & c@xmath18h & 23/2 j=17/2 - 15/2 f=8 - 7 e + 10058.257 & ch@xmath9oh & 4(3,1)-5(2,4 ) e & 23567.169 & c@xmath18h & 23/2 j=17/2 - 15/2 f=9 - 8 f + 10152.008 & hc@xmath16n & 9 - 8 & 23567.238 & c@xmath18h & 23/2 j=17/2 - 15/2 f=8 - 7 f + 10278.246 & hdo & 2(2,0)-2(2,1 ) & 23600.242 & sic2 & 1(0,1)-0(0,0 ) + 10458.639 & hc@xmath10n & 18 - 17 & 23657.471 & nh@xmath9 & 9(8)-9(8 ) + 10463.962 & h@xmath4cs & 4(1,3)-4(1,4 ) & 23687.898 & hc@xmath16n & 21 - 20 + 10648.419 & ch@xmath9cho & 4(1,3 ) - 4(1,4 ) a-+ & 23692.9265 & nh@xmath9 & 1(1)-1(1 ) f , f1=1/2,1 - 1/2,0 + 10650.563 & hc@xmath14n & 4 - 3 f=3 - 2 & 23692.9688 & nh@xmath9 & 1(1)-1(1 ) f , f1=3/2,1 - 1/2,0 + 10650.654 & hc@xmath14n & 4 - 3 f=4 - 3 & 23693.8722 & nh@xmath9 & 1(1)-1(1 ) f , f1=1/2,1 - 3/2,2 + 10650.686 & hc@xmath14n & 4 - 3 f=5 - 4 & 23693.9051 & nh@xmath9 & 1(1)-1(1 ) f , f1=3/2,1 - 5/2,2 + 11119.445 & ccs & 1,0 - 0,1 & 23693.9145 & nh@xmath9 & 1(1)-1(1 ) f , f1=3/2,1 - 3/2,2 + 11280.006 & hc@xmath16n & 10 - 9 & 23694.4591 & nh@xmath9 & 1(1)-1(1 ) f , f1=1/2,1 - 1/2,1 + 11561.513 & cccs & 2 - 1 & 23694.4700 & nh@xmath9 & 1(1)-1(1 ) f , f1=1/2,1 - 3/2,1 + 12162.979 & ocs & 1 - 0 & 23694.4709 & nh@xmath9 & 1(1)-1(1 ) f , f1=3/2,2 - 5/2,2 + 12178.593 & ch@xmath9oh & 2(0,2)-3(-1,3 ) e & 23694.4803 & nh@xmath9 & 1(1)-1(1 ) f , f1=3/2,2 - 3/2,2 + 12408.003 & hc@xmath16n & 11 - 10 & 23694.5014 & nh@xmath9 & 1(1)-1(1 ) f , f1=3/2,1 - 1/2,1 + 12782.769 & hc@xmath10n & 22 - 21 & 23694.5060 & nh@xmath9 & 1(1)-1(1 ) f , f1=5/2,2 - 5/2,2 + 12848.48 & unidentified & & 23694.5123 & nh@xmath9 & 1(1)-1(1 ) f , f1=3/2,1 - 3/2,1 + 12848.731 & hc@xmath20n & 38 - 37 & 23694.5153 & nh@xmath9 & 1(1)-1(1 ) f , f1=5/2,2 - 3/2,2 + 13043.814 & so & 1(2)-1(1 ) & 23695.0672 & nh@xmath9 & 1(1)-1(1 ) f , f1=3/2,2 - 3/2,1 + 13116.451 & unidentified & & 23695.0782 & nh@xmath9 & 1(1)-1(1 ) f , f1=3/2,2 - 3/2,1 + 13116.569 & unidentified & & 23695.1132 & nh@xmath9 & 1(1)-1(1 ) f , f1=5/2,2 - 3/2,1 + 13186.46 & unidentified & & 23696.0297 & nh@xmath9 & 1(1)-1(1 ) f , f1=1/2,0 - 1/2,1 + 13186.853 & hc@xmath20n & 39 - 38 & 23696.0406 & nh@xmath9 & 1(1)-1(1 ) f , f1=1/2,0 - 3/2,1 + 13186.98 & unidentified & & 23697.9 & unidentified & + 13313.312 & hc@xmath14n & 5 - 4 & 23718.325 & hc13ccccn & 9 - 8 + 13363.801 & hc@xmath10n & 23 - 22 & 23720.575 & nh@xmath9 & 2(2)-2(2 ) f1=1 - 2 + 13434.596 & oh & 23/2 j=7/2 f=3 - 3 & 23721.336 & nh@xmath9 & 2(2)-2(2 ) f1=3 - 2 + 13441.4173 & oh & 23/2 j=7/2 f=4 - 4 & 23722.6323 & nh@xmath9 & 2(2)-2(2 ) f1=2 - 2 + 13535.998 & hc@xmath16n & 12 - 11 & 23722.6336 & nh@xmath9 & 2(2)-2(2 ) f1=3 - 3 + 13778.804 & h@xmath11co & 2(1,1)-2(1,2 ) & 23722.6344 & nh@xmath9 & 2(2)-2(2 ) f1=1 - 1 + 13880.54 & unidentified & & 23723.929 & nh@xmath9 & 2(2)-2(2 ) f1=2 - 3 + 13944.832 & hc@xmath10n & 24 - 23 & 23724.691 & nh@xmath9 & 2(2)-2(2 ) f1=2 - 1 + 14488.4589 & h@xmath15co & 2(1,1)-2(1,2 ) f=1 - 1 & 23727.162 & hcccc13cn & 9 - 8 + 14488.4712 & h@xmath15co & 2(1,1)-2(1,2 ) f=1 - 2 & 23804.5 & unidentified & + 14488.4801 & h@xmath15co & 2(1,1)-2(1,2 ) f=3 - 3 & 23811.0 & unidentified & + 14488.4899 & h@xmath15co & 2(1,1)-2(1,2 ) f=2 - 2 & 23817.6153 & oh & 23/2 j=9/2 f=4 - 4 + 14525.862 & hc@xmath10n & 25 - 24 & 23822.265 & hc@xmath10n & 41 - 40 + 14663.993 & hc@xmath16n & 13 - 12 & 23826.6211 & oh & 23/2 j=9/2 f=5 - 5 + 14782.212 & 13ch@xmath9oh & 2(0,2)-3(-1,3 ) e & 23867.805 & nh@xmath9 & 3(3)-3(3 ) f1=2 - 3 + 14812.002 & c - c@xmath9h & 1(1,0)-1(1,1 ) j=3/2 - 1/2 f=2 - 1 & 23868.450 & nh@xmath9 & 3(3)-3(3 ) f1=4 - 3 + 14877.671 & c - c@xmath9h & 1(1,0)-1(1,1 ) j=3/2 - 3/2 f=2 - 1 & 23870.1279 & nh@xmath9 & 3(3)-3(3 ) f1=3 - 3 + 14893.050 & c - c@xmath9h & 1(1,0)-1(1,1 ) j=3/2 - 3/2 f=2 - 2 & 23870.1296 & nh@xmath9 & 3(3)-3(3 ) f1=4 - 4 + 14895.243 & c - c@xmath9h & 1(1,0)-1(1,1 ) j=3/2 - 3/2 f=1 - 1 & 23870.1302 & nh@xmath9 & 3(3)-3(3 ) f1=2 - 2 + 15106.892 & hc@xmath10n & 26 - 25 & 23871.807 & nh@xmath9 & 3(3)-3(3 ) f1=3 - 4 + 15248.225 & c@xmath18h & 23/2 j=11/2 - 9/2 f=6 - 5 f & 23872.453 & nh@xmath9 & 3(3)-3(3 ) f1=3 - 2 + 15248.359 & c@xmath18h & 23/2 j=11/2 - 9/2 f=5 - 4 f & 23922.3132 & 15nh@xmath9 & 6(6)-6(6 ) + 15249.064 & c@xmath18h & 23/2 j=11/2 - 9/2 f=6 - 5 e & 23939.089 & hcc13cccn & 9 - 8 + 15249.198 & c@xmath18h & 23/2 j=11/2 - 9/2 f=5 - 4 e & 23941.99 & hccc13ccn & 9 - 8 + 15687.921 & hc@xmath10n & 27 - 26 & 23959.5 & unidentified & + 15791.986 & hc@xmath16n & 14 - 13 & 23963.901 & hc@xmath14n & 9 - 8 + 15975.966 & hc@xmath14n & 6 - 5 & 23987.5 & unidentified & + 16268.950 & hc@xmath10n & 28 - 27 & 23990.2 & unidentified & + 16849.979 & hc@xmath10n & 29 - 28 & 23996.7 & unidentified & + 16886.312 & dcccn & 2 - 1 f=2 - 1 & 24004.5 & unidentified & + 16886.405 & dc@xmath9n & 2 - 1 f=3 - 2 & 24023.2 & unidentified & + 16919.979 & hc@xmath16n & 15 - 14 & 24037.1 & unidentified & + 17091.742 & ch@xmath9cch & 1(0)-0(0 ) & 24048.5 & unidentified & + 17342.256 & cccs & 3 - 2 & 24139.4169 & nh@xmath9 & 4(4)-4(4 ) + 17431.006 & hc@xmath10n & 30 - 59 & 24205.287 & nh@xmath9 & 10(9)-10(9 ) + 17632.685 & h@xmath19cccn & 2 - 1 f=2 - 2 & 24296.491 & ch@xmath9ocho & 2(0,2)-1(0,1 ) e + 17633.844 & h@xmath19cccn & 2 - 1 f=3 - 2 & 24298.481 & ch@xmath9ocho & 2(0,2)-1(0,1 ) a + 17647.479 & c@xmath8d & 5/2 - 3/2 f=5/2 - 3/2 & 24325.927 & ocs & 2 - 1 + 17647.526 & c@xmath8d & 5/2 - 3/2 f=3/2 - 1/2 & 24375.2 & unidentified & + 17647.716 & c@xmath8d & 5/2 - 3/2 f=7/2 - 5/2 & 24428.652 & ch3c@xmath8h & 6(1)-5(1 ) + 17666.995 & hccc15n & 2 - 1 & 24428.886 & ch3c@xmath8h & 6(0)-5(0 ) + 17683.961 & c@xmath8d & 3/2 - 1/2 f=5/2 - 3/2 & 24532.9887 & nh@xmath9 & 5(5)-5(5 ) + 17684.662 & c@xmath8d & 3/2 - 1/2 f=3/2 - 1/2 & 24788.541 & ch3cccn & 6(1)-5(1 ) + 17736.75 & unidentified & & 24788.780 & ch3cccn & 6(0)-5(0 ) + 17788.570 & h@xmath4cccc & 2(1,2)-1(1,1 ) & 24815.878 & hc@xmath16n & 22 - 21 + 17863.803 & h@xmath4cccc & 2(0,2)-1(0,1 ) & 24928.715 & ch@xmath9oh & 3(2,1)-3(1,2 ) e + 17937.956 & h@xmath4cccc & 2(1,1)-1(1,0 ) & 24933.468 & ch@xmath9oh & 4(2,2)-4(1,3 ) e + 17945.85 & unidentified & & 24934.382 & ch@xmath9oh & 2(2,0)-2(1,1 ) e + 17951.95 & unidentified & & 24959.079 & ch@xmath9oh & 5(2,3)-5(1,4 ) e + 17965.09 & unidentified & & 24984.302 & hc@xmath10n & 43 - 42 + 17974.01 & unidentified & & 24991.19 & sic2 & 8(2,6)-8(2,7 ) + 18012.033 & hc@xmath10n & 31 - 30 & 25018.123 & ch@xmath9oh & 6(2,4)-6(1,5 ) e + 18012.46 & unidentified & & 25023.792 & nh@xmath4d & 4(1,4)-4(0,4 ) + 18017.337 & nh@xmath9 & 7(3)-7(3 ) & 25056.025 & nh@xmath9 & 6(6)-6(6 ) + 18020.574 & c@xmath18h & 23/2 j=6.5 - 5.5 f=7 - 6 e & 25124.872 & ch@xmath9oh & 7(2,5)-7(1,6 ) e + 18020.644 & c@xmath18h & 23/2 j=6.5 - 5.5 f=6 - 5 e & 25249.938 & c5n & 21/2 n=9 - 8 j=9.5 - 8.5 + 18021.752 & c@xmath18h & 23/2 j=6.5 - 5.5 f=7 - 6 f & 25260.649 & c5n & 21/2 n=9 - 8 j=8.5 - 7.5 + 18021.818 & c@xmath18h & 23/2 j=6.5 - 5.5 f=6 - 5 f & 25294.417 & ch@xmath9oh & 8(2,6)-8(1,7 ) e + 18021.86 & unidentified & & 25329.441 & dc@xmath9n & 3 - 2 + 18047.969 & hc@xmath16n & 16 - 15 & 25421.036 & dc5n & 10 - 9 + 18119.029 & hc@xmath19ccn & 2 - 1 f=2 - 1 & 25541.398 & ch@xmath9oh & 9(2,7)-9(1,8 ) e + 18120.773 & hcc@xmath19cn & 2 - 1 f=2 - 1 & 25715.182 & nh@xmath9 & 7(7)-7(7 ) + 18120.865 & hcc@xmath19cn & 2 - 1 f=3 - 2 & 25878.266 & ch@xmath9oh & 10(2,8)-10(1,9 ) e + 18154.884 & sis & 1 - 0 & 25911.017 & ccs & 2,2 - 1,1 + 18186.652 & c@xmath17h & 23/2 15.5 - 15.5 e & 25943.855 & hc@xmath16n & 23 - 22 + 18186.782 & c@xmath17h & 23/2 15.5 - 15.5 f & 26337.414 & c@xmath18h & 23/2 j=19/2 - 17/2 f=10 - 9 f + 18194.9206 & hcccn & 2 - 1 f=2 - 2 & 26337.463 & c@xmath18h & 23/2 j=19/2 - 17/2 f=9 - 8 f + 18195.3176 & hcccn & 2 - 1 f=1 - 0 & 26339.924 & c@xmath18h & 23/2 j=19/2 - 17/2 f=10 - 9 e + 18196.2183 & hcccn & 2 - 1 f=2 - 1 & 26339.973 & c@xmath18h & 23/2 j=19/2 - 17/2 f=9 - 8 e + 18196.3119 & hcccn & 2 - 1 f=3 - 2 & 26363.491 & hcccc@xmath19cn & 10 - 9 + 18197.078 & hcccn & 2 - 1 f=1 - 2 & 26450.598 & h@xmath19cccn & 3 - 2 + 18198.3756 & hcccn & 2 - 1 f=1 - 1 & 26500.462 & hccc@xmath21n & 3 - 2 + 18222.65 & unidentified & & 26518.981 & nh@xmath9 & 8(8)-8(8 ) + 18285.434 & nh@xmath9 & 10(7)-10(7 ) & 26602.181 & hccc@xmath19ccn & 10 - 9 + 18294.20 & unidentified & & 26626.533 & hc@xmath14n & 10 - 9 + 18299.5 & unidentified & & 26682.814 & h@xmath4cccc & 3(1,3)-2(1,2 ) + 18306.3 & unidentified & & 26795.635 & h@xmath4cccc & 3(0,3)-2(0,2 ) + 18320.7 & unidentified & & 26847.205 & ch@xmath9oh & 12(2,10)-12(1,11 ) e + 18343.144 & c - c@xmath9h@xmath4 & 1(1,0)-1(0,1 ) & 26906.891 & h@xmath4cccc & 3(1,2)-2(1,1 ) + 18360.50 & unidentified & & 27071.824 & hc@xmath16n & 24 - 23 + 18363.045 & unidentified & & 27084.348 & c - c@xmath9h@xmath4 & 3(3,0)-3(2,1 ) + 18363.142 & unidentified & & 27178.511 & hc@xmath19ccn & 3 - 2 + 18363.306 & unidentified & & 27181.127 & hcc@xmath19cn & 3 - 2 + 18363.406 & unidentified & & 27292.903 & hcccn & 3 - 2 f=3 - 3 + 18368.0 & unidentified & & 27294.078 & hcccn & 3 - 2 f=2 - 1 + 18379.6 & unidentified & & 27294.295 & hcccn & 3 - 2 f=3 - 2 + 18383.3 & unidentified & & 27294.347 & hcccn & 3 - 2 f=4 - 3 + 18391.562 & nh@xmath9 & 6(1)-6(1 ) & 27296.235 & hcccn & 3 - 2 f=2 - 2 + 18396.7252 & ch@xmath9cn & 1(0)-0(0 ) f=1 - 1 & 27472.501 & ch@xmath9oh & 13(2,11)-13(1,12 ) e + 18397.9965 & ch@xmath9cn & 1(0)-0(0 ) f=2 - 1 & 27477.943 & nh@xmath9 & 9(9)-9(9 ) + 18399.8924 & ch@xmath9cn & 1(0)-0(0 ) f=0 - 1 & 28009.975 & hnccc & 3 - 2 + 18413.822 & c - h13ccch & 1(1,0)-1(0,1 ) & 28169.437 & ch@xmath9oh & 14(2,12)-14(1,13 ) e + 18422.00 & unidentified & & 28199.804 & hc@xmath16n & 25 - 24 + 18485.07 & unidentified & & 28199.805 & hc@xmath16n & 25 - 24 + 18494.1 & ch3sh & 18(2)-17(3 ) a+ & 28316.031 & ch@xmath9oh & 4(0,4)-3(1,2 ) e + 18499.390 & nh@xmath9 & 9(6)-9(6 ) & 28440.980 & ch@xmath4chcn & 3(0,3)-2(0,2 ) + 18513.316 & ch@xmath4chcn & 2(1,2)-1(1,1 ) f=3 - 2 & 28470.391 & hc@xmath10n & 49 - 48 + 18586.06 & unidentified & & 28532.31 & c@xmath8h & 7/2 - 5/2 f=3 - 2 + 18593.060 & hc@xmath10n & 32 - 31 & 28532.46 & c@xmath8h & 7/2 - 5/2 f=4 - 3 + 18638.616 & hc@xmath14n & 7 - 6 & 28542.284 & c@xmath8h & j=5/2 - 5/2 f=3 - 3 + 18650.308 & hcccho & 2(0,2)-1(0,1 ) & 28571.37 & c@xmath8h & 5/2 - 3/2 f=3 - 2 + 18673.312 & hnccc & 2 - 1 & 28571.53 & c@xmath8h & 5/2 - 3/2 f=2 - 1 + 18698.16 & unidentified & & 28604.737 & nh@xmath9 & 10(10)-10(10 ) + 18729.12 & unidentified & & 28903.688 & cccs & 5 - 4 + 18793.92 & unidentified & & 28905.787 & ch@xmath9oh & 15(2,13)-12(1,14 ) e + 18802.235 & h@xmath4cccccc & 7(1,7)-6(1,6 ) & 28919.931 & ch@xmath9cccn & 7(1)-6(1 ) + 18807.888 & nh@xmath4d & 3(1,3)-3(0,3 ) & 28920.209 & ch@xmath9cccn & 7(0)-6(0 ) + 18808.507 & nh@xmath9 & 8(5)-8(5 ) & 28969.954 & ch@xmath9oh & 8(2,7)-9(1,8 ) a + 18817.66 & unidentified & & 28974.781 & h@xmath15co & 3(1,2)-3(1,3 ) f=2 - 2 + 18864.65 & unidentified & & 28974.804 & h@xmath15co & 3(1,2)-3(1,3 ) f=4 - 4 + 18884.695 & nh@xmath9 & 6(2)-6(2 ) & 28974.814 & h@xmath15co & 3(1,2)-3(1,3 ) f=3 - 3 + 18907.54 & unidentified & & 28999.814 & hcccc@xmath19cn & 11 - 10 + 18918.50 & unidentified & & 29051.403 & hc@xmath10n & 50 - 49 + 18961.79 & unidentified & & 29109.644 & c@xmath18h & 23/2 j=21/2 - 19/2 f=11 - 10 f + 18965.588 & ch@xmath4chcn & 2(0,2)-1(0,1 ) f=1 - 0 & 29109.66 & c@xmath18h & 23/2 j=21/2 - 19/2 f + 18966.535 & ch@xmath4chcn & 2(0,2)-1(0,1 ) f=2 - 1 & 29109.686 & c@xmath18h & 23/2 j=21/2 - 19/2 f=10 - 9 f + 18966.616 & ch@xmath4chcn & 2(0,2)-1(0,1 ) f=3 - 2 & 29112.709 & c@xmath18h & 23/2 j=21/2 - 19/2 f=11 - 10 f + 18968.48 & unidentified & & 29112.73 & c@xmath18h & 23/2 j=21/2 - 19/2 e + 18986.20 & unidentified & & 29112.750 & c@xmath18h & 23/2 j=21/2 - 19/2 f=10 - 9 f + 19014.7204 & c@xmath8h & 5/2 - 3/2 f=2 - 1 & 29138.877 & ch@xmath4chcn & 3(1,2)-2(1,1 ) f=3 - 2 + 19015.1435 & c@xmath8h & 5/2 - 3/2 f=3 - 2 & 29139.215 & ch@xmath4chcn & 3(1,2)-2(1,1 ) f=4 - 3 , 2 - 1 + 19025.107 & c@xmath8h & 5/2 - 3/2 f=2 - 2 & 29258.834 & hcc@xmath19cccn & 11 - 10 + 19039.50 & unidentified & & 29289.159 & hc@xmath14n & 11 - 10 + 19043.0 & unidentified & & 29304.09 & c@xmath18h & 21/2 j=21/2 - 19/2 e + 19044.760 & c@xmath8h & 3/2 - 1/2 f=1 - 1 & 29310.5 & unidentified & + 19054.4762 & c@xmath8h & 3/2 - 1/2 f=2 - 1 & 29327.776 & hc@xmath16n & 26 - 25 + 19055.9468 & c@xmath8h & 3/2 - 1/2 f=1 - 0 & 29332.45 & c@xmath18h & 21/2 j=21/2 - 19/2 f + 19099.656 & c@xmath8h & 3/2 - 3/2 f=1 - 1 & 29333.3 & unidentified & + 19119.764 & c@xmath8h & j=3/2 - 3/2 f=2 - 2 & 29337.57 & hc@xmath14n & 11 - 10 v11=1 = 1c + 19174.086 & hc@xmath10n & 33 - 32 & 29342.0 & unidentified & + 19175.958 & hc@xmath16n & 17 - 16 & 29353.8 & unidentified & + 19218.465 & nh@xmath9 & 7(4)-7(4 ) & 29363.15 & hc@xmath14n & 11 - 10 v11=1 = 1d + 19243.521 & ccco & 2 - 1 & 29365.0 & unidentified & + 19262.140 & ch@xmath9cho & 1(0,1 ) - 0(0,0 ) e & 29477.704 & ccs & 2,3 - 1,2 + 19265.137 & ch@xmath9cho & 1(0,1 ) - 0(0,0 ) a++ & 29632.406 & hc@xmath10n & 51 - 50 + 19316.70 & unidentified & & 29632.413 & hc@xmath10n & 51 - 50 + 19325.20 & unidentified & & 29636.920 & ch@xmath9oh & 16(2,14)-12(1,15 ) e + 19336.10 & unidentified & & 29676.14 & cccn & 3 - 2 j=7/2 - 5/2 f=7/2 - 5/2 + 19361.50 & unidentified & & 29676.28 & cccn & 3 - 2 j=7/2 - 5/2 f=9/2 - 7/2 + 19418.661 & c - c@xmath9hd & 1(1,0)-1(0,1 ) f=1 - 1 & 29678.882 & @xmath22 so & 1(0)-0(1 ) + 19418.686 & c - c@xmath9hd & 1(1,0)-1(0,1 ) f=2 - 1 & 29694.99 & cccn & 3 - 2 j=5/2 - 3/2 f=3/2 - 1/2 + 19418.712 & c - c@xmath9hd & 1(1,0)-1(0,1 ) f=1 - 2 & 29695.14 & cccn & 3 - 2 j=5/2 - 3/2 f=7/2 - 5/2 + 19418.724 & c - c@xmath9hd & 1(1,0)-1(0,1 ) f=0 - 1 & 29806.963 & hccnc & 3 - 2 + 19418.740 & c - c@xmath9hd & 1(1,0)-1(0,1 ) f=2 - 2 & 29914.486 & nh@xmath9 & 11(11)-11(11 ) +

the simultaneous treatment of disorder and interactions in strongly correlated electron systems has always formed a knotty challenge ; this is because of the dearth of manageable analytical techniques that can deal with disorder and interactions at the same time.@xcite the strongly correlated system of interest in this work is the two - dimensional electron gas ( 2deg ) in a perpendicular magnetic field at landau level filling factor @xmath0 , whose ground state is commonly known as the _ quantum hall ferromagnet_. the quantum hall ferromagnet is the spin - polarized ground state of the 2deg at @xmath0 in which all electrons completely fill the lowest landau level with spin up polarization . such configuration minimizes the coulomb energy for fermionic systems . in general , it is a competition between kinetic and coulomb energies , which determines the ground state . in the case of the quantum hall ferromagnet having @xmath0 the kinetic energy is frozen and does not change with spin flip , thus , the ground state is ferromagnetic , even with zero zeeman splitting . typically , the zeeman splitting in the gaas heterojunctions turns out to be roughly 70 times smaller than the spacing between landau levels and an order of magnitude smaller than the coulomb energy per particle . the neutral elementary excitations are spin wave excitations , also called magnons . the spin waves can be described by the action of the spin lowering operator @xmath1 , projected to the lowest landau level , on the ferromagnetic ground state . it turns out , that the projected operator creates an exact excited eigenstate of the hamiltonian . in the regime of low momenta , the magnon s dispersion is quadratic and the coefficient of the quadratic term represents a phenomenological constant known as the _ spin stiffness_. the spin stiffness provides a measure of the free - energy increment associated with twisting the direction of the spins . a significant spin stiffness indicates that the system lies in the ferromagnetic phase , while a paramagnetic state corresponds to a vanishing spin stiffness . the spin wave dispersion at very large momenta saturates at a constant value given by the sum of the coulomb and zeeman energies . thus , at large momenta , the value corresponds to the energy of separate quasiparticle and quasihole excitations . one approach that has successfully dealt with strongly correlated electron systems is the so - called _ bosonization _ procedure . bosonization is a non - perturbative approximation scheme that essentially treats the electron - hole excitation , known as _ exciton _ , as a bosonic single - particle ; consequently , a fermionic hamiltonian can be recast into a bosonic one . in 1950 , tomonaga revealed , in a ground - breaking paper,@xcite that the application of the bosonization formalism to a one - dimensional electron gas ( 1deg ) yielded an exactly - solvable hamiltonian . the reason is that the electron and the hole propagate with nearly the same group velocity in the low - energy region . however , that is not the case in two dimensions . at a given momentum @xmath2 , the particle - hole pair excitation holds a continuous range of energies . therefore , it is less straightforward to construct a coherently propagating bosonic entity in two dimensions . the first attempt to extend the bosonization procedure for higher dimensions was done by luther @xcite and then revised by haldane.@xcite castro neto and fradkin,@xcite as well as houghton and marston,@xcite developed a bosonization technique for a fermi liquid in any number of dimensions . as regards the interacting 2deg subject to an external perpendicular magnetic field , westfahl jr . _ _ et al.__@xcite constructed a formalism that treated the elementary neutral excitations of the system , the _ magnons _ , in a bosonic framework such that the fermionic hamiltonian of the system was transmuted into a quadratic bosonic hamiltonian . the drawback is that this method is valid in the limit of weak magnetic fields , which amounts to large landau level filling factors @xmath3 . doretto _ _ et al.__@xcite extended the methodology of westfahl jr . _ _ et al.__@xcite to the case of the 2deg at @xmath0 ( i.e. for a very strong magnetic field ) . given that the system is now restricted to one landau level , the task greatly simplifies , since the landau level quantum degree of freedom can then be disregarded . projecting the original fermionic interacting hamiltonian of the system into the lowest landau level , which is completely filled ( @xmath0 ) , allows one to expand it in magnon states.@xcite it then turns out remarkably that the dispersion relation of the free magnons coincides with the result derived by kallin and halperin@xcite within the fermionic description at the _ random phase approximation _ ( rpa ) level and the quartic interacting part of the magnon hamiltonian might be related to the skyrmion - antiskyrmion neutral excitations of the hall ferromagnet.@xcite moreover , in the vicinity of the ground state , without magnon - magnon interactions , magnons behave like bosons . this allows to treat magnons approximately as bosons in the so - called single - mode approximation . @xcite here , we intend to calculate a quantum phase transition in the quantum hall ferromagnet driven by disorder , accounting for the coulomb interactions between electrons . we will use the bosonization technique allied to the usual self - consistent born - approximation for the disorder averaging procedure . before presenting the results obtained in this paper , it is worth getting acquainted with the current status of research related to the field . to begin with , green @xcite propounded that the vanishing of the renormalized spin stiffness at a threshold value of the disorder strength signifies the occurrence of a depolarization transition from the ferromagnetic phase to a paramagnetic one . his finding is based upon a previous result established by fogler and shklovskii , @xcite who proffered the same idea in the case of higher landau levels . green established this proposition in the framework of non - linear sigma model , used as an effective low - energy theory in the regime of weak disorder . the other quantity that green computed is the disorder contribution to the optical conductivity , which he found to be unmeasurably small . finally , green established that the quantization of the hall conductivity is not affected by the presence of weak disorder in the system.@xcite another work was carried out by sinova , macdonald and girvin,@xcite who established the occurrence of a phase transition from the paramagnetic state to the partially - polarized ferromagnetic one and then finally to the fully - polarized ferromagnetic one as the interaction strength increases relative to the disorder strength . they determined this result by computing the average value of the spin polarization as a function of the interaction strength relative to the disorder strength . et al.__@xcite did consider coulomb interactions within the framework of the hartree - fock approximation . moreover , the transition from the paramagnetic phase to the ferromagnetic one was found to take place when the coulomb energy scale is about twice as large as the landau - level - broadening disorder energy scale . as a final point , the authors inferred that no phase transition can take place in the strong disorder limit . the last germane paper was published by rapsch , lee and chalker.@xcite they established the occurrence of a phase transition from the ferromagnetic state to the so - called spin glass phase . this result was obtained by calculating the magnetization , the magnetic susceptibility and the spin stiffness as functions of the disorder strength . they assumed the disorder potential to be gaussian distributed and described the system in terms of a _ semiclassical _ spin model . in their model , they took into account coulomb interactions within the hartree - fock approximation but modelled them as being short - ranged . like green,@xcite rapsch _ _ et al.__@xcite computed the disorder contribution to the optical conductivity and found as well that it is undetectable . finally , they calculated the dielectric susceptibility of both the partially - polarized ferromagnetic phase and the spin glass one and they concluded that both regimes display an insulating behavior at low momenta and a metallic behavior at large momenta . let us now put our work in perspective . our objective is to establish the behavior of the renormalized spin stiffness as a function of the disorder strength in order to ascertain a potential quantum phase transition driven by disorder to a non - ferromagnetic state . indeed , if the spin stiffness vanishes for a critical value of the disorder strength , then this signals an instability in the ferromagnetic phase.@xcite on the other hand , the appearance of an imaginary component of the spin stiffness , which might be interpreted as a spin wave damping,@xcite at a certain disorder strength , might indicate the appearance of localized spin waves and a spin - glass phase transition . another important characteristic is the pauli susceptibility , which diverges at the point of the phase transition from a non - ferromagnetic to a ferromagnetic state , indicating spontaneous magnetization . we consider a fully _ quantum _ model , include a short - range weak disorder potential up to the @xmath4 order born approximation and treat the true _ long - range coulomb _ interactions up to the rpa level . the method that we employ consists of five steps . first of all , a bosonized expression of the total hamiltonian , which includes a contribution from disorder , is sought for . the dispersion relation of the free bosons corresponds to the one computed by macdonald _ _ @xcite and more explicitly by doretto _ et al._,@xcite which entails interactions between electrons up to the rpa level . the second step consists in obtaining the full green s function , and precisely its disorder self - average . in our case , because the impurities are randomly distributed throughout the system , the disorder self - average can also be taken by averaging over the impurity positions . the third stage is then to determine the self - energy of that disorder self - averaged green s function through the use of the dyson s equation . the self - energy is determined in the low - impurity density and weak disorder scattering approximations . as a result , the self - energy corresponds to a single diagram with one propagator line and two disorder potential lines . the propagator line is evaluated within two further possible approximations : the bare approximation , which consists in using the bare bosonic propagator , and the self - consistent approximation , which uses instead the full disorder self - averaged green s function . one must bear in mind that both propagators take into account interactions between electrons up to the rpa level . furthermore , the bare approximation is first taken in the long wavelength limit , which keeps the lowest order terms in momenta , and then in the general case , where all the momenta terms are taken into account . the fourth step consists in obtaining the renormalized dispersion in these approximations : bare and self - consistent approximations . the final stage is then to determine the spin stiffness in the approximations by taking the coefficient of the quadratic term in the renormalized dispersion . it is found that a naive extrapolation of the bare approximation to the regime of finite disorder strength predicts vanishing of the renormalized spin stiffness at a certain disorder strength @xmath5 , indicating a paramagnetic phase transition . a more realistic self - consistent approximation , however , predicts even faster decrease of the renormalized spin stiffness with growing disorder strength up to a certain critical value @xmath6 of the disorder . at this point , the renormalized spin stiffness drastically changes its behavior : it becomes nonanalytic , acquires an imaginary part , and the real part saturates at a certain positive value without reaching zero . such nonanalytic behavior can not be accessed by any finite number of perturbative corrections . in addition , our calculations show a strong indication that the pauli susceptibility also diverges at the same critical point @xmath6 , suggesting a phase transition , presumably to a spin glass phase . the outline of this paper is the following : in section ii we present the model and in section iii we derive the expression for the self - energy . then , we first solve the problem using the bare green s function in section iv . we present our numerical and analytical results for the self - consistent solution of the dyson equations in section v and draw our conclusions in section vi . the 2deg in the presence of both a perpendicular magnetic field ( @xmath7 ) at @xmath0 and disorder is described by the fermionic hamiltonian @xmath8 , with @xmath9 and @xmath10 here , @xmath11 and @xmath12 are , respectively , the fermionic creation and annihilation operators in coordinate space , @xmath13 denotes the effective mass of the electron , @xmath14 is the vector potential , @xmath15 stands for the effective land @xmath16-factor and @xmath17 is the bohr magneton . in addition , @xmath18 denotes the coulomb potential , with @xmath19 being the dielectric constant of the host semiconductor , and @xmath20 stands for the impurity potential , with @xmath21 being the random position of an impurity . the first step consists in obtaining a @xmath4 quantized version of the magnon hamiltonian of the system . in our model we consider only single magnon processes , which allow us to use a bosonic description . it was shown in ref . [ ] that the bosonized hamiltonian of the system in the absence of disorder is ( neglecting a constant term ) @xmath22 where @xmath23 and @xmath24 are , respectively , the bosonic creation and annihilation operators in @xmath25 space and the bosonic dispersion relation is given by @xmath26.\ ] ] here , @xmath27 stands for the coulomb energy scale ( @xmath28 being the magnetic length ) , @xmath29 denotes the modified bessel function of the first kind , and @xmath30 . it must be stressed that although the interaction between magnons is omitted from the discussion , the coulomb interaction between electrons up to rpa level is taken into account by the bosonic dispersion relation @xmath31.@xcite we now focus on the impurity part of the hamiltonian . we begin with the fermionic expression of the @xmath4 quantized impurity hamiltonian , @xmath32 here , @xmath33 denotes the fourier transformed density function @xmath34 for the impurities and @xmath35 and @xmath36 are , respectively , the fermionic creation and annihilation operators in @xmath25 space . in order to obtain the bosonic form of the above , the fourier - transformed electronic density operator must be used . it is given by @xmath37 the electronic field operators are related to the single - electron operators by @xmath38 where @xmath39 is the area of the system . substituting the above back into eq.([arshavin1 ] ) gives @xmath40 then , substituting eq . ( [ eq7 ] ) back into eq.([arshavin2 ] ) yields @xmath41 the bosonized version of the electron density operator reads @xcite @xmath42 where @xmath43 is the landau level degeneracy and @xmath44 . the disorder hamiltonian then becomes @xmath45.\end{aligned}\ ] ] the constant term @xmath46 is now omitted since the quantity of interest is the green s function . the bosonized impurity hamiltonian is then finally written as @xmath47 where @xmath48 labelling @xmath49 the full bosonized hamiltonian of the quantum hall ferromagnet in the presence of impurities is then expressed as @xmath50 let us now say a few words on the dimensions of the disorder potential . there are two sources of disorder present in the system : impurities positioned at a certain distance away from the 2deg and impurities present in the 2deg . in the case of gaas heterostructures , @xcite most of the disorder potential is spawned by the coulomb interaction between the electrons and the impurities located away from the 2deg . these impurities correspond to ionized donor atoms situated in the @xmath51-type region , which itself is detached from the 2deg by an insulating layer of thickness @xmath52 . in the present calculations , the disorder potential will be taken as an effective two - dimensional potential . having obtained the bosonized hamiltonian in the presence of impurities , one is now able to determine the expression for the self - energy . in the same spirit as ref [ ] , one first looks for the green s function , @xmath53 | 0 \rangle.\ ] ] here , @xmath54 stands for the bosonic vacuum state , which is none other than the quantum hall ferromagnet : i.e. @xmath55 . thus , one has @xmath56b^{\dagger}_{\textbf{p } ' } ( 0 ) \right ] | 0 \rangle , \ ] ] where @xmath57 is the time ordering operator . now , using eq.([arda1 ] ) , one easily finds that @xmath58 = \omega_{\textbf{p } } b_{\textbf{p } } ( t ) + \sum_{\textbf{q } } u^{e } ( \textbf{q } , \textbf{p } - \textbf{q } ) g_{\textbf{q } } b_{\textbf{p } - \textbf{q } } ( t),\ ] ] such that one obtains for the second term in eq.([arda2 ] ) @xmath59b^{\dagger}_{\textbf{p } ' } ( 0 ) \right ] | 0 \rangle = \omega_{\textbf{p } } g ( \textbf{p } ' , \textbf{p};t ) \\ & + \sum_{\textbf{q } } u^{e}(\textbf{q } , \textbf{p } - \textbf{q } ) g_{\textbf{q } } g ( \textbf{p } ' , \textbf{p } - \textbf{q};t).\end{aligned}\ ] ] hence , the equation of motion of @xmath60 is written as @xmath61 the zero - order approximation to the solution of eq.([arda2bis ] ) yields @xmath62 where @xmath63 stands for the bare bosonic green s function . we now look for the expression for @xmath63 . firstly , one needs to find the heisenberg bosonic operator in the absence of the disorder potential . starting with @xmath64 = \omega_{\textbf{p } } b_{\textbf{p } } ( t ) $ ] , one then obtains @xmath65 . therefore , for the case @xmath66 , the free green s function is @xmath67 whereas for @xmath68 , it turns out to be @xmath69 this solution is indeed identical to the electronic one . now , the cynosure is on the generic solution of the differential equation ( [ arda2bis ] ) . by coupling the latter with the boundary equation ( [ arda3 ] ) yields the integral equation @xmath70 by fourier - transforming the time in eq . to frequency and shifting @xmath71 one finds @xmath72 here , the bare green s function reads @xmath73 where @xmath74 and @xmath75 is given by eq .. in the same way as for fermions , the solution of eq.([arda5 ] ) is obtained by iteration . one gets the so - called _ born series _ : @xmath76 where @xmath77 and for @xmath78 , @xmath79 expansion of eq.([saqibismyfriend1 ] ) then yields @xmath80 due to disorder self - averaging in the limit of very large number of impurities @xmath81 , with constant density @xmath82 , the full bosonic one - particle green s function approaches its average value @xmath83 which is @xmath84 in the thermodynamic limit @xmath85 : @xmath86 moreover , one has @xmath87 substituting eqs . and eq.([firstorder ] ) into the expression for @xmath88 shows that the translational invariance is recovered after the averaging @xmath89 , where @xmath90 therefore , there is no @xmath91 order born scattering contribution to the bosonic self - energy . moreover , it is possible to show that all odd order contributions to the self - energy vanish ( see appendix a ) . this result is expressed diagrammatically in fig.[diaag9 ] . it was shown@xcite that the disorder averaged green s function can also be expressed as @xmath92 hence , the self - energy must now be computed . the low - density weak scattering approximation will be used throughout the calculations . low density means that the number of disorder atoms present in the system is taken to be much lower than the number of electrons , while the weak - scattering approximation signifies that the scattering potential induced by a given impurity atom is weak , such that only the first and second - order born scatterings are accounted for . the problem then reduces to solving the diagrammatic expression shown in fig.[diaag10 ] . the self - energy can be evaluated in two different manners : 1 ) the bare approximation that uses the bare propagator @xmath93 and 2 ) the self - consistent approximation that uses the full disorder self - averaged green s function @xmath94 . therefore , for generality we will use the propagator @xmath95 , which is going to be specified further for each particular case . this yields algebraically @xmath96 where eq.([qzdr1 ] ) was substituted in the second line . in this work , the impurity potential is assumed to be short - range , i.e. @xmath97 . an uniform potential in momentum space is attained from a delta function interaction in real space , @xmath98 ( such that @xmath99 ) . thus , this model assumes that the bosons ( and therefore the electrons ) collide directly with the impurity atoms that constitute the effective disorder potential ; in reality , most of the impurities are located away from the 2deg . thus , one has @xmath100 g ( \textbf{q } , \omega ) \left [ 2 i e^{- |\ell ( \textbf{p } - \textbf{q})|^2 / 4 } \sin \left ( \frac{(\textbf{p } - \textbf{q } ) \wedge \textbf{q}}{2 } \right ) \right ] \\ & = 4 n_{\rm imp } \left ( \frac{u}{a } \right)^2 \sum_{\textbf{q } } e^{- |\ell ( \textbf{q } - \textbf{p})|^2 / 2 } \sin ^2 \left ( \frac{\textbf{q } \wedge \textbf{p}}{2 } \right ) g ( \textbf{q } , \omega ) . \end{split}\label{fghw1}\ ] ] one then expands the argument of the exponential : @xmath101 here , @xmath102 denotes the angle between vectors @xmath25 and @xmath103 . furthermore , the summation is transmuted into an integration through the use of the formula , @xmath104 the angle @xmath105 is taken arbitrarily on the plane containing the vector @xmath25 , therefore , one is free to set @xmath106 . the sine squared term in eq.([fghw1 ] ) can be re - written as @xmath107 / 2 = [ 1 - \cos ( \ell^2 q p \sin \phi ) ] / 2 $ ] . we also assume rotation invariance of the green s function @xmath108 . hence , the self - energy is also rotation invariant and can be expressed as @xmath109,\ ] ] where @xmath110 stands for the impurity density . after a straightforward calculation ( see appendix b ) , we find @xmath111.\ ] ] rescaling the momenta by @xmath112 simplifies the self - energy to @xmath113 g ( q , \omega ) , \ ] ] where the various pre - factors , including the disorder potential strength and the impurity density , can be re - grouped into a single convenient parameter : @xmath114 which will be dubbed the _ disorder strength_. thus , @xmath115 is a dimensionless parameter that measures the disorder interaction strength relative to the coulomb interaction , @xmath116 . the above self - energy expression will be evaluated in two different ways : i ) first order corrections in @xmath115 and ii ) self- consistently . in the bare approximation the self - energy ( [ fghw4 ] ) becomes @xmath117 g^0 ( q , \omega ) .\ ] ] after substituting eq.([berbatovatmanutd1 ] ) into the above , we obtain @xmath118 making use of the identity ( for @xmath74 ) , @xmath119 we find the real and imaginary parts of the self - energy @xmath120 \pi \delta ( \omega - \omega_{q } ) . \label{saaidismybro2bisbis}\end{aligned}\ ] ] the above equations can be evaluated analytically in the long wavelength approximation , which is done in appendix c. here , one uses the complete bosonic dispersion relation given by eq .. as a result , one can only solve the imaginary self - energy numerically ; that task is not performed here . we concentrate , instead , on the real part . the renormalized energy of the bosons ( including the disorder contribution ) is obtained by looking at the poles of the full disorder self - averaged green s function in eq . , @xmath121 , such that the renormalized dispersion relation is determined from eq.([saaidismybro2 ] ) : @xmath122 + \frac{u}{4 } \epsilon_b^2 e^{-p^2 / 2 } \mathcal{p } \int^{\infty}_0 { \mathrm{d}}q \ q e^{-q^2 / 2 } \frac{i_0 ( q p ) -1}{\omega - \left [ g + \epsilon_b \left ( 1 - e^{- q^2 / 4 } i_0 \left ( \frac{q^2}{4 } \right ) \right ) \right ] } .\ ] ] the corresponding plot is illustrated on fig.[scottmurray3](a ) . one can notice that at not too large momenta ( i.e. near @xmath123 ) there exists already a substantial difference between the bare ( long wavelength ) and bare ( full @xmath2 ) approximations . approximation ( blue ) , in units of the coulomb energy @xmath124 , contrasted with the one in the bare long wavelength approximation ( red ) , both as functions of the momentum @xmath125 and at @xmath126 . ( b ) renormalized spin stiffness in the bare full @xmath2 approximation ( blue ) and in the bare long wavelength approximation ( red ) . notice that using the bare green s function @xmath127 we find a transition from a ferromagnetic to a paramagnetic phase , whereas using @xmath128 in the self - consistent approximation we find a transition into a spin glass phase ( see next section).,title="fig : " ] approximation ( blue ) , in units of the coulomb energy @xmath124 , contrasted with the one in the bare long wavelength approximation ( red ) , both as functions of the momentum @xmath125 and at @xmath126 . ( b ) renormalized spin stiffness in the bare full @xmath2 approximation ( blue ) and in the bare long wavelength approximation ( red ) . notice that using the bare green s function @xmath127 we find a transition from a ferromagnetic to a paramagnetic phase , whereas using @xmath128 in the self - consistent approximation we find a transition into a spin glass phase ( see next section).,title="fig : " ] now , the renormalized spin stiffness is sought for . for the sake of convenience , one begins by introducing the variables @xmath129 and re - writing eq.([kirilishere1 ] ) as @xmath130 } .\ ] ] one then expands the above in powers of p , @xmath131 } , \end{split}\ ] ] and one takes only the @xmath132 terms , @xmath133 } + \ldots \\ & = \frac{p^2}{4 } \left[1 - \frac{u}{4 } ( 5.72 ) \right ] + \dots \end{split}\ ] ] the renormalized spin stiffness then reads @xmath134 eq.([mainbare ] ) is the main result of this section . the above expression was derived in the bare approximation , which takes into account only the lowest order corrections in @xmath115 . such assumption is only true in the realm of weak - disorder scattering.@xcite it can be seen that the renormalized spin stiffness decreases linearly in this approximation . a naive extrapolation of this dependence to the region of finite and strong disorder strength shows that there is a certain value @xmath135 , for which the renormalized spin stiffness vanishes ( in the long wavelength approximation @xmath136 ) , see fig.[scottmurray3](b ) . green @xcite explains that a vanishing renormalized spin stiffness at a threshold disorder strength means that the 2deg at @xmath0 undergoes a quantum phase transition from a ferromagnetic state to a paramagnetic one . thus , one can infer that the quantum hall ferromagnet undergoes a disorder - driven quantum phase transition to a paramagnetic state at critical disorder strength @xmath137 . it is also interesting to remark that green established this general finding in the domain of the weak disorder limit ( though in the context of a different model ) . the results obtained in this section can not be directly compared quantitatively with those of green @xcite , sinova _ et al . _ @xcite and rapsch _ et al . _ @xcite in addition to the fact that the model used in the studies of green is different , he does not complement his proposition on the vanishing of the renormalized spin stiffness with some quantitative results . sinova _ _ et al.__@xcite use a disparate variable in the ratio of the interaction strength to the landau - level broadening disorder energy scale . finally , rapsch _ _ et al.__@xcite perform their numerical calculations on a semiclassical spin model . in the next section we evaluate the self - energy using the so - called self - consistent approximation and show that the renormalized spin stiffness drastically changes its behavior , which leads to completely different conclusions about the phase transition . the self - consistent approximation means that the self - energy is evaluated with the total disorder averaged green s function ( [ berabatamanutd4 ] ) instead of the bare one . therefore , one has ( see eq.([fghw4 ] ) ) @xmath138 \langle g_u ( q , \omega ) \rangle_{\rm imp},\ ] ] now , by referring to the computations carried out in the previous section and substituting eq.([berabatamanutd4 ] ) , one gets @xmath139 using that @xmath140 one has @xmath141 thus , one can write @xmath142 with @xmath143 using such expansion allows one to promptly get a numerical solution by iterations ( see fig.[diaag11 ] ) . the convergence of the iterative solution is rather good up to some value of the disorder concentration @xmath6 . however , when @xmath144 , we find that @xmath145 . therefore , it would be desirable to derive an analytical solution in the neighborhood of @xmath6 . for convenience , we omit the arguments of @xmath146 in our notation in the next part . in general , @xmath147 or equivalently @xmath148 introducing for simplicity @xmath149 ^ 2}\ ] ] one finds @xmath150 introducing a matrix notation @xmath151 eq . reads @xmath152 its solution is found by computing the inverse matrix to eq . and has the form @xmath153 substituting this result into @xmath154 yields @xmath155 where @xmath156 with @xmath157 and @xmath158 where @xmath159 ^ 3}.\ ] ] suppose that @xmath160 when @xmath161 . in this case @xmath162 remains finite , as well as @xmath163 . this suggests that @xmath164 when @xmath165 , since the other terms are finite . moreover , if @xmath163 is a smooth function around @xmath6 , such that @xmath166 for some @xmath167 from the neighborhood of @xmath6 , then according to eq . there holds @xmath168=\sqrt{\kappa(u_c)(u_c - u)}+{\cal o}(u_c - u).\ ] ] it follows then from eq . that @xmath169 ^ 2/\kappa(u_0)$ ] as long as @xmath170 . however , the analysis of the infinite dimensional matrix @xmath171 and its determinant is quite complicated , which forces us to use an approximate solution , where we keep only the first @xmath172 terms in the expansion , thus reducing the dimension of the matrices to @xmath173 . in the absence of zeeman splitting ( @xmath174 ) , for @xmath175 , and @xmath176 one finds , setting @xmath177 , that @xmath178=0.0551776 $ ] and @xmath179 , which yields @xmath180 in excellent agreement with the numerical solution . the approximation also allows to check the validity of eq . , which yields @xmath181 at the point @xmath176 ( here the prime stands for the partial derivative with respect to @xmath115 ) . on the other hand , the numerical solution for the two points @xmath176 and @xmath182 yields @xmath183 , which agrees reasonably well with the previous result . the main difference stems from the fact that @xmath176 is rather close to the critical point @xmath6 , where the derivative diverges , so the value @xmath184 is still rather large and , of course , computational errors and approximation with finite number of terms make the result not very precise . furthermore , it follows from eq . that @xmath185 remains finite with @xmath165 . thus , @xmath186\sigma'_n(u_0)}{\sqrt{\kappa(u_0)}}\int^{u}_{u_0 } { \mathrm{d}}v(u_c - v)^{-1/2},\end{aligned}\ ] ] which leads to @xmath187\frac{\det[b(u)]}{\det[b(u_0)]}\ ] ] after performing the integration , where @xmath188 . from this analytic solution one may observe that @xmath146 and , consequently , @xmath189 acquires an imaginary part when @xmath190 . in particular , considering @xmath191 , for the case at hand @xmath192 and @xmath193 . defining @xmath194}{\sqrt{\kappa(u)}},\ ] ] the value of @xmath195 can be evaluated without any fitting parameters directly from eqs . and , which yields @xmath196 . it follows directly from the above that the renormalized spin stiffness now obeys @xmath197 where @xmath198 . both numerical and analytic results for @xmath199 are plotted in fig.[scottmurray5 ] , which shows that the analytic solution remains in excellent agreement with the numerical one even for those values of @xmath115 , which are far from the critical point @xmath6 . the whole behavior of the renormalized spin stiffness is very similar to the one obtained by chalker _ et al._,@xcite describing a spin glass phase transition . moreover , such dependence of the renormalized spin stiffness as a square root function of a control parameter was already observed previously by shender,@xcite as well as by avgin _ _ et al.__@xcite they considered the two- and tree- dimensional @xmath200 heisenberg spin glass model in a ferromagnetic ground state due to a strong external magnetic field . they found that for a certain value of the control parameter , @xmath199 acquires an imaginary part . the real part of @xmath199 is proportional to the spin wave stiffness , whereas the imaginary part is proportional to the damping of the spin wave excitations , thus signalling localization . it was argued that when the frequency of the spin - wave excitation @xmath201 multiplied by its lifetime @xmath202 is @xmath203/{\rm im}[\rho_s(u)]<1 $ ] , then the spin waves are completely localized . as we can see from the fig.[scottmurray5 ] , the condition of localization is already satisfied for the values of the disorder strength starting from @xmath204 . the calculations presented in the appendix d contain a strong indication that the pauli susceptibility diverge at the point @xmath205 , suggesting a phase transition from a ferromagnetic ground state to a spin glass state,@xcite since the spin waves become localized . our discussion was mainly concerned with the static case @xmath175 . however , our approach allows to find @xmath206 for any given @xmath201 . the dispersion spectrum in the self - consistent approximation then satisfies @xmath207 . as functions of the disorder strength @xmath115 . ( square - analytic solution , star - numerical ) ] this paper accounts for the presence of both disorder and interactions in a 2deg at landau level filling factor @xmath0 , whose ground state constitutes the well - known quantum hall ferromagnet . the bosonization technique developed by doretto @xcite was employed in order to facilitate the treatment of both disorder and interactions in this strongly correlated system . the bosonization procedure consists in treating the spin wave ( magnon ) excitation as a boson such that the fermionic hamiltonian of the system can be approximately recast into a hamiltonian expressed in terms of bosonic operators . as a consequence , the interaction between electrons up to rpa level was incorporated within the bare propagator that represents the free boson . the intent was then to identify a disorder - driven quantum phase transition to a non - ferromagnetic state by analyzing the behavior of the renormalized spin stiffness as a function of the disorder strength , which itself corresponds to the ratio squared of the disorder energy scale to the coulomb energy one . to achieve this aim , firstly , we derived the bosonic expression for the hamiltonian of the system . in the second stage , the focus was on seeking out the disorder self - averaged green s function , which is the full bosonic green s function averaged over the impurity positions . then , by using the dyson s equation , we obtained a diagrammatic representation of the self - energy . the latter was subsequently computed within the framework of the low - density weak - scattering approximation . low density means that the number of disorder atoms present in the system is taken to be much lower than the number of electrons , while the weak - scattering approximation signifies that the scattering potential induced by a given impurity atom is weak , such that only the first and second - order born scatterings are accounted for . as a result , the self - energy corresponded to a single diagram . furthermore , the self - energy was evaluated in three different approximations : 1 ) the bare ( long wavelength ) approximation , which consists in using the bare bosonic propagator and keeping the lowest order terms in momenta , 2 ) the bare ( full @xmath2 ) approximation , which uses as well the bare bosonic propagator but with all the momenta terms kept in the calculation and , finally , 3 ) the self - consistent approximation , which uses the full disorder averaged green s function instead of the bare one in the self - energy diagram . then , the renormalized spin stiffness was determined by extracting the coefficient of the quadratic term in the dispersion relation together with the contribution from the self - energy . in the case of the bare ( long wavelength ) approximation , the spin stiffness was found to vanish linearly at the disorder strength @xmath136 . for the bare ( full @xmath2 ) scheme , the spin stiffness also vanished linearly , but at the disorder strength @xmath137 . these results suggest the occurrence of a disorder - driven quantum phase transition from the ferromagnetic phase to a paramagnetic one at the critical value @xmath137 . lastly , the self - consistent calculation revealed a completely different behavior : the real part of the renormalized spin stiffness also initially decreases with increasing the disorder strength @xmath115 , but then it saturates without reaching zero beyond a critical value @xmath6 , at which it ( and the self - energy ) acquires an imaginary component . according to the shender criterium,@xcite the spin waves become completely localized when the imaginary part of the renormalized spin stiffness becomes larger than the real part , which occurs in our system for @xmath208 ( see fig.[scottmurray5 ] ) . the physical mechanism behind a phase transition from the ferromagnetic ground state can be understood by considering electrons completely filling the lowest landau level ( @xmath0 ) in the presence of some inhomogeneous electrostatic background ( disorder ) . then , for sufficiently strong impurity potential , by adjusting the electron density to the electrostatic background , the system would gain more energy than is needed to rearrange the spin configuration . in this case the ferromagnetic state does not minimize the total energy of the system and a phase transition should take place . this quantum phase transition could be detected by calculating the behavior of the magnetic susceptibility as a function of the disorder strength . a sharp peak is anticipated at the transition point . in particular , if the energy cost for exciting a spin wave is less than the gain in the electrostatic energy , then the renormalized spin stiffness becomes negative and the system undergoes a phase transition to a paramagnetic state with zero local magnetization . on the other hand , as it was argued by rapsch _ et al . _ , @xcite in the case of a smoothly varying impurity potential , keeping nonzero local magnetization is still energetically favorable and the electrostatic energy is lowered by screening the impurity potential due to the formation of spin textures . at strong disorder such phase would correspond to a spin glass and the spin textures might be considered as the localized spin waves . thus , the character of the phase transition might depend on the nature of the disorder . the calculations performed within our model indicate that the pauli susceptibility diverges at the same critical point of the disorder strength @xmath6 , where an imaginary part of the renormalized spin stiffness appears , thus suggesting a phase transition to a spin glass phase . our approach can be extended for the case of bilayer systems in the presence of disorder . in fact , fertig and murthy @xcite have already considered such systems . thus , it would be interesting to apply our formalism to the case of a bilayer system with the total filling factor @xmath209 and compare the results . we acknowledge insightful discussions with t. giamarchi and l.cugliandolo . we also would like to thank r. l. doretto for proposing us this interesting problem . this work was partially supported by the netherlands organization for scientific research ( nwo ) . let us now evaluate the @xmath210 order diagram contribution to the self - energy . its diagrammatic representation is shown in fig . 6.algebraically , we have @xmath211 [ diaag12 ] here again , the impurity potential is short - range @xmath212 . by replacing all the functions defined previously , we find @xmath213 g ( \textbf{q } , \omega ) \\ & \qquad \times \left [ 2 i e^{- |\ell ( \textbf{q } ' - \textbf{q})|^2 / 4 } \sin \left ( \frac{(\textbf{q } ' - \textbf{q } ) \wedge \textbf{q}}{2 } \right ) \right ] g ( \textbf{q } ' , \omega ) \left [ 2 i e^{- |\ell ( \textbf{p } - \textbf{q}')|^2 / 4 } \sin \left ( \frac{(\textbf{p } - \textbf{q } ' ) \wedge \textbf{q}'}{2 } \right ) \right ] \\ & = n_{\rm imp } \sum_{\textbf{q } , \textbf{q } ' } \left ( \frac{u}{a } \right)^3 ( 2 i)^3 \underset{\textbf{term a}}{\underbrace { e^{- \textbf{q}')|^2 / 4 } e^{- |\ell ( \textbf{q } ' - \textbf{q})|^2 / 4 } } } \underset{\textbf{term b}}{\underbrace{\sin \left ( \frac{\textbf{q } \wedge \textbf{p}}{2 } \right ) \sin \left ( \frac{\textbf{p } \wedge \textbf{q}'}{2 } \right ) } } \\ & \qquad \times \underset{\textbf{term c}}{\underbrace { \sin \left ( \frac{\textbf{q } ' \wedge \textbf{q}}{2 } \right ) } } g ( \textbf{q } , \omega ) g ( \textbf{q } ' , \omega ) . \end{split}\ ] ] it is clear that terms a and b are symmetric under the interchange @xmath214 while term c is antisymmetric . thus , one has @xmath215 this result holds true in both the full bare and self - consistent approximations . as a matter of fact , due to the antisymmetric property of the wedge product within the sine term , it turns out that all odd order terms vanish . we prove here the expressions and . we begin with the expression of the self - energy given by eq.([fghw2 ] ) @xmath216 . \label{a1}\ ] ] one first deals with the polar integral , @xmath217 = \frac{1}{2 } \int^{2 \pi}_0 \frac{{\mathrm{d}}\phi } { 2 \pi } \ e^{\ell^2 q p \cos \phi } - \frac{1}{2 } \int^{2 \pi}_0 \frac{{\mathrm{d}}\phi } { 2 \pi } \ e^{\ell^2 q p \cos \phi } \cos ( \ell^2 q p \sin \phi).\ ] ] the two terms are evaluated separately . for the first term , one must note that @xcite @xmath218 such that @xmath219 the second term @xmath220= \frac{1}{2}\ ] ] substituting eqs.([ghigbigb1 ] ) and ( [ ghigbigb2 ] ) back into eq.([ghigbigb3 ] ) then yields the simpler expression , @xmath221 = \frac{1}{2 } [ i_0 ( \ell^2 q p)-1 ] .\ ] ] now , substituting the above back into eq.([a1 ] ) we find @xmath222.\ ] ] to evaluate the self - energy within the long - wavelength approximation , we must return to eq.([fghw1 ] ) . firstly , one remarks that the sine squared term in eq.([fghw1 ] ) greatly simplifies , @xmath223 ^ 2 = \frac{1}{4 } ( \ell^2 |\textbf{q } \times \textbf{p}|)^2 = \frac{1}{4 } \ell^4 |\textbf{q}|^2 |\textbf{p}|^2 \sin^2 \phi.\ ] ] then , substituting eq.([longsine ] ) into eq.([fghw1 ] ) yields : @xmath224 the polar integral then turns out to match @xcite @xmath225.\ ] ] the series expansion for eq.([intbessel ] ) gives ( @xmath226 ) @xmath227 - \left [ \frac{x^2 } { 8 } + \ldots \right ] \\ & = 1 + \frac{x^2}{8 } + \ldots \end{split}\ ] ] eq.([stefanos1 ] ) already holds a @xmath228 term and therefore a @xmath229 term is not needed in the long wavelength approximation . thus , one assumes that @xmath230 moreover , the momenta are rescaled as @xmath231 . as a result , eq.([stefanos1 ] ) simplifies to @xmath232 where we replaced @xmath233 by its definition ( see eq . ) . one can then make use of the identity @xcite @xmath119 where @xmath234 symbolizes the cauchy principal value of the integral . consequently , one has @xmath235 let us first examine the real part of the self - energy , which actually denotes the physical self - energy . it has been shown@xcite that in the long wavelength approximation the bosonic dispersion relation for @xmath25 can be written as @xmath236 the physical self - energy then becomes @xmath237 let us then work temporarily with the new quantities @xmath238 such that the self - energy is re - written as @xmath239 now , one performs a change of variable in the @xmath240 momentum : @xmath241 . one must note that @xmath242 and that the integration limits are not altered . consequently , one gets @xmath243 a further change of the integration variable is performed @xmath244 , leading to @xmath245 . \end{aligned}\ ] ] term a corresponds to the definition of the exponential integral function;@xcite @xmath246 whereas term b can be straightforwardly integrated , @xmath247 where the shift of variable @xmath248 was used in the first step . thus , the physical self - energy becomes @xmath249 . \end{split}\ ] ] the renormalized energy of the bosons is obtained by looking at the poles of the full disorder self - averaged green s function , @xmath250 consequently , in the long wavelength approximation , the renormalized dispersion relation takes the form @xmath251 . \end{split}\ ] ] it is straightforward to notice that the renormalized spin stiffness , which corresponds to the coefficient of the @xmath132 term , is given by @xmath252 we now turn to the imaginary part of the self - energy given by eq .. in the long wavelength approximation , the dirac delta function becomes @xmath253 by performing a change of variable in the @xmath240 momentum , @xmath241 and replacing eq . into eq . one gets @xmath254 finally , the scattering time , which amounts to the lifetime of the bosonic excitation , is given by @xmath255 it is clear that @xmath256 when @xmath257 , i.e. low energy quasiparticles are long - lived , with finite lifetime induced by disorder . the pauli susceptibility in case of linear response is given by the kubo formula : @xmath258 using the fourier transformation @xmath259 the susceptibility can be written as @xmath260 on the other hand , the operators @xmath261 can be written in the bosonized form@xcite @xmath262 where @xmath263 . thus , after substitution @xmath264 evaluation of the expectation value yields @xmath265 using the notation defined earlier in eq.([greendoniach ] ) . expanding the green s function @xmath266 into the born series and performing the disorder averaging one recovers the translational invariance @xmath267 . moreover , performing the fourier transformation in the time variable @xmath268 and introducing @xmath269 the susceptibility is in the self consistent approximation the function @xmath271 obeys@xcite @xmath272.\ ] ] we are interested mostly in the static susceptibility @xmath273 . thus , in particular @xmath274.\ ] ] a spherically symmetric solution satisfies @xmath275.\ ] ] let us introduce a new function @xmath276 then eq.([pdef ] ) can be rewritten as @xmath277 or explicitly @xmath278 ^ 2 } h(q,\omega).\ ] ] notice that eq.([intgequ ] ) has the same form as eq.([selfderiv ] ) but with @xmath279 instead of @xmath280 , which is known to diverge @xmath281 when @xmath161 . in the next part we will demonstrate that @xmath282 also diverges , @xmath283 when @xmath165 . we are looking for a solution in the form @xmath284 substitution of eq.([solform ] ) into eq.([intgequ ] ) yields an expression , which looks similar to the equation previously obtained ( see eq.([matreq ] ) ) , @xmath285 where the function @xmath286 was defined earlier by eq.([fndef ] ) and @xmath287 ^ 2}.\ ] ] notice that @xmath288 and @xmath289 equivalently @xmath290 where @xmath291 was defined in eq.([matr ] ) . the solution is found by computing the inverse matrix to eq . and has the form @xmath292 therefore , @xmath293 and @xmath294 if the integral is convergent . otherwise , it has to be regularized , which we wo nt consider here . this leads to @xmath295 which can be further simplified by means of some algebraic transformations , @xmath296 despite the simplifications , the above expression is difficult to evaluate analytically , as well as numerically . however , since most of the terms there involve the inverse matrix , it is reasonable to suppose that if @xmath175 it diverges with @xmath297 as @xmath298^{-1}.\ ] ] on the other hand the susceptibility is given by @xmath299 thus , considering @xmath300 , we see that the integrand is divergent at @xmath175 with @xmath165 , which is definitely not enough to infer the divergence of the integral itself , but can be considered as an indication to such possibility .

[ sec : intro ] an impressive amount of data on backward hadron production in p+c interactions has been collected over the past four decades . a literature survey reveals no less than 19 experiments which have contributed a total amount of more than 3500 data points covering wide areas in projectile momentum , lab angle and lab momentum . looking at the physics motivation and at the distribution in time of these efforts , two distinct classes of experimental approaches become evident . 15 experiments cluster in a first period during the two decades between 1970 and 1990 . all these measurements have been motivated by the nuclear part of proton - nucleus collisions , in particular by the width of the momentum distributions in the nuclear rest system which reach far beyond the narrow limits expected from nuclear binding alone . these studies have ceased in the late 1980 s with the advent of relativistic heavy ion collisions and their promise of `` new '' phenomena beyond the realm of classic nuclear physics . a second class of very recent measurements has appeared and is being pursued after the turn of the century , with publications starting about 2008 . here the motivation is totally different . it is driven by the necessity of obtaining hadronic reference data for the study of systematic effects in cosmic ray and neutrino physics , in particular concerning atmospheric and long base line experiments as well as eventual novel neutrino factories . the main aim of these studies is the comparison to and the improvement of hadronic production models models which are to be considered as multi - parameter descriptions of the non - calculable sector of the strong interaction , with very limited predictive power . this new and exclusive aim has led to the strange situation that if all recent publications contain detailed comparisons to available production models , no comparison to existing data is attempted . it remains therefore unclear how these new results compare to the wealth of already available data and whether they in fact may over - ride and replace the existing results . in this environment the studies conducted since 15 years by the na49 experiment at the cern sps have a completely different aim . here it is attempted to trace a model - independent way from the basic hadron - nucleon interaction via hadron - nucleus to nucleus - nucleus collisions . this aim needs precision data from a large variety of projectile and target combinations as well as a maximum phase space coverage . as the acceptance of the na49 detector is limited to lab angles below 45 degrees , it is indicated to use existing backward data in the sps energy range in order to extend the acceptance coverage for the asymmetric proton - nucleus interactions . this requires a careful study of the dependence on cms energy and of the reliability of the results to be used . in the course of this work it appeared useful and even mandatory to provide a survey of all available data over the full scale of interaction energies , the more so as no overview of the experimental situation is available to date . this means that the present study deals with projectile momenta from 1 to 400 gev / c , for a lab angle range from 10 to 180 degrees , and for lab momenta from 0.2 to 1.2 gev / c . [ sec : var_kin ] most available data have been obtained as a function of the lab momentum @xmath8 ( or kinetic energy @xmath9 ) at constant lab angle @xmath10 . in this publication all given yields are transformed to the double differential invariant cross section @xmath11 in this context the term `` backward '' needs a precise definition . one possibility would be to define as `` backward '' the region of lab angles @xmath12 90 degrees . the present paper uses instead a definition which refers to the cms frame with the basic variables feynman @xmath13 and transverse momentum @xmath14 , defining as `` backward '' the particle yields at @xmath15 0 . this allows a clear separation of the projectile fragmentation region at positive @xmath13 with a limited feed - over into negative @xmath13 and the target fragmentation region at negative @xmath13 with a limited feed - over into positive @xmath13 . at the same time the notion of `` kinematic limit '' in participant fragmentation is clearly brought out at @xmath13 = @xmath16 1 and the contributions from intranuclear cascading may be clearly visualized and eventually separated . and @xmath10 in the cms frame spanned by feynman @xmath13 and @xmath14 for protons and pions at two different projectile momenta , a ) protons at 158 gev / c , b ) protons at 3 gev / c , c ) pions at 158 gev / c and d ) pions at 3 gev / c , width=510 ] the correlation between the two pairs of variables is presented in fig . [ fig : kinematics ] which shows lines of constant @xmath8 and @xmath10 in the coordinate frame of @xmath13 and @xmath14 for protons and pions for the two values of projectile momentum at 158 and 3 gev / c which are representative of the typical range of interaction energies discussed in this paper . several comments are due in this context . the definition of feynman @xmath13 has been modified from the standard one , @xmath17 to @xmath18 with @xmath19 the proton mass . this takes care of baryon number conservation and regularises the kinematic borders at low interaction energies . the @xmath20 dependence in fig . [ fig : kinematics ] is small to negligible for lab angles above about 50 degrees both for pions and protons but becomes noticeable at small @xmath10 . if at sps energy the full range of lab momenta up to 1.4 gev / c and angles above 10 degrees is confined to the backward region both for protons and pions , the coverage for pions extends to positive @xmath13 at low lab angles and low beam momenta . another remark concerns the overlap between target fragmentation and nuclear cascading . for protons , at all lab angles above about 70 degrees the kinematic limit for fragmentation of a target nucleon at rest in the lab system is exceeded . for pions on the other hand this is not the case as their @xmath13 value for @xmath8 = 0 is at @xmath21 this means that over the full range of lab angles and up to large @xmath8 values the contribution from target participants mixes with the nuclear component . the separation of the two processes therefore becomes an important task , see sect . [ sec : separation ] of this paper . a last remark is due to the limits of experimental coverage . all existing experiments run out of statistics at cross section levels of about 10 @xmath22b , that is about 4 orders of magnitude below the maximum yields . as visible from the momentum ranges indicated in tables 1 and 2 , this corresponds to a typical upper momentum cut - off in the region of 1 gev / c . [ sec : exp_sit ] the backward phase space coverage in p+c interactions is surprisingly complete if compared with the forward direction and even with the available data in the elementary p+p collisions . this is apparent from the list of experiments given in tables [ tab : prot_data ] and [ tab : pion_data ] with their ranges in beam momentum , lab angle , and lab momentum . although some effort has been spent to pick up all published results , this list is not claimed to be exhaustive as some results given as `` private communication '' , in conference proceedings or unpublished internal reports might have escaped attention . .data sets for proton production in p+c and n+c collisions from seven experiments giving the ranges covered in projectile momentum , lab angle , and lab momentum , the number of measured data points and errors [ cols="^,^,^,^,^,^,^,^ " , ] for secondary protons , table [ tab : prot_data ] , the important amount of low energy n+c data by franz et al . @xcite has been added to the survey as the isospin factors for the transformation into p+c results have been studied and determined with some precision , see sect . [ sec : prot ] . for secondary pions , table [ tab : pion_data ] , the situation is somewhat complicated by the fact that two independent sets of results have been published by the harp - cdp @xcite and the harp @xcite groups , based on identical input data obtained with the same detector . an attempt to clarify this partially contradictory situation is presented in sect . [ sec : harp ] of this paper . unfortunately , no commonly agreed scale in the three basic variables @xmath10 , @xmath8 and @xmath23 of the double - differential cross sections has been defined by the different collaborations providing the data contained in tables [ tab : prot_data ] and [ tab : pion_data ] . this leads to the fact that not a single couple out of the more than 3500 data points contained in these tables may be directly compared . the application of an interpolation scheme as described in sect . [ sec : comp ] is therefore an absolute necessity . ideally the thus obtained interpolated cross sections would form an internally consistent sample of results which would be coherent within the given experimental errors . as will become apparent in the following data comparison , this assumption is surprisingly well fulfilled for the majority of the experiments . only four of the 20 quoted groups of results fall significantly out of this comparison ; those will be discussed in sect . [ sec : dev_exp ] of this paper . in this sense the overall survey of the backward proton and pion production results in a powerful constraint for the comparison with any new data sample . [ sec : comp ] as stated above the main problem in bringing the wealth of available data into a consistent picture is given by the generally disparate position in phase space and interaction energy of the different experiments . the triplet of lab variables given by the beam momentum @xmath23 , the lab momentum @xmath8 and the lab angle @xmath10 has been used for the establishment of the following interpolation scheme . in addition and of course , the statistical and systematic errors have to be taken into account in the data comparison . [ sec : errors ] the last columns of tables [ tab : prot_data ] and [ tab : pion_data ] contain some information about the statistical and systematic errors of the different experiments . the given numbers are to be regarded as mean values excluding some upward tails as they are inevitable at the limits of the covered phase space in particular for the statistical uncertainties . in some cases only rudimentary information about the systematic errors is available or the systematic and statistical errors are even combined into one quantity . in the latter cases these values are given in between the respective columns of table [ tab : pion_data ] . inspection of these approximate error levels reveals a rather broad band of uncertainties ranging from about 4% to about 20% , the latter limit being generally defined by overall normalization errors . the presence of extensive data sets well below the 10% range of both statistical and systematic errors gives however some hope that a resulting overall consistency on this level might become attainable by the extensive use of data interpolation . the term `` interpolation '' is to be regarded in this context as a smooth interconnection of the data points in any of the three phase space variables defined above . this interconnection is generally done by eyeball fits which offer , within the error limits shown above , sufficient accuracy . if the distributions in @xmath10 and interaction energy are anyway not describable by straight - forward arithmetic parametrization , the @xmath8 dependences are , as discussed in sect . [ sec : pldep ] below , in a majority of cases approximately exponential . in these cases exponential fits have been used if applicable . as additional constraint physics asks of course for smoothness and continuity in all three variables simultaneously . therefore the resulting overall data interpolation has to attempt a three - dimensional consistency . if the data interpolation helps , by the inter - correlation of data points , to reduce the local statistical fluctuations , it does of course not reduce the systematic uncertainties . it is rather on the level of systematic deviations that the consistency of different experimental results is to be judged . it will become apparent from the detailed discussion described below that the majority of the quoted experiments allows for the establishment of a surprisingly consistent overall description in all three variables . [ sec : sdep ] as the data discussed here span an extremely wide range of cms energy from close to production threshold to the upper range of fermilab energies , a suitable compression of the energy scale has been introduced in order to be able to present the results in a close - to - equidistant fashion against energy . the form chosen here is the variable @xmath24 . this choice is suggested by the considerable amount of work invested in studying the approach of hadronic cross sections to the scaling limit at high energy in the 1970 s @xcite . in fact the regge parametrization suggested a smooth dependence of the cross sections as @xmath25 , with @xmath26 = 0.250.5 depending on the choice of trajectories involved . such behaviour was indeed found experimentally . in the present study the cross sections turn out to have only a mild @xmath24 dependence for @xmath27 5 gev , a dependence which is however different for pions and protons . this dependence is strongly modified below @xmath28 2.5 gev due to threshold effects . [ sec : thdep ] a convenient and often used scale for the lab angle dependence is given by @xmath29 . this scale has the advantage of producing shapes that are again to zero order exponential . of course , continuity through @xmath10 = 180 degrees imposes an approach to 180 degrees with tangent zero . as the data samples are generally not measured at common values of @xmath10 , a fixed grid of angles has been defined based on the @xmath10 values of the harp - cdp experiment @xcite dominating the range from 25 to 112 degrees . measured values down to 10 degrees and in the higher angular range at 137 , 160 , and 180 degrees have been added . measurements not corresponding to these grid values are interpolated using the @xmath29 distributions specified below . [ sec : pldep ] all data discussed here have been transformed into invariant cross sections ( [ eq : ddcs ] ) . this facilitates the presentation in different coordinate systems and eliminates the trivial approach of the phase space element to zero with decreasing momentum . in addition , most of the invariant @xmath8 distributions are close to exponential within the measured @xmath8 range . there are notable deviations mostly at low momentum and in the lower ( higher ) range of lab angles for pions and protons , respectively , as well as in the approach to threshold . in these cases an eyeball fit has been used which can be reliably performed within the error margins indicated above . at low lab momenta physics requires a deviation from the exponential shape as the invariant cross sections must approach @xmath8 = 0 with tangent zero . this limit appears in general at @xmath30 0.2 gev / c for pions and @xmath30 0.5 gev / c for protons . the data presented here fall practically all above these momentum limits . only the harp experiment @xcite gives results at @xmath8 = 0.125 gev / c for pions where indeed a substantial deviation from the exponential shape is visible . this is shown in fig . [ fig : first_point ] where the deviation from exponential fits at this @xmath8 is given in percent for all angles and beam momenta together with similar deviations observed in p+p interactions @xcite . = 0.125 gev / c from the exponential fits for @xmath31 and @xmath32 at all angles and beam momenta , b ) deviation of @xmath31 cross sections at @xmath14 = 0.125 gev / c from exponential fits to the higher @xmath14 region in p+p interactions as a function of @xmath13,width=453 ] a number of examples of momentum distributions for protons and pions is given in the following figs . [ fig : pldist_harpcdp ] and [ fig : plabex ] which show the invariant cross sections as a function of @xmath8 and the corresponding exponential fits @xmath33 which are , whenever necessary , supplemented by hand interpolations into the non - exponential regions . a first group of distributions in the medium angular range at 45 and 97 degrees is presented in fig . [ fig : pldist_harpcdp ] for the harp - cdp data concerning protons and pions , including exponential fits . and @xmath32 as a function of @xmath8 at @xmath10 = 45 and 97 degrees . full lines : exponential fits . broken line : hand - interpolation into the non - exponential region , width=529 ] evidently the exponential shape is within errors in general a good approximation to the momentum dependence . more quantitative information is contained in the normalized residual distributions of the data points , @xmath34 where @xmath35 is the difference between data and fit and sigma the statistical error of the given data point . should the fit describe the physics and should systematic effects be negligible , the distribution of @xmath36 is expected to be gaussian with rms equal to unity . the @xmath36 distributions are given for the totality of the harp - cdp data in fig . [ fig : diff_dist ] . these distributions are well described by centred gaussians . the resulting rms values are however somewhat bigger than one signalling systematic experimental effects or a deviation of physics from the simple exponential parametrization . in view of the statistical errors of 4% to 6% given by harp - cdp ( tables [ tab : prot_data ] and [ tab : pion_data ] ) these deviations are on the level of a few percent which is below the error margin to be anyway expected from the present general data survey . further examples of @xmath8 distributions from other experiments are given in fig . [ fig : plabex ] for a selection of particle type , beam momenta and angles . for a variety of particle type , lab angle and beam momenta including exponential fits ( full lines ) and , when necessary , eyeball fits into the non - exponential regions of @xmath8 ( broken lines).,width=529 ] again the basically exponential shape of these distributions is evident . characterizing the exponential fits by their inverse slopes @xmath37 a smooth and distinct dependence on lab angle and beam momentum becomes visible as shown in fig . [ fig : inverse ] . as a function of @xmath10 for four beam momenta from 3 to 158 gev / c , a ) for protons , b ) for @xmath31 and c ) for @xmath32 . the full lines are drawn to guide the eye ] compared to the strong dependence of b on @xmath10 which ranges from 0.3 to 0.05 gev / c , the only modest dependence on @xmath23 of @xmath380.03 gev / c for beam momenta from 3 to 158 gev / c is noticeable . following the above data parametrization a generalized grid of @xmath8 values between 0.2 and 1.2 gev / c , in steps of 0.1 gev / c , may now be established . concerning the lower and upper limits of this grid , an extrapolation beyond the limits given by the experimental values has been performed in some cases . this extrapolation does not exceed the bin width of the respective data lists and is therefore defendable in view of the generally smooth , gentle and well - defined @xmath8 dependences . [ sec : ph ] in the absence of theoretical predictability in the soft sector of the strong interaction , any attempt at bringing a multitude of experimental results into a common and consistent picture has to rely on a minimal set of model - independent physics constraints . in fact a `` democratic '' averaging of eventually contradictory data sets would only add confusion instead of clarity . [ sec : ph1 ] two examples of the continuity constraint have already been mentioned above : invariant @xmath8 distributions have to approach zero momentum horizontally that is with tangent zero . the same is true for angular distributions in their approach to 180 degrees . [ sec : ph2 ] it is a matter of experimental experience in the realm of soft hadronic interactions that in general distributions in any kind of kinematic variable tend to be `` smooth '' in the sense of absence of abrupt local upwards or downwards variations . the widespread use of simple algebraic parametrizations has its origin in this fact , specifically in the absence of local maxima and minima , with the eventual exception of threshold behaviour of which some examples will become visible below . [ sec : ph3 ] charge conservation has of course to be fulfilled by any type of experimental result . this means for instance that for the interaction of a positively charged projectile ( proton ) with an isoscalar nucleus ( carbon ) the @xmath39 ratio has to be greater or equal to unity over the full phase space invoking isospin symmetry ( and of course the experience from a wide range of experimental results ) . the presence of data with @xmath40 1 therefore immediately indicates experimental problems . the inspection of @xmath39 ratios has the further advantage that a large part of the systematic uncertainties , notably the overall normalization errors , cancel in this ratio . [ sec : ph4 ] it has been shown that in proton induced nuclear collisions the yields of the secondary protons and neutrons are related by a constant factor of about 2 which is in turn related to the ratio of the basic nucleon - nucleon interaction @xcite . similarly , when rotating the projectile isospin from proton to neutron , it has been predicted that the yield ratio of secondary protons from proton and neutron projectiles @xmath41 should be constant and equal to 2.5 for light nuclei @xcite . the extensive and precise low - energy data set of franz et al . @xcite from n+c interactions has therefore been included in the present survey . these data present a welcome extension of the @xmath24 scale into the region 0.47 to 0.49 which is not covered for most of the angular range with proton projectiles . as shown below , these data fit indeed very well , after re - normalization , into the general @xmath24 dependence of secondary protons where the low energy data by frankel et al . @xcite and komarov et al . @xcite at angles between 112 and 180 degrees provide an independent control of the normalization . [ sec : ph5 ] with these constraints in mind , and having established the parametrization and interpolation of the @xmath8 distributions as discussed above , one may now proceed to the attempt at sorting the 19 available experiments into a consistent global data set . it would of course be rather surprising if all experiments would fit into this global picture within their respective error limits . in fact it turns out that this procedure establishes a very strong constraint for possible deviations , as a large majority of results is creating a perfectly consistent picture both for protons and for pions . only four of the 19 data sets can not be brought into consistency with all other experiments without gravely affecting and contradicting the above constraints . these data are not included in the following global interpolation scheme . they will be discussed separately in sect . [ sec : dev_exp ] below . [ sec : prot ] [ sec : prot_sdep ] the invariant proton cross sections are shown in fig . [ fig : oneoversq ] as a function of @xmath24 for a grid of ten lab angles between 25 and 180 degrees and constant lab momenta between 0.3 and 1.2 gev / c . the interpolated data points in each panel are identified by symbols corresponding to the different experiments . at fixed @xmath8 and @xmath10 . the interpolated data points are indicated by symbols corresponding to the respective experiments in each panel . the solid lines represent the interpolation of the data , width=268 ] at fixed @xmath8 and @xmath10 . the interpolated data points are indicated by symbols corresponding to the respective experiments in each panel . the solid lines represent the interpolation of the data , width=268 ] the solid lines are eyeball interpolations through the data points . a first remark concerning this figure concerns the smoothness and continuity of the @xmath24 dependences . the achieved overall consistency of all data is rather impressive even if single points are deviating in some areas of phase space . the salient features of the physics contained in these plots may be summarized as follows : * a strong yield suppression between 1/@xmath28 0.45 and the elastic limit at @xmath24 = 0.53 is evident . * the n+c data @xcite are well consistent with the p+c results in the overlap regions ; they define a broad maximum of the cross sections at 1/@xmath28 0.46 at medium angles and low @xmath8 . * there is a well - defined asymptotic behaviour of the cross sections for @xmath24 below about 0.2 or beam momenta above about 12 gev / c . * for the lower @xmath10 region and/or low @xmath8 the asymptotic region is approached from above . the latter point is reminiscent of the behaviour of the proton yields in p+p interactions , as shown in fig . [ fig : oneoversq_pp ] . for p+p interactions at the four lab angles a ) 25 , b ) 35 , c ) 45 and d ) 55 degrees for @xmath8 values from 0.4 to 1 gev / c . the data are interpolated from blobel @xcite and na49 @xcite . the lines are drawn to guide the eye ] another feature of fig . [ fig : oneoversq ] is the systematic droop of the cross sections from harp - cdp at their highest beam momentum of 15 gev / c or @xmath24 = 0.18 , demonstrating the discriminative power of the approach . this decrease is quantified in fig . [ fig : harp15_prot ] where the ratio @xmath43 between the measured invariant cross sections and the data interpolation is shown as a function of @xmath8 for the complete angular range from 25 to 97 degrees . here deviations of up to 50% are visible . between the interpolated invariant proton cross sections from harp - cdp @xcite and the global interpolation as a function of @xmath8 for the angular range 25 @xmath44 97 degrees ] the abruptness of this decrease would necessitate a rather violent variation of the cross sections with increasing energy including a minimum between ps and sps energies . a final clarification of this situation is given by the proton data from serpukhov @xcite which , although suffering from a different and independent problem , at least exclude such variations in the region between 17 and 67 gev / c beam momentum , see sect . [ sec : belyaev ] below . [ sec : prot_thdep ] in addition to the description of the energy dependence , the global interpolation has of course also to result in a smooth and continuous verification of the angular dependence presenting the third dimension of the present study . this constraint has to be fulfilled at any value of @xmath24 . in a first example the situation at @xmath24 = 0.05 is shown in fig . [ fig : pc_prot_theta ] . this value lies in between the fermilab @xcite and na49 @xcite data in the region of negligible @xmath20-dependence . it therefore allows for the direct comparison of the two experiments in their respective angular regions which have no overlap . = 0.05 as a function of @xmath46 combining the fermilab and na49 data for @xmath8 between 0.2 and 1.4 gev / c . the global interpolation is shown as full lines . the measured cross sections in the angular ranges from 70 to 160 degrees @xcite and from 10 to 40 degrees @xcite are given on the vertical broken lines , width=472 ] several observations are in place here : * the two experimental results connect perfectly through the gap between the na49 ( @xmath47 40 degrees ) and the fermilab ( @xmath48 70 degrees ) data . * there is at most a few percent variation of the cross sections between the angles of 160 and 180 degrees taking into account the constraint of continuity through 180 degrees discussed in sect . [ sec : comp ] above . this allows the combination of results in this angular region as it is applied in the determination of the @xmath24 dependence , fig . [ fig : oneoversq ] . * the angular distributions are smooth and close to exponential in shape . in particular , no instability in the region around 90 degrees is visible where an eventual diffractive peak from target fragmentation would appear , see also @xcite . further angular distributions at four @xmath24 values between 0.1 and 0.4 gev@xmath49 are given in fig . [ fig : costheta ] . in fact such distributions at arbitrary values of @xmath24 may be obtained from the global interpolation as it is presented in numerical form at the na49 web page @xcite . for four values of @xmath24 : a ) 0.1 , b ) 0.2 , c ) 0.3 , d ) 0.4 gev@xmath49 and for @xmath8 values between 0.4 and 1.2 gev / c . the standard grid of 10 angles , fig . [ fig : oneoversq ] , is indicated by the vertical broken lines ] evidently the angular distributions maintain their smooth and continuous shape , specifically through 90 degrees , at all interaction energies . with the approach to low beam momenta however , a progressive rounding of the shape towards higher lab angles manifests itself . [ sec : pip ] the global interpolation of the @xmath31 data is presented in this section in close analogy to the preceding section for protons . [ sec : pip_sdep ] the invariant @xmath31 cross sections are shown in fig . [ fig : sqs_pip ] as a function of @xmath24 for the standard grid of ten lab angles between 25 and 180 degrees and for constant lab momenta between 0.2 and 1.2 gev / c . the interpolated data points in each panel are identified by symbols corresponding to the different experiments . in p+c collisions as a function of @xmath24 at fixed @xmath8 and @xmath50 . the interpolated data points are indicated by symbols corresponding to the respective experiments in each panel . the solid lines represent the global data interpolation , width=257 ] in p+c collisions as a function of @xmath24 at fixed @xmath8 and @xmath50 . the interpolated data points are indicated by symbols corresponding to the respective experiments in each panel . the solid lines represent the global data interpolation , width=257 ] in p+c collisions as a function of @xmath24 at fixed @xmath8 and @xmath50 . the interpolated data points are indicated by symbols corresponding to the respective experiments in each panel . the solid lines represent the global data interpolation , width=257 ] in p+c collisions as a function of @xmath24 at fixed @xmath8 and @xmath50 . the interpolated data points are indicated by symbols corresponding to the respective experiments in each panel . the solid lines represent the global data interpolation , width=257 ] in p+c collisions as a function of @xmath24 at fixed @xmath8 and @xmath50 . the interpolated data points are indicated by symbols corresponding to the respective experiments in each panel . the solid lines represent the global data interpolation , width=257 ] in p+c collisions as a function of @xmath24 at fixed @xmath8 and @xmath50 . the interpolated data points are indicated by symbols corresponding to the respective experiments in each panel . the solid lines represent the global data interpolation , width=257 ] the solid lines represent the global interpolation by eyeball fits of both the energy and the angular dependences . again the @xmath24 dependence is in general smooth and continuous , with an impressive overall consistency of all data with only few exceptions discussed below . there are some general trends to be pointed out : * at the lowest lab momentum , the pion cross sections are practically @xmath20-independent , with variations of only 1020% in the range from 1 to 400 gev / c beam momentum . * this fact suggests @xmath31 production at low momentum transfer in the nuclear cascade . * for all lab momenta , the approach to high energies is very flat for @xmath51 0.2 or beam momenta above 12 gev / c . * the high energy cross sections are approached for all angles and beam momenta from below . there are two areas of deviation from the global interpolation which are both connected to the harp - cdp data @xcite . at their lowest angle of 25 degrees , the cross sections are systematically low by up to a factor of two below @xmath52 0.5 gev / c and @xmath24 above 0.2 . this is in contradiction to the available low energy data from other experiments also shown in fig . [ fig : sqs_pip ] . the second area concerns , as for the protons , the data at 15 gev / c beam momentum where a characteristic pattern of deviations is visible : at low angles and low @xmath8 , the data tend to overshoot the interpolation , whereas at angles above 45 degrees a progressive droop with increasing lab momentum is evident . this is quantified by the ratio @xmath43 between the harp - cdp data and the global interpolation shown in fig . [ fig : harp15_pip ] . between the interpolated invariant @xmath31 cross sections from harp - cdp @xcite and the global interpolation as a function of @xmath8 for the angular range 25 @xmath53 112 degrees ] these deviations are rather consistent with the ones found for protons . also in this case a rapid variation of the cross sections with increasing beam momentum can be excluded by the comparison with the pion data from the serpukhov experiment @xcite between 17 and 67 gev / c beam momentum , see sect . [ sec : belyaev ] below . [ sec : pip_thdep ] as already shown in sect . [ sec : prot_thdep ] for protons , the angular distributions at @xmath24 = 0.05 , in between the fermilab @xcite and na49 @xcite energies , are presented in fig . [ fig : pc_pip_theta ] . this allows the comparison of the two data sets and their connection across the gap in lab angles between 40 and 70 degrees which represent the upper and lower limit of the respective experiment . cross sections at @xmath24 = 0.05 as a function of @xmath46 combining the fermilab @xcite and na49 @xcite data for @xmath8 between 0.2 and 1.2 gev / c . the global interpolation is shown as full lines . the measured cross sections in the angular ranges from 70 to 160 degrees ( @xcite ) and from 10 to 40 degrees ( @xcite ) are given on the vertical broken lines , width=472 ] further angular distributions at four @xmath24 values between 0.1 and 0.4 gev@xmath49 are given in fig . [ fig : costheta_pip ] . cross sections as a function of @xmath46 for four values of @xmath24 : a ) 0.1 , b ) 0.2 , c ) 0.3 , d ) 0.4 gev@xmath49 and for @xmath8 values between 0.2 and 1.2 gev / c . the standard grid of 10 angles , fig . [ fig : oneoversq ] , is indicated by the vertical broken lines ] the angular distributions are characterized by a smooth , close to exponential shape . at backward angles , the @xmath8 dependence is very steep with four orders of magnitude already between @xmath8 = 0.2 and 0.8 gev / c . in forward direction this dependence is much reduced with less than one order of magnitude between @xmath8 = 0.2 and 1.2 gev / c . this is due to the prevailance of target fragmentation in this region , see sect . [ sec : separation ] for a quantitative study of this phenomenology . [ sec : pin ] this section follows closely the discussion of the @xmath31 cross sections in the preceding section . [ sec : pin_sdep ] the invariant @xmath32 cross sections are shown in fig . [ fig : sqs_pin ] as a function of @xmath24 for the standard grid of ten lab angles between 25 and 180 degrees and for constant lab momenta between 0.2 and 1.2 gev / c . the interpolated data points in each panel are identified by symbols corresponding to the different experiments . in p+c collisions as a function of @xmath24 at fixed @xmath8 and @xmath50 . the interpolated data points are indicated by symbols corresponding to the respective experiments in each panel . the solid lines represent the global data interpolation , width=268 ] in p+c collisions as a function of @xmath24 at fixed @xmath8 and @xmath50 . the interpolated data points are indicated by symbols corresponding to the respective experiments in each panel . the solid lines represent the global data interpolation , width=268 ] in p+c collisions as a function of @xmath24 at fixed @xmath8 and @xmath50 . the interpolated data points are indicated by symbols corresponding to the respective experiments in each panel . the solid lines represent the global data interpolation , width=268 ] in p+c collisions as a function of @xmath24 at fixed @xmath8 and @xmath50 . the interpolated data points are indicated by symbols corresponding to the respective experiments in each panel . the solid lines represent the global data interpolation , width=268 ] the solid lines represent the global interpolation by eyeball fits to the data , with several features which are worth noticing : * all the different data sets form a consistent ensemble without the systematic deviations visible in some regions of the proton and @xmath31 results . * the approach to large beam momenta happens from below for all @xmath8 . * the @xmath20-dependence is in general stronger than for @xmath31 , fig . [ fig : sqs_pip ] . if it is again flat up to @xmath54 0.2 at low @xmath8 , it becomes more pronounced both towards higher @xmath8 and in the approach to the production threshold at large @xmath24 indicating a marked increase of the @xmath39 ratio . * this effect has as physics origin the progressive change of the production mechanism from pion exchange at low energy to gluon or pomeron exchange at sps energy . this will be discussed in relation to the charge ratios in sect . [ sec : p2n ] . it is again interesting to compare the energy dependence to the one observed in p+p interactions as presented in fig . [ fig : oneoversq_pp_piminus ] . cross sections as a function of @xmath24 for p+p interactions at the two lab angles a ) 45 , and b ) 55 degrees for @xmath8 values from 0.2 to 1.2 gev / c . the data are interpolated from blobel @xcite and na49 @xcite . the lines are drawn to guide the eye ] although for both reactions the asymptotic high energy region is approached from below , this comparison shows a stronger @xmath20-dependence , at the same lab angle , in p+c than in p+p collisions . this is due to the component of nuclear cascading which contributes , in the given angular range , with equal strength than the target fragmentation to the total yield ( see sect . [ sec : separation ] below ) . [ sec : pin_thdep ] as for protons and @xmath31 in figs . [ fig : pc_prot_theta ] and [ fig : pc_pip_theta ] , the @xmath32 cross sections from the fermilab @xcite and na49 @xcite experiments are compared and combined as a function of @xmath46 in fig . [ fig : pc_pin_theta ] . cross sections at @xmath24= 0.05 as a function of @xmath46 combining the fermilab @xcite and na49 @xcite data for @xmath8 between 0.2 and 1.2 gev / c . the global interpolation is shown as full lines . the measured cross sections in the angular ranges from 70 to 160 degrees ( @xcite ) and from 10 to 40 degrees ( @xcite ) are given on the vertical broken lines , width=415 ] further angular distributions at four @xmath24 values between 0.1 and 0.4 gev@xmath49 are given in fig . [ fig : costheta_pin ] . cross sections as a function of @xmath46 for four values of @xmath24 : a ) 0.1 , b ) 0.2 , c ) 0.3 , d ) 0.4 gev@xmath49 and for @xmath8 values between 0.2 and 1.2 gev / c . the standard grid of 10 angles , fig . [ fig : oneoversq ] , is indicated by the vertical broken lines , width=529 ] concerning smoothness and continuity these distributions are similar to the @xmath31 data , including the large asymmetry between the forward and backward directions . the reduction of the cross sections for @xmath32 with respect to @xmath31 with increasing @xmath24 is however very apparent . this will be quantified in the following section on @xmath39 ratios . [ sec : p2n ] as already evoked in sect . [ sec : ph3 ] above , the study of @xmath39 ratios has two main advantages . firstly , in this ratio a major fraction of the experimental systematic uncertainties cancels . secondly , the ratio is constrained by very fundamental and model - independent physics arguments like charge conservation and isospin symmetry . in addition , its @xmath20-dependence is governed by the hadronic meson exchange process which leads to a power - law behaviour that will be shown to be common to a wide range of interactions . in the following argumentation the ratio between the global data interpolation for @xmath31 and @xmath32 as described in the preceding sects . [ sec : pip ] and [ sec : pin ] will be used : @xmath56 as a by - product , the fluctuation of this ratio as a function of angle and interaction energy will allow for the estimation of the local precision of the interpolation procedure . [ sec : p2n_high ] it has been established by numerous experimental results that at collision energies in the sps / fermilab range and above the hadronic interactions are characterized by the absence of charge and flavour exchange . it has also been shown that the feed - over of pions from the projectile hemisphere into the backward region of @xmath13 is sharply limited to the range of @xmath57 -0.05 , see @xcite for a detailed discussion . this range is outside the coverage in @xmath10 and @xmath8 considered in this publication . it is therefore to be expected that the backward production of pions off an isoscalar nucleus should be charge - symmetric at high energy . this is indeed verified by the results on pion production shown in the preceding sections . it is quantified in fig . [ fig : costheta_pin ] which shows the @xmath39 ratio at @xmath24 = 0.04 or 330 gev / c beam momentum for all lab angles and lab momenta treated in this publication . this number distribution has a mean value of 1.0125 with an rms deviation of 3.2% . this rms value may be seen as a first estimate of the local precision of the three - dimensional interpolation scheme at this energy which has been established independently for both pion charges . /@xmath32 ratio @xmath58 at @xmath24 = 0.04 gev@xmath49 for 25 @xmath59 162 degrees and 0.2 @xmath60 1.2 gev / c ] [ sec : p2n_endep ] with decreasing interaction energy or increasing @xmath24 the @xmath39 ratio develops a strong increase at all lab momenta and lab angles . this is shown in fig . [ fig : rpm_plab ] which gives @xmath61 as a function of @xmath24 for four lab momenta . the ratio of the global data interpolation is given in steps of 0.02 in @xmath24 . at each value of @xmath24 the number of points corresponds to the standard grid of angles available at this energy . as a function of @xmath24 for four values of @xmath8 . the dots represent the ratio of the global data interpolation for @xmath31 and @xmath32 in steps of 0.02 in @xmath24 and for the chosen grid of 10 angles . a ) @xmath8 = 0.2 , b ) @xmath8 = 0.4 , c ) @xmath8 = 0.6 , d ) @xmath8 = 0.8 gev / c ] several features of fig . [ fig : rpm_plab ] are noteworthy : * considering the wide range of lab angles , @xmath61 is at each value of @xmath24 confined to a narrow band indicating an approximative angle independence . * large @xmath61 values in excess of 5 are reached at the upper limit of the available scale in @xmath24 . * there is a systematic increase of @xmath61 with @xmath8 . [ sec : p2n_high1 ] the features pointed out above may be quantified and at the same time the local systematic fluctuations of the interpolation may be estimated by establishing the mean values @xmath62 averaged over the angular range at each @xmath24 . these mean values are well defined as shown in fig . [ fig : p2n_angdiff ] which presents the normalized distribution of the point - by - point deviations from the mean in percent , @xmath63 for four values of @xmath24 , summing the four @xmath8 values used in fig . [ fig : rpm_plab ] . from the mean over the angular range , a ) @xmath24 = 0.1 , b ) @xmath24 = 0.2 , c ) @xmath24 = 0.3 , d)@xmath24 = 0.4 . the four @xmath8 values shown in fig . [ fig : rpm_plab ] are summed up , width=283 ] these distributions are of gaussian shape with an rms which increases with @xmath24 as indicated in fig . [ fig : sigsqs ] . as a function of @xmath24 . the full line represents a hand interpolation . broken line : corresponding error margin of the mean values @xmath62 . dotted lines : corresponding errors for @xmath31 and @xmath32 separately ] the observed energy dependence of the rms deviations is due to the fact that the invariant pion cross sections decrease , after a relatively flat behaviour up to @xmath54 0.15 , progressively steeper towards the production threshold , see figs . [ fig : sqs_pip ] and [ fig : sqs_pin ] . this leads inevitably to larger variations in the corresponding energy interpolation . from the rms values given in fig . [ fig : sigsqs ] the error of @xmath62 may be derived which varies between 1% and 5% for the highest and lowest interaction energy , respectively ( broken line ) . also the corresponding error margins for the mean pion yields may be extracted as indicated by the dotted lines in fig . [ fig : sigsqs ] . from these plots it appears that the global interpolation induces fluctuations which increase from a few percent in the high @xmath20 region to about 10% in the approach to the pion threshold . [ sec : p2n_high2 ] the dependence of @xmath61 on @xmath10 is shown in fig . [ fig : angdep ] for four values of @xmath24 and four values of @xmath8 . as a function of @xmath10 for four values of @xmath24 , a ) @xmath8 = 0.2 , b ) @xmath8 = 0.4 , c ) @xmath8 = 0.6 , and d ) @xmath8 = 0.8 gev / c . the mean values @xmath62 are indicated as the horizontal broken lines in each panel , width=510 ] evidently no systematic @xmath10 dependence is visible over the complete angular range within the quoted errors . [ sec : p2n_high3 ] in the absence of angular dependence of @xmath61 as shown above , the mean values @xmath62 may now be used in order to establish a precise view of the @xmath24 dependence for different @xmath8 values . this dependence is presented in fig . [ fig : p2n_meanang ] . as a function of @xmath24 for five values of @xmath8 between 0.2 and 1.0 gev / c . the full lines are hand interpolations through the data points ] within the errors of @xmath62 extracted above , a clear @xmath8 dependence is evident superposing itself to the strong common increase of @xmath62 with @xmath24 . this increase may be parametrized up to @xmath54 0.3 by the functional from @xmath64 which is , as discussed below , typical of meson exchange processes . indeed the exponent beta varies from 2 to 1.2 for @xmath8 increasing from 0.2 to 0.8 gev / c . [ sec : p2n_epdep ] the strong increase of @xmath61 with @xmath24 merits a detailed study as it is directly connected to the basic hadronic production mechanisms in p+a interactions . the fact that the pion yields in the complete backward fragmentation region of an isoscalar nucleus remember the isospin of the projectile is clearly incompatible with charge and flavour independent exchange processes . instead a meson exchange mechanism may be invoked which has indeed been used successfully in a wide range of work at low projectile momenta , see for instance @xcite and references therein . close to the pion production threshold in the nuclear hemisphere , single excitation processes via pion exchange of the type @xmath65 only allow @xmath31 and @xmath66 production , whereas @xmath32 production needs double excitation like @xmath67 with in general an additional penalty for @xmath32 due to the isospin clebsch gordan coefficients . all meson exchange mechanisms are characterized by a strong decrease with projectile energy . this energy dependence and its interplay with processes governing the high energy sector is studied here for the first time in p+a collisions using the @xmath39 ratio . in this context it seems mandatory to first refer to the study of exclusive charge exchange reactions in elementary nucleon - nucleon collisions as the complete energy range discussed here has been covered there by a number of experiments @xcite . [ sec : p2n_epdep1 ] charge exchange processes may be cleanly isolated experimentally in nucleon - nucleon interactions by studying the following exclusive channels : * charge exchange scattering of the elastic type + & + + & * single dissociation with pion production + [ eq : single ] & + + ^++ + ( + ^+ ) & * double dissociation with pion production + [ eq : double ] & + ( + ^- ) + ( + ^+ ) & these channels are characterized by a very steep energy dependence . this is to be confronted with non - charge - exchange exclusive channels like : * elastic scattering + & + + & * single dissociation + [ eq : single0 ] & + + ( + ^+ + ^- ) & * double dissociation + [ eq : double0 ] & + ( + ^+ + ^- ) + ( + ^+ + ^- ) & which show a constant or logarithmically increasing @xmath20-dependence . charge exchange scattering has been measured by five experiments in the range of neutron beam momenta from 3 to 300 gev / c . this is exactly covering the energy range discussed in this paper . the single and double dissociation has been studied at the cern isr by two experiments @xcite extending the energy scale to @xmath20 = 3700 gev@xmath68 . the two isr experiments may be directly compared to the charge exchange measurements after appropriate re - normalization of the cross sections in the overlap region at the lowest isr energy . the resulting @xmath20-dependence at a momentum transfer @xmath69 = 0.032 gev@xmath68 is presented in fig . [ fig : chargeexch ] . at a momentum transfer @xmath69 = 0.032 gev@xmath68 . the full line represents an interpolation of the data points . the insert gives the local slope @xmath70 in the parametrization @xmath71 as a function of @xmath20,width=491 ] several features of fig . [ fig : chargeexch ] are of interest : * there is a decrease of about 4 orders of magnitude in cross section between the lowest and highest @xmath20 value . this decrease is to be compared to the constant or logarithmically increasing elastic and nucleon diffraction cross sections . the charge exchange contribution is therefore negligible compared to the inclusive baryon yields already at sps energy . * there is a steady decrease of the local slope @xmath72 with energy , from about 3.6 at 3 gev / c to about 1.1 above 80 gev / c beam momentum . * a characteristic change of slope manifests itself at around 30 gev / c beam momentum . these features have been interpreted in the 1970 s when the relevant experiments were performed , in the framework of regge theory which predicts an @xmath20-dependence of the form @xmath73 where @xmath26 is the intercept of the leading trajectory . this should in the case of one - meson exchange at low energy be given by the pion trajectory with zero intercept . the actual beta values above 3 at low @xmath20 seem to contradict however this expectation . here threshold effects may play a role which are not included in the parametrization ( [ eq : regge ] ) . with increasing energy the slopes move through the region of pion exchange with @xmath74 2 down to values of about 1.1 at high energy which could be connected to @xmath75 and a@xmath76 exchange with correspondingly higher intercepts @xmath26 in the region of 0.5 . at isr energy the ratio of @xmath77 contributions has indeed been estimated to be about 2 @xcite . anyway the simple parametrization given by ( [ eq : regge ] ) should not be expected to hold over the full energy scale . what is interesting here is rather the strong decline of the charge exchange cross sections with energy and the experimentally rather precisely determined slope variation . [ sec : p2n_epdep2 ] the single ( [ eq : single ] ) and double ( [ eq : double ] ) dissociation processes defined above are determined by the formation of @xmath35 resonances in the final states . they therefore constitute a source of direct @xmath35 production in nucleon - nucleon interactions . these channel cross sections decrease rapidly to the @xmath22barn level at sps energies . in contrast , the non - charge exchange channels like ( [ eq : single0 ] ) and ( [ eq : double0 ] ) have no @xmath20-dependence and stay on the mb level of cross sections . their final states have been shown to be governed by n@xmath78 resonances @xcite which may be excited by pomeron exchange . moreover , the p+@xmath31 combination of the p+@xmath31+@xmath32 final states has been shown to be dominated by @xmath79 @xcite . this is an indirect source of @xmath35 resonances as a decay product of n@xmath78 states which have large decay branching fractions into @xmath35+@xmath80 and @xmath35+@xmath75 . it is therefore questionable if , at sps energies and above , any direct @xmath35 production is persisting . this is an interesting question for the majority of microscopic models which produce final states by string fragmentation . in the baryonic sector , diquark fragmentation is generally invoked with a prevailing direct production of @xmath35 resonances which by isospin counting will dominate over n@xmath78 . indeed in practically all such models there is no or only negligible n@xmath78 production . as shown below , the decrease of charge exchange processes can be traced well into the non - diffractive , inelastic region of particle production . the multi - step , cascading decay of primordial n@xmath78 resonances into @xmath35 resonances and final state baryons should therefore be seriously considered , in particular also concerning the consequences for the evolution of final state energy densities with time . [ sec : p2n_epdep3 ] the very characteristic decrease of @xmath62 with increasing @xmath20 derived from the global data interpolation , fig . [ fig : p2n_meanang ] , offers a tempting possibility of comparison to the phenomenology discussed above for the elementary nucleon - nucleon sector . indeed , two components should contribute to the observed @xmath39 ratios : at high energy this ratio should approach unity due to the absence of charge and flavour exchange in this region . at low energy on the contrary it should be governed by meson exchange with its strong @xmath20-dependence . these two components may be tentatively separated by using instead of @xmath62 the quantity @xmath81 in order to extract the meson exchange contribution . this quantity is plotted in fig . [ fig : p2n_meanang1 ] as a function of @xmath20 for four @xmath8 values from 0.2 to 0.8 gev / c . as a function of @xmath20 for different values of @xmath8 , a ) @xmath8 = 0.2 , b ) @xmath8 = 0.4 , c ) @xmath8 = 0.6 , and d ) @xmath8 = 0.8 gev / c . the elastic limit is indicated by the arrows ] a very characteristic pattern emerges which resembles the @xmath20-dependence for the charge exchange in elementary interactions described above , see fig . [ fig : chargeexch ] . in general @xmath82 follows a power law dependence on @xmath20 @xmath83 with local slopes @xmath84 which are in turn a function of @xmath20 . three different regions with distinct local slopes can be identified in fig . [ fig : p2n_meanang1 ] : * a first region with large slopes is located at @xmath20 below about 6 gev@xmath68 . this region is strongly influenced by threshold effects as the threshold for inelastic production is placed at the elastic limit @xmath85 indicated in fig . [ fig : p2n_meanang1 ] . in the approach to pion threshold the @xmath39 ratio has to diverge as @xmath32 is progressively suppressed , see above . with increasing @xmath8 this suppression will of course be more pronounced . * an intermediate region between about 8 and 40 gev@xmath68 with an @xmath20 dependence decreasing with increasing @xmath8 . * a third region with flattening @xmath20-dependence above about 40 gev@xmath68 . at the lowest @xmath8 value of 0.2 gev / c corresponding to the lowest momentum transfer , the similarity to the charge exchange process in nucleon - nucleon interactions , fig . [ fig : chargeexch ] , is absolutely striking . this concerns both the detailed shape and the overall suppression factors . with increasing @xmath8 , the @xmath20 dependence is modified in a systematic way by a general reduction of slopes , with the exception of the threshold enhancement this is quantified in fig . [ fig : beta ] which shows the local slopes as a function of @xmath20 for @xmath8 values between 0.2 and 1 gev / c . of the @xmath20-dependence of @xmath82 as a function of s : a ) @xmath8 = 0.2 , b ) @xmath8 = 0.4 , c ) @xmath8 = 0.6 , d ) @xmath8 = 0.8 , and e ) @xmath8 = 1.0 gev / c . the shaded regions mark the error margins ] with the exception of the threshold region , the slopes are confined to the region between 2 and 1 typical of meson exchange processes . the dependence on @xmath8 is given in fig . [ fig : beta_pldep ] where the slopes in the three regions of @xmath20 specified above are presented . 6 gev@xmath68 ( broken line ) , 6 @xmath86 40 gev@xmath68 ( full line ) and @xmath87 40 gev@xmath68 ( dotted line ) as a function of @xmath8 between 0.2 and 1 gev / c ] this figure shows clearly the different nature of the low @xmath20 enhancement where the slopes increase strongly with @xmath8 . the two other regions , full and dotted lines , are compatible with a regge parametrization with trajectory intercepts which increase with @xmath8 . this is insofar interesting as the region of measurements regarded here covers the complete backward angular range and the corresponding interactions are by no means confined to diffractive or low momentum transfer collisions . it is shown in sect . [ sec : separation ] of this paper that in the backward hemisphere the pion yields from nuclear cascading and target fragmentation are comparable . if the nuclear component is characterized by low momentum transfer reactions @xcite the target fragmentation is manifestly inelastic and non - diffractive . it governs the total yield at all angles below about 70 degrees . in conclusion of this study of @xmath39 ratios in p+c interactions the following points should be stressed : * the global data interpolation leads to a precise and consistent description of the behaviour of the @xmath39 ratios in the full backward hemisphere , thus offering an additional tool for the discrimination of experimental deviations . * the inspection of the detailed @xmath20-dependence of the ratios opens a new window on the underlying exchange processes . * in particular the comparison to the elementary nucleon - nucleon collisions establishes a close relation between apparently disjoint sectors of the different hadronic interactions . [ sec : dev_exp ] as mentioned in sect . [ sec : ph ] four of the 19 investigated data sets are incompatible with the attempt at generating an overall consistent description of the experimental situation . these data will be shortly discussed below . [ sec : geaga ] these data have been obtained at the bevalac using beam momenta of 1.75 , 2.89 and 5.89 gev / c , spanning a lab momentum range from 0.3 to 0.9 gev / c at a lab angle of 180 degrees . the resulting cross sections trace the shape of the @xmath24 dependence rather precisely but are consistently about a factor of two below the global interpolation as shown in fig . [ fig : bevalac ] . here the full lines correspond to the global interpolation and the broken lines give the interpolation divided by a factor of two . for three values of @xmath24 ; a ) 0.427 , b ) 0.366 and c ) 0.281 gev@xmath49 . the full lines give the global data interpolation , the open circles the data from @xcite , width=585 ] as the angular bin from 160 to 180 degrees is mostly covered by data around 160162 degrees , a steep angular dependence in this region can not a priori be excluded . the smooth and gentle angular dependence of the interpolated data shown in fig . [ fig : geaga ] for the angular range from 82 to 180 degrees and for the three @xmath24 values of ref . @xcite , together with the constraint of the approach to 180 degrees with tangent zero , excludes however a drop of the cross sections by a factor of two between 160 and 180 degrees . [ sec : belyaev ] a sizeable set of data on proton @xcite and pion @xcite production has been obtained at the serpukhov accelerator spanning the range of beam momenta between 17 and 57 gev / c . this fills the gap between the ps and sps energies where no other data are available . the data cover the @xmath8 range from 0.25 to 1.2 gev / c at @xmath10 = 159 degrees . they are presented in fig . [ fig : serp_sdep ] in comparison to the global data interpolation . at @xmath10 = 159 degrees ( open circles ) in comparison to the global data interpolation at 160 degrees ( full lines),width=604 ] several features of this comparison are noteworthy : * the shape of the @xmath24 dependences complies precisely with the global interpolation . this is compatible with the absence of rapid variations of the cross sections with energy in the region between ps and sps . * there is a pronounced suppression of these data with respect to the interpolation with increasing @xmath8 reaching factors of three at the upper ranges for protons and pions . * the @xmath31 and @xmath32 data show an identical behaviour . * the proton data are tracing the interpolation up to @xmath8 = 0.4 gev / c whereas the pion data are already suppressed in this @xmath8 range . * the suppression factors are generally bigger for the pions at equal @xmath8 . * these features might be compatible with a momentum scale error . in addition to the reproduction of the shape of the @xmath24 dependence , also the @xmath39 ratio complies exactly with the one extracted from the global interpolation , fig . [ fig : serp_p2n ] , up to @xmath8 = 0.7 gev / c . above this value there is a sharp drop of @xmath61 reaching unphysical values at the upper limit of @xmath8 . this drop of about 20% has however to be compared to a drop of 300% of the invariant cross sections at this limit . from @xcite as a function of @xmath8 averaged over beam momentum between 17 and 57 gev / c . the full line gives the result of the global interpolation averaged over the same beam momentum scale , sect . [ sec : p2n],width=340 ] [ sec : harp ] these results cover a range from 3 to 12 gev / c beam momentum at @xmath10 between 25 and 117 degrees and 0.125 @xmath60 0.75 gev / c . they are thus directly comparable to the ones from @xcite which are part of the global data interpolation . their differences to this interpolation are presented in fig . [ fig : harpdiffdist ] for all beam momenta and the standard grid of @xmath8 and @xmath10 values . , panel b ) @xmath31 ] if the mean values of the differences are close to zero , their number distributions show wide spreads especially for @xmath31 . this is exemplified in fig . [ fig : pion_sdep ] where a typical comparison to the global interpolation ( full lines ) is given as a function of @xmath24 at @xmath10 = 67 degrees for four @xmath8 values . = 67 degrees and four @xmath8 values as a function of @xmath24 , ( open circles ) in comparison to the global data interpolation ( full lines ) . panel a ) @xmath31 , panel b ) @xmath32,width=585 ] a comparison of @xmath39 ratios is given in fig . [ fig : harp_p2n_5 ] as a function of @xmath8 for four values of @xmath10 at @xmath23 = 5 gev / c . from @xcite as a function of @xmath8 for four lab angles and a beam momentum of 5 gev / c . the full lines give the results of the data interpolation from @xcite , the full circles correspond to the ratios of the measured cross sections . the broken lines give the result of the global interpolation ] this figure demonstrates the importance of using , in addition to the invariant cross section proper , the particle ratios which are strongly constrained by physical arguments , see sect . [ sec : p2n ] . [ sec : shine ] these data have been obtained at a beam momentum of 31 gev / c in a @xmath10 range from 0.6 to 22.3 degrees and @xmath8 from 0.2 to 18 gev / c . if a large part of the given angular and momentum coverage falls outside the backward region regarded here , the low momentum range up to @xmath52 0.5 gev / c for all angles and the range 0.6 @xmath60 1 gev / c for angles above about 9 degrees corresponds to negative @xmath13 and can therefore be considered here . the complete @xmath24 dependence established in the preceding sections has a lower angular limit at 25 degrees corresponding to the lowest value of the standard grid of angles . this angle is close to the highest angle of @xcite at 22.3 degrees allowing for a safe interpolation . this is shown in fig . [ fig : na61 ] where the global interpolation is compared to the data of @xcite at their highest angles between 12 and 22 degrees for two @xmath8 values both for @xmath31 and for @xmath32 . at @xmath24 = 0.13 gev@xmath49 in the @xmath10 region from 12 to 35 degrees for @xmath8 = 0.4 and 0.6 full lines : global data interpolation . full dots and broken lines : data from @xcite ] as the global interpolation is limited to @xmath12 25 degrees , another way of comparison is offered by the combined na49 and fermilab results at 158/400 gev / c where the former data cover the complete angular range of @xcite . the ratio of the available high energy data to the results at 31 gev / c beam momentum is shown , as a function of @xmath10 , in fig . [ fig : na61_ratio ] . for @xmath8 = 0.4 and 0.6 gev / c . the full circles correspond to the global data interpolation , the open circles to the direct ratio @xcite/@xcite . the broken lines give the mean ratio over the complete range of the global data interpolation ] apparently this cross section ratio is within errors angle independent over the full range of the global data survey , with well defined averages below 1.05 for @xmath31 and 1.1 for @xmath32 . in contrast , the cross section ratio between na49 and ref . @xcite shows values in the region of 1.4 increasing with decreasing @xmath10 . in conclusion to this section it may be stated that the global data interpolation between 15 different experiments attempted in this paper proves to be a useful tool for the detection of deviating data sets . further details concerning the above comparisons can be found in an internal report on ref . [ sec : separation ] hadronic production in the backward direction of p+a collisions has two components : the fragmentation of the target nucleons which have been hit by the projectile proton , and the propagation of momentum transfer into the nucleus by secondary nucleon - nucleon interaction which follow , on a longer time scale , the initial excitation process . both processes are governed by the mean number of collisions @xmath89 suffered by the projectile on his trajectory through the nucleus . as only the sum of these two separate mechanisms is experimentally accessible , a minimum assumption about the fragmentation of the target nucleons is needed in order to allow the separation of the components in an otherwise model - independent fashion . this minimal assumption consists in assuming that the fragmentation process of the hit nucleons is equal to the basic nucleon - nucleon interaction , taking full account of course of isospin symmetry . in addition and only valid for the relatively small value of @xmath89 in the carbon nucleus , it will be assumed that successive collisions result in hadronization at full interaction energy of the corresponding elementary interactions . as far as the value of @xmath89 is concerned , this has been determined for pion production in some detail in @xcite using the forward and the backward region at @xmath90 -0.1 where no intra - nuclear cascading is present , see below . this determination used three independent approaches : * a monte - carlo calculation using the measured nuclear density distributions . * the relation between the inelastic cross sections of p+p and p+c interactions . * the approach to @xmath13 = -0.1 of the ratio of pion densities in p+c and p+p collisions . the two former methods have to make the assumption that the inelastic interaction cross sections are independent of the number of subsequent collisions @xmath91 . in @xcite a similar approach is used concerning the production of protons and anti - protons , again in the regions where there is no contribution from nuclear cascading as well as in the full backward hemisphere . all methods mentioned above result in a consistent estimate of @xmath89 = 1.6 in p+c collisions , with a relative systematic uncertainty of the order of a few percent . in the following argumentation a prediction of the mean pion density of target fragmentation in the backward hemisphere at @xmath92 = 17.2 gev will be used which is relying on the published pion data from na49 @xcite and the estimated mean number of collisions , @xmath89 . the invariant pion cross sections are divided by the inelastic cross section to yield the quantity @xmath93 per inelastic event which establishes isospin symmetry , and @xmath94 this prediction is transformed into the appropriate coordinates @xmath8 and @xmath10 and divided by the measured invariant p+c cross sections @xmath95 per inelastic event yielding the ratio @xmath96 this ratio is shown in fig . [ fig : rpred_plab ] as a function of @xmath8 for the lab angles 10 , 20 , 30 , 40 and 45 degrees . as a function of @xmath8 for the five angles 10 , 20 , 30 , 40 and 45 degrees ] it is evident that the ratio is close to one for the three lowest angles at all @xmath8 and for the region below 0.8 gev / c for 40 and 45 degrees . this is quantified in fig . [ fig : rpred_hist ] which gives the distribution of the ratio for the mentioned @xmath8 ranges . for the lab angles 10 , 20 , 30 degrees and 40 and 45 degrees at @xmath30 0.8 gev / c . full line : gauss fit to the distribution with a mean value at 1.01 and a relative rms of 4.3% ] the results show that indeed the measured pion cross sections correspond for lab angles up to 45 degrees precisely to the prediction from elementary collisions . this indicates that there is no contribution from intra - nuclear cascading in this region , in accordance with the results of @xcite . a drop of the ratio becomes however visible in the higher @xmath8 range at 40 and 45 degrees . this marks the onset of a nuclear component which becomes clearly visible in the ratios at larger angles shown in fig . [ fig : rpred_plab_high ] . as a function of @xmath8 for the angles of 45 , 55 , 70 , 90 , 112 , 137 and 160 the full lines are local interpolations , width=396 ] it is interesting to note that the target fragmentation governs the pion density up to the highest lab angles at low @xmath8 , with @xmath97 values of more than 50% . the ratio decreases however steadily with increasing @xmath8 and reaches zero at distinct momentum values indicating the approach to @xmath13 = -1 in the plots of fig . [ fig : kinematics ] . this kinematic effect is more clearly brought out in fig . [ fig : rpred_kin ] showing that the fraction of target fragmentation is essentially a function of @xmath13 and rather independent on lab angle . and @xmath10 for constant values of @xmath98 between 0.1 and 0.9 , b ) correlation between @xmath8 and @xmath10 for fixed values of @xmath13 between -0.1 and -1.0 and c ) @xmath98 as a function of @xmath13 ] the correlation between @xmath8 and @xmath10 for fixed values of @xmath97 shown in panel a traces rather exactly the kinematic correlation between the same variables for fixed values of @xmath13 , panel b. this allows to establish a direct dependence of @xmath97 on @xmath13 which is to first order angle - independent , panel c. the invariant densities @xmath100 per inelastic event as predicted from the fragmentation of the participant target nucleons is presented in fig . [ fig : fpred_theta ] . per inelastic event as a function of @xmath29 for fixed values of @xmath8 between 0.1 and 1.2 gev / c . the full lines represent data interpolations , width=415 ] this density may be subtracted from the pion density @xmath101 measured in p+c interactions which is within errors equal for @xmath31 and @xmath32 , see figs . [ fig : pc_pip_theta ] and [ fig : pc_pin_theta ] . the resulting invariant density @xmath102 is shown in fig . [ fig : nucl_theta ] . from intra - nuclear cascading as a function of @xmath29 for fixed values of @xmath8 between 0.1 and 1.2 gev / c . the full lines represent data interpolations , width=415 ] this subtraction procedure becomes of course uncertain in the small angle region where the nuclear component is on the few percent level and below with respect to the target fragmentation , see figs . [ fig : rpred_plab ] and [ fig : rpred_plab_high ] . the invariant angular distributions shown in figs . [ fig : fpred_theta ] and [ fig : nucl_theta ] may be converted into number distributions following : @xmath103 and @xmath104 integrating these distributions over @xmath8 , the number distributions @xmath105 and @xmath106 are obtained which are shown in fig . [ fig : dndcostheta ] together with the ratio @xmath107 as a function of @xmath108 ( full line ) , @xmath109 as a function of @xmath108 ( broken line ) , b ) the ratio @xmath110 as a function of @xmath108,width=453 ] evidently the nuclear component of pion production stays comparable to the target fragmentation in the full backward hemisphere of @xmath10 . it decreases rapidly for @xmath10 below about 60 degrees and vanishes below @xmath10 25 degrees . integration of @xmath111 over @xmath108 results in the total single pion yield from nuclear cascading @xmath112 per inelastic event . the predicted integrated yield from target fragmentation is @xmath113 with @xmath114 and @xmath115 from p+p interactions as measured by na49 , @xcite . this means that for p+c interactions the nuclear component of pion production amounts to 4.9% of the pions originating from the fragmentation of the hit target nucleons . applying isospin symmetry on the isoscalar c nucleus with @xmath116 the total pion yields are 6.45 from target fragmentation and 0.315 from nuclear cascading . making use of the kinematic relation between the coordinate pairs @xmath8 , @xmath10 and @xmath13 , @xmath14 , see fig . [ fig : kinematics]c , the double differential yields for the nuclear component as functions of @xmath13 and @xmath14 @xmath117 may be obtained where @xmath118 , ( [ eq : xf ] ) , and @xmath119 are cms quantities . the resulting pion density distributions are shown in fig . [ fig : dndxfdpt ] as a function of @xmath13 for @xmath14 values from 0.05 to 0.7 gev / c . as a function of @xmath13 for a ) 0.05 @xmath120 0.3 gev / c and b ) 0.4 @xmath120 0.7 gev / c , width=453 ] a peak at low @xmath14 and @xmath13 = -0.15 is apparent which corresponds to the location of pions with small lab momentum , see fig . [ fig : kinematics ] . with increasing @xmath14 the maximum density decreases and shifts in @xmath13 to lower values which is again in accordance with the kinematic correlation visible in fig . [ fig : kinematics ] . integration over @xmath14 results in the single differential density @xmath121 shown in fig . [ fig : dndxf ] together with the predicted density distribution @xmath122 from target fragmentation and with the ratio of the two densities . as a function of @xmath13 ( full line ) . the predicted density distribution from target fragmentation @xmath123 is shown as the broken line ; b ) ratio @xmath124 as a function of @xmath13,width=453 ] the @xmath14 integrated pion density @xmath121 shows a peak at @xmath125 -0.2 and vanishes at @xmath125 -0.08 . as shown by the density ratio with the predicted target fragmentation @xmath122 in fig . [ fig : dndxf]b , the nuclear component reaches 10% of the target fragmentation at @xmath13 = -0.15 and exceeds this contribution for @xmath15 -0.55 . the nuclear pion component extracted above is used in @xcite in conjunction with the complementary nuclear proton component to obtain the percentage of cascading protons which are accompanied by pion emission . [ sec : conclusion ] this paper presents a survey of available data concerning backward proton and pion production in minimum bias p+c interactions , including new and extensive data sets obtained at the cern ps and sps . the backward direction being defined as the complete phase space at negative feynman @xmath13 , the data cover , for projectile momenta from 1 to 400 gev / c , the ranges from 0.2 to 1.2 gev / c in lab momentum @xmath8 and from 10 to 180 degrees in lab angle @xmath10 . the paper attempts an interconnection of the different data sets by a detailed three - dimensional interpolation scheme in the variables @xmath24 , @xmath8 , and @xmath108 . this attempt allows a precise control of the internal data consistency as well as the study of the evolution of the invariant inclusive cross sections in all three variables . a literature search has provided a set of 19 different experiments with a total of more than 3500 data points . these measurements were obtained over 40 years of experimentation by collaborations employing widely different experimental techniques . in this respect it may be stated as a first positive result that the majority of the data may be combined into a surprisingly self - consistent ensemble . this global interpolation scheme results in a considerable discriminative power against the systematic deviation of particular data sets . only 4 of the 19 quoted experiments show in fact deviations which clearly mark them as systematically diverging . these experiments are inspected in detail one by one in an attempt to clearly bring out the discrepancies . in some of the cases , possible experimental error sources are pointed out . the underlying physics provides for additional constraints concerning basic quantities like charge conservation and isospin symmetry as well as the necessity of smoothness and continuity of the observed cross sections . whenever possible , contact to the complementary elementary nucleon - nucleon interactions is established . this concerns in particular the evocation of mesonic exchange processes for the description of @xmath39 ratios and the prediction of the target fragmentation from elementary interactions and its separation from the component of nuclear cascading . as far as the dependences of the invariant cross sections on the three basic variables @xmath8 , @xmath10 and @xmath24 is concerned , a well constrained phenomenology emerges . the @xmath8 dependences are exponential or close to exponential over a major part of the phase space with some exceptions mostly towards low interaction energies . this fact results in an important constraint for the data interpolation . the @xmath108 dependences are not far from exponential and smooth and continuous through all lab angles . in particular there is no indication of an instability around 90 degrees for the proton yields . the @xmath24 dependences converge , after strong variations close to production threshold , smoothly to asymptotic behaviour in the sps energy range . this region is approached from above by the protons and from below for the pions . this convergence is confirmed by the @xmath39 ratios which show , being governed by meson exchange at low @xmath92 with large values marked by the projectile isospin , a smooth decline with energy towards unity as expected from the underlying elementary exchange processes . this work was supported by the polish state committee for scientific research ( p03b00630 ) , the polish national science centre ( on the basis of decision no . dec-2011/03/b / st2/02634 ) the bulgarian national science fund ( ph-09/05 ) , the eu fp6 hrm marie curie intra - european fellowship program , the hungarian scientific research fund otka ( t68506 ) and the hungarian otka / nkth a08 - 77719 and a08 - 77815 grants . [ 20 ] y. d. bayukov et al . , phys . rev . * c20 * , ( 1979 ) 764 b. baatar et al . , arxiv:1207.6520v1 [ hep - ex ] i. m. belyaev et al . , * jinr - p1 - 90 - 551 * ( 1990 ) ( in russian ) a. bolshakova et al . , j. * c70 * ( 2010 ) 573 n. a. burgov et al . phys . * 30 * ( 1979 ) 371 , yad . fiz . * 30 * ( 1979 ) 720 y. d. bayukov et al . . phys . * 18 * ( 1974 ) 639 + y. d. bayukov et al . * 19 * ( 1974 ) 648 j. v. geaga et al . * 45 * ( 1980 ) 1993 s. frankel et al . lett . * 36 * ( 1976 ) 642 v. i. komarov et al . , phys . lett . * 69b * ( 1977 ) 37 j. franz et al . , nucl . phys . * a472 * ( 1987 ) 733 n. a. nikiforov et al . , phys . rev . * c22 * ( 1980 ) 700 c. alt et al . j. * c49 * ( 2007 ) 897 i. m. belyaev et al . , sov . * 49 * ( 1989 ) 295 n. abgrall et al . , phys . rev . * c84 * ( 2011 ) 034604 m. g. catanesi er al . , phys . rev . * c77 * ( 2008 ) 055207 n. a. burgov et al . * 32 * ( 1980 ) 219 , yad * 32 * ( 1980 ) 423 a. m. baldin et al . , * 20 * ( 1975 ) 629 d. r. f. cochran , phys * d6 * ( 1972 ) 3085 j. f. crawford , phys . rev . * c22 * ( 1980 ) 1184 j. whitmore et al . , phys . lett * 60b * ( 1976 ) 211 + t. ferbel , phys . * 29 * ( 1972 ) 448 + a. h. mueller , phys d2 * ( 1970 ) 2963 c. alt et al . j. * c45 * ( 2006 ) 343 b. d. anderson et al . lett . * 46 * ( 1981 ) 226 d. h. boal , phys . rev . * c25 * ( 1982 ) 3068 v. blobel et al . , nucl .phys . * b135 * ( 1978 ) 379 t. anticic et al . j. * c65 * ( 2010 ) 9 http://cern.ch/spshadrons v. blobel et al . , nucl . phys . * b69 * ( 1974 ) 454 g. barr et al . j. * c49 * ( 2007 ) 919 e. l. miller et al . * 26 * ( 1971 ) 984 m. n. kreisler et al . , nucl . phys . * b84 * ( 1975 ) 3 v. bhmer et al . * b110 * ( 1976 ) 205 a. babaev et al . , nucl . phys . * b110 * ( 1976 ) 189 h. r. barton et al . , phys . * 37 * ( 1976 ) 1656 h. de kerret et al . * 69b * ( 1977 ) 372 g. goggi et al . , phys . lett . * 72b * ( 1977 ) 265 g. goggi et al . , phys . lett . * 79b * ( 1978 ) 165 c. conta et al . , nucl . phys . * b175 * ( 1980 ) 97

the study of random combinatorial objects is an amazing success story , which grew from ad hoc problems and simple ideas to a flourishing field of its own , with many astonishing results , advanced tools and important applications ( see e.g. @xcite ) . since the early work of erds and collaborators , it has long been known that the `` typical '' objects often have an interesting and unexpected structure often worth exploring . some such random combinatorial objects such as various classes of random graphs are understood to a remarkable degree ( see e.g. @xcite ) , while others , such as random finite groups remain largely mysterious ( see e.g. @xcite ) . _ random integer partitions _ occupy the middle ground , with a large mix of known results and open problems ( cf . section [ final : remarks ] ) , and are a subject of this paper . _ partition theory _ is a classical subject introduced by euler nearly three hundred years ago , which has strong connections and applications to a variety of areas ranging from number theory to special functions , from group representation theory to probability and statistical physics . a _ partition _ of @xmath0 is an integer sequence @xmath1 , such that @xmath2 and @xmath3 . much of this paper is concerned with asymptotic analysis of various classes of partitions , some known and some new . let us begin by noting that already the number @xmath4 of integer partitions of @xmath0 is a notoriously difficult sequence ; it is not even known if it is equidistributed modulo @xmath5 . there are several specialized tools , however , which allow a detailed information about random partitions . first , the analytic tools by hardy ramanujan and rademacher give a precise asymptotics of @xmath4 , see e.g. @xcite . second , the _ boltzmann sampling _ , in this case invented earlier by fristedt @xcite , allows a uniform sampling of random partitions of @xmath0 with a high degree of independence between part sizes . this paper introduces a new combinatorial tool , which is best used in conjunction with these . our main emphasis will be not on enumeration of various classes of partitions , but on the limit shapes of random such partitions . it is a well known phenomenon that random combinatorial objects , e.g. random graphs , tend to have unusually interesting parameters , of interest in both theory and applications . when represented geometrically , these random objects illuminate the underlying 0/1 laws of their asymptotic behavior ; examples included various _ arctic circle _ shapes for domino tilings and perfect matchings ( see e.g. @xcite ) , and for alternating sign matrices ( see @xcite ) . note that there are many different notions of the _ limit shape _ , which differ depending on the context and the geometric representation . first , one needs to be careful in the choice of the _ scaling _ , by which combinatorial objects of different sizes are compared to each other . in this paper we use different scaling for different classes of partitions . second , there is more than one notion of convergence to the limit shape , some stronger than others . roughly speaking , they describe how close are the random objects to the limit shape . here we work with two different notions , convergence in expectation , which we refer to as the _ weak notion of the limit shape _ , and convergence in probability , which we refer to as the _ strong notion of the limit shape _ , which is implied by a stronger large deviation result as well as a local central limit theorem ( see below ) . as we mentioned earlier the main result of the paper is the computation of limit shapes for several well known classes ( _ ensembles _ ) of random partitions for which this was neither known nor even conjectured before . most our results are not obtainable by previously known techniques . we postpone the precise formulation of our results until the next subsection ; for now let us briefly survey the long history of ideas leading to this work . arguably , partition theory began with euler s work establishing equality for the number of partitions in different classes by means of algebraic manipulation of generating functions . the next crucial step was made by sylvester and his students which introduced _ partition bijections _ and used them to prove many results by euler and others . roughly speaking , a partition is represented as a _ young diagram _ and the squares are rearranged accordingly . we refer to @xcite for a detailed survey of this approach . the study of the number @xmath6 of partitions of @xmath0 with at most @xmath7 parts goes back to cayley , sylvester and macmahon . hardy and ramanujan obtained a remarkable asymptotic formula , further refined by others . the first result on limit shapes is the _ erds szekeres theorem _ , which can be formulated as follows : @xmath8 \. = \. \beta \sqrt{n}\ts \bigl(1+o(1)\bigr ) , \quad \text{where } \ \ e^{-\alpha c } + e^{-\beta c } \.= \.1 \ \ \ , \text{and } \ \ c = \frac{\pi}{\sqrt{6}}\ .. \ ] ] these results were strengthened and extended in a number of papers , see e.g. @xcite . there are two prominent types of partition classes considered in the literature , which are essentially conjugate to each other . the classes are defined by restrictions on parts @xmath9 , and those with restriction on the number of times @xmath10 part @xmath11 appear in the partition , with no restrictions relating parts to each other . for example , the set of partitions into odd numbers fits into the latter class , which corresponds to a restriction of the form @xmath12 for all @xmath11 even . sometimes this is less obvious , for example , for partitions into distinct parts , the restriction is that each part appears at most once , i.e. , @xmath13 , rather than the fact that they are distinct , i.e. , @xmath14 . there is an extensive body of research related to limit shapes of integer partitions for which the restrictions are solely placed on the multiplicities @xmath10 . in this paper we consider restrictions on part sizes , such as the _ convex partitions _ , which satisfy @xmath15 these are equinumerous with partitions into triangular parts @xmath16 , i.e. , @xmath12 whenever @xmath11 is not a triangular number , which have a well understood limit shape . the bijection between these two classes of partitions is also well known @xcite , and we demonstrate how to obtain one limit shape from the other . the idea of this paper is to use partition bijections to transfer the limit shape results from one class of partitions onto another . making this formal is rather subtle . in @xcite , the second author introduces a necessary formalism essentially allowing such transfer , with the aim of proving that there is no natural proof of the classical _ rogers ramanujan identities_. this paper continues this approach , but can be read independently of @xcite and other earlier work . @xmath17 to understand the notion of the limit shape , we start with a geometric interpretation of an integer partition . each integer partition can be visualized as a curve in the first quadrant of @xmath18 , with a height function determined by the part sizes ; see figure [ young : diagram ] . by simultaneously plotting all of the young diagrams of every integer partition of a very large fixed size @xmath0 , for each @xmath0 one observes empirically a curve which is close to most of the diagram functions , namely , @xmath19 normalizing the @xmath20- and @xmath21-axes by @xmath22 , a single curve emerges in the limit as @xmath23 . this curve is a law of large numbers for the joint distribution of part sizes in a random integer partition . our definition of a limit shape , made precise in [ diagram functions ] , is a limit _ in probability _ as @xmath23 , of appropriately scaled part sizes of a uniformly random integer partition of size @xmath0 . the main technique for finding and proving the form of the limit shape is via a probabilistic model . the joint distribution of part sizes in a random integer partition of size @xmath0 is approximated by a certain joint distribution of _ independent _ random variables , which produces a random integer partition of random size . a weak notion of a limit shape is the convergence in expectation of the joint distribution of the independent random variables . it is easy to calculate the limit shape in this setting , as the calculation simplifies to a riemann sum . we show in section [ sect prob ] that the weak notion of a limit shape agrees with the usual limit shape in the context of integer partitions . this is also known as the _ equivalence of ensembles_. we start with two pivotal results in the field of limit shapes of integer partitions : the classical limit shape , and the limit shape of partitions into distinct parts . there is a long and rather involved history involving limit shapes , which is confounded by the weak and strong notions of limit shapes ; see [ history ] for a more detailed account . we denote the limit shape of unrestricted integer partitions by @xmath24 . [ unrestricted : limit : shape ] let @xmath25 denote the limit shape of integer partitions . then : @xmath26 a similar result is true for the limit shape of integer partitions into distinct parts , which we denote by @xmath27 . let @xmath28 denote the limit shape of integer partitions into distinct parts . then : @xmath29 one can also place certain types of restrictions on the part sizes in an integer partition , and ask whether or not a limit shape exists . we assume the part sizes are restricted to a set @xmath30 , where @xmath31 for some @xmath32 and @xmath33 ( subject to other technical assumptions , see [ credit ] ) . then , analogously as before , most partitions of size @xmath0 are close to a certain curve , which satisfies @xmath34 and @xmath35 is a constant which makes the area under the curve equal to one . in this case , the scaling of the @xmath20-axis is by @xmath36 and the scaling of the @xmath21-axis is by @xmath37 . [ rb : theorem ] let @xmath38 denote the limit shape of integer partitions with part sizes restricted to @xmath39 , with @xmath31 for some @xmath32 and @xmath33 . denote by @xmath40 the derivative of @xmath21 with respect to @xmath20 . then : @xmath41 where @xmath35 is given by equation . a similar generalization holds for partitions into distinct parts from a set @xmath39 . an alternative parameterization is by andrews @xcite , who defined the restrictions on part sizes of an integer partition by a sequence @xmath42 , where the multiplicity of the parts of size@xmath43 is restricted to be strictly less than @xmath44 . for example , the set of unrestricted partitions corresponds to the sequence @xmath45 , and partitions into distinct parts corresponds to @xmath46 . when parts sizes are restricted to @xmath39 . , this corresponds to the sequence @xmath47 , @xmath48 , and @xmath49 for @xmath50 . there are , however , certain restrictions on integer partitions for which a limit shape does not exist . for example , if we consider the sequence @xmath51 , @xmath48 , and @xmath52 if @xmath53 is not a power of 2 , then there is exactly one integer partition for each @xmath0 ; namely , the base-2 representation of the number @xmath0 . for another example , let @xmath54 for @xmath55 , and @xmath56 for @xmath57 for some positive @xmath58 . this sequence corresponds to partitions of @xmath0 with largest part at most @xmath58 . for @xmath58 fixed , the number of such partitions only grows polynomially with @xmath0 , and there is no limit shape . in contrast , the case @xmath59 was handled by romik @xcite , as well as vershik and yakubovich @xcite , and the limit shape does indeed exist . there are other classes of partitions which can not be described by andrews s parameterization . in particular , for the example of convex partitions described above , it is not immediately clear whether or not a limit shape exists , or if it does , what is the proper scaling . nevertheless , we show in [ convex ] that indeed the limit shape of convex partitions exists and demonstrate how to compute it . the goal of this paper is to obtain new limit shapes for integer partitions with restrictions which do not fit into the usual framework described above . the form of the restrictions given by andrews is a subset of restrictions called _ multiplicative restrictions _ by vershik @xcite , owing to the form of the generating function as a product of terms ; it also refers to the fact that under the weak form of the limit shape , part sizes are treated as independent , and so probabilities factor according to part sizes . our first example is an application of the various geometric transformations defined in @xcite . let @xmath60 denote the set of partitions @xmath61 with consecutive parts @xmath62 for even part sizes @xmath63 , and @xmath64 for all odd part sizes @xmath63 , @xmath65 let @xmath66 denote the limit shape of @xmath60 , and let @xmath67 , and @xmath68 . then the limit shape satisfies @xmath69 let @xmath70 denote the set of partitions into distinct parts which are congruent to @xmath71 or @xmath5 modulo @xmath72 . we demonstrate in section [ lebesgue ] how to apply these transformations to obtain the limit shape of @xmath60 . moreover , once we have the limit shape , certain statistics immediately follow . [ number : of : parts : cor ] let @xmath73 denote the number of parts in a random partition of size @xmath0 in @xmath60 . we have @xmath74 where @xmath75 denotes convergence in probability . the durfee square of a partition @xmath76 is the number of @xmath77 such that @xmath78 . it is a well - known statistic which plays a fundamental role in some of the bijections exploited in this paper , and is also described as the largest square which can fit entirely inside the young diagram of an integer partition . [ durfee : square : cor ] let @xmath79 denote the size of the largest durfee square in a random partition of size @xmath0 in @xmath60 . @xmath80 denote the real valued solution to @xmath81 then we have : @xmath82 where @xmath75 denotes convergence in probability . there are , however , certain bijections which are not covered by pak s geometric transformation approach . recall the example of convex partitions , which are partitions which satisfy @xmath83 let @xmath84 denote the set of convex partitions of size @xmath0 . let @xmath85 denote the set of partitions of size @xmath0 with parts in the set @xmath86 , @xmath87 , with unrestricted multiplicities . [ andrews : convex ] we have @xmath88 for all @xmath89 . the limit shape of @xmath90 is a special case of theorem [ rb : theorem ] , using @xmath91 and @xmath92 . it is thus natural to ask if it is possible to obtain the limit shape of @xmath93 from the limit shape of @xmath90 . the answer is affirmative , as witnessed in section [ sect r ] , _ due to the form of the bijection_. in fact , corollary [ convex : corollary ] below is a special case of a much larger class of partition bijections which map limit shapes continuously , which we describe in [ transfer : theorems ] . [ convex : corollary ] let @xmath94 denote the limit shape of @xmath90 , and let @xmath95 denote the the limit shape of @xmath93 . then we have @xmath96 where @xmath97 is the compositional inverse of @xmath98 . romik s unpublished manuscript @xcite contains an example which uses a single geometric transformation to obtain the limit shape of the set of partitions whose parts differ by at least 2 . it is a specialization of pak s geometric transformations , and requires a further refinement , in that it maps partitions of size @xmath0 with exactly @xmath58 parts . let @xmath99 denote the set of partitions whose parts differ by at least 2 , and denote its limit shape by @xmath100 . for @xmath101 , let @xmath102 , @xmath103 , and @xmath104 , where @xmath105 , and @xmath35 is a normalizing constant . then , the limit shape satisfies @xmath106 the rest of the paper is organized as follows . we start with three motivating examples : partitions with no consecutive parts , self - conjugate partitions , and convex partitions . then we make precise all of the necessary notions of limit shapes , bijections , and asymptotic stability in section [ sect notation ] , and state the new limit shape theorems in section [ sect special cases ] . we then present several examples demonstrating the application of the theorems . in section [ sect r ] , we present a generalization of convex partitions . section [ sect odd ] presents various applications of the theorems to euler s classical odd - distinct bijection , including a generalization due to stanton . sections [ lebesgue ] and [ diff d ] demonstrate step - by - step how to apply pak s geometric transformations in two important cases . section [ sect further ] contains examples where the number of summands is restricted . section [ sect prob ] contains the formal probabilistic setting and proofs of the main lemmas , including a large deviation principle . section [ sect main ] contains the proof of the main results . finally , section [ sect remarks ] contains historical remarks , and section [ final : remarks ] is a collection of final remarks . we first consider romik s example , which demonstrates the utility of considering bijections between classes of partitions . [ romik : theorem ] let @xmath107 denote the set of partitions such that no parts differ by exactly 1 , and there are no parts of size 1 . let @xmath108 , and let @xmath109 denote the limit shape of @xmath107 . then we have @xmath110 the _ existence _ of the limit shape of @xmath107 in theorem [ romik : theorem ] follows by considering the set of partitions which are conjugate to those in @xmath107 . let @xmath111 denote the set of partitions of size @xmath0 such that no parts differ by exactly 1 , and there are no parts of size 1 . let @xmath112 denote the set of partitions of size @xmath0 such that no part has multiplicity 1 . the conjugation map gives a bijection between partitions @xmath113 and @xmath114 for each @xmath89 . the limit shape of @xmath99 , the set of partitions such that no part has multiplicity 1 , has a restriction which is multiplicative , and its form follows in a straightforward manner from theorem [ special unrestricted ] . let @xmath99 denote the set of partitions such that no part has multiplicity 1 . let @xmath108 . then : @xmath115 as was noted by romik @xcite , even though the limit shapes satisfy @xmath116 , @xmath117 , it is not straightforward to compute the explicit form of @xmath109 using algebraic techniques , even with a computer algebra system . a general approach of finding limit shapes via a class of continuous transformations was developed in @xcite , where many well - known partition bijections are presented as certain geometric transformations acting on young diagrams , which also act continuously on the corresponding limit shapes . an example where the limit shape is apparent in a weak sense , but lacks a rigorous argument , is the set of _ self - conjugate partitions _ @xmath118 , consisting of all partitions @xmath76 such that @xmath119 implies also @xmath120 . [ sylvester ] let @xmath118 denote the set of partitions which are self - conjugate . the limit shape of @xmath118 satisfies equation . one can reason heuristically that the limit shape of @xmath118 should coincide with the limit shape of unrestricted integer partitions , however , we are unaware of any simple , rigorous methods to do so . the first difficulty is that the number of self - conjugate partitions of size @xmath0 is of an exponentially smaller proportion than the number of unrestricted partitions of size @xmath0 . the second difficulty is that we lose the asymptotic independence between part sizes . there are several different ways to obtain a rigorous proof of theorem [ sylvester ] , starting with the following bijection . let @xmath111 denote the number of partitions of @xmath0 into odd , distinct parts . let @xmath121 denote the number of self - conjugate partitions of @xmath0 . we have @xmath122 for all @xmath89 . one construction of the bijection is to break the principle hooks ( see ( * ? ? ? * figure 13 ) ) and stretch them out . when the partition @xmath76 is self - conjugate , each of the principle hooks must necessarily consist of an odd number of squares , and each principle hook consists of at least two squares more than the previous principle hook . in this instance , we have an explicit formula for the map which sends partitions @xmath119 to partitions @xmath61 into distinct , odd parts , namely , @xmath123 where @xmath124 is the size of the largest durfee square in the partition @xmath76 . a formal procedure to obtain theorem [ sylvester ] from this bijection is as follows , see ( * ? ? ? * section 7 ) . for the remainder of this section , @xmath125 . let @xmath126 denote the limit shape of partitions into odd , distinct parts , which is given by theorem [ rb : theorem ] using @xmath127 and @xmath128 . we shall need the explicit form of the limit shape in the calculations that follow , which is given by @xmath129 recall also that the limit shape of unrestricted integer partitions is given by the explicit formula @xmath130 denote by @xmath131 the limit shape of self conjugate partitions , which exists by ( * ? ? ? 7.1 ) . after an appropriate scaling by @xmath22 , by equation we have @xmath132 by rearranging and solving for @xmath133 , we obtain @xmath134 for @xmath135 . the value @xmath136 is precisely the value for which @xmath137 , @xmath138 , which cuts the limit shape in half via the line @xmath139 . thus , for @xmath140 , we can invoke the symmetry of self conjugate partitions about the line @xmath139 and conclude that @xmath141 for all @xmath138 . this solution is uncharacteristically simple , and most bijections are not as obliging as to allow for elementary algebraic manipulations of their part sizes . fortunately , this bijection can also be defined by a natural geometric bijection , and we next demonstrate each step in the process of mapping the limit shape from one set of partitions to another . figure [ parallel : mapping : young ] demonstrates how a geometric bijection from @xcite acts on the young diagrams , and figure [ parallel : mapping : limit ] demonstrates how a geometric bijection acts on the limit shapes . 0.3 @xmath142 \mbox{\begin{tabular}{c}\includegraphics[scale=0.282]{phi1.pdf } \\ \mbox{\large $ \phi(t)$ } \end{tabular } } \ar[dr ] & \\ \ar[d]\mbox{\begin{tabular}{c}\includegraphics[scale=0.282]{phi2b.pdf}\\\mbox{\large $ p\phi(t)$ } \end{tabular } } & & \mbox{\begin{tabular}{c}\includegraphics[scale=0.282]{phi2a.pdf}\\\mbox{\large $ q\phi(t)$}\end{tabular } } \ar[d ] \\ \ar[ddr]\mbox{\begin{tabular}{c } \mbox{\large $ p\phi(t)-t$ } \\ \includegraphics[scale=0.282]{phi3b.pdf}\\ \mbox{\large $ \phi_1(t)$}\end{tabular } } & & \mbox{\begin{tabular}{c } \mbox{\large $ \left((q\phi(t))^{\langle -1\rangle } - t\right)^{\langle -1\rangle}$ } \\ \includegraphics[scale=0.282]{phi3a.pdf } \\ \mbox{\large $ \phi_2(t)$ } \end{tabular } } \ar[ddl ] \\ & \mbox{\large $ \phi_1(t ) + \phi_2(t)$ } & \\ & \mbox{$\begin{array}{c } \includegraphics[scale=0.282]{phi4.pdf } \\ \mbox{\large $ \frac{1}{\sqrt{2}}\psi\left(\frac{t}{\sqrt{2}}\right)$ } \end{array}$ } & } } \ ] ] .9 in 0.3 @xmath143 \mbox{\begin{tabular}{c}\ydiagram[*(red ) ] { 1 + 0,2 + 0,2 + 0,2 + 0,2 + 1,1 + 5,0 + 7 } * [ * ( blue)]{1,2,2,2,2,2,1 } \\ \\ \mbox{\huge cut } \end{tabular } } \ar[dr ] & \\ \ydiagram[*(blue)]{1,2,2,2,2,1 } \ar[ddd]|-{\mbox{\huge shift } } & & \ydiagram[*(red)]{2 + 1,1 + 5,7 } \ar[ddd]|-{\mbox{\huge shift } } \\ \\ \\ \ydiagram[*(blue)]{1,1,2,2,2,2 } \ar[ddddd ] & & \ \ \ \ \ \ \ \ydiagram[*(red)]{1,5,7}\ar[dddd]|-{\mbox{\huge transpose } } \hskip .85 in \\ \\ \\ \\ & & \ydiagram[*(red)]{1,1,2,2,2,2,3 } \ar@*{[thicker]}[dll ] \\ \mbox{\begin{tabular}{cc } \ydiagram[*(red ) ] { 0 + 1,0 + 1,0 + 1,0 + 1,0 + 2,0 + 2,0 + 2,1 + 1,1 + 1,2 + 0,2 + 0,2 + 0,2 + 1 } * [ * ( blue)]{0,0,0,0,0,0,0,1,1,2,2,2,2 } & \mbox{\huge add } \end{tabular } } & & } } \ ] ] we summarize the steps of figure [ parallel : mapping : limit ] by the following set of mappings : @xmath144 the first map splits the curve into two curves of the same shape but scaled to half the original area . the next map shifts the coordinates up so that the functions evaluated at the point @xmath145 are equal to @xmath146 . the third map reflects the first coordinate about the line @xmath139 , so that the curves now lie in complementary regions in the first quadrant . finally , the last map pastes together the functions at @xmath147 the mappings defined above are presented in a way which emphasizes how the height functions are transformed with respect to the axes , whereas they are actually transforming regions in the plane . from a technical point of view , the mappings should be considered as transformations on the plane , as they are presented in @xcite , but for the purpose of finding explicit formulas for limit shapes , one can very often more simply track how the mappings transform the height functions and their inverses directly . our last example of this section exemplifies another class of bijections , those which can be realized as a linear transformation of multiplicities of part sizes . consider the set @xmath148 of all _ convex _ partitions , i.e. , partitions whose parts @xmath149 also satisfy the convexity condition : @xmath150 . this set of partitions does not exclude any particular set of sizes , nor does there appear to be any simple , a priori transformation , e.g. , conjugation , to such a set . nevertheless , we find the limit shape of convex partitions . [ conj convex ] let @xmath95 denote the limit shape of convex partitions @xmath148 . then : @xmath151 computing this limit shape requires several steps . first , recall andrews s theorem [ andrews : convex ] , that there is a bijection between partitions into parts of sizes @xmath86 , @xmath152 and convex partitions . simply knowing a bijection exists is not sufficient , however , and it is in fact the particular form of this bijection which allows us to find the limit shape . recall @xmath90 denotes the set of partitions into parts with sizes in @xmath39 , with @xmath153 . let @xmath154 denote the number of parts of size @xmath155 in a partition @xmath156 , @xmath152 . the following map @xmath157 was defined in @xcite . @xmath158 map @xmath159 is a bijection between @xmath160 and @xmath161 for each @xmath89 . each @xmath162 , with @xmath163 parts of size @xmath164 , corresponds to a @xmath165 with part sizes given by @xmath166 we can immediately write the weak form of the limit shape by appealing to riemann sums . in this case , the number of parts of size @xmath11 , @xmath48 , is generated from independent geometric random variables @xmath167 with parameter @xmath168 , where @xmath169 . the constant @xmath35 is given by equation with @xmath170 , @xmath92 , and @xmath171 . in this case , however , we set @xmath172 if @xmath173 . it will also be apparent in the calculation below that we need two different , but complementary , scaling functions . we have : @xmath174 & \ , = \ , \frac{\b}{n}\ , \sum_{j \ts\geq\ts \b \ts x } \. ( j - x\,\b+1)\ , \e\left[z_{\binom{j+1}{2}}\right ] \\ & \ , = \ , \frac{\b}{n}\ , \sum_{j \ts\geq\ts \b \ts x } \ , \bigl(j - x\,\b+1\bigr ) \ , \frac{\exp\left(-c\ , \frac{\binom{j+1}{2}}{\a}\right)}{1-\exp\left(-c\ , \frac{\binom{j+1}{2}}{\a}\right)}\ .. \end{aligned}\ ] ] we let @xmath175 satisfy @xmath176 . hence , @xmath177 , @xmath178 , and @xmath179 the final limit is valid if and only if @xmath180 , and @xmath181 , thus we obtain scaling factors @xmath171 and @xmath182 . there is one final step , however , since the calculation above only implies the formula given in equation satisfies _ the weak notion of a limit shape_. we need some form of concentration result , i.e. , large deviations , to claim that the above formula is indeed the limit shape for the set of partitions . as we shall see , however , the same concentration results valid for the limit shape of @xmath183 also apply immediately to the image @xmath148 , see [ concentration ] . the bijection @xmath159 is a special case of a more general example discussed in section [ sect r ] . in addition , this example is typical of a general method to find limit shapes described in [ transfer : theorems ] . we denote the set of positive real numbers by @xmath184 . the set of all probability measures absolutely continuous with respect to lebesgue measure on the positive real line is denoted by @xmath185 . let @xmath186 denote the set of real valued operators on the banach space of real valued , countable sequences . we think of transformations in @xmath186 as infinite dimensional , real valued matrices . the expression @xmath187 means @xmath188 we denote the positive part of @xmath20 by @xmath189 the ceiling function , or smallest integer larger or equal to @xmath20 , is denoted by @xmath190 let @xmath191 denotes the _ gamma function _ , and @xmath192 the _ riemann zeta function_. define the family of constants @xmath193^{r/(1+r ) } \quad \text{for all } \ \. a \geq 2 , \ r \geq 1 , \ b>0\ts.\ ] ] we also define @xmath194 the letter @xmath35 is not restricted to a single value , but will be defined locally . given a sequence @xmath195 , @xmath196 , we denote the forward difference operator by @xmath197 we also define the @xmath198 difference recursively as @xmath199 the row vector @xmath200 has a transpose given by a column vector and is denoted by @xmath201 . from this point on , we adopt the notation that matrix indices are always integers , i.e. , @xmath202 for a given limit shape @xmath203 , we denote its compositional inverse by @xmath204 , or simply @xmath205 when the context is clear . in our setting , the domain and range of a limit shape is always the positive real numbers , and limit shapes are always monotonically decreasing , hence @xmath204 is also a function . an integer partition of a positive integer @xmath0 is an unordered list of nonnegative integers whose sum is @xmath0 . it is standard notation to list the parts in descending order , @xmath149 , and we denote the number of parts by @xmath206 . we say that @xmath207 is an integer partition of _ size _ @xmath0 if @xmath208 . each partition @xmath76 has a unique _ conjugate _ partition @xmath209 that satisfies @xmath210 , @xmath211 . we denote by @xmath212 the set of partitions of size @xmath0 , and @xmath213 denotes the set of partitions of all sizes . a subset of partitions @xmath214 is defined similarly , with @xmath215 . each partition @xmath216 has a corresponding ferrers diagram , which is a collection of points on a two - dimensional lattice corresponding to parts in the partition ; see figure [ ferrers ] and see also @xcite for a further explanation . we note , in particular , that the ferrers diagram for @xmath209 is a reflection of the ferrers diagram for @xmath76 about the line @xmath217 . let @xmath218 denote the multiplicity of parts of size @xmath11 , where @xmath48 . there is a natural one - to - one correspondence between partitions @xmath219 and sequences @xmath220 with @xmath221 . when there are no restrictions on the multiplicities of parts , we call such a partition _ unrestricted_. we define @xmath222 to be the set of partitions where parts of size @xmath11 can occur at most @xmath223 times , where @xmath224 , @xmath225 $ ] , @xmath48 . the case @xmath226 for all @xmath48 corresponds to distinct part sizes , and the case @xmath54 for all @xmath48 corresponds to unrestricted integer partitions . we use the term _ andrews class partitions _ to refer to sets @xmath227 . 0.3 @xmath228 0.3 @xmath17 we consider sequences @xmath229 where @xmath230 are strictly monotone increasing and are polynomials in @xmath58 ; that is , @xmath231 we can , in fact , consider more general sequences , however , there are added technical conditions which unnecessarily complicate the statement of our theorems , and since all of our examples take the simpler form in , we forego this added generality . see [ ingham form ] for a discussion on the consequences of such generalizations . when gcd@xmath232 , then in what follows , by @xmath233 we mean taking @xmath0 to infinity through integer multiples of @xmath234 . [ ingham : theorem ] assume the set @xmath235 satisfies , for some @xmath236 , and let @xmath237 and @xmath238 . let @xmath239 denote the number of partitions of @xmath0 into part sizes in @xmath235 . then we have @xmath240}{c_1\ , n^{\frac{b\,r + e}{b(r+1)}+\frac{1}{2}}}\ts,\ ] ] where @xmath241 is a constant , and where @xmath242 denotes the negatives of the roots of @xmath243 . [ rothszekeres ] assume the set @xmath235 satisfies for some @xmath32 , and let @xmath237 and @xmath238 . let @xmath244 denote the number of partitions of @xmath0 into distinct parts from @xmath235 . then we have @xmath245}{c_2\ , n^{\frac{b\,r + e}{2b(r+1)}+1}}\ts,\ ] ] where @xmath246 is a constant . whereas only the leading two coefficients of @xmath243 are relevant for theorem [ rothszekeres ] above , the values of the roots of @xmath243 are also needed in theorem [ ingham : theorem ] . however , in terms of limit shapes only the exponent @xmath7 and leading coefficient @xmath237 will play a fundamental role . a set of partitions @xmath247 taking parts only from the set @xmath235 will be called _ with parameters @xmath7 and @xmath248 _ if the multiplicities of parts is unrestricted and @xmath235 satisfies with @xmath249 . a set of partitions @xmath247 taking parts only from the set @xmath235 will be called _ restrictedly smooth with parameters @xmath7 , @xmath248 and @xmath250 _ if the multiplicity of each part size is strictly bounded from above by @xmath250 and @xmath235 satisfies with @xmath251 . for each @xmath216 , the _ diagram function _ of @xmath76 is a right - continuous step function @xmath252 defined as @xmath253 define a _ scaling function _ to be any positive , monotonically increasing function @xmath254 , such that @xmath255 as @xmath233 . we define the _ @xmath256-scaled diagram function _ of @xmath76 as @xmath257 our definition of limit shape is from @xcite , and is as follows . let @xmath258 denote a family of measures on subsets of partitions @xmath259 , where @xmath89 . in this paper we consider only @xmath260 , the uniform distribution . suppose there exists a scaling function @xmath261 and a piecewise continuous function @xmath262 such that @xmath263 . suppose further that for any finite collection @xmath264 of continuity points of @xmath265 , and any @xmath266 , @xmath267 we say that @xmath265 is the _ limit shape of _ @xmath268 under scaling function @xmath261 . similarly , define the _ conjugate diagram function _ @xmath269 , such that @xmath270 we similarly define the _ @xmath256-scaled conjugate diagram function _ @xmath271 suppose there exists a scaling function @xmath261 and a piecewise continuous function @xmath262 such that @xmath272 . suppose further , that for any finite collection @xmath273 of continuity points of @xmath265 , and any @xmath266 , we have @xmath274 then we say that @xmath275 is the _ conjugate limit shape of _ @xmath268 under scaling function @xmath261 . the scaling function @xmath261 in the definition of limit shape is not unique , since it can be replaced with any scaling function @xmath276 such that @xmath277 . one could consider an equivalence class of scaling functions in order to uniquely specify such a class of functions , but we do not pursue this further . in either case , when the limit shape of @xmath278 exists , it is unique since it must integrate to 1 . following @xcite , define the space @xmath279 of triples @xmath280 , and nonnegative monotonically decreasing functions @xmath281 , such that @xmath282 one can think of @xmath283 as the set of measures on @xmath284 of the form @xmath285 , where @xmath124 and @xmath286 are point masses at 0 and @xmath287 . note that @xmath288 for all @xmath289 and scaling functions @xmath261 , and similarly , @xmath290 for all @xmath289 and scaling functions @xmath261 . the space @xmath283 is closed , so all limits of diagram functions exist in @xmath283 , even when limit shapes do not exist . [ vanishing : mass]sometimes the limit shape does not exist , even though there is a scaling for which a limit _ curve _ exists . consider , as in ( * ? ? ? * example 3.4 ) , the set of partitions @xmath291 , @xmath152 , consisting of a @xmath292 square block and a @xmath293 tail . for each @xmath0 of the form @xmath294 , there is exactly one partition in @xmath268 of size @xmath0 . in fact , one can take the limit of the scaled diagram function through values of @xmath0 indicated , and obtain the limit _ curve _ @xmath295 which does not have unit area . this corresponds to the element @xmath296 . all other scaling functions would send the scaled diagram function to either @xmath297 or @xmath298 in the limit . the theorems below indicate the types of limit shape results already known , which we shall use to obtain new ones . they are specializations of ( * ? ? ? . 8) . [ special unrestricted ] suppose @xmath247 is with parameters @xmath299 and @xmath33 . let @xmath300 . then the limit shape of @xmath247 under scaling function @xmath301 exists and is denoted by @xmath302 , and we have @xmath303 theorem [ special unrestricted ] above covers a large class of examples of interest in our present setting . it is also the starting point for finding limit shapes of partitions with non - multiplicative restrictions in [ transfer : theorems ] . we will also need another particular specialization , which is when all multiplicities of allowable part sizes are strictly bounded from above by some fixed number @xmath304 . typically the value @xmath305 , i.e. , distinct parts , is sufficient , but [ od general ] contains an example where we need further generality . [ special distinct ] suppose @xmath247 is with parameters @xmath306 @xmath307 and @xmath304 . let @xmath308 . then the limit shape of @xmath107 under scaling function @xmath301 exists and is denoted by @xmath309 , and we have @xmath310 in particular , by taking @xmath311 and @xmath312 , respectively , in ( * ? ? ? * th . 8) , we recover theorem [ special unrestricted ] and theorem [ special distinct ] . two subsets of partitions @xmath313 are called equinumerous if @xmath314 for all @xmath89 . for sets @xmath315 , @xmath316 @xmath317 , we consider one - to - one correspondences @xmath318 which can be expressed as a real valued linear transformations . we define ( multiplicities to parts ) to be a one - to - one correspondence that maps , using a linear transformation , multiplicities of parts of partitions @xmath319 to parts of partitions @xmath320 for all @xmath0 . the transformation has the form @xmath321 for some set of coefficients @xmath322 such that @xmath323 to avoid excessive subscripting we have opted for the functional notation of indices , i.e. , @xmath324 instead of @xmath325 . similarly , we define ( multiplicities to multiplicities ) to be a one - to - one correspondence @xmath326 that maps , using a linear transformation , multiplicities of parts of partitions @xmath319 to multiplicities of parts of partitions @xmath320 . define the vector @xmath327 , with @xmath328 denote the multiplicities of @xmath61 by @xmath329 , then the transformation has the form @xmath330 and the vector @xmath331 is a left eigenvector of the matrix @xmath332 with eigenvalue 1 . we define @xmath333 when the correspondence @xmath334 is an or with coefficients @xmath332 , to emphasize the role of @xmath332 in the bijection , and to distinguish s and s from general one - to - one correspondences between sets of partitions ( see for example @xcite ) . we next extend the notion of and to countable sequences . in this setting , the matrix of coefficients @xmath335 lies in @xmath186 , which we recall is the set of real valued operators on the banach space of real valued , countable sequences . define also the vector @xmath336 , with @xmath337 for s , we have @xmath338 similarly , for s , we have the vector @xmath339 is a left eigenvector of matrix @xmath332 with eigenvalue 1 . note that it is not necessary for every entry of @xmath340 to be nonnegative . suppose @xmath214 denotes a set of partitions which is with parameters @xmath341 @xmath342 we denote the limit shape of @xmath268 by @xmath343 and , furthermore , we define @xmath344 as the continuous function @xmath345 which satisfies @xmath346 suppose there exists an or between a set of partitions @xmath268 and some other set of partitions @xmath347 , with coefficients @xmath332 . since partitions in @xmath268 only have parts in @xmath235 , the nonzero values of coefficients @xmath332 are of the form @xmath348 , @xmath48 , @xmath152 . we require a condition on the coefficients @xmath332 as follows . we let @xmath349 so that @xmath350 we say the coefficients @xmath332 are if there exists a bounded and piecewise continuous function @xmath351 such that for all @xmath138 and @xmath352 we have @xmath353 similarly , we say the coefficients @xmath332 are if there exists a bounded and piecewise continuous function @xmath351 such that for all @xmath138 and @xmath352 we have @xmath354 we say the coefficients @xmath332 are if there exists a bounded and piecewise continuous function @xmath351 such that for all @xmath138 and @xmath352 we have @xmath355 similarly , we say the coefficients @xmath332 are if there exists a bounded and piecewise continuous function @xmath351 such that for all @xmath138 and @xmath352 we have @xmath356 when an @xmath159 is defined by an set of coefficients @xmath332 , we say that bijection @xmath159 is , and we denote the limit shape of @xmath357 by @xmath358 . similarly , when an @xmath159 is defined by an set of coefficients @xmath332 , we say that bijection @xmath159 is , and we denote the limit shape of @xmath357 by @xmath359 . a bijection defined in terms of linear combinations of multiplicities of part sizes acts as a markov operator on the diagram function ; see for example ( * ? ? ? * section 2.3.39 ) . recall that a markov operator p is a linear operator that sends functions @xmath360 to @xmath361 and is such that 1 . @xmath362 for @xmath363 , and 2 . @xmath364 . the coefficients @xmath332 in equation correspond to an explicit form of the operator p. note that while we must have @xmath365 for all @xmath366 , there is no restriction that each entry @xmath324 be nonnegative . most noteworthy , however , is the fact that markov operators are linear contractions , and hence preserve a form of the large deviation principle , which we revisit in [ concentration ] . not all bijections give a means to compute the limit shape . for example , it is not apparent how to utilize dyson s iterated map for odd - distinct parts to compute limit shapes @xcite . a more natural example would be the _ garsia milne involution principle _ bijections of rogers ramanujan identities @xcite ( see also @xcite ) . both sides have a limit shape , however , we are not aware of any way to draw conclusions about one limit shape from the other using the involution , as it would seem to be difficult to do in general @xcite . we now state our key technical theorems , which specify conditions under which the limit shapes map continuously , and provide the means for computing an explicit formula . [ mt unrestricted ] suppose @xmath247 is with parameters @xmath299 and @xmath33 . let @xmath367 . 1 . if there exists a set of partitions @xmath368 and an @xmath369 that is , then the limit shape of @xmath357 under scaling function @xmath370 is given by @xmath371 2 . if there exists a set of partitions @xmath368 and an @xmath369 that is , then the conjugate limit shape of @xmath357 under scaling function @xmath370 is given by @xmath372 note that the integrand function @xmath373 in the theorem is not necessarily multiplicative with respect to its final argument , as it is in the case of convex partitions in [ convex ] ; see also section [ sect r ] . it is more generally a distribution ( in the analysis sense ) on @xmath374 , and we present an explicit example demonstrating the need for this generality in [ sect ohara ] . an analogous result holds for partitions with multiplicities all strictly bounded by some @xmath304 . [ thm dist ] suppose @xmath247 is with parameters @xmath306 @xmath33 , @xmath375 let @xmath376 . 1 . if there exists a set of partitions @xmath377 and an @xmath369 , defined by a set of coefficients @xmath378 that is , then @xmath379 is the limit shape under scaling function @xmath370 . 2 . if there exists a set of partitions @xmath377 and an @xmath369 , defined by a set of coefficients @xmath378 that is , then @xmath380 is the conjugate limit shape under scaling function @xmath370 . the case @xmath305 is most common , corresponding to partitions into distinct parts , but the analysis is the same for any @xmath304 , and we shall utilize the full generality in [ od general ] . in @xcite , a general paradigm for representing partition bijections is presented involving the composition of transformations acting on the young diagram of an integer partition . these transformations map points in the plane via a bijection @xmath381 in such a way that under certain conditions , the limit shape along with certain other statistics of interest are mapped continuously via these transformations . this property is referred to as _ asymptotic stability _ , and the limit shape is the statistic of the young diagram corresponding to the boundary . we briefly recount the transformations as they pertain to the mapping of the limit shape , and defer the interested reader to the more comprehensive treatment in @xcite . for example , _ stretch and paste _ corresponds to combining several components , say with boundaries @xmath382 , which exist on the same domain and with the same area , by taking the sum with an appropriate scaling to keep the sum of the resulting expression constant . we think of this operation as a shuffle , which corresponds to @xmath383 _ conjugation _ is equivalent to taking a functional inverse : @xmath384 . _ move _ adds a constant : @xmath385 . _ shift _ adds a linear function @xmath386 : @xmath387 . _ shred _ and move acting together on @xmath7 components effectively splits up the limit shape into @xmath7 pieces , each having shape scaled by @xmath7 with area @xmath388 , which corresponds to @xmath389 a _ union _ , or _ sort _ , between two components corresponds to the addition @xmath390 . the _ + _ operator between two components corresponds to the inverse to the addition of the inverses : @xmath391 . a _ cut _ is somewhat more complicated , and requires treating the limit shape as defining a region in the positive quadrant of @xmath392 , which can be split in two pieces , one of which is a piecewise function , the other being the inverse of a piece - wise function ; see for example , figure [ parallel : mapping : limit ] . it partitions the curve into two curves , say via projections @xmath393 and @xmath394 , as @xmath395 a paste _ by itself _ corresponds to combining two curves whose domains partition some region in @xmath392 into a single curve . under appropriate conditions , paste acts as the inverse to a cut , i.e. , @xmath396 . [ transformations ] the geometric transformations on integer partitions correspond to the following transformations on limit shapes . + [ cols="<,<",options="header " , ] throughout this section , @xmath7 is any fixed integer , @xmath398 . [ nonneg ] let @xmath399 denote the set of partitions with parts that have nonnegative @xmath397 differences ; i.e. , @xmath400 , @xmath48 for each @xmath401 . the conjugate limit shape of @xmath399 under scaling function @xmath370 is given by @xmath402 with explicit formula @xmath403 when @xmath91 , we obtain corollary [ conj convex ] . a similarly defined set of partitions with multiplicative restrictions is utilized to obtain this result . [ r - th triangular ] let @xmath404 denote the set of partitions into parts of sizes given by @xmath405 , @xmath406 . the limit shape of @xmath404 under scaling function @xmath407 is given by @xmath408 . we now define the bijection @xmath409 in @xcite , which is a straightforward generalization to the one in [ convex ] . for each @xmath410 , let @xmath411 denote the number of parts of size @xmath412 in @xmath76 . then @xmath413 is given by @xmath414 [ rth lemma ] the bijection @xmath415 is an . moreover , @xmath415 is @xmath416 , with @xmath417 for @xmath418 and @xmath419 , we have @xmath420 , and 0 otherwise . we have @xmath421 with @xmath422 , @xmath423 , we have @xmath424 so @xmath425 hence , @xmath426 where @xmath427 , and @xmath428 which implies the result . let @xmath429 denote the set of partitions with all parts odd and @xmath430 denote the set of partitions with all parts distinct . we show in this section that glaisher s bijection @xmath431 is an , and present several extensions . recall the following classical result . for each @xmath89 , we have @xmath432 . glaisher s bijection @xmath157 is defined as follows . start with @xmath433 . replace each part of size @xmath434 with @xmath435 parts of size @xmath436 , for all @xmath437 and @xmath436 odd . define the sets @xmath438odd parts with multiplicities @xmath439 and @xmath440distinct parts not divisible by @xmath441 . the bijection @xmath431 , restricted to @xmath442 and @xmath443 , shows that @xmath444 for all @xmath299 . the limit shapes of @xmath443 and @xmath442 exist , and the bijection between them was shown to be a geometric bijection @xcite . an informal calculation in @xcite derives the limit shape of @xmath429 using the limit shapes @xmath442 for all @xmath299 . below we formalize this calculation by showing that glaisher s bijection @xmath157 is an . glaisher s bijection @xmath445 is an for all @xmath446 . furthermore , glaisher s bijection @xmath431 is an . for clarity of exposition , we first show cases @xmath447 then we show in full generality . in the case @xmath448 , @xmath449 and @xmath450 , the identity transformation . the transformation matrix is given by @xmath451 for all @xmath452 , and 0 otherwise . in the case @xmath91 , @xmath453 consists of parts of odd size and even parts not divisible by 4 . the transformation is given by @xmath454 the general case @xmath455 has transformation given by @xmath456 it follows that @xmath457 is a transformation matrix for an for all @xmath152 . using pointwise convergence in @xmath186 , we have @xmath458 as @xmath459 , where @xmath460 it is easy to verify that @xmath461 has a left eigenvector of @xmath462 corresponding to eigenvalue 1 . thus , glaisher s bijection @xmath431 is an . we will refer to the transformation @xmath461 from equation as _ glaisher s linear transformation_. a generalization to euler s theorem was given by andrews @xcite ( see also @xcite ) . for vectors @xmath463 and @xmath464 , let @xmath315 and @xmath316 denote the andrews class partitions @xmath227 and @xmath465 of size @xmath0 , respectively . define @xmath466 to be the set of all @xmath48 such that @xmath467 . we say that @xmath468 if there exists a bijection @xmath469 such that @xmath470 for all @xmath471 . [ andrews theorem ] if @xmath468 , then @xmath314 for all @xmath89 . we have : @xmath472 where we assume the convention that @xmath473 . the following algorithm , see @xcite , performs the bijection given by andrews classes of partitions , and is known as _ s algorithm_. for sets of partitions @xmath227 and @xmath465 , suppose @xmath474 . start with @xmath475 . replace @xmath44 parts of size @xmath11 with @xmath476 parts of size @xmath53 , where as before @xmath471 , @xmath65 repeat until @xmath477 for all @xmath65 in its full generality , ohara s algorithm is not asymptotically stable , and may require exponentially many steps , see @xcite . in the special case below , however , the algorithm is in a sense monotonic , and gives a valid alternative approach to obtaining glaisher s linear bijection . consider @xmath478 , @xmath479 and @xmath480 , @xmath481 , with @xmath482 . ohara s algorithm exchanges each even part for two parts of half the size . this is repeated until all parts are odd . denote by @xmath483 the linear transformation of one step in ohara s algorithm . then @xmath484 each step in ohara s algorithm corresponds to multiplication of the vector of multiplicities by the matrix @xmath483 . let @xmath485 denote @xmath58 steps in ohara s algorithm . we have : @xmath486 using pointwise convergence in @xmath186 , we have : @xmath487 where @xmath488 let @xmath489 denote the integrand for the limit shape of partitions into distinct parts . [ glaisher : relation ] glaisher s linear transformation is for @xmath490 with @xmath491 moreover , the limit shapes satisfy @xmath492 note first that @xmath493 , so @xmath494 next , we split the terms @xmath495 according to the integer @xmath58 for which we have . this gives @xmath497 note that @xmath461 corresponds to a markov operator p acting on the diagram function @xmath498 , with @xmath499 let @xmath500 for all @xmath501 , so that @xmath502 , and take the limit of the expectation , as @xmath233 . we obtain : @xmath503 \ , = \ , \sum_{i\geq 1 } \. \frac{\a^2}{2n } \. \sum_{j : y_j \geq t/2^{i-1 } } 2^{i-1 } \e z_{y_j \a 2^{i-1 } } \delta y_j \ , \. \longrightarrow \ , \. \frac{1}{2 } \. \sum_{i \geq 1 } \psi\mts\left(t/ 2^{i-1}\right ) , \quad t>0\. , \ ] ] which is finite for each @xmath138 if and only if @xmath504 . finally , note that also directly we have @xmath503 \to \sqrt{2 } \. \phi\left(t\.\sqrt{2}\right),\ ] ] which is the standard expression for the limit shape of partitions into odd parts . this example also demonstrates that the function @xmath373 in equation is not always multiplicative with respect to its final argument . another way to understand the connection between the limit shapes of @xmath505 and @xmath506 is via the construction of a geometric bijection . in this case , the bijection is more complicated , see ( * ? ? ? * figure 16 ) , but the same rules apply to the limit shape . a summary of the mappings is below . note that instead of mapping the diagram function at each stage , we have instead chosen to demonstrate how the bijection acts on the conjugate diagram function . we have : @xmath507where the final equality comes from a straightforward rearrangement of terms . in this section , @xmath32 , and @xmath508 are integers . the following is a generalization of theorem [ glaisher : relation ] . [ stanton : relation ] let @xmath268 denote the set of partitions into perfect @xmath397 powers with multiplicity at most @xmath509 , with limit shape given by @xmath510 . let @xmath357 denote the set of partitions into perfect @xmath397 powers not divisible by @xmath511 , with limit shape given by @xmath512 then we have @xmath513 note that the limit shapes of @xmath268 and @xmath357 in theorem [ stanton : relation ] follow from theorem [ special unrestricted ] and theorem [ special distinct ] , respectively . the relation in equation is obtained from the following generalization of euler s theorem . [ q prop]let @xmath514 be a sequence of distinct positive integers , and let @xmath515 be an integer . if @xmath516 is an integer for all @xmath58 , then the number of integer partitions of @xmath0 into parts of sizes avoiding @xmath517 is equal to the number of integer partitions of @xmath0 into parts of size @xmath514 , where @xmath243 has multiplicity at most @xmath518 . theorem [ q prop ] also follows by andrews s theorem [ andrews theorem ] using the mapping @xmath519 . the generalization of glaisher s bijection in this case is replacing parts of size @xmath520 into @xmath521 parts of size @xmath243 . when @xmath522 and @xmath523 , we obtain glaisher s bijection . stanton s @xmath524-bijection _ @xmath525 as follows . replace parts of size @xmath526 into @xmath527 parts of size @xmath528 , for all @xmath152 not divisible by @xmath511 . stanton s @xmath524-bijection @xmath529 is an . stanton s @xmath524-bijection can be written using the transformation matrix @xmath530 with entries @xmath531 for all @xmath152 not divisible by @xmath511 and @xmath532 . theorem [ stanton : relation ] then follows from equation in a similar fashion as the proof of theorem [ glaisher : relation ] . in this section , we use a bijection due to bressoud , see @xcite , to obtain the limit shape below . [ lebesgue : theorem ] let @xmath533 . let @xmath534 denote the set of partitions @xmath61 into parts congruent to @xmath535 or @xmath536 mod @xmath58 such that parts differ by at least @xmath58 , and parts congruent to @xmath536 mod @xmath58 differ by at least @xmath537 . the limit shape of @xmath534 is given by @xmath538 , \quad t \geq 0\. .\ ] ] let @xmath315 denote the set of partitions of size @xmath0 into distinct parts which are congruent to @xmath539 or @xmath5 modulo @xmath72 . let @xmath540 denote the set of partitions @xmath61 of size @xmath0 with consecutive parts @xmath62 for even part sizes @xmath63 and @xmath64 for all odd part sizes @xmath63 , @xmath48 . recall _ bressoud s bijection _ @xmath541 illustrated in figure [ bressoud ] ( see @xcite ) . the bijection @xmath157 is used in @xcite to prove lebesgue s identity , given below . we have : @xmath542 , the set of partitions @xmath61 with consecutive parts @xmath62 for even part sizes @xmath63 and @xmath64 for all odd part sizes @xmath63 , @xmath366 . ] the proof of theorem [ lebesgue : theorem ] , using @xmath543 and @xmath544 for simplicity , is explicitly worked out below , and a plot appears in figure [ lebesgue : plot ] . whereas the value of @xmath536 is an important aspect of the bijection , the limit shape is invariant to the particular value of @xmath536 chosen . let @xmath268 denote the set of partitions into distinct parts which are congruent to @xmath539 or @xmath5 modulo @xmath72 . let @xmath60 denote the set of partitions @xmath61 with consecutive parts @xmath545 for even part sizes @xmath63 and @xmath64 for all odd part sizes @xmath63 , @xmath366 . by theorem [ special distinct ] , the limit shape of @xmath268 is given by @xmath546 since the bijection to @xmath60 is given by a geometric bijection , see ( * ? ? ? * corollary 10.2 ) , the limit shape of @xmath60 exists , and we now derive its explicit form . we are unaware of its explicit form appearing in the literature . in figure [ bressoud ] , each non - blue square is the equivalent of two dots side by side , and each blue square is the equivalent of one dot . asymptotically , this distinction is negligible , but for the bijection it is essential . the first step in the bijection is to separate all of the parts divisible by three ; call this partition @xmath547 , and the partition consisting of the remaining parts by @xmath548 . the first step of the bijection , see remark [ stable : remark ] below , is the map @xmath549 which corresponds to separating the limit curves of @xmath550 and @xmath547 , respectively , in figure [ bressoud ] . note that since @xmath551 , we have @xmath552 . . 0.3 0.3 0.3 the next step in the bijection is to separate out partitions @xmath256 and @xmath553 from @xmath548 . the bijection places all parts in @xmath554 which are @xmath555 in @xmath553 , and leaves the rest in @xmath256 . since @xmath556 is an asymptotically stable statistic , with @xmath557 , we may further separate limit curves @xmath558 corresponding to @xmath559 as follows : @xmath560 where @xmath561 @xmath562 for @xmath563 , and @xmath564 we also need the following inverses : @xmath565 the limit curve of @xmath566 is given by @xmath567 the limit curve of @xmath568 is the line @xmath569 , for @xmath570 . we let @xmath571 now we subtract out the staircase by defining @xmath572 and @xmath573 . denote the limit curve of @xmath574 by @xmath575 and the limit curve of @xmath576 by @xmath577 . we have @xmath578 ( red ) , @xmath568 ( black ) , and @xmath574 ( blue ) . ] next , we sort the parts of partitions @xmath574 and @xmath576 , which corresponds to a union of parts ; call this partition @xmath579 , with corresponding limit curve @xmath580 . first , we have @xmath581 and @xmath582 note that @xmath583 for @xmath584 . therefore , @xmath585 the last step of the bijection adds partitions @xmath579 and @xmath568 , and corresponds to the conjugate limit shape of @xmath60 . namely , @xmath586 we can also invert to obtain the limit shape of @xmath60 . to obtain the inverse , we solve for @xmath587 in @xmath588 where @xmath589 , and @xmath590 . the positive solution for positive @xmath587 is of the form @xmath591 the limit shape of @xmath60 is thus given by @xmath592 this completes the proof of theorem [ lebesgue : theorem ] . [ stable : remark]note that the limit shapes of @xmath547 and @xmath593 do not follow from any of the transformations defined in [ geometric transformations ] . instead , the limit shapes exist in the stated form by section [ sect prob ] . first , the expected number of parts divisible by three over any collection of @xmath594 parts , @xmath595 , in the joint distribution of independent random variables exists and is given by @xmath596 . appealing again to section [ sect prob ] , we see that the pointwise convergence of expectation of the independent random variables is sufficient to guarantee convergence in probability for the set of integer partitions . from the limit shape given in equation , we can now obtain various statistics . for example , the expected number of parts in a random partition in @xmath60 is given by @xmath597 , which corresponds to equation in corollary [ number : of : parts : cor ] . to obtain corollary [ durfee : square : cor ] , we solve either for @xmath598 or @xmath599 . in this case , solving for @xmath600 using equation is the simpler calculation , from which equation follows . in this section , let @xmath601 denote a fixed positive integer , i.e. @xmath602 . [ diff : d : theorem ] let @xmath603 denote the set of partitions @xmath61 in which @xmath604 ; that is , parts are at least @xmath601 apart and there is no restriction on the number of parts . let @xmath605 denote the limit shape of @xmath603 . then we have @xmath606 , \qquad x>0\.,\ ] ] where @xmath607 where @xmath608 is the unique solution to @xmath609 in the interval @xmath610 , and @xmath35 is the solution to @xmath611 here @xmath612 is the usual _ dilogarithm function _ , which for positive real - valued inputs @xmath20 is defined as @xmath613 it is easy to obtain a minimal difference @xmath601 partition : take any partition @xmath614 and apply the transformation @xmath615 , for all @xmath616 . this mapping , however , is not size - preserving , as it is between unrestricted partitions of @xmath0 into exactly @xmath58 parts , and minimal difference @xmath601 partitions of @xmath617 into exactly @xmath58 parts . geometrically , we are adding a right triangle with slope @xmath601 whose right angle lies at the origin to the _ conjugate _ limit shape . [ theorem diff d k ] let @xmath618 denote the set of partitions @xmath61 into exactly @xmath58 parts , in which @xmath604 for @xmath619 . suppose @xmath620 . , and @xmath602 . let @xmath621 denote the limit shape of @xmath618 . then we have @xmath622 , \qquad x>0\. , \ ] ] where @xmath623 . let @xmath624 denote the set of minimal difference @xmath601 partitions of @xmath0 into exactly @xmath58 parts . let @xmath625 denote the set of unrestricted partitions of @xmath0 into exactly @xmath58 parts . we have @xmath626 , so we must tilt the distribution of the random size of the partition to have expected value @xmath627 and exactly @xmath58 parts . by conjugation , each partition in @xmath625 corresponds to an unrestricted partition with largest part at most @xmath58 . hence , for @xmath628 @xmath629 , we may generate the conjugate limit shape of @xmath630 which we denote by @xmath631 , directly using the method of section [ sect prob ] . let @xmath632 denote the limit shape of @xmath625 for @xmath633 , @xmath634 . we have @xmath635 where @xmath35 is the constant defined by the equation @xmath636 in @xcite , it is shown that the constant @xmath35 satisfies @xmath637 however , rather than tilt the distribution of a random partition to have expected size @xmath0 , we instead tilt the distribution so that the expected size of the partition is @xmath638 . this changes the limit shape from @xmath632 to a limit curve , say @xmath639 , given by @xmath640 where @xmath641 is such that @xmath642 the final step is to add the triangle of slope @xmath601 , i.e. , the curve @xmath643 , to @xmath639 . rearranging the terms yields equation . theorem [ diff : d : theorem ] then follows by the result below , c.f . [ romik : mistake ] . a random partition in @xmath603 has number of parts asymptotic to @xmath644 , where @xmath547 is given by equation . this section presents a bijection which requires one additional transformation before applying the conjugation transformation , see figure [ stretch ] . , the set of partitions into even parts such that the largest part has size @xmath646 , and @xmath647 , the set of partitions into even parts such that the number of parts is @xmath645 . ] [ ank][bb : theorem ] let @xmath647 denote the set of partitions into even parts such that the number of parts is @xmath645 . suppose @xmath648 , and let @xmath649 . then the limit shape of @xmath650 is given by @xmath651 where @xmath35 is the constant such that @xmath652 . let @xmath653 denote the set of partitions into even parts such that the largest part has size @xmath646 . the map @xmath654 given in ( * ? ? ? * section 5 ) is defined as follows : divide each part of @xmath655 by 2 , double the resulting multiplicities , and take the conjugate . for each @xmath89 and @xmath656 , the map @xmath334 is a bijection between sets @xmath653 and @xmath647 . bijection @xmath334 is an . the bijection @xmath334 corresponds to the transformation matrix @xmath657 furthermore , this transformation @xmath483 can also be realized as a composition of two transformations , @xmath658 , where @xmath659 @xmath660 we have @xmath661 is the conjugate transformation , and @xmath662 is the transformation @xmath663 defined above . suppose @xmath664 . let @xmath665 and let @xmath666 . the bijection @xmath667 is with @xmath448 , @xmath668 , scaling function @xmath504 , and @xmath669 for @xmath670 and 0 otherwise . [ ab : theorem ] suppose @xmath648 , and let @xmath671 . the limit shape of @xmath672 under scaling function @xmath504 is given by @xmath673 where @xmath35 is the constant such that @xmath674 . in fact , as written , the constant @xmath35 in theorem [ ab : theorem ] and the constant @xmath35 in theorem [ bb : theorem ] coincide . for any @xmath508 and @xmath32 , let @xmath675 denote the set of partitions of @xmath0 into parts from the set @xmath235 with @xmath676 , with largest part having size at most @xmath677 let @xmath678 denote the set of partitions of @xmath0 into parts from the set @xmath235 with @xmath676 and number of summands @xmath645 . the bijection @xmath679 is as follows : divide each part of @xmath680 by @xmath511 , multiply the resulting multiplicities by @xmath511 , and take the conjugate . define @xmath681 and similarly @xmath682 . suppose @xmath683 . let @xmath681 and similarly let @xmath682 . 1 . ( cf . [ history ] ) the limit shape of @xmath684 under scaling function @xmath685 is given by @xmath686 where @xmath35 is a constant such that @xmath687 . the bijection @xmath688 is with @xmath689 , scaling function @xmath685 , and @xmath690 for @xmath691 and 0 otherwise . the limit shape of @xmath692 under scaling function @xmath693 is given by @xmath694 where @xmath35 is the constant such that @xmath695 . in this section we introduce the probabilistic tools used in the proofs of the theorems . denote by @xmath696 a random partition of size @xmath0 , where each partition is equally likely . similarly , denote by @xmath697 the number of parts of size @xmath11 in the random partition @xmath696 , so that @xmath698 is satisfied with probability 1 . the diagram function ( see [ diagram functions ] ) for a random partition of size @xmath0 is given by @xmath699 [ convergence lemma ] let @xmath700 denote a sequence of _ independent _ , but not identically distributed , geometrically distributed random variables with parameters @xmath168 , @xmath211 , where @xmath701 , @xmath702 , @xmath703 . we write @xmath704 and @xmath705 when the parameters are understood by context . the diagram function @xmath706 of equation is constrained to have fixed total area for a given @xmath0 , corresponding to a partition of fixed size @xmath0 , for @xmath707 . the random variables @xmath708 and @xmath709 are related by , for all @xmath703 , @xmath710 see @xcite ; see also @xcite . here , equality means in distribution . we next introduce a sequence of random variables @xmath711 , where @xmath712 is defined as the value of @xmath713 conditional on @xmath714 , @xmath48 , @xmath715 , @xmath703 ; that is , @xmath716 we also allow for the case when @xmath717 , in which case we have @xmath718 . denote by @xmath719 a random partition of fixed size @xmath0 from an unrestrictedly smooth or restrictedly smooth set of partitions with parameters @xmath32 and @xmath33 . let @xmath720 denote the number of parts of size @xmath11 in the random partition @xmath719 , @xmath211 . then @xmath721 for @xmath722 and @xmath723 , @xmath724 . an analogous result to holds , namely , for all @xmath703 , @xmath725 for a given scaling function @xmath261 and constant @xmath35 , define the _ tilting parameter _ we define the _ independence diagram function _ by @xmath727 it has the same form as the diagram function @xmath728 , except it is a sum of independent random variables . the tilting parameter @xmath729 tilts the distribution , and we choose @xmath35 so that the expected area of @xmath730 is @xmath0 . in this particular case , the tilt given in equation also maximizes the probability that the area of @xmath730 is @xmath0 ( see ( * ? ? ? * section 5 ) ) . we also define the _ scaled independence diagram function _ as @xmath731 in this section , we formalize the intuition that the weak notion of a limit shape , i.e. , the pointwise convergence in expectation of the scaled independence diagram function , coincides with the limit shape of a certain set of integer partitions . [ distribution is sufficient ] let @xmath732 denote the scaled independence diagram function corresponding to a set of partitions @xmath247 which is either restrictedly smooth or unrestrictedly smooth . suppose there exists a piecewise continuous function @xmath733 , such that , @xmath734\ \longrightarrow\ \phi(t ) \qquad \text { for each continuity point \,\. $ t>0$}.\ ] ] then the limit shape of @xmath268 is given by @xmath265 . the same kind of result holds in our setting as well . note that an or @xmath735 induces a continuous , bijective map @xmath736 . for diagram functions , we have @xmath737 [ distribution : sufficient : bijection ] let @xmath732 denote the scaled independence diagram function corresponding to a set of partitions @xmath247 which is either restrictedly smooth or unrestrictedly smooth with parameters @xmath32 and @xmath33 . suppose there exists a piecewise continuous function @xmath733 , such that , @xmath738\ \longrightarrow\ \phi(t ) , \quad \text { for each continuity point \,\. $ t>0$}.\ ] ] suppose further there exists a set of partitions @xmath357 and an or @xmath739 that is . suppose there exists a piecewise continuous function @xmath740 , such that , @xmath741\ \longrightarrow\ \phi_k(t ) , \quad \text { for each continuity point \,\. $ t>0$}.\ ] ] then the limit shape of @xmath357 is given by @xmath742 . the proof is straightforward , but requires several technical lemmas , so we present it in the section below . the analogous theorem , valid for geometric transformations , is stated below . [ geo distr ] let @xmath732 denote the scaled independence diagram function corresponding to a set of partitions @xmath247 which is either restrictedly smooth or unrestrictedly smooth . suppose there exists a piecewise continuous function @xmath733 , such that , @xmath743\ \longrightarrow\ \phi(t ) \qquad \quad \quad \text { for each continuity point \,\. $ t>0$}.\ ] ] suppose further there exists a set of partitions @xmath357 and a geometric bijection @xmath744 . then the limit shape of @xmath357 exists and can be found via the application of transformations in theorem [ transformations ] . in addition , each geometric transformation is asymptotically stable . each geometric transformation is a continuous injection in the space @xmath185 , and thus the contraction principle applies to each transformation individually . the proof has two steps . the first establishes an exponential concentration , and the second is an example of overpowering the conditioning " ( see ( * ? ? ? * section 10 ) ) ; that is , that the conditioning on the event @xmath745 is not too strong . let @xmath746 denote the pointwise limit of the expected value of a scaled independence diagram function . a well - known and key calculation is the following : @xmath747 this formalizes the intuition that it is sufficient to have a large deviation which is little - o of the probability of hitting the target . first , we note that it is sufficient to find a concentration inequality for the _ independence _ diagram , as long as we have some bound on the probability of hitting the target of interest . fortunately , in the case of integer partitions , much is already known . [ large deviation ] suppose @xmath748 is the independence diagram function for a set of partitions @xmath247 that is either unrestrictedly smooth or restrictedly smooth with parameters @xmath299 and @xmath33 . assume @xmath749 for all continuity points @xmath138 of @xmath265 . then for every @xmath266 , there exists @xmath750 such that for all @xmath0 large enough we have @xmath751 where @xmath752 is a constant that does not depend on @xmath0 . [ local bound ] suppose @xmath235 is with parameters @xmath32 and @xmath33 , and @xmath753 ; or , @xmath235 is with parameters @xmath32 and @xmath33 for some @xmath754 , and @xmath755 . then , there exists an @xmath756 such that @xmath757 where @xmath758 is a constant that does not depend on @xmath0 . under the assumptions of theorem [ distribution : sufficient : bijection ] , for every @xmath266 , there exists @xmath759 such that for all @xmath0 large enough , we have @xmath760 where @xmath761 is a constant that does not depend on @xmath0 . equation follows by the contraction principle for continuous bijections , see ( * ? ? ? 4.2.1 ) . in @xcite , a large deviation principle ( ldp ) is presented for a general class of measures over a process of _ independent coordinates _ , which is stronger than theorem [ large deviation ] . in particular , the authors of @xcite let @xmath762 denote a sequence of non - negative integers , and say that @xmath763 is of type @xmath764 whenever @xmath765 in this more general context , @xmath766 denotes the number of different kinds of parts of size @xmath58 ; for example , we can give each part of size 1 any of @xmath767 colors , each part of size 2 can be any of @xmath768 colors , etc . in this paper , we consider only the case @xmath769 for all @xmath58 . in particular , for a sequence @xmath770 , this corresponds to @xmath771 then , as in @xcite , we let @xmath772 denote the positive , @xmath773finite measure on @xmath774 which has density @xmath775 . then the corresponding limit shapes given in @xcite , namely , @xmath776 and @xmath777 , correspond to our @xmath778 and @xmath779 . in our setting , the ldp is an asymptotic expression , both upper and lower bounds , governing the exponential rate at which a sequence of measures , say @xmath780 , assigns probability to a given measurable event as @xmath781 . for example , a typical event is of the form @xmath782 , and we say that measure @xmath780 satisfies the ldp with rate @xmath261 and good rate function @xmath783 if it has the form @xmath784 this is equivalent to @xmath785 and is stronger than the inequality in theorem [ large deviation ] . there are further details which pertain to the topology of convergence , a weakening of the limit as a @xmath786 and @xmath787 , and the precise form of the rate function @xmath788 . these details are relevant in order to distinguish between limit shapes of andrews partitions with unrestricted multiplicities of allowable part sizes , where the expected number of parts grows like @xmath789 , and andrews partitions with bounded multiplicities of allowable part sizes , where the expected number of parts grows like @xmath790 . in the former case , we only obtain pointwise convergence . in the latter case , the ldp is obtained in the topology of uniform convergence . we are only able to prove a ldp when the allowable part sizes have bounded multiplicity , see theorem [ our ldp ] below . let @xmath791 denote an interval , and let @xmath792 be the space of all functions @xmath793 that are left - continuous and have right limits . let @xmath794 denote the subset of @xmath795 of non - increasing absolutely continuous functions @xmath796 which satisfy @xmath797 , and let @xmath798 } \subset \mathcal{ac}_{\infty}$ ] denote the set of functions with derivatives which are lebesgue - a.e . in the interval @xmath799 $ ] . we write the lebesgue decomposition of @xmath796 as @xmath800 , where @xmath801 denotes the absolutely continuous part of @xmath796 , and @xmath802 denotes the singular part . let @xmath803 denote the derivative of a function @xmath804 with respect to @xmath805 . we write @xmath601 instead of @xmath806 just for equation below . define @xmath807 for all @xmath808 and @xmath809 , otherwise . here @xmath810 denotes the relative entropy between functions @xmath796 and @xmath804 . [ our ldp ] suppose the set of partitions @xmath247 is with parameters @xmath299 and @xmath33 and @xmath754 . then the diagram functions @xmath811 satisfy the large deviation principle in @xmath812 ( equipped with the topology of uniform convergence ) , with speed @xmath813 and good rate function @xmath814 , given by equation . for partitions without any restrictions on allowable part sizes , there is an area transformation that allows one to conclude that the ldp for the independent process also applies to the dependent process . when restrictions like @xmath770 are imposed , this area transformation is no longer defined . we now supply an appropriate area transformation and prove the above claim , _ which only covers the case when the multiplicity of each part size is bounded_. our treatment closely mimics @xcite . let @xmath815 denote the size of the partition @xmath76 . we define the transformation @xmath816 by 1 . if @xmath817 , then @xmath818 . 2 . if @xmath819 , then 1 . if there exists a set of partitions of size @xmath58 consisting of parts from @xmath820 , with each part size having multiplicity strictly less than @xmath250 , for @xmath821 then @xmath816 adds the parts of the partition which has the least number of parts to @xmath76 ; in the case of a tie , we choose the partition which is lower in lexicographic order ; 2 . if there does not exist a set of partitions of size @xmath58 consisting of parts from @xmath820 , then @xmath816 is the partition of @xmath0 in @xmath247 with the least number of parts . we now note the following properties of the transformation @xmath822 : 1 . for some fixed @xmath823 that does not depend on @xmath0 , and for @xmath0 large enough , we have @xmath824 for all @xmath825 . 2 . for any @xmath76 with @xmath817 , we have @xmath826 3 . let @xmath827 , which we will choose later . for any @xmath828 with @xmath829 , and for @xmath0 sufficiently large , we have @xmath830 where @xmath831 is some real valued constant which does not depend on @xmath0 . in addition , @xmath832 + where @xmath833 is some real valued constant which does not depend on @xmath0 . next , we define the sets @xmath834 define @xmath835 . then @xmath0 can be taken sufficiently large , so that we have @xmath836 thus , for such sufficiently large @xmath0 , we have @xmath837 \. \exp\bigl(c\ , \delta\ , n^{1/(r+1)}\bigr ) \. \pr(n^{-1}n(\lambda)=1 , \dhat_\lambda \in b_{\phi,\delta_r } ) , \end{aligned}\ ] ] and @xmath838 \. \exp(c\ , \delta\ , n^{\eta - r/(r+1 ) } ) \. \pr(n^{-1}n(\lambda)=1 , \dhat_\lambda \in b_{\phi,\delta_r } ) . \end{aligned}\ ] ] thus , for any @xmath827 , we have @xmath839 where @xmath840 is given by equation . the rest of the proof is straightforward ; the only ingredient previously missing was the area transformation @xmath841 and the inequality given by equation , which holds under the assumption that @xmath235 is with parameters @xmath842 , @xmath33 , and any @xmath843 . finally , we note that our transformation @xmath841 does not apply for integer partitions into unrestricted part sizes , as our proof relies on the total number of parts in a typical partition to be @xmath844 as @xmath781 . for @xmath845 and @xmath304 , let @xmath846 @xmath366 , @xmath847 , and @xmath848 be defined as in [ convergence lemma ] . define @xmath849 $ ] , which has explicit form given by @xmath850 note that @xmath851 . similarly , when multiplicities are unrestricted , we define @xmath852 we are now ready to state and prove our general theorem regarding limit shapes . we assume for the rest of the section that the set @xmath235 is such that @xmath247 is or with parameters @xmath299 , @xmath33 , @xmath304 . let @xmath195 be such that as @xmath23 , we have @xmath853 , @xmath854 . then as @xmath233 , we have @xmath855 & = \frac{n^{r/(1+r)}}{n } \sum_{k\in u : k\geq t \a } \e [ z_k ] \\ & \sim \ , \begin{cases } \sum_{y_k \geq ( t / b)^{1/r } } e\left(e^{-c\ , b\ , y_k^r}\right ) \delta y_k \to \phi(t ; r , b ) , & \text{for theorem~\ref{special unrestricted } } , \\ \sum_{y_k \geq ( t / b)^{1/r } } e_a\left(e^{-c\ , b\ , y_k^r}\right ) \delta y_k \to \phi(t ; r , b , a ) , & \text{for theorem~\ref{special distinct}}. \end{cases}\end{aligned}\ ] ] by theorem [ distribution is sufficient ] , this calculation is sufficient to establish the limit shape . we continue to use the notation from [ proof special distinct ] . in the case of an , we have @xmath856 taking expectation , and applying equation and equation , as @xmath23 we have @xmath857 & = \frac{\b}{n } \sum_{y_k \geq 0 } v({\b t,\a b\ , y_k^r})\ , \mathbb{e}\left[z_{b\th y_k^r \a}\right ] \\ & \longrightarrow \begin{cases } \int_0^\infty k\left(t , y , e\left(e^{-c\ , b\ , y^r}\right ) \right)\ , dy , & \text{for theorem~\ref{mt unrestricted } , } \\ \int_0^\infty k\left(t , y , e_a\left(e^{-c\ , b\ , y^r}\right ) \right)\ , dy , & \text{for theorem~\ref{thm dist}. } \end{cases}\end{aligned}\ ] ] in the case of an , similarly by equation and equation , as @xmath23 we have @xmath858 & = \frac{\b}{n } \sum_{i \geq t \b } \sum_{k \geq 1 } v ( { \b t , u_k})\ , \e z_{u_k } \\ & \longrightarrow \begin{cases } \int_0^\infty k\left(t , y , e\left(e^{-c\ , b\ , y^r}\right)\right)\ , dy , & \text{for theorem~\ref{mt unrestricted } , } \\ \int_0^\infty k\left(t , y , e_a\left(e^{-c\ , b\ , y^r}\right)\right)\ , dy , & \text{for theorem~\ref{thm dist}. } \end{cases } \end{aligned}\ ] ] by theorem [ distribution : sufficient : bijection ] , the proof is complete . [ [ history ] ] ~ the probabilistic model utilized for integer partitions goes back to khinchin @xcite . the explicit form of the limit shape curve @xmath746 for unrestricted integer partitions was derived via heuristic arguments by temperley @xcite . he applied the method of steepest descents to the generating function for @xmath4 , the number of partitions of size @xmath0 , in order to describe the equilibrium state of crystalline structures . several decades later , kerov and vershik @xcite found and proved the limit shape for integer partitions under the _ plancherel measure _ , which weights each partition of size @xmath0 by the squared dimension @xmath859 of the corresponding irreducible representation of @xmath860 . its formula ( rotated by @xmath861 ) is given by @xmath862 we refer to @xcite for a thorough and well written proof of this result and its applications . let us mention that the final section of @xcite states without proof that the approach used also gives the limit shape of integer partitions under the uniform measure . a followup paper by vershik @xcite states the strong notion of a limit shape for uniformly random integer partitions , but leaves out the essential details which imply the strong version from the weak one , sometimes referred to as equivalence of ensembles " in the statistical mechanics literature . in parallel to the probabilistic notions of a limit shape , there was an early interest in understanding the joint distribution of part sizes , which appears to begin with the work of erds and lehmer on the largest part size @xcite , later extended by erds and szalay @xcite to the largest @xmath805 part sizes , where @xmath138 is fixed . a significant improvement occurred with the work of fristedt @xcite , who extended many but not all of the results to the largest @xmath863 part sizes . pittel fully resolved the strong notion of a limit shape in @xcite . the proof starts by confirming a conjecture of arratia and tavare @xcite governing the total variation distance between the joint distribution of component sizes and the joint distribution of appropriately chosen independent random variables , precisely those defined in section [ sect prob ] , and essentially extending fristedt s results to the largest @xmath864 part sizes . the _ total variation distance _ between two distributions , say @xmath865 and @xmath866 , is defined as @xmath867 where @xmath868 is a borel subset of @xmath869 . there are two initial , complementary results , governing the smallest part sizes and the largest part sizes . ? * th . 1 ) ) in the definitions from section [ sect prob ] , let @xmath870 , with @xmath871 , and suppose @xmath872 is a geometric random variable with parameter @xmath168 , for @xmath366 , where all @xmath873 are mutually independent . let @xmath697 denote the number of parts of size @xmath11 in a random integer partition of size @xmath0 . then : @xmath874 thus , one may treat both small and large part sizes as approximately independent , which was utilized in @xcite , for example , to resolve several conjectures for integer partitions involving statistics localized on the largest part sizes . pittel also showed in ( * ? ? ? * th . 1 ) , that if we include in the joint distribution of part sizes those parts which are at most @xmath875 , for some @xmath138 , then this joint distribution converges to a non - trivial limit which is the total variation distance between two appropriately defined normal distributions . however , total variation distance is a considerably strong metric , since one can take any measurable function of the random variables and the resulting total variation distance is guaranteed not to increase , whereas a limit shape is one particular statistic . thus , by a more detailed analysis of the component sizes , specifically for the limit shape statistic , pittel was able to show that the limit shape of random integer partitions is indeed close in probability to the limit shape @xmath265 indicated by the weaker notions of convergence , thus formally establishing the equivalence of ensembles . the reason why the weaker notion of the limit shape , i.e. , convergence in expectation of random - size partitions with independent part sizes , coincides with the strong notion of the limit shape , i.e. , convergence in probability of integer partitions of a fixed size , is due to the exponential concentration of the limit shape statistic . this has traditionally been handled either via a martingale technique , or via more direct methods like markov s inequality , and demonstrates an unsurprising exponential concentration around the expected value which could be considered implicit in the work of vershik @xcite . as noted in [ sect : ld ] , an ldp is not just an upper bound on the rate of exponential concentration , as is often all that is needed to prove the strong notion of a limit shape from the weak notion , it is also an asymptotic expression for the rate of concentration . while pittel s work proved the limit shape of unrestricted partitions , the large deviation results by dembo , vershik , and zeituni in @xcite provided the limit shape of integer partitions into distinct parts , and made their connection with the ldp explicit . in addition , ( * ? ? ? * section 6.2 ) states the form of the limit shape for partitions with restrictions of the form @xmath876 and provides the ldp for the _ independent _ process in both cases of unrestricted multiplicities and distinct part sizes . combining the ldp with classical asymptotic enumeration formulas , e.g. , those in for example @xcite , it is straightforward to derive the equivalence of ensembles under these restrictions via the inequality in equation . we have provided the ldp for the _ dependent _ process with bounded multiplicities , i.e. , the process governing random partitions of fixed size @xmath0 , in [ sect : ld ] . [ [ credit ] ] ~ it is difficult to credit limit shape results like theorem [ special unrestricted ] precisely , owing to the connection between asymptotic enumeration and concentration , and the precise conditions which are assumed and utilized . the weak form of the limit shape in theorem [ special unrestricted ] is easily calculated via the statistical models in khinchin @xcite , and more explicitly by kerov and vershik @xcite ; the same ones utilized by fristedt @xcite and pittel @xcite . the limit shape then follows by concentration and a local central limit theorem , or equivalently , asymptotic enumeration , via the inequality in equation ; see also @xcite . ingham ( * ? ? ? 2 ) provided quite general asymptotic enumeration formulas for partitions satisfying the following conditions : let @xmath877 denote the number of elements in @xmath235 less than or equal to @xmath878 . suppose there are constants @xmath879 , @xmath880 , and function @xmath881 such that @xmath882 and for some @xmath883 and @xmath884 we also have @xmath885 define @xmath886^{1/\beta}.\ ] ] then , _ as long as @xmath239 is monotonically increasing _ , we have @xmath887 similarly , under the slightly weaker assumption that @xmath888 and again assuming _ @xmath889 is monotonically increasing _ , with @xmath890 , we have @xmath891 erds and bateman in ( * ? ? 6 ) simplified the monotonically increasing assumption for @xmath239 to be more simply that the greatest common divisor of @xmath235 is 1 . it is unknown whether or not a corresponding condition exists for partitions into distinct parts , and it would be interesting to find such necessary and sufficient conditions . roth and szekeres provide a partial answer , at least one detailed enough for our purposes . they demand the elements @xmath243 satisfy more generally @xmath892 in addition , they also assume the following condition , which is sufficient for the asymptotic enumeration results to hold : @xmath893 ^ 2 \right\ } \to \infty \ \ { \rm as } \ \ k\to \infty,$}\end{aligned}\ ] ] where @xmath894 $ ] denotes the closest integer to @xmath20 and the infinum is taken over @xmath895 $ ] . they also show that when @xmath7 is small , condition @xmath896 is not necessary . @xmath897 there are also several classes of sequences @xmath235 presented in the introduction in @xcite which cover many cases of interest ; namely , _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ * @xmath898 where @xmath899 is the @xmath198 prime number . * @xmath900 , where @xmath901 is a polynomial which takes only integral values for integral @xmath20 and has the property that corresponding to every prime @xmath902 there exists an integer @xmath20 such that @xmath903 . * @xmath904 , where @xmath901 is a polynomial which takes only integral values for integral @xmath20 and has the property that corresponding to every prime @xmath902 there exists an @xmath20 such that @xmath905 . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ in particular , they specialize their main theorem in the case when @xmath906 and @xmath243 satisfies ( ii ) , which prevents any gaps in @xmath244 ; this is theorem [ rothszekeres ] . we have dropped the final condition that for every prime @xmath902 there exists an integer @xmath20 such that @xmath903 , requiring instead that should such a @xmath902 exist then we take @xmath0 to infinity through multiples of the gcd of @xmath235 , as is required for the corresponding limit shape . one could also argue that theorem [ special unrestricted ] follows from pittel s analysis @xcite , in that it requires no new ideas other than generalizing the analysis appropriately . however , there are certain estimates utilized which must be carefully adapted , and a priori require some kind of technical conditions similar to the ones initially imposed by ingham and others . one may also be tempted to attribute theorem [ special unrestricted ] to dembo , vershik , and zeituni @xcite . in ( * section 6.2 ) , they state explicitly an ldp for an independent process of random variables , which is sufficient to prove concentration , but as stated the sequence @xmath235 consists of multiples of perfect powers . this is similar to the work in goh and hitczenko @xcite , whose extra technical assumption is the existence of four @xmath907 s which are relatively prime . canfield , corteel , and wilf @xcite provide all of the ingredients for theorem [ special unrestricted ] , even including the restriction that @xmath243 is the image of a polynomial , which is the special case we have adopted . [ [ section ] ] we have decided to describe the growth conditions on sets @xmath235 in terms of the growth of @xmath243 , where @xmath196 . meinardus @xcite , granovski , stark and erlihson @xcite , hwang @xcite , and bogachev @xcite , for example , impose conditions directly on the corresponding generating functions . yakubovich @xcite , goh and hitczenko @xcite , and ingham @xcite , on the other hand , have described growth conditions on the inverse function @xmath908 . our choice is natural for this setting since our sets of partitions and bijections have traditionally been written in the form of allowable part sizes . [ [ ingham form ] ] ~ the original form of the growth conditions in ingham s ( * ? ? ? * th . 2 ) is particularly elegant , even though it is not ideal for our setting . rather than requiring a condition on the greatest common divisor of allowable part sizes , the result is instead stated in terms of sums of partition numbers . formally , let @xmath239 denote the number of partitions of @xmath0 into parts from the set @xmath235 . ingham defines for real @xmath909 @xmath910 then , he proves an asymptotic result for @xmath911 _ and _ @xmath912 under the condition that @xmath913 is increasing for every fixed @xmath914 . when @xmath915 , we can choose @xmath916 . then , _ for every _ @xmath909 , the number @xmath913 is precisely one of @xmath917 for some @xmath918 , and , since @xmath919 , it thus captures the asymptotic rate of growth of @xmath239 , without needing to specify , as we have done , that @xmath23 through multiples of @xmath250 . [ [ section-1 ] ] the conditions of meinardus @xcite and yakubovich @xcite are related to condition , and apply more generally to partitions where each part size can have multiple versions ( for example , three different types of 1s ) . these conditions are used by hwang @xcite , for example , where the distribution of the number of summands ( taken without multiplicities ) follows a central and local limit theorem . there is also a central limit theorem proved by madritsch and wagner in @xcite , concerning partitions whose base-@xmath234 representation only contains digits from some given set . this type of condition is natural given the fact that the bits in a geometric random variable are independent , a property which was also exploited in , e.g. , @xcite for the purpose of random sampling of integer partitions . [ [ fixed : summands ] ] ~ the asymptotic number of partitions with part sizes restricted to be less than some @xmath920 , was studied by romik @xcite . the limit shape of the conjugate set of partitions , i.e. , the set of partitions with a fixed number of summands @xmath875 , was studied by yakubovich and vershik @xcite . this is another example , see [ romik : example ] , where a bijection involving conjugation is used to obtain the limit shape of a non - multiplicative set of restrictions . a more detailed saddle point analysis counting the number of partitions of @xmath0 into exactly @xmath436 summands under further restrictions like those in [ further : example : generalization ] are contained in @xcite . [ [ romik : mistake ] ] ~ another proof of theorem [ theorem diff d k ] appears in the last section of the unpublished preprint @xcite . this was used to obtain ( * ? ? ? * th . 1 ) . unfortunately , there appears to be a mistake , as the formula given for the limit shape does not have unit area . [ [ section-2 ] ] ~ in this paper we consider only @xmath260 , the uniform distribution ; see , however , @xcite for limit shapes under plancherel measure , and e.g. @xcite for other types of measures on partitions . [ [ natural : generalization ] ] ~ the next natural generalization of our theorems would be the one considered by canfield and wilf @xcite : partitions of size @xmath0 with part sizes in set @xmath235 and with multiplicities in set @xmath921 . in our setting @xmath922 , and in [ romik : example ] we have @xmath923 . it would be interesting to investigate conditions on @xmath921 and @xmath235 such that the limit shape exists . our approach does not extend to compute the limit shape for classes of partitions for which no bijection to andrews class partitions is known . one notable example is the partitions into powers of @xmath5 , which are equinumerous with certain _ cayley compositions _ ( see @xcite and ( * ? ? ? * http://oeis.org/a000123[a000123 ] ) ) . in this case the usual notion of a limit shape does not exist and an alternative notion is required . fristedt s method @xcite in section [ sect prob ] defines a random sampling algorithm for andrews class partitions of a fixed size @xmath0 . namely , we sample independent geometric random variables @xmath924 , where @xmath873 has parameter @xmath168 , @xmath703 , @xmath48 , and check whether @xmath925 , repeating the sampling of @xmath924 until the condition is satisfied . the expected number of times that we resample is given by @xmath926 , where @xmath927 an application of probabilistic divide - and - conquer ( pdc ) , introduced in @xcite , see also @xcite , uses von neumann s rejection sampling @xcite to obtain a speedup . the algorithm is to sample repeatedly from @xmath928 and @xmath235 , uniform on @xmath610 , until @xmath929 the expected number of times that we resample this procedure is @xmath930 , see ( * ? ? ? * table 1 ) , a considerable speedup . more generally , consider the set of partitions with parts in the set @xmath235 , where @xmath931 . the pdc deterministic second half algorithm to generate a random integer partition of size @xmath0 with parts in the set @xmath235 is to sample repeatedly rom @xmath932 and @xmath235 , uniform on @xmath610 , until @xmath933 the speedup in this case is given by @xmath934 hence , the expected number of rejections before a single accepted sample via the pdc deterministic second half algorithm is asymptotically @xmath935 . there are also bijections defined as linear transformations by corteel and savage @xcite , and corteel , savage , and wilf @xcite , which correspond to partitions which satisfy various inequality conditions on the part sizes , in a sense generalizing the set @xmath936 of partitions with nonnegative @xmath397 differences in section [ sect r ] . a further generalization was demonstrated by pak @xcite ; most of those examples involve sets @xmath235 where @xmath243 grows exponentially in @xmath58 , and so our theorems do not apply . of particular note is that our linear transformations are only assumed to act as markov operators on the diagram function , so it is not necessary that the coefficients of the matrix transformation @xmath332 each be nonnegative ; matrix @xmath332 simply needs to map positive coordinates to positive coordinates . for self - conjugate partitions , it would be interesting to define a coupling over the set of random variables governing the part sizes , although this has many apparent difficulties . we hope to return to this problem in the future . we showed earlier in remark [ vanishing : mass ] that the limit shape of @xmath268 under scaling function @xmath22 does not exist , since the scaled diagram function converges to @xmath939 for @xmath940 , which does not have unit area . the limit shape of @xmath357 exists and is equal to @xmath941 for @xmath942 . consider the ( non - bijective ) transformation @xmath943 , which sends partitions in @xmath268 to @xmath357 . even though there is a transformation , and each set of partitions tends to some limiting curve , there is no connection implied about the limit curves since the transformation is not even a bijection . the authors are grateful to ken alexander , richard arratia , quentin berger , amir dembo , alejandro morales , greta panova , dan romik , bruce rothschild , richard stanley and damir yeliussizov for helpful discussions . the second author was partially supported by the nsf . d. romik , identities arising from limit shapes of constrained random partitions ; preprint ( 2003 ) , 15 pp . ; available at .

animal packing in social aggregations is of fundamental interest in ecology , and their conformations have been extensively studied @xcite . pioneer tank observations of pelagic fishes ( herring , sprat and mackerel ) were conducted by @xcite , and the `` persistently uniform density '' of a school was noted . from experiments on schools of saithe _ pollachius virens _ , herring _ clupea harengus _ and cod _ gadus morhua _ cruising in a ten meters circular gantry tank , @xcite validated that all the fish in a school occupy a volume proportional to @xmath2 , where @xmath3 is the number of fish and @xmath4 is the mean nearest - neighbor distance ( approximate to the mean fish - body length ) in the school . @xcite reported from field observations that the number densities of herring schools are almost independent of the dimensional size of school but are an order of magnitude lower than the packing densities observed when schooling in small tanks . tauti and colleagues @xcite had experimentally shown that a fish school can be viewed as a group of inorganic particles and treated with the methods of physics . @xcite explicitly applied physical equations to such fish schools . recently , a number of theoretical and numerical models of schooling and flocking behavior have been studied by biologists , mathematicians and physicists @xcite . as for the animal group geometry , many models for social aggregations , however , predict that densities increase as the group size ( in number ) increases [ overviews of such models are given in refs . @xcite mathematically tackled this problem of constant density and revealed the condition for a well - spaced group , i.e. what class of mutual interactions results in a relatively constant individual distance in the interior of the aggregate . more recently , by means of underwater acoustics , the school sizes ( in number or biomass of fish ) have been quantitatively measured for different values of the dimensional size of schools in the wild . precise data on conformations of large - sized schools of pelegic fishes became available rapidly , which were extremely helpful in elucidating a certain geometric law , i.e. the relation between dimensional and biomass sizes of pelagic fish schools , bringing about some important changes in our viewpoints . misund and colleagues @xcite found that the power - law scaling in dimension - to - biomass relationship exists , and is robust across a broad range of pelagic species as well as across diverse environments . they demonstrated that if the biomass @xmath3 in a school is , say , doubled , the cross - sectional area of a school is increased by a factor @xmath5 , i.e. @xmath6 and that the exponent @xmath7 looks universal , reading @xmath8 ( from the field data @xmath7 ranges from @xmath9 to @xmath10 ) . the geometric law they found implies that the mean density of a school scales as @xmath11 in three dimensions of space and the conformation of social aggregations swells ( i.e. @xmath12 ) . such a relationship as eq.([eqn : misund - school - geometry ] ) with @xmath13 has been utilized for the dimensions to biomass conversion ( e.g. transforming the school diameter @xmath14 to the number @xmath3 of fish ) in analyzing school - size distributions @xcite . it might come as a surprise that packing densities decrease as the group size increases , contrary to previous observations and predictions . laboratory observations for school geometry and internal structure have been made exclusively in small tanks , which generally show the constant density of fish that the school volume is proportional to the number of individuals . in all set - ups , the factors resulting in homogeneous , cohesive school may be especially pronounced , and it therefore seems that the quantified structure is skewed . @xcite claimed that homogeneity in schools has been over - emphasized . _ in - situ _ observations of herring and sprat schools with a high resolution sonar revealed that the packing structure within the schools is rather heterogeneous @xcite . this has been confirmed by measurements of free - swimming schools using photography and high - resolution echo integration , which showed that the packing density distribution in capelin _ mallotus villosus _ and clupeoid schools varies considerably @xcite . regions of high density are usually found within the schools , and even empty vacuoles have been recorded . @xcite observed by repeated echo integration that there were large variations in internal packing density of herring schools ( i.e. high - density regions or empty lacunas within a school ) , and that the packing density structure was quasi - stationary . besides artificial environments in small tanks , the discrepancy in former observations could have been caused by too small numbers of fish in the schools . for instance , in @xcite , @xmath3 takes a few tens of fish . since such a geometric law above is always defined only in a certain limit @xcite , the scaling in the dimension - to - biomass relationship is expected to hold for large - sized schools of pelagics . in this paper , the exponent @xmath7 is estimated according to the established universal scaling law in the school - size distribution of pelagic fishes @xcite : choose the suitable value of @xmath7 to achieve the best data - collapse on the size distributions in terms of the school dimension . notice that the dimension - to - biomass relationship is a property of the single school and the scaling exponent @xmath7 is determined by the behavioral algorithm of fish schooling at individual level , while the scaling in the school - size distribution emerges from the inter - school interactions at population level ( i.e. a global property of the interacting school system ) . animal group size is a focal issue in ecology that , in contrast to scaling , has introduced a single preferred size ( i.e. optimal or compromise size ) for any organism living in groups @xcite . figure [ fig:1 ] shows an example of the histogram of school dimensions of japanese sardine _ sardinops melanostictus_. a peak frequency and a right skew are typical of pelagic fishes @xcite . their linear dimension , e.g. the vertical thickness , of the school falls into a certain range below a few tens of meters . one fish may not be the right atomic unit in schooling , since field observations suggest that no school exists under a certain minimal size . this may cause binned data of school sizes to exhibit the fake peak frequency . @xcite and @xcite have ignored slowly decaying ( or fat - tailed ) distributions , including scaling laws , in such data of school sizes short ranged with a fake peak frequency ( the possibility of power laws was already presented in their models but not exploited ) . a traditional , widely used gauss statistics says that , for the data from @xcite shown in fig.[fig:1 ] , finding sardine schools ranging from 18 to 20 meters in vertical thickness should only occur about once every @xmath15 detections of schools [ for detail , consult @xcite ] . in other words , it is not the real world ! aquatic observations actually say that finding such schools occurs about once every 500 detections . the probability that such schools are found is @xmath16 times as large ! ) may be stray ones . , width=245 ] the possibility of scaling in such distributions was found by @xcite . fat - tailed distributions have been found to quite generally characterize the size heterogeneity of pelagic fish schools @xcite . only lately , @xcite showed that the school - size distribution of pelagic fishes has the universal functional form of a power - law decay and a crossover towards the exponential distribution . scaling laws have been found to characterize many patterns in ecological systems @xcite . there is an empirical rule about the relationship between the areas ( @xmath17 ) of island and the numbers ( @xmath18 ) of species on them : whether counting birds , butterflies , plants or copepods in ponds , a tenfold smaller area contains roughly half the species . this can be fitted by a power - law function @xmath19 with the exponent @xmath20 @xcite . the species - area relationship is widely used in rough assessments of likely future rates of species extinction ( because observed rates of tropical deforestation are equated to loss of habitat area , we can assess the annual production of species committed to extinction ) . so , a more secure understanding of such relationships has great practical importance @xcite . @xcite applied finite - size scaling @xcite to test the hypothesis of scaling invariance in the species - area relationship resting on a model proposed by @xcite for the species abundance distribution . following the conventional notation in the physics literature , the system sizes are hereafter denoted by @xmath17 . the finite - size scaling ( fss ) hypothesis assumes that the fraction of objects with size @xmath21 for a finite system of size @xmath17 is written , with a scaling exponent @xmath22 , as @xmath23 where @xmath24 is a universal scaling function . @xmath25 reads , for instance , the species - abundance distribution , which defines the probability that any given species on a census patch of area @xmath17 has @xmath21 individuals , where @xmath26 is the expected number of species with @xmath21 individuals . @xmath27 is a measure of the width of the probability distribution @xmath25 , e.g. the mean or the standard deviation of the distribution . equation ( [ eqn : fss - hypothesis ] ) expresses the principle that the behavior of the finite system , i.e. @xmath25 , is controlled by the ratio @xmath28 . we test the fss hypothesis by data collapsing : when @xmath29 is plotted versus @xmath30 , if the distributions across different @xmath17 s fall on a single curve ( e.g. collapsing distinct abundance distributions measured in different areas and with different total numbers of individuals onto one scaling curve ) , then we should be able to identify a scaling function @xmath31 . in order to determine the exponent @xmath22 , we try a best - fitting procedure such that the family of curves @xmath25 collapses onto a single curve as well as possible . we then obtain other scaling exponents ( e.g. @xmath32 of the species - area relationship ) resting on the scaling relation between scaling exponents @xcite . by noting that the total number of individuals of all species is equal to @xmath33 , the species - area relationship leads to a scaling relation @xmath34 assuming that the total number of individuals of all species is proportional to area @xmath17 @xcite . the universal scaling is fundamental to data collapse . the power - law scaling emerges as ubiquitous properties in ecology . in statistical analysis in ecology , data collapsing across different environments or species has been , however , observed only lately in dynamics of breeding bird populations @xcite , food web structure @xcite , microbial body - mass spectra in marine ecosystems @xcite , school - size distributions of pelagic fishes @xcite , and in the context of ecological economics , capture fisheries productions of countries @xcite . in order to understand the geometric properties of school configurations , the school biomass @xmath3 must be measured for different values of dimensional size @xmath14 , and we must compare them . misund and colleagues extensively performed the simultaneous observations of the values of @xmath14 and @xmath3 for pelagics by the following two methods : ( i ) two - dimensional ( cross - sectional ) acoustic measurements ( unit in square meters ) and subsequent purse seine capture ( unit in tonnes ) of schools @xcite , and ( ii ) acoustic measurements of the three - dimensional structures and backscattered echo energy @xcite . the surveys were conducted on stocks of anchovy _ engraulis capensis _ , herring _ clupea harengus _ , horse mackerel _ trachurus trachurus capensis _ , mackerel _ scomber scombrus _ , pilchard _ sardinops ocellatus _ , round herring _ etrumeus whiteheadi _ , saithe _ pollachius virens _ , sardine _ sardinops sagax _ , and sprat _ sprattus sprattus _ , in different seasons and geographic regions ( the barents sea , the north sea , the norwegian sea , the northeastern atlantic and off namibia , and off the coast of south africa ) . they found that there is some biological universality in the dimensions - to - biomass relationships , eq.([eqn : misund - school - geometry ] ) . their finding implies that the radius @xmath14 of the school scales as @xmath35 in a statistical sense . @xmath3 , denoting the school biomass , is hereafter defined by the number of fish in a school . now i propose another way to establish the scaling law for school dimensions , eq.([eqn : scaling - school - geometry ] ) , by applying universal scaling law in fish school - biomass distributions @xcite . the biomass distributions @xmath36 follow a power law with exponent @xmath37 up to a cut - off size @xmath38 , @xmath39 where @xmath40 is a crossover scaling function with a strong drop for @xmath41 , and the cut - off size ( crossover size from power - law to exponential decay ) is calculated from the biomass histogram data @xmath42 , @xmath43 where histogram bins are chosen with width @xmath44 . the dimension data of fish schools are binned with width @xmath45 , giving the set of frequencies @xmath46 . from eqs.([eqn : scaling - school - geometry ] ) and ( [ eqn : biomass - distribution ] ) , the distribution of geometric dimensions of fish schools is represented as @xmath47 where @xmath48 and @xmath49 therefore , the school - dimension distribution follows a power law with the same exponent `` @xmath50 '' as the school - biomass distribution . the power - law distribution @xmath51 is truncated at a cut - off size @xmath52 , which is also calculated from histogram data of school geometric dimensions , @xmath53^{\nu}. \label{eqn : depend - cutoff - nu}\ ] ] the following normalizations are adopted for the scaling function @xmath54 and the histogram data of geometric dimensions of fish schools , because the cut - off size @xmath38 is proportional to the total number of fish in the school system [ @xcite ; see also eq.([eqn : cutoff - population ] ) ] : @xmath55 and @xmath56 respectively . fat - tailed school - size distributions are necessarily truncated because the population is finite . since the size @xmath57 depends on the exponent @xmath7 following eq.([eqn : depend - cutoff - nu ] ) , so that we can determine the value of @xmath7 by evaluating the location of the cut - off in the power - law distribution @xmath51 . the ordinary least squares regressions might , however , lead to a `` wrong '' estimation of the exponent @xcite . i make use of the data collapse to extract the `` right '' exponent . from eqs.([eqn : dimension - distribution ] ) and ( [ eqn : dimension - distribution - normalization ] ) , when @xmath58 is plotted against @xmath59 with correct parameter @xmath7 , all the empirical data should collapse onto each other . the power - law exponent of school - dimension distributions , @xmath7 , is then evaluated through data collapse . let us search for the value of @xmath7 that places all the points most accurately on a single curve . we use a set of histogram data of vertical dimension of japanese sardine _ sardinops melanostictus _ schools , from 22 acoustic surveys by @xcite off southeastern hokkaido for seven years , 19811987 . @xcite reported that japanese sardine migrate as a huge - sized school in number from a few hundreds of thousands to a few million of fish . to obtain the best data collapse , the @xmath60-axis is divided into bins ( fig.[fig:2]a ) , and for each bin two - dimensional variance @xmath61 is calculated , where @xmath62 and @xmath63 denote the standard deviation of the mean @xmath64 and @xmath65 , respectively . the parameter @xmath7 is then estimated at value that minimize the mean of two - dimensional variance for the bins ( fig.[fig:2]b ) . the mean of two - dimensional variance , @xmath66 , is a measure to determine the goodness of collapse @xcite . a good data collapse can be obtained by using the value @xmath67 . the resulting plot of empirical school data is shown in fig.[fig:2]a . experimentally fitting the parameter @xmath7 to achieve a good data collapse , `` @xmath0 '' , is reminiscent of the flory value of the exponent in a power - law dependence of the coil radius of a polymer chain ( in three - dimensional solutions ) on the degree of polymerization @xcite . . ( a ) scaled distribution of geometric dimensions of sardine schools . @xmath68 is plotted versus @xmath69 with @xmath70 on double - logarithmic scale . the bins are chosen equally spaced on a logarithmic scale as @xmath71 with @xmath72 . the rectangle in gray reads the interval @xmath73 , i.e. the error @xmath63 on the mean @xmath65 ( indicated by the slit ) for each bin . the solid line is a prediction of the mean - field theory @xcite . ( b ) the mean of two - dimensional variance , @xmath66 , versus the power - law exponent @xmath7 . although @xmath66 shows noisy fluctuations , it takes a minimum around @xmath74 . data from @xcite are analyzed . , title="fig:",width=245 ] . ( a ) scaled distribution of geometric dimensions of sardine schools . @xmath68 is plotted versus @xmath69 with @xmath70 on double - logarithmic scale . the bins are chosen equally spaced on a logarithmic scale as @xmath71 with @xmath72 . the rectangle in gray reads the interval @xmath73 , i.e. the error @xmath63 on the mean @xmath65 ( indicated by the slit ) for each bin . the solid line is a prediction of the mean - field theory @xcite . ( b ) the mean of two - dimensional variance , @xmath66 , versus the power - law exponent @xmath7 . although @xmath66 shows noisy fluctuations , it takes a minimum around @xmath74 . data from @xcite are analyzed . , title="fig:",width=245 ] here we see that the power - law regime of the distribution is too short , which misled @xcite and @xcite into overlooking the power - law distributions of school sizes . notice that many power laws that appeared in the ecology literature span less than two orders of magnitude of scale @xcite . the power law range of too few scales is not unique to ecology ; the largest numbers of power laws reported in the physical science are for small ranges @xcite . the acoustic - survey data are converted into a school - biomass histogram as follows @xmath75 we now crosscheck the value of @xmath7 through finite - size scaling ( fss ) analysis of school - biomass distribution @xcite . since the finite population size causes the truncation of power - law distribution @xmath76 , there is a well - defined quantity @xmath77 which depends on the system population size . in order to characterize the finite size effects , fss hypothesis is used : the distribution function depends on @xmath3 only through the ratio @xmath78 , @xmath79 where @xmath31 is a universal function independent of fish population size . the prefactor @xmath80 is required to ensure the normalization @xmath81 where @xmath82 $ ] now represents the probability distribution of school - biomass sizes . from the fss hypothesis , it is expected that when @xmath83 is plotted against @xmath78 with correct parameters @xmath22 and @xmath84 all the data collapse onto a single curve . the power - law exponent of biomass distributions , @xmath85 , is then evaluated through fss analysis . the value of @xmath86 is the estimate of the power - law exponent @xmath87 let us analyze a set of 22 acoustic - survey data of sardine schools ( same as fig.[fig:2 ] ) converted into biomass histograms by using eq.([eqn : biomass - dimension - convert ] ) with @xmath88 . in a simultaneous best - fitting procedure ( fig.[fig:3 ] ) , a good data collapse can be obtained by using the values @xmath89 and @xmath90 . the power - law exponent derived from the fss collapse is @xmath91 . the resulting plot is shown in fig.[fig:3]a . the school - biomass distribution follows a power - law decay with exponent @xmath50 , and is truncated at the cut - off size of eq.([eqn : cut - off - size ] ) . the fss collapse confirms the scaling laws for school sizes , eq.([eqn : scaling - school - geometry ] ) with @xmath88 and eq.([eqn : biomass - distribution ] ) with @xmath37 . ) are converted by eq.([eqn : biomass - dimension - convert ] ) with @xmath88 . here @xmath92 is plotted versus @xmath93 with @xmath94 . two - dimensional variances [ same as eq.([eqn:2d - var ] ) ] are calculated for bins chosen equally spaced on a logarithmic scale as @xmath95 with @xmath96 . the rectangle in gray is same as fig.[fig:2 ] . the solid line is a prediction of the mean - field theory @xcite . ( b ) the region of the @xmath97-plane in which the minimum of the mean of two - dimensional variance exists . a measure of data collapse for scaling , i.e. the mean of two - dimensional variance , @xmath66 , takes a minimum @xmath98 for the right choice of @xmath99 . the minimum is found with the precision i.e. width of the minimum , @xmath100 in black region ( @xmath101 ) . the values of the parameters lie in the intervals @xmath102 and @xmath103 , and therefore @xmath104 . , title="fig:",width=245 ] ) are converted by eq.([eqn : biomass - dimension - convert ] ) with @xmath88 . here @xmath92 is plotted versus @xmath93 with @xmath94 . two - dimensional variances [ same as eq.([eqn:2d - var ] ) ] are calculated for bins chosen equally spaced on a logarithmic scale as @xmath95 with @xmath96 . the rectangle in gray is same as fig.[fig:2 ] . the solid line is a prediction of the mean - field theory @xcite . ( b ) the region of the @xmath97-plane in which the minimum of the mean of two - dimensional variance exists . a measure of data collapse for scaling , i.e. the mean of two - dimensional variance , @xmath66 , takes a minimum @xmath98 for the right choice of @xmath99 . the minimum is found with the precision i.e. width of the minimum , @xmath100 in black region ( @xmath101 ) . the values of the parameters lie in the intervals @xmath102 and @xmath103 , and therefore @xmath104 . , title="fig:",width=245 ] finite - size scaling techniques have been applied to ecology and shown to predict scaling relations between scaling exponents in relative abundance of species [ eq.([eqn : scaling - ecology ] ) ] @xcite , in dynamics of breeding bird populations @xcite , in biomass - size distributions of seston @xcite , and in exploitation of fish stocks @xcite . in school - size statistics , we expect to have a scaling relation . we now choose the normalization @xmath105 where @xmath106 denotes the total fish population in the school system ( @xmath107 gives the total number of schools ) . since eq.([eqn : sys - size - school ] ) implies @xmath108 with a scaling exponent @xmath109 , the fss relation for the school - biomass distribution is written as @xmath110 accordingly , the normalization of eq.([eqn : norm - phi ] ) yields the scaling relation @xmath111 the best - fitting value in data collapse , @xmath112 , gives @xmath113 . as a consequence , the location of the cut - off in the power - law distribution of school sizes simply reads @xmath114 which is verified by means of extensive numerical simulations @xcite . let us now investigate cohesive motion of schools in a three - dimensional space from the viewpoint of the behavioral algorithms which govern their formation and dynamics : attraction between neighbors maintains cohesion of the school ; a tendency to align with neighbors produces collective motion of the school . the minimal model of cohesion is a linkage of neighbors consisting of harmonic spring , because the inter - fish distance follows a gaussian distribution @xcite . using the relative coordinates to the center of school , the equation of motion of fish ( as noisy self - propelled particles ) in a large school of size @xmath3 is written by a one - body approximation as the following langevin equation @xcite : @xmath115 providing that individuals are sufficiently sensitive to behavior of their neighbors , where @xmath116 is the cohesive force ; @xmath117 denotes the strength of alignment ; and @xmath118 is a random perturbation of the velocity with strength @xmath119 and a @xmath120-correlated time dependence . the attractive interaction force acting on one body due to a system of bodies is the neighbor - joining harmonic spring . one essential approximation is to replace the many - fish problem by the problem of solving the motion of one fish in a certain self - consistent field . the cohesive force is then written in the following form by the one - body approximation : @xmath121 where @xmath122 is the effective spring constant for the springs strung out from the center of the school to a fish : the number of consecutive springs joining the fish to the center of the school via other companions is proportional to @xmath123 . the total `` elastic energy '' of inter - fish bonds in the school depends linearly on number @xmath3 of individuals in the school . the overall elastic energy after integration over a sphere of radius @xmath14 ( denoting the average radius of the school ) results in @xmath124 , which is derived from the ideal harmonic spring , eq.([eqn : cohesion ] ) . therefore , the exponent @xmath125 will be determined self - consistently , @xmath126 the solution of the langevin equation ( [ eqn : langevin ] ) with eq.([eqn : cohesion ] ) takes the following asymptotic forms @xcite : @xmath127 \approx { 3\epsilon_v}/j , \label{eqn : fluctuation - dissipation}\ ] ] @xmath128 where the root - mean - square @xmath129 gives the average radius of the school , @xmath14 . from eq.([eqn : self - consistent - spring ] ) together with eq.([eqn : mean - square - radius ] ) , the self - consistent value of the exponent @xmath125 is obtained : @xmath130 as a consequence , the self - consistent cohesive force reads @xmath131 where @xmath4 denotes the effective inter - fish distance ( a constant independent of @xmath3 ) : @xmath132 . the average radius of the school is then given by @xmath133 . note that eq.([eqn : fluctuation - dissipation ] ) is an example of a more general principle called the fluctuation - dissipation theorem @xcite . notice that the above model has been developed in the absence of the excluded volume interactions . social cohesion by harmonic spring interactions between neighboring fish reminds us of the classical picture of a polymer chain based on the bead - spring model @xcite . the exponent @xmath7 close to the value @xmath0 may then be understood by taking the excluded volume effect @xcite . such a school without excluded volume effect may be called the `` gaussian '' school in line with polymer physics . from eq.([eqn : langevin ] ) , the probability of the school radius being between @xmath134 and @xmath135 is given by the following @xcite : @xmath136 in a stationary state in three dimensions ( i.e. the position vector @xmath137 follows a gaussian distribution ) . the gaussian school model considered above permits fish to occupy the same region in space . of course this is a physical impossibility since each fish possesses its own finite volume . therefore , in the `` excluded volume '' school , there are a number of gaussian school configurations which are disallowed due to the steric effect . let @xmath138 be the probability that a gaussian school configuration , as counted in eq.([eqn : prob - gaussian - radius ] ) , is also allowable under the excluded volume condition . we now calculate the probability that no overlaps occur when we place @xmath3 fish within a region of volume ( @xmath139 ) , which will lead to an estimation for @xmath138 . the approach is due to @xcite . letting @xmath140 be the volume which is effectively excluded to one fish by the presence of another ( @xmath141 ) , the probability that one particular fish will not overlap with another is then given by @xmath142 . since there are @xmath143 possible combinations of pairs , the probability that no overlap occurs in all of these combinations is given by @xmath144 where @xmath145 and @xmath146 . therefore , the probability distribution of the school radius @xmath134 can then be written as @xmath147 for the excluded volume school . both @xmath148 and @xmath149 have a maximum at certain values of @xmath134 . let us estimate the average size of the school radius in each model by calculating the positions of these maxima . the maximum of @xmath148 occurs at @xmath150 . the maximum of @xmath149 occurs at @xmath14 , which satisfies the following equation obtained by differentiating the logarithm of eq.([eqn : prob - excluded - radius ] ) : @xmath151 if @xmath152 , the second term on the left - hand side of eq.([eqn : diff - log ] ) can be neglected and hence @xmath153 the exponent @xmath0 is the very value extracted through the data collapse . the scaling in the relationship between geometric dimensions and biomass of pelagic fish schools is analogous to that developed in polymer physics @xcite . the result suggests that the dynamics are unexpectedly `` simple '' and depend primarily on common cohesive motion in these animate and inanimate systems . though the above is a very rough theory of school conformation with the excluded volume effect , the overall statistical properties do not depend on the details of the model , which is a consequence of universality @xcite . as indicated in the introduction , large internal variations in packing density occur within a school . @xcite pointed out that a source of substantial variation in internal school structure is the formation of subgroups . such subgroups have been observed in saithe schools @xcite , and in schools of minnow _ phoxinus phoxinus _ @xcite and herring @xcite . relatively independent movements of such clusters of individuals can open up empty spaces and cause large variation in school volume . the polymer - chain analogue of the subgroup in school conformation is the `` blob '' @xcite : the polymer chain behaves as a series of blobs . based on the blob concept , for small value of the excluded volume @xmath140 , a subgroup within a school , with a number of @xmath154 of fish , must be nearly gaussian . we see this from eq.([eqn : prob - excluded - radius ] ) when we find no effect of @xmath140 if @xmath155 . there is a certain value of @xmath156 $ ] , beyond which excluded volume effects become important . a single school will appear gaussian at scales @xmath157 where @xmath158 while at scales @xmath159 it will show excluded volume effects . according to the `` blob '' approach , an @xmath3-sized school can be described as a cluster of @xmath160 subgroups . inside the subgroup the core repulsion ( by excluded volume effects ) is a weak perturbation leading to a gaussian state . the school radius is then written as @xmath161 in the native state . notice the plasticity of school geometry and internal structure . the compact , dense packing , as observed in artificial environments like small tanks , may be caused by the very strong stress . when the school is confronted by danger such as predators , the inter - fish distance decreases rapidly and all the vacuoles within the school collapse quickly @xcite . this change may be closely related to the coil - globule transition in a polymer solution @xcite . in real schools the nature of the short range interaction is quite complicated like van der waals - type inter - molecular forces . accordingly , all fish within a school interact via the two - body potential deformed from a parabolic potential of harmonic spring [ grouping forces have been expressed as the gradient of a potential function in previous model studies @xcite ] . the potential change , denoted by @xmath162 , is given as the correction term to cohesive interaction potential between the @xmath163-th and the @xmath164-th fish : @xmath162 will include steric effects and also may involve weak attractions . the correction factor to the probability distribution @xmath149 of the school radius @xmath134 , in place of eq.([eqn : prob - no - overlap-1 ] ) , is then given by the configuration integral @xmath165 where @xmath166 denotes the position vector of the @xmath163-th fish , and @xmath167 is integrated over the school configuration space . we can estimate @xmath138 by means of the cluster expansion @xcite . introducing a function of separation between pairs , @xmath168 , @xmath169 we integrate eq.([eqn : prob - no - overlap ] ) in the mayer cluster expansion over diagrams up to the second irreducible cluster , yielding @xmath170 with the excluded volume parameter @xmath171 and the other parameter defined by @xmath172 the actual configuration is decided by maximizing the distribution function @xmath173 , and this maximization condition leads to the equation determining the school radius @xmath14 , @xmath174 notice that , when the first irreducible cluster integral in mayer s expansion is only taken into account , the configuration integral , eq.([eqn : prob - no - overlap ] ) , is evaluated by eq.([eqn : prob - no - overlap-1 ] ) . in case that @xmath162 consists of a hard core repulsion and a short - range weak attraction ( i.e. at the distance of near - collision ( @xmath175 ) the attraction gradient between fish becomes steeper ) , eq.([eqn : second - virial ] ) is estimated as @xmath176 , expressing the dependence of @xmath140 on the velocity - fluctuation @xmath119 of fish , where @xmath177 and @xmath178 are constants @xcite . during predation threats or laboratory observations , the value of excluded volume parameter @xmath140 will change sign from positive to negative , provided that the velocity fluctuation @xmath119 is suppressed below a certain value @xmath178 . decreasing @xmath119 below @xmath178 , the dimensional size of the school becomes much smaller than that of a gaussian school , as depicted in fig.[fig:4 ] . according to eq.([eqn : diff - log-2 ] ) the so called `` expansion factor '' @xmath179 is determined not by @xmath140 but by @xmath180 , and so if @xmath3 is large only a small change in @xmath119 will cause a big change in dimensional size . for example , for school of @xmath16 fish , a small variation in @xmath119 will induce a dramatic change in the school radius . equation ( [ eqn : diff - log-2 ] ) states that when @xmath181 , @xmath179 is proportional to @xmath182 and hence the solution is as follows : @xmath183 which shows the closest packing of fish in a school , i.e. the constant density independent of the number of individuals . this change may be called the swollen - dense packing transition of schools . ) ] . here @xmath184 is plotted as a function of a combined variable @xmath185 . shown is the transition region @xmath186 . at @xmath187 the excluded volume parameter @xmath140 equals @xmath188 . the repulsive excluded volume effect then balances the attractive forces between fish , and the school behaves as a gaussian school . the property of the swollen - dense packing transition is quite different in the two regions separated by the critical value @xmath189 . the transition between the the swollen conformation and the compact packing is gradual when @xmath190 ( dotted line ) , while it becomes extremely sharp and the transition is discontinuous when @xmath191 ( gray line ) . the solid line shows the marginal transition behavior , and the crossover takes place when @xmath192 ( indicated by the arrow ) . notice that the solution of eq.([eqn : diff - log-2 ] ) is multivalued [ depicted by a sigmoid curve ( white cutout ) ] , showing a first - order discrete phase change @xcite , when @xmath191 . , width=245 ] the approach of treating fish schools as interacting particle systems naturally leads to the idea of applying successful methods of statistical physics to the description of moving together without a leader @xcite . in statistical physics , the presence of non - trivial scaling is usually taken to mean that the dynamics are largely governed by simple geometric properties of the system and do not depend strongly on detailed properties of the system components @xcite . as for the school geometry , the basic units are the effective inter - fish length @xmath4 and the number of fish , @xmath3 . we assemble @xmath193 neighboring fish in a school into @xmath194 groups , e.g. in the sense of the voronoi - dirichlet diagram @xcite . the length between centers of neighboring subgroups is defined by @xmath195 with an exponent @xmath196 . after such a scale change as @xmath197 the macroscopic quantities which determine the overall properties of the school satisfy @xmath198 where @xmath199 is an exponent which depends on the quantity under consideration . for this to hold true for arbitrary @xmath193 , the function @xmath200 must take the scaling form @xmath201 where a new function @xmath202 is introduced . this is the widom - kadanoff scaling law in critical phenomena @xcite . for example , the average radius of the school , @xmath14 , should be unaltered under the transformation of eq.([eqn : scale - change ] ) : @xmath203 because of eq.([eqn : scaling - school - geometry ] ) , @xmath204 and we must have @xmath205 where @xmath206 in three dimensions . as an example of the application of the above scaling law , let us consider the pair correlation function inside a school , where fish movement take place in a three - dimensional space . a pair correlation function @xmath207 is defined as follows . we pick one fish at random in the school , and we place it at origin . then we ask , what is the number density of other fish at a ( directional ) distance @xmath137 from the first , and we average the result over all choices of the first fish ( and of directions ) . the function @xmath207 has an integral which is just the total number of fish per school , @xmath208 . hence , from dimensional analysis , we can write @xmath209 under the transformation that neighboring @xmath193 fish are grouped to form one subunit in a school , @xmath207 will be reduced by @xmath210 , since the pair correlation function is proportional to the number density of fish . therefore @xmath211 then , the function @xmath207 obeys a simple scaling rule @xmath212 where @xmath213 is a dimensionless universal function , and @xmath214 is the dimensional factor . focusing on the limit @xmath215 , we can reach the form of @xmath207 by a simple argument . in a sphere of radius @xmath134 we have a certain number of fish , @xmath216 , related to @xmath134 by the excluded volume exponent : @xmath217 . the function @xmath207 scales like the density of fish in the sphere , @xmath218 which gives @xmath219 . the possible observable property , eq.([eqn : pair - correlation - scaling ] ) , tells us that if we were to measure @xmath207 for fish schools of different size @xmath3 , there would be superposition of the curves obtained by plotting @xmath220 against @xmath221 . thus this kind of scaling relation may be verified experimentally on large schools of pelagic fish . the power - law scaling generally exists in dimension - to - biomass relationship of pelagic fish schools in nature . here i have estimated the power - law exponent @xmath7 of the geometric relation , based on the dependence of the power - law regime of the school - dimension distribution @xmath51 on @xmath7 , i.e. eq.([eqn : depend - cutoff - nu ] ) . we have tested whether the distributions @xmath51 are self - similar ( i.e. exhibit scaling ) , relying on both the power law for the dimension - to - biomass relationship of eq.([eqn : scaling - school - geometry ] ) and the fss relation in the power - law school - biomass distribution of eq.([eqn : biomass - distribution ] ) . plotting the scaled histogram - data from the 22 _ in - situ _ observations , we have found that the 22 curves do indeed collapse onto each other ( fig.[fig:2]a ) , suggesting that @xmath51 follows a universal functional form [ eq.([eqn : dimension - distribution ] ) ] . we have extracted the power - law exponent of dimension - to - biomass scaling relationship via a minimization of a measure to quantify the nature of scaling collapse . next i have explained the value of the exponent @xmath7 , proposing the gaussian school model for the fish with excluded volume @xmath140 . by exploiting the analogy between fish - school and polymer conformations , we have examined the behavioral algorithm governing the swollen conformation of large - sized schools . we have seen that the exponent is modified strongly by the steric effect . what is universal in eq.([eqn : scaling - school - geometry ] ) is the exponent @xmath222 : it is independent of species as well as environmental conditions , and the same for all schools . the constant that multiplies @xmath223 in eq.([eqn : scaling - school - geometry ] ) is non - universal [ @xmath224 with effective inter - fish distance @xmath4 , predicted from eq.([eqn : rouse - excluded - volume - effect ] ) ] , and depends on the details of interactions between fish which may vary with species and environmental conditions @xcite . to understand the scaling law for school geometry , eq.([eqn : scaling - school - geometry ] ) with @xmath88 , it is essential to see what the value of @xmath7 represents . obviously , for a regular object embedded into a @xmath225-dimensional euclidean space , eq.([eqn : scaling - school - geometry ] ) would have the form @xmath226 expressing the fact that the volume of a @xmath225-dimensional object grows with its linear size @xmath14 as @xmath227 . contrary to an integer dimensionality , it has been shown that the packing structure within the schools is characterized by a non - integer ( i.e. fractal ) dimensionality of @xmath228 : the number of fish in a school of radius @xmath14 scales as @xmath229 . during the last decades of the twentieth century it has widely been recognized by researchers working in diverse areas of science that many of the structures commonly observed possess a rather special kind of geometrical complexity ; the name `` fractal '' was coined by benoit mandelbrot for these complex shapes @xcite . objects of biological origin are many times fractal - like @xcite . fss plots of data lead to possibly important and practical insights in ecology . in the context of the fisheries mission , the demonstrated geometric relation between dimensions and biomass of pelagic fish schools , when applied to mapping pelagic schooling fish , will largely improve the precision in the fish stock assessment . because of the linear dependence of the fish population on @xmath38 , as given by eq.([eqn : cutoff - population ] ) , the fish stock abundance can be inferred from an index ( i.e. cut - off size ) that can be determined directly from observations [ see eq.([eqn : cut - off - size ] ) ] . 99 aji , v. , goldenfeld , n. , 2001 . fluctuations in finite critical and turbulent systems . 86 , 10071010 . , doi:10.1103/physrevlett.86.1007 . anderson , j. j. , 1981 . a stochastic model for the size of fish schools . fish . bull . 79 , 315323 . aoki , i. , 1980 . an analysis of the schooling behavior of fish : internal organization and communication process . ocean res . tokyo 12 , 165 . aoki , i. , inagaki , t. , 1988 . photographic observations on the behavior of japanese anchovy _ engraulis japonica _ at night in the sea . 43 , 213221 . azovsky , a. i. , 2000 . concept of scale in marine ecology : linking the words or the worlds ? web ecol . 1 , 2834 . banavar , j. r. , green , j. l. , harte , j. , maritan , a. , 1999 . finite size scaling in ecology . 83 , 42124214 . beecham , j. a. , farnsworth , k. d. , 1999 . animal group forces resulting from predator avoidance and competition minimization . 198 , 533548 . binder , k. , heermann , d. w. , 1988 . monte carlo simulation in statistical physics . springer - verlag , berlin . bonabeau , e. , dagorn , l. , 1995 . possible universality in the size distribution of fish schools . e 51 , r5220r5223 . bonabeau , e. , dagorn , l. , fron , p. , 1998 . space dimension and scaling in fish school - size distributions . a : math . gen . 31 , l731l736 . bonabeau , e. , dagorn , l. , fron , p. , 1999 . scaling in animal group - size distribution . usa 96 , 44724477 . breder , jr . , c. m. , 1954 . equation descriptive of fish schools and other animal aggregations . ecology 35 , 361370 . camacho , j. , guimer , r. , amaral , l. a. n. , 2002 . robust patterns in food web structure . 88 , 228102 . , doi:10.1103/physrevlett.88.228102 . camazine , s. , deneubourg , j .- l . , franks , n. r. , sneyd , j. , theraulaz , g. , bonabeau , e. , 2001 . self - organization in biological systems . princeton university press , princeton . chave , j. , levin , s. , 2003 . scale and scaling in ecological and economic systems . resource econ . 26 , 527557 . coetzee , j. , 2000 . use of a shoal analysis and patch estimation system ( shapes ) to characterise sardine schools . living resour . 13 , 110 . cushing , d. h. , 1977 . observations of fish schools with the arl scanner . rapp . p .- v run mer 170 , 1520 . de gennes , p .- g . , 1975 . collapse of a polymer chain in poor solvents . j. de physique 36 , l55l57 . de gennes , p .- g . , 1979 . scaling concepts in polymer physics . cornell university press , ithaca , new york . dill , l. m. , dunbrack , r. l. , major , p. f. , 1981 . a new stereophotographic technique for analyzing the three - dimensional structure of fish schools . biol . fish . dirichlet , g. l. , 1850 . ber die reduktion der positiven quadratischen formen mit drei unbestimmten ganzen zahlen . j. reine angew . 40 , 209227 . doi , m. , 1996 . introduction to polymer physics . oxford university press , new york . doi , m. , edwards , s. f. , 1986 . the theory of polymer dynamics . oxford university press , new york . fisher , m. e. , barber , m. n. , 1972 . scaling theory for finite - size effects in the critical region . 28 , 15161519 . flory , p. , 1953 principles of polymer chemistry . cornell university press , ithaca , new york . fron , p. , gerlotto , f. , soria , m. , 1992 . changes in school structure according to external stimuli : description and influence on acoustic assessment . fish . 15 , 4566 . fron , p. , misund , o. a. , 1999 . dynamics of pelagic fish distribution and behaviour : effects on fisheries and stock assessment . fishing news books , blackwell , osney mead , oxford . goldenfeld , n. , 1992 . lectures on phase transitions and the renormalization group . vol.85 of frontiers in physics . addison - wesley , reading , massachusetts . graves , j. , 1976 . photographic method for measuring spacing and density within pelagic fish schools at sea . u.s . 75 , 230234 . hamburger , d. a. , biham , o. , avnir , d. , 1996 . apparent fractality emerging from models of random distributions . e 53 , 33423358 . halley , j. m. , hartley , s. , kallimanis , a. s. , kunin , w. e. , lennon , j. j. , sgardelis , s. p. , 2004 . uses and abuses of fractal methodology in ecology . . lett . 7 , 254271 . , doi:10.1111/j.1461 - 0248.2004.00568.x . hara , i. , 1984 . distribution and school size of japanese sardine in the waters off the southeastern coast of hokkaido on the basis of echo sounder surveys . tokai reg . fish . res . lab . 113 , 6778 . hara , i. , 1986 . stock assessment of japanese sardine in the waters off the southeast coast of hokkaido using line transect method . sci . fish . 52 , 6973 . hara , i. , 1990 . stock assessment and migration of japanese sardine schools in the waters off the southeastern coast of hokkaido . ph.d . thesis , kyoto university , kyoto , japan . harte , j. , kinzig , a. , green , j. , 1999 . self - similarity in the distribution and abundance of species . science 284 , 334336 . higashi , m. , yamamura , n. , 1993 . what determines animal group size ? insider - outsider conflict and its resolution . 142 , 553563 . hoare , d. j. , couzin , i. d. , godin , j .- j. , krause , j. , 2004 . context - dependent group size choice in fish . 67 , 155164 . , doi:10.1016/j.anbehav.2003.04.004 . hori , j. , 1977 . langevin equation . iwanami shoten , publ . , keitt , t. h. , stanley , h. e. , 1998 . dynamics of north american breeding bird populations . nature 393 , 257260 . keitt , t. h. , amaral , l. a. n. , buldyrev , s. v. , stanley , h. e. , 2002 . scaling in the growth of geographically subdivided populations : invariant patterns from a continent - wide biological survey . b 357 , 627633 . , doi:10.1098/rstb.2001.1013 . kubo , r. , 1966 . the fluctuation - dissipation theorem . 29 ( 1 ) , 255284 . lillo , f. , farmer , j. d. , mantegna , r. n. , 2002 . single curve collapse of the price impact function for the new york stock exchange . preprint cond - mat/0207428 at @xmath230 lillo , f. , farmer , j. d. , mantegna , r. n. , 2003 . master curve for price - impact function . nature 421 , 129130 . macarthur , r. h. , wilson , e. o. , 1967 . the theory of island biogeography . princeton university press , princeton . mandelbrot , b. b. , 1982 . the fractal geometry of nature . w. h. freeman and co. , new york . may , r. m. , 1999 . unanswered questions in ecology . b 354 , 19511959 . mayer , m. g. , mayer , j. e. , 1940 . statistical mechanics . john wiley & sons , new york . misund , o. a. , 1990 . sonar observations of schooling herring : school dimensions , swimming behaviour , and avoidance of vessel and purse seine . p .- v run . int . explor . mer 189 , 135146 . misund , o. a. , 1993a . dynamics of moving masses : variability in packing density shape and size among herring sprat and saithe schools . ices j. mar . 50 , 145160 . misund , o. a. , 1993b . abundance estimation of fish schools based on a relationship between school area and school biomass living resour . 6 , 235241 . misund , o. a. , aglen , a. , beltestad , a. k. , dalen , j. , 1992 . relationship between the geometric dimensions and biomass of schools . ices j. mar . 49 , 305315 . misund , o. a. , aglen , a. , frns , e. , 1995 . mapping the shape , size , and density of fish schools by echo integration and a high - resolution sonar . ices j. mar . 52 , 1120 . misund , o. a. , aglen , a. , hamre , j. , ona , e. , rttingen , i. , skagen , d. , valdemarsen , j. w. , 1996 . improved mapping of schooling fish near the surface : comparison of abundance estimates obtained by sonar and echo integration . ices j. mar . 53 , 383388 . misund , o. a. , coetzee , j. , 2000 . recording fish schools by multi - beam sonar : potential for validating and supplementing echo integration recordings of schooling fish . 47 , 149159 . , doi:10.1016/s0165 - 7836(00)00166 - 1 . misund , o. a. , coetzee , j. , fron , p. , gardener , m. , olsen , k. , svellingen , i. , hampton , i. , 2003 . schooling behaviour of sardine _ sardinops sagax _ in false bay , south africa . 25 , 185193 . misund , o. a. , floen , s. , 1993 . packing density structure of herring schools . 196 , 2629 . mogilner , a. , edelstein - keshet , l. , bent , l. , spiros , a. , 2003 . mutual interactions , potentials , and individual distance in a social aggregation . 47 , 353389 . , doi:10.1007/s00285 - 003 - 0209 - 7 . morgan , m. j. , 1988 . the effect of hunger , shoal size and the presence of a predator on shoal cohesiveness in bluntnose minnows , _ pimephales notatus _ rafinesque . 32 , 963971 . niwa , h .- s . , 1994 . self - organizing dynamic model of fish schooling . 171 , 123136 . niwa , h .- s . , 1996a . newtonian dynamical approach to fish schooling . 181 , 4763 . niwa , h .- s . , 1996b . mathematical model for the size distribution of fish schools . computers math . 32 ( 11 ) , 7988 . niwa , h .- s . , 1998 . school size statistics of fish . 195 , 351361 . niwa , h .- s . , 2003 . power - law versus exponential distributions of animal group sizes . 224 , 451457 . , doi:10.1016/s0022 - 5193(03)00192 - 9 . niwa , h .- s . space - irrelevant scaling law for fish school sizes . 228 , 347357 . , doi:10.1016/j.jtbi.2004.01.011 . niwa , h .- s . , 2004b . exploitation dynamics of fish stocks . meeting abstracts of the physical society of japan 59 ( 2 ) , part 2 , p.201 . okubo , a. , chiang , h. c. , 1974 . an analysis of the kinematics of swarming of _ anarete pritchardi _ kim ( diptera : cecidomyiidae ) . 16 , 142 . okubo , a. , levin , s. a. , 2001 . diffusion and ecological problem : modern perspectives . springer - verlag , new york . parr , a. e. , 1927 . a contribution to the theoretical analysis of the schooling behavior of fishes . bulletin of the bingham oceanographic collection 1 ( 1 ) , 132 . parrish , j. k. , hamner , w. m. ( eds . ) , 1997 . animal groups in three dimensions . cambridge university press , cambridge . partridge , b. l. , 1980 . the effect of school size or the structure and dynamics of minnow schools . 28 , 6877 . partridge , b. l. , 1981 . internal dynamics and interrelations of fish in schools . 144 , 313325 . partridge , b. l. , 1982 . the structure and function of fish schools . 246 , 9099 . pitcher , t. j. , 1973 . the three - dimensional structure of fish schools in the minnow , _ phcxinus phcxinus_. anim . 21 , 673686 . pitcher , t. j. , parrish , j. k. , 1993 . functions of shoaling behaviour in teleosts , in : pitcher , t. j. ( ed . ) , behaviour of teleost fishes . chapman & hall , london , ch.12 , pp.363439 . pitcher , t. j. , partridge , b. l. , 1979 . fish school density and volume . 54 , 383394 . preston , f. w. , 1962 . the canonical distribution of commonness and rarity : part i , and part ii . ecology 43 , 185215 , and 410432 . ptitsyn , o. b. , kron , a. k. , eizner , y. y. , 1968 . the models of the denaturation of globular proteins . i. theory of globula - coil transitions in macromolecules . c 16 , 35093517 . pulliam , h. r. , caraco , t. , 1984 . living in groups : is there an optimal group size ? , in : krebs , j.r . , davies , n.b . ( eds . ) , behavioural ecology . an evolutionary approach . blackwell science , oxford , pp . 122147 . rinaldo , a. , maritan , a. , cavender - bares , k. k. , chisholm , s. w. , 2002 . cross - scale ecological dynamics and microbial size spectra in marine ecosystems . b 269 , 20512059 . , doi : 10.1098/rspb.2002.2102 . rouse , p. e. , 1953 . a theory of the linear viscoelastic properties of dilute solutions of coiling polymers . 21 , 12721280 . stanley , h. e. , 1995 . power laws and universality . nature 378 , 554 . symons , p. e. k. , 1971 . spacing and density in schooling threespine sticklebacks ( _ gasterosteus aculeatus _ ) and mummichog ( _ fundulus heteroclitus _ ) . 28 , 9991004 . tauti , m. , miyosi , k. , 1929 . on the movement of a fish - group . i. on the direction of a fish - group on passing the end of a wall along which it has moved . j. imperial fish . 24 , 7376 . tauti , m. , hudino , t. , 1929 . on the movement of a fish - group . the case when fishes were barred from their progress by a wall . j. imperial fish . inst . jpn . 24 , 7679 . tauti , m. , yasuda , h. , 1929 . on the movement of a fish - group . the case when fishes were barred from their progress by a net . j. imperial fish . 24 , 8086 . tauti , m. , yasuda , h. , 1930 . on the movement of a fish - group . the shape of the entrance of fishing net and the fish - group moving in and out . j. imperial fish . 25 , 4150 . tauti , m. , yasuda , h. , 1933a . the direction in which a fish - group goes on after having met with a net . 2 , 1719 . tauti , m. , yasuda , h. , 1933b . how will a fish - group be attracted ? 2 , 124127 . uhlenbeck , g. e. , ford , g. w. , 1962 . the theory of linear graphs with applications to the theory of the virial development of the properties of gases , in : de boer , j. , uhlenbeck , g. e. ( eds . ) , studies in statistical mechanics . north - holland , amsterdam , vol.1 , ch.2 , pp.119211 . vicsek , t. , 2001 . a question of scale . nature 411 , 421 . vicsek , t. ( ed . ) , 2001 . fluctuations and scaling in biology . oxford university press , oxford . voronoi , g. m. , 1908 . nouvelles applications des paramtres continus la thorie des formes quadratiques deuxime mmoire : rechrches sur les parallllodres primitifs . j. reine angew . 134 , 198287 . voronoi , g. m. , 1909 . deuxime mmoire : recherches sur les parallloedres primitifs . j. reine angew . 136 , 67181 . warburton , k. , lazarus , j. , 1991 . tendency - distance models of social cohesion in animal groups . 150 , 473488 . wilson , k. g. , 1983 . the renormalization group and the critical phenomena . 55 , 583 - 600 .

over the last decades , efforts have been made to get a better understanding of non - gaussian time series in the time and frequency domains . in particular , time series whose marginal distributions exhibit power law behavior have attracted a lot of attention . the need for such models arises from applications in areas as diverse as insurance , geophysics , finance and telecommunications . infinite fourth moments are not untypical for series of daily log - returns from exchange rates , stock indices , and other speculative prices , whereas infinite second moments can be observed in time series from insurance such as for windstorm , industrial fire and earthquake insurance @xcite . infinite first moments are typical for the marginal distribution of the magnitudes of earthquakes @xcite . infinite variances are observed for the sizes of teletraffic data in the world wide web @xcite ; see also the recent books @xcite . classical time series analysis deals with the second ( or higher ) order moment structure of a stationary sequence . heavy - tailed modeling requires , in addition , that one takes into account the interplay between the dependence structure and the tails of the series . an important task is to understand the classical statistical estimators and test procedures when big shocks to the underlying system are present . when the marginal distributions have infinite variance , the notions of autocovariance , autocorrelation and spectral distribution lose their meaning . however , various studies over the last twenty years have shown that the analysis of linear processes @xmath3 , @xmath4 , with heavy - tailed i.i.d . _ innovations _ @xmath5 and constant coefficients @xmath6 is very similar to classical ( finite variance ) time series analysis , where notions such as autocovariances and spectral density are defined only in terms of the @xmath7 s and the innovation variance @xmath8 . most estimators and test statistics from classical time series analysis can be modified insofar that one considers self - normalized ( or studentized ) versions of them and for these versions , an asymptotic theory exists which parallels the classical theory with gaussian limit processes . in contrast to the latter theory , the limits involve infinite variance stable distributions and processes @xcite . one of the main goals of classical ( finite variance ) time series analysis is to study the spectral properties of the linear process @xmath9 . in this context , the _ periodogram _ @xmath10 , \ ] ] plays a prominent role as an estimator of the spectral density . numerous estimation and test procedures are based on this statistic and integrated versions of the form @xmath11 for appropriate classes of real - valued functions @xmath12 on @xmath13 $ ] . in applications , the class @xmath14}\dvt x\in[0,\curpi]\}$ ] is most important . the resulting integrated periodogram is a process indexed by @xmath15 $ ] . under general conditions , @xmath16})$ ] converges uniformly with probability 1 to the function @xmath17 , where @xmath18 , \end{aligned}\ ] ] and @xmath19 is the corresponding _ power transfer function_. the latter is one of the essential building blocks of the _ spectral density _ of the stationary process @xmath9 : @xmath20 .\ ] ] this is the fourier series based on the _ autocovariance function _ @xmath21 since @xmath16})$ ] estimates the spectral distribution function of the stationary process @xmath9 , it has been used for a long time as the _ empirical spectral distribution function _ , both as an estimator and a basic tool for constructing goodness - of - fit tests for the underlying spectral distribution function @xcite . since the limit process of the properly centered and normalized process @xmath16})$ ] depends on the ( in general unknown ) spectral density @xmath22 , bartlett @xcite proposed to consider the integrated periodogram based on @xmath23}/f_x(x)\dvt x\in[0,\curpi]\}$ ] . under general conditions , this process converges uniformly , with probability 1 , to the function @xmath24 and the limit process can be shown to be independent of the coefficients of the linear process , but depends on the fourth moment of @xmath25 . more generally , weighted integrated periodograms based on the classes @xmath26 } g(x)\dvt x \in[0,\curpi]\}$ ] for suitable functions @xmath27 are used to estimate the spectral density or to perform various tests concerning the spectrum of the underlying stationary sequence @xcite . the weighted integrated periodogram is also the basis of one of the classical estimators for fitting arma and fractional arima models . this method goes back to early work by whittle @xcite . in this context , one considers the functional @xmath28 , @xmath29 , where @xmath30 is a class of spectral densities indexed by a parameter @xmath31 . the _ whittle estimator _ @xmath32 of the true parameter @xmath33 is the minimizer of @xmath28 over @xmath34 . this estimation technique is one of the backbones of quasi - maximum likelihood estimation in parametric time series modeling . the whittle estimator is known to be asymptotically equivalent to the corresponding least - squares and gaussian quasi - maximum likelihood estimators @xcite . when proving the asymptotic normality and consistency of @xmath32 , one has to study the properties of the sequence @xmath35 which , again , can be considered as a weighted integrated periodogram indexed by a class of functions . the above examples have in common that one always considers a stochastic process @xmath36 for some class of functions . in all cases , one is interested in the asymptotic behavior of the process @xmath37 , uniformly over the class @xmath38 . this is analogous to the case of the empirical distribution function indexed by classes of functions . general references in this context are the monographs @xcite . early on , this analogy was discovered by dahlhaus @xcite , who gave the uniform convergence theory for @xmath37 under entropy and exponential moment conditions . the almost sure and weak convergence theory under entropy and power moment conditions was given in @xcite . a recent survey of nonparametric statistical methods related to the empirical spectral distribution indexed by classes of functions is @xcite . it is the aim of this paper to develop an analogous weak convergence theory for heavy - tailed stationary processes . we will understand ` heavy - tailedness ' in the sense of infinite variance of the marginal distributions . our focus will be on linear processes @xmath9 with i.i.d . symmetric @xmath0-stable ( s@xmath0s ) innovations @xmath39 for some @xmath40 . recall that a random variabl @xmath41 has a _ symmetric stable distribution _ @xmath42 if there are parameters @xmath43 , and @xmath44 such that its characteristic function has the form latexmath:[$e \mathrm{e}^{\mathrm{i } t y_{\alpha } } = \mathrm{e}^{-\sigma ^\alpha of @xmath25 . for @xmath46 , the random variable @xmath25 is known to have infinite variance @xcite . much of the theory given below depends on tail estimates for random quadratic forms in i.i.d . infinite variance random variables . such results are available for i.i.d . stable sequences . although it seems feasible that the theory can be extended to the more general class of processes whose innovations have regularly varying tails , we do not attempt to achieve this goal . the price would be more technicalities and the gain would be negligible . we intend to show how the classical ( finite variance ) tools and methods have to be modified in the infinite variance stable situation which can be considered as a boundary case of the classical one when some of the innovations assume extremely large values . by now , there exists quite a clear picture concerning the asymptotic theory of the sample autocovariances , the periodogram and its integrated versions when the innovation sequencein a linear process has infinite variance ; see @xcite , chapter 7 . in addition to the latter reference , goodness - of - fit tests for heavy - tailed processes ( corresponding to the class @xmath47 ) were considered for short- and long - memory linear processes @xcite , and whittle estimation for infinite variance arma and farima processes was also studied @xcite . the paper is organized as follows . in section [ sec:1 ] , we introduce some useful notation for the integrated periodogram . our main goal is to prove the weak convergence of the integrated periodograms indexed by suitable classes of functions . we achieve this goal for an i.i.d . sequence in section [ sec : iid ] , first by showing the _ convergence of the finite - dimensional distributions _ ( section [ s1 ] ) , then the _ tightness_. the conditions and methods are rather different in the cases @xmath1 ( section [ s2 ] ) and @xmath2 ( section [ s3 ] ) . the case @xmath1 is treated in the more general context of random quadratic forms with toeplitz coefficient matrices satisfying some summability condition . the case @xmath2 requires entropy conditions and the corresponding techniques . in section [ sextra ] , we extend the limit theory for the integrated periodograms from an i.i.d . sequence to linear processes . the contains some auxiliary results concerning tail estimates of random quadratic forms in stable random variables . the weak convergence results of this paper might also be of separate interest in the context of infinite variance random quadratic forms . the theory for such quadratic forms is not well studied . we also refer to an extended version of this paper @xcite which covers the class of stochastic volatility processes with regularly varying marginal distributions . the following decomposition of the periodogram is fundamental : @xmath48 where @xmath49 , @xmath50 , denotes the _ sample autocovariance function _ of the sample @xmath51 . note that the definition of @xmath52 deviates slightly from the usual one where the @xmath53 s are centered by the sample mean . however , for the theory given below , this centering is not essential . centering with the sample mean @xmath54 is not the most natural choice when dealing with infinite variance processes . in what follows , we will frequently make use of the _ self - normalized periodogram _ @xmath55 where @xmath56 , @xmath50 , denotes the _ sample autocorrelation function _ of @xmath51 . in view of ( [ decomp ] ) , we can rewrite @xmath57 as follows : @xmath58 where @xmath59 are the _ fourier coefficients _ of @xmath60 . we also introduce the self - normalized version of @xmath61 : @xmath62 in this section , we study the limit behavior of the integrated periodograms @xmath61 indexed by classes of functions for an i.i.d . s@xmath0s sequence with @xmath40 . in section [ s1 ] , we consider the convergence of the finite - dimensional distributions . in sections [ s2 ] and [ s3 ] , we prove the tightness of the processes in the cases @xmath1 and @xmath2 , respectively . in the case @xmath63 we solve a more general weak convergence problem for random quadratic forms in the i.i.d . sequence @xmath39 ; the convergence of the integrated periodograms indexed by classes of functions is only a special case . the case @xmath2 is more involved . among others , entropy conditions will be needed and we only prove results on the weak convergence of the empirical spectral distribution , that is , we focus on random quadratic forms with toeplitz coefficient matrices given by the fourier coefficients @xmath64 defined in . a glance at decomposition ( [ decom2 ] ) is enough to see that the convergence of the finite - dimensional distributions of @xmath61 is essentially determined by the weak limit behavior of the sample autocovariances @xmath65 . for this reason , we recall a well - known result due to davis and resnick @xcite ; see also @xcite , section 13.3 . [ le1 ] for every @xmath66 , @xmath67 where @xmath68 denotes weak convergence , the @xmath69 s are independent , @xmath70 is @xmath71 and @xmath72 are i.i.d . @xmath73 for some @xmath74 , @xmath75 . in particular , @xmath76 the latter result is an immediate consequence of and the continuous mapping theorem . lemma [ le1 ] yields the weak convergence for any finite linear combination of the sample autocovariances and autocorrelations . it also suggests that the weak limit of the standardized process @xmath57 will be determined by the infinite series @xmath77 . but this also means that we need to require additional assumptions on the sequence @xmath78 . we will treat this problem in a more general context . consider a sequence @xmath79 that is , @xmath80 satisfies the summability condition @xmath81 . for such an @xmath80 , we define the sequences of processes @xmath82 , & \quad$y(\mathbf{a})= \displaystyle\sum_{k=1}^{\infty } a_k y_k , $ \vspace*{1pt}\cr \displaystyle\widetilde x_n(\mathbf{a})= ( n/\log n)^{1/\alpha } \sum_{k=1}^{n-1}a_k \rho_{n,\varepsilon}(k ) , & \quad$\widetilde y(\mathbf { a})=y(\mathbf{a})/y_0 .$ } \ ] ] here , @xmath83 are independent stable random variables , as described in lemma [ le1 ] . the 3-series theorem @xcite implies that @xmath84 is equivalent to the a.s . convergence of the infinite series @xmath85 in ( [ eqrto ] ) . however , for the weak convergence of @xmath86 and @xmath87 , we need a slightly stronger assumption : @xmath88 this assumption ensures the weak convergence of the random quadratic forms in ( [ eqrto ] ) ; see the proof of theorem [ pr1 ] below . assumptions of this type frequently occur in the literature on infinite variance quadratic forms ( e.g. , @xcite ) . they appear in a natural way in tail estimates for quadratic forms in i.i.d . stable random variables ; see the . we can now formulate our result concerning the convergence of the finite - dimensional distributions . [ pr1 ] for any @xmath40 , @xmath89 using a cramr wold argument , it suffices to prove the convergence of the one - dimensional distributions . from ( [ aqua ] ) and the continuous mapping theorem , it immediately follows that for every @xmath90 , @xmath91 \longrightarrow y_m(\mathbf{a})=\sum_{k=1}^{m } a_k y_k , \ ] ] where @xmath68 denotes weak convergence . since @xmath92 , as @xmath93 follows from the 3-series theorem . according to @xcite , theorem 4.2 , it remains to show that @xmath94\biggr|>\epsilon\biggr)=0\end{aligned}\ ] ] for every @xmath95 and @xmath96 . we write @xmath97 for the above probabilities . applying lemma [ le2 ] in the and the fact that @xmath98 , we conclude that @xmath99 ( the constant on the right - hand side depends on @xmath100 . ) this proves the theorem for @xmath86 ; the convergence of @xmath87 can be shown analogously by utilizing ( [ aqua1 ] ) . as an immediate corollary of theorem [ pr1 ] , we obtain the following result which solves the problem of finding the limits of the finite - dimensional distributions for the integrated periodogram @xmath61 in ( [ decom2 ] ) and its self - normalized version @xmath101 in ( [ decom2a ] ) . [ co1 ] let @xmath40 and @xmath102\dvt \mathbf { a}(f)=(a_1(f),a_2(f),\ldots)\in\ell^{\alpha } \log\ell\ } , \ ] ] where @xmath103 is specified in @xmath104 . we then have @xmath105_{f\in\mathcal{f}}&\stackrel{\mathit{fidi}}{\longrightarrow } & 2 [ y(\mathbf{a}(f))]_{f\in\mathcal{f } } , \\ ( n/\log n)^{1/\alpha}[\widetilde j_{n,\varepsilon } ( f)-a_0(f)]_{f\in\mathcal{f}}&\stackrel { \mathit{fidi}}{\longrightarrow } & 2 [ \widetilde y(\mathbf{a}(f))]_{f\in\mathcal{f } } .\end{aligned}\ ] ] [ rem:1 ] the condition @xmath106 is , in general , not easily verified . however , if @xmath60 represents the spectral density of a stationary process @xmath86 with absolutely summable autocovariance function @xmath107 , then , up to a constant multiple , @xmath60 is represented by the fourier series of @xmath107 , and the rate of decay of @xmath108 as @xmath109 is well known for numerous time series models . for example , if @xmath60 is the spectral density of an arma process , @xmath108 at an exponential rate and then @xmath110 is satisfied for every @xmath111 . conditions ensuring that @xmath112 can be found in the literature on fourier series , for example , in @xcite . theorem ( 3.10 ) on page 243 in volume i of that reference yields , for lipschitz continuous functions @xmath60 with exponent @xmath113 $ ] , that @xmath112 for @xmath114 , but not necessarily for @xmath115 . this means , in particular , that lipschitz continuous functions do not necessarily satisfy @xmath112 for small values @xmath116 . zygmund s theorem ( 3.13 ) , @xcite , page 243 , volume i , states that @xmath117 if @xmath60 is of bounded variation and lipschitz continuous with exponent @xmath113 $ ] such that @xmath118 , but this statement is not necessarily valid for @xmath119 . we also note that @xmath120 for @xmath121}$ ] , @xmath122 $ ] and @xmath123 . indeed , then @xmath124 , @xmath125 and @xmath126 diverges a.s . by the 3-series theorem and the 01 law hence , corollary [ co1 ] does not apply to the important class of indicator functions when @xmath116 . moreover , @xmath127 is not tight . indeed , it follows from the argument above and from @xcite , theorem 6.2.1 that for some @xmath128 , for every @xmath129 , @xmath130\biggr|>k\biggr ) \\ & \le & \mathit{const}\ , \liminf_{n\to\infty } p\bigl(n ( n \log n)^{-1/\alpha } |j_{n,\varepsilon } ( f)-a_0(f ) \gamma_{n,\varepsilon}(0)|>k\bigr ) .\end{aligned}\ ] ] in order to derive a full weak convergence counterpart of the convergence in terms of the finite - dimensional distributions in corollary [ co1 ] , it remains to establish tightness of the corresponding family of laws . we start , once again , in the more general context of random fields indexed by sequences in @xmath132 . since we are dealing with the weak convergence of infinite - dimensional objects , we may expect difficulties which are due to the geometric properties of the underlying path spaces . it is also not completely surprising that the case @xmath1 is the ` better one ' in comparison with @xmath2 ; see , for example , the results on boundedness , continuity and oscillations of @xmath0-stable processes in @xcite , chapter 10 . note , however , that the constraint @xmath133 is harder to satisfy for smaller @xmath0 than for larger @xmath0 ; see also remark [ rem:1 ] . in the present case @xmath1 , we introduce the function @xmath134 where @xmath135 is chosen so large that @xmath136 is concave on @xmath137 . note that @xmath138 and this set is a linear metric space when endowed with the metric @xmath139 assume that @xmath140 is a compact set of @xmath141 with the additional property that @xmath142 observe that @xmath140 is then also a compact subset of @xmath143 and @xmath144 is sample - continuous as a random element with values in @xmath145 , the space of continuous functions on @xmath140 equipped with the uniform topology ; see @xcite , section 10.4 . the following is our main result on the weak convergence of the sequences @xmath86 and @xmath87 of infinite variance random quadratic forms in the case @xmath1 . [ th1 ] assume @xmath1 . for a compact set @xmath140 of @xmath132 satisfying @xmath146 the following weak convergence result holds in @xmath145 : @xmath147 where @xmath148 , @xmath149 , @xmath150 and @xmath151 are defined in @xmath152 and the processes @xmath150 and @xmath151 are sample - continuous . we restrict ourselves to showing that @xmath153 . in view of theorem [ pr1 ] , it suffices to prove the tightness of the processes @xmath148 in @xmath154 , where @xmath155 is the restriction of @xmath156 to @xmath140 . we have , for positive @xmath100 and @xmath157 , @xmath158\\[-8pt ] & & \quad \le p\biggl(\sum_{k=1}^{n-1 } \sup_{d_{\mathcal{a}}(\mathbf { a},\mathbf{b})<\delta}|a_k - b_k| \bigl[n \gamma_{n,|\varepsilon = p_n(\epsilon,\delta ) .\nonumber\end{aligned}\ ] ] we want to show that @xmath159 can be made arbitrarily small for all @xmath160 , provided that @xmath157 is small . we solve this problem in a modified form : let @xmath161 be a sequence of i.i.d . @xmath162 random variables , independent of @xmath39 , and @xmath163 a double array of real numbers . we then have @xmath164 by virtue of this argument , it suffices to replace the products @xmath165 in the quadratic form in ( [ eq4a ] ) with the products @xmath166 . this means that it suffices to show that @xmath167 can be made arbitrarily small for all @xmath160 , provided that @xmath157 is small , where @xmath168 now , first apply lemma [ le3 ] to the @xmath169 s and then condition ( [ eq2 ] ) : @xmath170 theorem [ th1 ] provides the limit process for a very general class of random quadratic forms with infinite first moments . the coefficient matrices of these quadratic forms are given by toeplitz matrices . the conditions on the parameter set @xmath140 are nothing but restrictions on the infinite toeplitz matrices @xmath171 . when specified to the particular case of fourier coefficients , as in ( [ eq45 ] ) , theorem [ th1 ] yields the following . [ co2 ] assume that @xmath1 and let @xmath172\dvt \mathbf{a}(f)=(a_1(f),a_2(f),\ldots ) \in\mathcal{a}\ } , \ ] ] where @xmath140 is a compact set of @xmath132 satisfying @xmath173 and @xmath103 is specified in @xmath104 . we then have @xmath174_{f\in\mathcal{f}}\longrightarrow 2 [ y(\mathbf{a}(f))]_{f\in\mathcal{f } } , \vspace*{3pt}\cr ( n/\log n)^{1/\alpha}[\widetilde j_{n,\varepsilon } ( f)-a_0(f)]_{f\in\mathcal{f}}\longrightarrow 2 [ \widetilde y(\mathbf{a}(f))]_{f\in\mathcal{f } } , } \ ] ] where the convergence holds in @xmath175 . let @xmath176 be defined by @xmath177 . we claim that @xmath178 is closed , hence compact . indeed , if @xmath179 is such that @xmath180 converges in @xmath132 to some point @xmath181 , then ( as @xmath182 ) , the sequence @xmath183 , \ ] ] @xmath184 converges in @xmath185 $ ] to some function @xmath60 that must necessarily be in @xmath38 . therefore , @xmath186 , and the latter set is compact . the above argument shows that @xmath187 $ ] convergence in @xmath38 is equivalent to @xmath132 convergence in @xmath188 . since theorem [ th1 ] implies weak convergence of the left - hand side of to its right - hand side in @xmath145 ( when each function @xmath12 is identified with @xmath189 ) , we conclude that weak convergence in also holds in @xmath175.=1 establishing full weak convergence in the case @xmath2 is more difficult than in the case @xmath1 . indeed , for @xmath1 , we were allowed to switch from the random variables @xmath191 to their absolute values , due to the specific geometry of the spaces @xmath192 and , in particular , @xmath141 . the spaces @xmath192 , @xmath2 , have a much more complicated structure and , therefore , the particular geometry of these spaces will be need to be invoked in proving tightness for the random quadratic forms @xmath148 and @xmath149 . the requirements prescribed by the geometry are usually given by entropy conditions ; see @xcite for a general treatment of random elements with values in banach spaces . entropy conditions are typically needed when @xmath0-stable processes with @xmath2 appear ; see the discussion in @xcite , chapter 12 . in this section , we only consider vectors @xmath193 of the form ( [ eq45 ] ) , that is , they are the fourier coefficients of some functions @xmath60 . corollary [ co1 ] determines the structures of the limit processes of the quadratic forms @xmath61 via the convergence of their finite - dimensional distributions . it hence suffices to show the tightness in @xmath175 for suitable classes @xmath38 . @xcite considered the special case of the one - dimensional class @xmath194 of indicator functions on @xmath13 $ ] . we extend their approach to more general classes of functions , using an entropy condition . for @xmath195 , let @xmath196 each @xmath197 defines a pseudo - metric on @xmath38 . let @xmath198 recall that the @xmath100-covering number @xmath199 of @xmath200 is the minimal integer @xmath201 for which we can find functions @xmath202 such that @xmath203 [ thm3 ] assume that @xmath2 , define @xmath103 as in @xmath204 and let @xmath38 be a subset of @xmath187 $ ] satisfying : @xmath106 for all @xmath205 ; @xmath206 such that @xmath207 , \qquad \epsilon > 0 , k \ge0 .\end{aligned}\ ] ] the weak convergence result @xmath208 then holds in @xmath209 . in contrast to the finite variance case @xcite , the entropy condition is a rather strong one . indeed , in the papers mentioned , integrability of some power of @xmath210 in a neighborhood of the origin suffices . however , conditions such as are common in problems of continuity and boundedness for stable processes ; see @xcite , chapter 10 . proof of theorem [ thm3 ] the convergence of the finite - dimensional distributions follows from theorem [ pr1 ] . we restrict ourselves to proving tightness for @xmath61 , which follows by proving that @xmath211 where @xmath212 and @xmath213 , \qquad j=1,2,\ldots .\ ] ] as in @xcite , ( 6.4 ) , one can argue that it suffices in ( [ eq56 ] ) to consider @xmath201 and @xmath160 of some specific form . let @xmath214 be two positive integers , @xmath215 and @xmath216 , and consider numbers @xmath217 , @xmath218 with @xmath219 . for @xmath220 sufficiently small and @xmath221 large enough , we have @xmath222 consider an array @xmath223 of positive numbers such that @xmath224 as @xmath225 for each @xmath226 . then , @xmath227 where @xmath228 \widehat\gamma_{n,\varepsilon}(j ) \biggr| > 2^{-(l+1)}\epsilon_k \biggr ) .\end{aligned}\ ] ] by virtue of lemma [ le2 ] , we have , for all @xmath195 , @xmath229 \widehat\gamma_{n,\varepsilon}(j ) \biggr|>2^{-(l+1)}\epsilon_k \biggr ) \le \mathit{const}\ , b_{k , l } , \end{aligned}\ ] ] where @xmath230 b_{k , l}= \epsilon_k^{-\alpha}2^{\alpha l } \sum_{j=2^k}^{2^{k+1}-1}|a_j(f)-a_j(g)|^{\alpha } \bigl[1 + \log^+\bigl(1/|a_j(f)-a_j(g)|\bigr)\bigr ] .\end{aligned}\ ] ] assuming that @xmath231 , we have @xmath232 \\ % a similar argument yields % b_{k , l } & \le & \mathit{const}\ , \epsilon_k^{-\alpha } 2^{\alpha l } \epsilon_{k , l-1}^{\alpha } 2^{-k(\alpha-1 ) } [ 1+k \log^+\epsilon_{k , l-1}^{-1 } ] .\end{aligned}\ ] ] hence , we are left to consider @xmath233\nonumber \\ & & \quad= \sum_{k = a}^b 2^{-k(\alpha-1-\alpha\theta ) } \sum_{l=1}^{\infty } n(\epsilon_{k , l},\mathcal{f},\rho_k)\epsilon_{k , l-1}^{\alpha } [ 1+k \log^+\epsilon_{k , l-1}^{-1 } ] 2^{\alpha l } \\ & & \quad\le \mathit{const}\ , \sum_{k = a}^b 2^{-k(\alpha-1-\alpha\theta ) } \sum _ { l=1}^\infty\biggl[1 + \biggl ( \frac{2^k}{\epsilon_{k , l } } \biggr)^\beta\biggr]\epsilon_{k , l-1}^\alpha[1 + \log^+\epsilon _ { k , l-1}^{-1 } ] 2^{\alpha l } .\nonumber % 2^{-k(\alpha-1-\alpha\theta ) } % [ 1+k \log^+\epsilon_{k , l-1}^{-1 } ] 2^{\alpha l}\end{aligned}\ ] ] assume that @xmath220 is so small that ( [ neu ] ) holds . define the numbers @xmath234 , @xmath235 with @xmath236 such that @xmath237 and @xmath238 . for these parameter choices , it is not difficult to see that ( [ nanuchen ] ) converges to zero by first letting @xmath239 ( i.e. , @xmath240 ) and then @xmath93 ( i.e. , @xmath241 ) . this proves ( [ eq56 ] ) , hence the tightness of the processes considered in @xmath175 . in what follows , we give examples of function spaces @xmath38 satisfying condition ( ii ) of theorem [ thm3 ] . consider a space of indexed functions @xmath242 that are defined on @xmath243 $ ] such that @xmath244 is a compact metric space , the mapping @xmath245 is hlder continuous with exponent @xmath246 and constant @xmath129 , that is , @xmath247 and the number of balls ( in metric @xmath248 ) of radius at most @xmath249 necessary to cover @xmath34 is of the order @xmath250 for some @xmath251 . then , @xmath252 satisfies @xmath253 with @xmath254 . indeed , let @xmath255 . we can find @xmath256 balls of radius at most @xmath257 covering @xmath34 . call them @xmath258 , with centers @xmath259 . now , given @xmath260 , we have @xmath261 for some @xmath262 and @xmath263 the desired bound now follows since @xmath264 consider a vapnik ervonenkis ( vc ) class @xmath265 of functions defined on @xmath13 $ ] with vc index @xmath266 ; see @xcite , section 2.6.2 for more information on vc classes of functions . given @xmath267 and @xmath268 , we can find @xmath269 balls of radius at most @xmath270 that cover @xmath265 in the norm @xmath271 ; see , for example , @xcite , theorem 2.6.7 . therefore , there exist @xmath272 such that for any @xmath273 , @xmath274 we then have @xmath275 it follows that @xmath276 it is the aim of this section to show that the results for the case of an i.i.d . sequence @xmath39 translate to the linear process case . the following decomposition will be crucial : @xmath277 this decomposition is analogous to the decomposition @xmath278 of the spectral density @xmath22 of a linear process . we will show that the normalized integrated remainder term @xmath279 is negligible , uniformly over the class of functions @xmath280 , in comparison to the normalized main part @xmath281 which can be treated by the methods of the previous section . note that , for a given sequence of coefficients @xmath282 , the functions @xmath283 constitute just another class of functions on @xmath13 $ ] , @xmath284 say , and therefore we will study the process @xmath285 , @xmath286 , for suitable classes @xmath287 . [ remainder ] let @xmath288 be the remainder term appearing in the decomposition of the periodogram @xmath289 . suppose that the linear filter @xmath290 of the process @xmath291 satisfies @xmath292 and @xmath38 is a collection of real - valued functions defined on @xmath243 $ ] such that @xmath293 @xmath294 . we then have @xmath295 from @xcite , proposition 5.1 , substituting @xmath296 for @xmath297 , we have the following decomposition for @xmath288 : @xmath298 where @xmath299 is the transfer function as defined before and @xmath300 we first show that @xmath301 note that @xmath302 & \le & % |u_{nj}(x)|^2 dx\\ % & = & \mathit{const}\ , \biggl ( \sum_{j=-\infty}^{-1 } + \sum_{j=1}^\infty\biggr ) the convergence in ( [ izmir ] ) will follow if we can show that the suprema over @xmath303 of the two infinite sums in the last expression are bounded in probability as @xmath304 . we will prove this for the second sum ; the first one can be handled analogously . we have , by definition of the terms @xmath305 , the cauchy schwarz inequality and the fact that , by assumption , @xmath306 , @xmath307 & & \quad\le \sup_{f\in{\mathcal{f}}}\sum_{j=1}^n |\psi_j| \int_0^\curpi|f(x)| \biggl|\sum_{t=1-j}^0 \varepsilon_t \mathrm{e}^{-\mathrm{i}xt } - \sum_{t = n - j+1}^n \varepsilon_t \mathrm{e}^{-\mathrm{i}xt}\biggr|^2\ , \mathrm{d}x \\[-2pt ] & & \qquad { } + \sup_{f\in{\mathcal{f}}}\sum_{j = n+1}^\infty|\psi_j| \int _ 0^\curpi \sum _ { t=1}^n \varepsilon_t \mathrm{e}^{-\mathrm{i}xt}\biggr|^2\ , \mathrm{d}x \\[-2pt ] & & \quad\le c [ i_1(n ) + i_2(n ) + i_3(n ) + i_4(n ) ] , \end{aligned}\ ] ] where @xmath308 i_2(n)&= & \sum_{j=1}^n |\psi_j| \biggl(\int_0^\curpi\biggl|\sum _ { t = n - j+1}^n \varepsilon_t \mathrm{e}^{-\mathrm{i}xt}\biggr|^4\ , \mathrm{d}x \biggr)^{1/2 } , \\[-2pt ] i_3(n)&= & \sum_{j = n+1}^\infty|\psi_j| \biggl(\int_0^\curpi \biggl|\sum_{t=1-j}^{n - j } \varepsilon_t \mathrm{e}^{-\mathrm{i}xt}\biggr|^4 \,\mathrm{d}x \biggr)^{1/2 } , \\[-2pt ] i_4(n)&= & \sum_{j = n+1}^\infty|\psi_j| \biggl(\int_0^\curpi\biggl| \sum_{t=1}^n \varepsilon_t \mathrm{e}^{-\mathrm{i}xt}\biggr|^4 \,\mathrm{d}x \biggr)^{1/2 } .\end{aligned}\ ] ] it remains to show that each sequence @xmath309 , is tight . now , @xmath310 let @xmath267 . choose @xmath311 so large that the following holds , for @xmath312 : @xmath313 write @xmath314 then , for @xmath315 , we have , for @xmath157 chosen as above , @xmath316^{1/2 } \\ & & \quad \le \frac{c}{k } \sum_{j=1}^n |\psi_j|\biggl [ \biggl(\sum_{m=1}^j \bigl ( m^{1/\alpha}(1 + \log m)^{{1/\alpha } + \delta } \bigr)^{4 - \alpha } \biggr)^{1/2 } \\ & & { } \qquad + \biggl ( \biggl[\sum_{m=1}^j \bigl ( m^{1/\alpha } ( 1 + \log m)^{{1/\alpha } + \delta } \bigr)^{2 - \alpha } \biggr]^2 \biggr)^{1/2}\biggr ] \\ % & \le&\frac{c}{k } \sum_{j=1}^n |\psi_j| [ ( \sum_{m=1}^j m^ { % + \sum_{m=1}^j m^{\frac{2}{\alpha}-1}(1 + \log m)^{(2-\alpha)(\frac{1 } { % & \le & \frac{c}{k } \sum_{j=1}^n |\psi_j| j^{2/\alpha } [ ( 1 + \log % j)^{\frac{1}{2}(4-\alpha)(\frac{1}{\alpha } + \delta ) } % + ( 1 + \log j)^{(2-\alpha)(\frac{1}{\alpha } + \delta)}]\\ % & \le & \frac{c}{k } \sum_{j=1}^n |\psi_j| j^{2/\alpha } ( 1 + \log j)^ { & & \quad \le \frac{c}{k } \sum_{j=1}^\infty|\psi_j| j^{2/\alpha } ( 1 + \log j)^{{(4-\alpha)/(2\alpha ) } + \tau } .\end{aligned}\ ] ] by virtue of , the last expression can be made smaller than @xmath317 by choosing @xmath318 large enough , which proves the tightness of @xmath319 . similar arguments show that @xmath320 , @xmath321 are tight sequences as well . the convergence in ( [ izmir ] ) follows . by the decomposition ( [ vienna ] ) , the proof will be finished if we can also establish that @xmath322 we have , by the cauchy schwarz inequality and the identity @xmath323 , @xmath324 similar arguments as for ( [ izmir ] ) ensure the tightness of the sequence @xmath325 . the tightness of the term @xmath326 follows from @xcite theorem 2.1 ( for @xmath327 ) and proposition 3.1 ( for @xmath328 ) . thus , we conclude that ( [ prague ] ) holds , and lemma [ remainder ] is proved . a referee kindly pointed out that lemma [ remainder ] remains valid under the following condition , which is weaker than : assume that there exists a sequence @xmath329 of positive numbers such that @xmath330 condition follows by taking @xmath331 for some positive @xmath157 . by ( [ munich ] ) , we may write , for each @xmath60 , @xmath332 where @xmath333 stands for @xmath334 . combining this decomposition with lemma [ remainder ] , we can now state the following analogs to corollary [ co2 ] and theorem [ thm3 ] . assume that @xmath1 or @xmath2 and let @xmath335 be as defined as in corollary [ co2 ] or theorem [ thm3 ] , respectively . suppose that the set @xmath336 \rightarrow\mathbb{r}\dvt f|\psi|^2 \in\mathcal{f}\ } = \mathcal{f}_\psi$ ] satisfies @xmath337 and holds for some @xmath338 . we then have @xmath339_{f\in\mathcal{f}_\psi}\longrightarrow 2 [ y(\mathbf{a}(f|\psi|^2))]_{f\in\mathcal{f}_\psi } , \vspace*{3pt}\cr ( n/\log n)^{1/\alpha}[\widetilde j_{n , x}(f)-a_0(f|\psi 2 [ \widetilde y(\mathbf{a}(f|\psi|^2))]_{f\in\mathcal{f}_\psi } , } \vspace*{2pt}\ ] ] where the convergence holds in @xmath340 . for an array @xmath341 of real numbers , define the quadratic forms @xmath342 and @xmath343 the following lemma is a consequence of @xcite , theorem 3.1 ; see also @xcite . [ le2 ] for @xmath40 , there exists a positive constant @xmath344 such that for all @xmath345 , @xmath346 now , let @xmath161 be a sequence of i.i.d . @xmath162 random variables , independent of @xmath39 , and let @xmath347 be as above . the following lemma is a consequence of lemma [ le2].=1 [ le3 ] for @xmath1 , there exists a positive constant @xmath348 such that for all @xmath345 , @xmath349 apply lemma [ le2 ] to @xmath350 , conditionally on @xmath351 : @xmath352 because @xmath1 , we also have , for @xmath345 , @xmath353 and combining ( [ eq8 ] ) and ( [ eq9 ] ) , we thus obtain ( [ eq18 ] ) . thomas mikosch s research was supported in part by the danish research council ( fnu ) grant 272 - 06 - 0442 . gennady samorodnitsky s research was supported in part by the aro grant w911nf-07 - 1 - 0078 at cornell university and a villum kann rasmussen visiting professor grant at the university of copenhagen . the paper was written when sami umut can and gennady samorodnitsky visited the department of mathematical sciences at the university of copenhagen in 2008/2009 . they take pleasure in thanking the department for hospitality and financial support . the authors also thank the anonymous referees for suggestions that helped to improve the paper . can , s.u . , mikosch , t. and samorodnitsky , g. ( 2009 ) . weak convergence of the function - indexed integrated periodogram for infinite variuance processes . extended technical report . available at http://www.math.ku.dk/\textasciitilde mikosch[www.math.ku.dk/~mikosch ] . crovella , m. and bestavros , a. ( 1996 ) . self - similarity in world wide web traffic : evidence and possible causes . in _ proceedings of the 1996 acm sigmetrics international conference on measurement and modeling of computer systems _ * 24 * 160169 . dahlhaus , r. and polonik , w. ( 2002 ) . empirical processes and nonparametric maximum likelihood estimation for time series . in _ empirical process techniques for dependent data dehling , t. mikosch and m. srensen , eds . ) boston : birkhuser . davis , r.a . and mikosch , t. ( 2009 ) . extreme value theory for garch processes . in _ the handbook of financial time series _ ( t.g . andersen , r.a . kreiss and t. mikosch , eds . ) 187200 . heidelberg : springer . davis , r.a . and mikosch , t. ( 2009 ) . extremes of stochastic volatility models . in _ the handbook of financial time series _ andersen , r.a . davis , j .- kreiss and t. mikosch , eds . ) 355364 . heidelberg : springer . klppelberg , c. and mikosch , t. ( 1996 ) . self - normalized and randomly centred spectral estimates . in _ proceedings of the athens international conference on applied probability and time series , vol . 2 : time series _ ( c.c . heyde , yu.v . prokhorov , r. pyke and s.t . rachev , eds . ) 259271 . berlin : springer . mikosch , t. ( 1998 ) . periodogram estimates from heavy - tailed data . in _ a practical guide to heavy tails : statistical techniques for analysing heavy - tailed distributions _ ( r.a . adler , r. feldman and m.s . taqqu , eds . ) 241258 . boston : birkhuser . mikosch , t. ( 2003 ) . modelling dependence and tails of financial time series . in _ extreme values in finance , telecommunications and the environment _ ( b. finkenstdt and h. rootzn , eds . ) 185286 . london : chapman & hall . willinger , w. , taqqu , m.s . , sherman , r. and wilson , d. ( 1995 ) . self - similarity through high variability : statistical analysis of ethernet lan traffic at the source level . proceedings of the acm / sigcomm95 , cambridge , ma . _ computer communications review _ * 25 * 100113 .

understanding the deconfining phase transition of qcd is a long - standing problem in high energy physics . the situation is somewhat clearer in pure su(n ) yang - mills theory , where it is understood that on length scales larger than @xmath2 the relevant effective degrees of freedom are su(n ) spin variables ( thermal wilson loops ) in three dimensions , obtained by compactifying the euclidean time direction . in this picture , the deconfining phase transition manifests as spontaneous breaking of the z(n ) center symmetry @xcite . numerical simulations show that the interaction measure , @xmath3 , times @xmath4 , is approximately flat from @xmath5 to @xmath6 , for up to six colors @xcite . similar behavior is observed in three space - time dimensions , for @xmath7 times @xmath8 @xcite . this shows that the leading corrections to an `` ideal '' gas term , @xmath9 in four dimensions , and @xmath10 in three dimensions , are terms @xmath11 , in both four and three dimensions . this differs from corrections due to a bag pressure as in the mit bag model , which would be independent of temperature . this behavior can be reproduced within matrix models of the thermal wilson loop @xcite with coupling constants which are rather simple functions of @xmath12 . to understand this phenomenon better , it is useful to study the dynamics of the eigenvalues of the wilson loops in different representations of the gauge group . the purpose of this work is to study the effective potential of bare wilson loops and of their eigenvalues , in the fundamental and adjoint representations of su(2 ) , through ab - initio simulations in discretized space - time . our goal is to shed some light on the validity of the perturbative ansatzes for the effective action , which are used in the construction of effective theories . effective potentials for variables related to the phases of the eigenvalues of thermal wilson loops have also been obtained from functional methods such as the functional renormalization group or dyson - schwinger equations typically using background landau - dewitt gauges @xcite , and our results may be useful to benchmark these non - perturbative continuum quantum field - theoretic computations . also , we wish to verify that the ansatz used to parametrize the results obtained in refs . @xcite for the classical potential of a three dimensional matrix model of wilson loops in the fundamental representation is valid also for the full gauge theory . we simulate pure su(2 ) gauge theory using the standard wilson action s= _ ( 1- re tr * u * _ ) , [ slatt ] where the sum runs over all plaquettes of the four dimensional lattice . lattice configurations are generated using an exact heatbath algorithm @xcite which generates link variables according to the local boltzmann distribution given by the staple matrix . su(2 ) matrices are stored in the quaternionic representation u = a_0 * 1 * + i _ j=1 ^ 3 a_j _ j , _ j=0 ^ 3 a_i^2=1 , [ eq : quater1 ] where @xmath13 and @xmath14 are the pauli matrices . we construct thermal wilson loops @xmath15 in the fundamental representation in the usual way by multiplying temporal links : * l*(*x*)^f=_t=0^n_t-1u_4(*x*,t ) .[eq : ploop1 ] loops in higher representations can be generated from the fundamental loop @xcite . we currently only also consider the adjoint representation which is constructed as * l*_ij^a=2tr(t_i * l*^f t_j * l*^f ) , [ eq : adj1 ] where @xmath16 are the three generators in the fundamental representation . we are interested in the per - site effective potential @xmath17 for the polyakov loop @xmath18 and for the eigenvalues of the wilson line @xmath19 , in the fundamental or adjoint representations . in other words , @xmath20 describes fluctuations about the volume averaged mean field @xmath21 . the latter , in turn , determines the couplings @xmath22 , @xmath23 , @xmath24 in @xmath20 , see below . in our work , `` sites '' are understood as the lattice - points of the three dimensional spin - system of the thermal wilson loops . our approach differs from the computation of a potential for volume averaged fields as in refs . . we also comment on the difference of our approach to that of ref . @xcite . in this work they take the wilson line to be constant in time . this is allowed , and in the continuum , corresponds to a static gauge , @xmath25 . there is still a residual gauge freedom to fix spatial gauge transformations at a given time . for example , one could fix to coulomb gauge , @xmath26 . instead , ref . @xcite computes an effective potential by assuming that the thermal wilson line is constant in space . in the continuum this corresponds to setting @xmath27 for all three spatial directions , and is not an allowable gauge condition . because of this , their results are very different from ours or from refs . @xcite . in particular , @xcite finds a potential which is completely flat in the confined phase . to obtain @xmath17 we first compute the per - site probability distribution @xmath28 by histogramming each observable . from these distributions we obtain its classical or `` constraint '' effective potential @xmath29 . the effective potential is then obtained from the legendre transform of the moment - generating function @xmath30 : w(h)= dx(-v_0(x)+hx ) . [ eq : momgen ] here , @xmath31 stands for the respective observable and the integral runs over the entire range of @xmath31 . we obtain v_()= - ( ) , where ( ) = w(h())-h ( ) , = . [ eq : legendre1 ] in most cases considered here it is simpler to carry out the legendre transformation via ( ) = _ h ( h - w(h ) ) .[eq : legendre2 ] this is our method of choice . , for each data - set we take a proper parametrization of @xmath32 as input . for a given @xmath33 we change @xmath34 from @xmath35 to @xmath36 in steps of @xmath37 . for each @xmath34 , @xmath30 is computed from eq . ( [ eq : momgen ] ) using a @xmath38 point gaussian quadrature with @xmath39 . we obtain the maximum @xmath40 with respect to @xmath34 . we repeat this process for @xmath41 equidistant values of @xmath33 in the domain of @xmath42 . we find that this procedure yields results which are stable under further increase of the resolution . ] we use eq . ( [ eq : legendre1 ] ) only in some special cases where analytical approximations to @xmath30 can be easily obtained . we use the string tension to fix the lattice spacing and thus the temperature @xmath43 in physical units . the exact procedure is discussed at length in appendix [ sec : physunits ] ( in all figures , the quoted @xmath8 are understood to have a relative uncertainty of less than @xmath44 ) . we change the temperature by varying @xmath45 . the advantage of this fixed scale approach is that it allows us to clearly disentangle the effects of temperature changes and renormalization . in the following we present a few formulae which shall be used throughout this work . using eqs . ( [ eq : quater1 ] ) and ( [ eq : adj1 ] ) with @xmath46 one can easily obtain the compact expression * l*_ij^a=2(a_i a_j + a_0 _ k=1 ^ 3 _ ijk a_k + _ ij(a_0 ^ 2- ) ) , [ eq : funadjoint_entries ] which relates the elements of the adjoint matrices to the quaternionic parameters @xmath47 . from eq . ( [ eq : funadjoint_entries ] ) it immediately follows that tr*l*^a=|tr*l*^f|^2 - 1 .[eq : funadjoint ] from ( [ eq : quater1 ] ) it follows that the eigenvalues @xmath48 of the fundamental loop are given by _ 1,2=a_0 . the eigenvalues form a pair of complex conjugates which lie on the unit circle . also , they are related to the polyakov loop @xmath49 by ^f = tr*l*^f = a_0 = ( _ 1 + _ 2 ) . likewise , for the adjoint wilson loop ( which is a real 3-by-3 matrix ) one can easily show that the eigenvalues @xmath50 are _ 1=1 , _ 2,3=(tr*l*^a -1 ) . the non - unit eigenvalues @xmath51 again form a pair of complex conjugates with @xmath52 . thus for both representations a single phase @xmath53 uniquely fixes all eigenvalues . the @xmath19 are related by a similarity transform to @xmath54 and @xmath55 respectively , so that @xmath56 also fixes the trace of @xmath19 through the relations tr*l*^f= 2 ( ^f ) , tr*l*^a= 1 + 2 ( ^a ) .[eq : wilsontrafo ] we simulate @xmath57 gauge theory at @xmath58 and @xmath59 , which corresponds to @xmath0 at @xmath60 . the lattice spacing is @xmath61 and @xmath62 , respectively . we use time - like lattice sizes of @xmath63 . for every lattice size and @xmath64 value several hundreds of independent lattice configurations were sampled from the equilibrium distribution . bin sizes for the histograms were chosen such that the results appear as smooth lines at the resolutions of the figures presented here . all distributions @xmath28 presented here are normalized . there is a subtle issue regarding ergodicity of simulations in the deconfined phase where the @xmath65 center symmetry is spontaneously broken : in principle , at any finite lattice volume , an ergodic algorithm which runs forever will tunnel infinitely often between the two groundstates . hence , all probability densities @xmath28 will display a mixing of these groundstates and show no signs of a broken symmetry . the physics presented here however is concerned with the thermodynamic limit , for which the potential barrier between the groundstates becomes infinitely large . tunneling events should therefore not occur , which strictly speaking violates ergodicity . from a practical point of view it is justified to ignore this problem as long as the lattice volumes considered are sufficiently large so that tunneling is strongly suppressed . one should take care however to ensure that this is truly the case for a given temperature , since such tunneling events occur more frequently as one approaches @xmath0 from above . one can achieve this , for instance , by monitoring the volume average of the polyakov loop and looking for sign flips . the results presented here are unaffected by this issue . the lowest temperatures in the deconfined phase which we consider here are well above @xmath0 , so that even on a @xmath66 lattice tunneling is practically impossible and the infinite volume limit is well approximated . for each dataset discussed in this section simulations were done on lattices with @xmath67 and @xmath68 . we find that both cases give identical results ( indistinguishable histograms ) , which confirms that finite - volume effects play no role . likewise , this confirms that statistical errors are negligible given the resolutions of the presented figures . without restricting generality we chose to represent the symmetry broken phase as the state with @xmath69 . we begin with the probability density @xmath70 of the polyakov loop @xmath71 in the fundamental representation . [ fig : su2fun1 ] shows @xmath70 for @xmath58 and @xmath59 at @xmath63 , which corresponds to temperatures ranging from slightly below @xmath0 to roughly @xmath72 ( smaller @xmath64 corresponds to lower @xmath12 for fixed @xmath45 ) . ( top ) and @xmath59 ( bottom ) . the different temperatures correspond to @xmath63 respectively . [ fig : su2fun1],title="fig : " ] ( top ) and @xmath59 ( bottom ) . the different temperatures correspond to @xmath63 respectively . [ fig : su2fun1],title="fig : " ] one can clearly see the effect of spontaneous breaking of the @xmath65 ( @xmath73 ) symmetry : right at @xmath0 the distributions are symmetric around @xmath74 . in fact , we have confirmed that they are identical for any @xmath75 $ ] at the respective @xmath76 . at higher temperature @xmath70 is skewed and it develops a peak at non - zero @xmath49 which moves towards @xmath77 as one raises the temperature . in fig . [ fig : su2fun2 ] one can see that the corresponding constraint effective potential @xmath78 gets tilted , which is exactly what one expects at a second order phase transition . ( top ) and @xmath59 ( bottom ) . the different temperatures correspond to @xmath63 respectively . [ fig : su2fun2],title="fig : " ] ( top ) and @xmath59 ( bottom ) . the different temperatures correspond to @xmath63 respectively . [ fig : su2fun2],title="fig : " ] what is striking is that the lattice spacing affects the distribution . comparing the simulations at the two different values of the coupling , @xmath58 ( corresponding to a larger lattice spacing ) clearly approaches the perturbative vacuum ( @xmath79 ) much faster with increasing temperature . it is well known that the expectation value of the bare polyakov loop is scale dependent and vanishes in the continuum limit ( hence the need to construct renormalized loops @xcite ) . however , here we show that in fact the entire distribution is affected in a profound way . in the continuum @xmath80 limit @xmath28 is deformed into the distribution at @xmath81 . one way of thinking of this is that the @xmath81 distribution is an attractive ultra - violet fixed point of renormalization group transformations . what is also noteworthy is that the distribution appears not to change at @xmath82 . which we are not sensitive to . such would be consistent with the findings of other authors @xcite . ] to make these statements quantitative we model the distributions using appropriate parametrizations . it turns out that the @xmath81 distribution is exactly p^(t_c)()= , [ eq : vand1 ] which corresponds to a random walk on the su(2 ) group manifold with the appropriate measure . the corresponding constraint potential is the vandermonde potential @xcite v^(t_c)_0()=-(1-^2 ) - ( ) .[eq : vand2 ] at @xmath83 , for the entire range of temperatures considered here , the following ansatz reproduces the simulated curves accurately : v_0()=v^(t_c)_0 ( ) + a(t)-b(t)+c(t)^2 .[eq : vand3 ] this @xmath84 corresponds to the potential for a single `` local spin '' @xmath85 given that @xmath86 . if instead one adopts the convention that @xmath87 by performing a z(2 ) transformation then @xmath84 is obtained from the above by letting @xmath88 . the potential ( [ eq : vand3 ] ) corresponds to the distribution p()=(-a(t)+b(t)-c(t)^2 ) . [ eq : vand4 ] we obtain the parameters @xmath89 from a @xmath90 fit . we do not plot the fits as they are indistinguishable from the simulation results . the temperature dependence of the parameters for both values of @xmath64 are shown in fig . [ fig : su2evfun5 ] . ( top ) and @xmath59 ( bottom ) . @xmath22 and @xmath24 are scaled up by factors of @xmath91 . the connecting lines are there to guide the eye only but do not represent actual models . [ fig : su2evfun5],title="fig : " ] ( top ) and @xmath59 ( bottom ) . @xmath22 and @xmath24 are scaled up by factors of @xmath91 . the connecting lines are there to guide the eye only but do not represent actual models . [ fig : su2evfun5],title="fig : " ] it appears that the term linear in @xmath49 is dominant for the temperature range investigated here . it is also precisely this term which drives breaking of the @xmath65 symmetry ( it in some sense describes the interaction of a particular site with the external field generated by the symmetry broken state of the lattice as a whole ) . the ansatz ( [ eq : vand3 ] ) is exactly of the form which was used in refs . @xcite to parametrize the constraint effective potential for @xmath49 in a @xmath92 effective matrix theory of @xmath57 wilson lines , with a similar behavior of the parameters @xmath89 . ( top ) and @xmath59 ( bottom ) . the quoted temperatures correspond to @xmath63 respectively . [ fig : su2fun2eff],title="fig : " ] ( top ) and @xmath59 ( bottom ) . the quoted temperatures correspond to @xmath63 respectively . [ fig : su2fun2eff],title="fig : " ] we obtain the effective potential @xmath20 by numerically solving eq . ( [ eq : legendre2 ] ) , using the parametrization ( [ eq : vand3 ] ) of @xmath84 as input . obtaining the supremum numerically is straightforward . the integral in eq . ( [ eq : momgen ] ) is computed numerically via gaussian quadrature . the results are shown in fig . [ fig : su2fun2eff ] . we have confirmed that the expectation value = _ -1 ^ 1d p ( ) , coincides with the point at which @xmath20 is minimal , as it should . ( top ) and @xmath59 ( bottom ) . the quoted temperatures correspond to @xmath63 respectively . [ fig : su2evfun3],title="fig : " ] ( top ) and @xmath59 ( bottom ) . the quoted temperatures correspond to @xmath63 respectively . [ fig : su2evfun3],title="fig : " ] ( top ) and @xmath59 ( bottom ) . the quoted temperatures correspond to @xmath63 respectively . [ fig : su2evfun4],title="fig : " ] ( top ) and @xmath59 ( bottom ) . the quoted temperatures correspond to @xmath63 respectively . [ fig : su2evfun4],title="fig : " ] from the same ensembles we obtain the per - site distribution of the phase @xmath56 of the eigenvalues of the wilson loop @xmath19 . [ fig : su2evfun3 ] and [ fig : su2evfun4 ] show the distributions @xmath93 and the potentials @xmath94 respectively . one can see that right at @xmath0 the distribution is peaked around @xmath95 , which is the state of maximal eigenvalue repulsion . increasing the temperature drives the peak towards @xmath96 though eigenvalue repulsion persists even above @xmath0 . such an effect is known to occur also in matrix models of the wilson line @xcite . we can transform the parametrization of @xmath70 given in eq . ( [ eq : vand4 ] ) to @xmath93 . ( [ eq : wilsontrafo ] ) provides a one - to - one mapping from @xmath97 $ ] to @xmath98 $ ] ; we can thus use the standard integral transformation law to obtain @xmath93 from our models ( [ eq : vand1 ] ) and ( [ eq : vand4 ] ) . @xmath99 $ ] maps into the same distribution @xmath70 , thus the restriction to positive @xmath56 is legitimate . using _ -1 ^ 1dp ( ) = _ ( = -1)^(=1 ) p ( ( ) ) d=_^0 p()d , we arrive at p^(t_c)()=^2= ( 1+(2 + ) ) , [ eq : vand5 ] for the distribution right at @xmath0 and p ( ) & = & ^2 [ eq : vand6 ] + & & ( -a(t)+b(t)()-c(t)^2 ( ) ) , for the general case . the parameters @xmath89 are unchanged . ( [ eq : vand5 ] ) and ( [ eq : vand6 ] ) again describe the simulated results to such accuracy that the lines are indistinguishable and thus the fit - curves are left out of the figures . we again use our model function v_0()= & & [ eq : vand7 ] + & & -1.2 cm -(^2 ) + a(t)-b(t)()+c(t)^2 ( ) , to obtain @xmath100 . the legendre transformation ( [ eq : legendre2 ] ) is carried out numerically , as before . the results are shown in fig . [ fig : su2evfun4eff ] . ( top ) and @xmath59 ( bottom ) . the different temperatures correspond to @xmath63 respectively . [ fig : su2evfun4eff],title="fig : " ] ( top ) and @xmath59 ( bottom ) . the different temperatures correspond to @xmath63 respectively . [ fig : su2evfun4eff],title="fig : " ] in order to make our results portable ( to compare them with other calculations or use them as input for effective theories ) we would like to obtain analytical expressions for @xmath100 and @xmath20 . one possible approach is to again use polynomial models ( together with the vandermonde contribution ) and adjust the parameters to fit the data . it turns out , however , that this approach is not satisfactory as it requires high - order polynomials and many parameters in both cases . given how well our models ( [ eq : vand4 ] ) and ( [ eq : vand6 ] ) described the simulated results for @xmath70 and @xmath93 , it would therefore be nice if the legendre transformation ( [ eq : legendre1 ] ) could be carried out exactly . unfortunately this is not possible in closed form . nevertheless , an approximate solution can still be obtained using the saddle point approximation . the next - to - leading order result is easily obtained , and we will compare this result to our data in the following . consider a stationary point @xmath101 of the exponential appearing in eq . ( [ eq : momgen ] ) such that v_0(x_0)=h .[eq : saddle1 ] we can expand the exponent in eq . ( [ eq : momgen ] ) around this point . neglecting terms containing derivatives of higher than second order we can carry out the integral : w(h ) & = & _ -^dx ( -v_0(x)+hx ) + & & -1.4 cm _ -^ dx ( -v_0(x_0)+hx_0 -v_0(x_0)(x_0-x)^2 ) + & = & -v_0(x_0)+hx_0 + ( ) . it holds that = x_0 + o ( v_0 ) , and thus the nlo effective potential is given by v_(x)v_0(x)+ ( ) . the leading order is just the classical potential . this formula can be applied to the potentials for both @xmath49 and @xmath56 . from eqs . ( [ eq : vand7 ] ) and ( [ eq : vand3 ] ) we obtain the second derivatives v_0()= 2c(t)+ , and v_0()= + b(t)()+2 c(t)(^2()-^2 ( ) ) . in figs . [ fig : su2fun2eff_spcompare ] and [ fig : su2evfun4eff_spcompare ] we compare the nlo saddle point approximation to the exact results for both @xmath20 and @xmath100 . the approximate curves are corrected by an additive shift to account for the loss of normalization of @xmath28 which occurs . the shift is such that @xmath17 is moved to zero at its minimum . in both cases @xmath102 is chosen as an example . it appears that the approximation is quite good . ( top ) and @xmath59 ( bottom ) . @xmath102 simulation result compared to nlo saddle point expansion . the sp curve has been shifted vertically to bring the minimum to zero . [ fig : su2fun2eff_spcompare],title="fig : " ] ( top ) and @xmath59 ( bottom ) . @xmath102 simulation result compared to nlo saddle point expansion . the sp curve has been shifted vertically to bring the minimum to zero . [ fig : su2fun2eff_spcompare],title="fig : " ] ( top ) and @xmath59 ( bottom ) . @xmath102 result compared to nlo saddle point expansion . the sp curve has been shifted vertically to bring the minimum to zero . [ fig : su2evfun4eff_spcompare],title="fig : " ] ( top ) and @xmath59 ( bottom ) . @xmath102 result compared to nlo saddle point expansion . the sp curve has been shifted vertically to bring the minimum to zero . [ fig : su2evfun4eff_spcompare],title="fig : " ] lastly , we investigate how well the classical potentials @xmath84 and @xmath94 can be approximated by taylor polynomials around the confined ( @xmath103 ) or deconfined ( @xmath104 ) vacua . expanding about @xmath105 we obtain v_0,1 ( ) & & - ( ) + a(t ) + b(t ) ( - ) + & & + ( c(t)+1)(- ) ^2 - ( - ) ^3 + & & + ( - ) ( - ) ^4 .[eq : vand8 ] likewise , about @xmath106 v_0,2 ( ) & & -(^2 ( ) ) + a(t ) -b(t ) + c(t ) + & & + ( -c(t ) ) ^2 + & & + ( - ) ^4 .[eq : vand9 ] the logarithmic term has not been expanded since it diverges at @xmath106 . finally , an expansion around @xmath107 yields v_0,3 ( ) & & -(^2 ( ) ) + a(t ) + b(t ) + c(t ) + & & + ( -c(t ) ) ( -)^2 + & & + ( - ) ( -)^4 .[eq : vand10 ] note that ( [ eq : vand10 ] ) is nothing but the z(2 ) transform of ( [ eq : vand9 ] ) corresponding to @xmath108 and @xmath88 . [ fig : su2evfun_taylor ] compares these expansions to the simulation results for @xmath102 , which corresponds to the highest temperature for each respective @xmath64 . it can be seen that eq . ( [ eq : vand8 ] ) works well , even if one terminates the expansion at @xmath109 . the approximation smoothly approaches the simulation results when at higher orders , and @xmath110 reproduces the exact result to high accuracy . convergence is slower for eq . ( [ eq : vand9 ] ) and very slow for ( [ eq : vand10 ] ) . the latter is not extremely suprising since @xmath107 is the wrong groundstate . ( top ) and @xmath59 ( bottom ) . @xmath102 results compared to taylor expansions . [ fig : su2evfun_taylor],title="fig : " ] ( top ) and @xmath59 ( bottom ) . @xmath102 results compared to taylor expansions . [ fig : su2evfun_taylor],title="fig : " ] a similar expansion can be constructed for the logarithmic contribution to @xmath84 . expanding eq . ( [ eq : vand3 ] ) to @xmath111 about @xmath74 we get v_0,1 ( ) & & - ( ) + a(t)- b(t ) + & & + ( c(t)+)^2 + ^4+^6 .[eq : vand11 ] fig . [ fig : su2fun_taylor ] shows the convergence of this expansion towards the simulated curves for @xmath102 . high orders are required to accurately reproduce the data . ( top ) and @xmath59 ( bottom ) . @xmath102 results compared to taylor expansions . [ fig : su2fun_taylor],title="fig : " ] ( top ) and @xmath59 ( bottom ) . @xmath102 results compared to taylor expansions . [ fig : su2fun_taylor],title="fig : " ] from the same ensembles , we can also construct distributions @xmath112 ( here without any normalization factors before the trace ) for the adjoint loop as well as for its eigenvalues . the results are shown in figs . [ fig : su2adj1 ] and [ fig : su2evadj1 ] , respectively . figs . [ fig : su2adj2 ] and [ fig : su2evadj2 ] show the corresponding potentials @xmath29 . what is striking is that temperature effects appear to be small . nevertheless , a slight suppression of @xmath93 around @xmath107 can be observed with increasing temperature . ( top ) and @xmath59 ( bottom ) . the different temperatures correspond to @xmath63 respectively . [ fig : su2adj1],title="fig : " ] ( top ) and @xmath59 ( bottom ) . the different temperatures correspond to @xmath63 respectively . [ fig : su2adj1],title="fig : " ] ( top ) and @xmath59 ( bottom ) . the different temperatures correspond to @xmath63 respectively . [ fig : su2adj2],title="fig : " ] ( top ) and @xmath59 ( bottom ) . the different temperatures correspond to @xmath63 respectively . [ fig : su2adj2],title="fig : " ] ( top ) and @xmath59 ( bottom ) . the different temperatures correspond to @xmath63 respectively . [ fig : su2evadj1],title="fig : " ] ( top ) and @xmath59 ( bottom ) . the different temperatures correspond to @xmath63 respectively . [ fig : su2evadj1],title="fig : " ] ( top ) and @xmath59 ( bottom ) . the different temperatures correspond to @xmath63 respectively . [ fig : su2evadj2],title="fig : " ] ( top ) and @xmath59 ( bottom ) . the different temperatures correspond to @xmath63 respectively . [ fig : su2evadj2],title="fig : " ] we parametrize these results in a way similar to those for the fundamental representation . we begin by transforming the parametrization of @xmath93 given in eq . ( [ eq : vand6 ] ) to the adjoint representation . once this is achieved , @xmath113 can be easily obtained . to approach this problem we require a one - to - one mapping @xmath114 from the fundamental to the adjoint phase . to construct such a mapping a few subtleties arise . ( [ eq : funadjoint ] ) implies that ^2(^f ) = ( 1+(^a ) ) .[eq : funadjointphase ] completing one full rotation in the fundamental phase @xmath115 ( from @xmath116 to @xmath117 ) generates two full cycles of the square cosine on the left - hand side of eq . ( [ eq : funadjointphase ] ) , whereas one rotation of the adjoint phase @xmath118 is only one cycle of the right - hand side . both sides are symmetric around @xmath106 , thus it is sufficient to consider only @xmath119 $ ] which can be mapped to one full rotation of the adjoint phase . second , the left - hand side possesses an inflection point at @xmath120 . thus the mapping must be constructed piece - wise in the intervals @xmath121 $ ] and @xmath122 $ ] . taking positive roots in eq . ( [ eq : funadjointphase ] ) and considering that @xmath123\\ -\cos\left(\frac{x}{2}\right ) & : x \in [ \pi,2\pi ] \end{array } \right.\ ] ] we obtain @xmath124\\ -\cos\left(\frac{\phi^a}{2}\right ) & : \phi^f \in [ \pi/2,\ldots,\pi ] \end{array } \right.\ ] ] and thus @xmath125\\ \pi - \frac{\phi^a}{2 } & : \phi^f \in [ \pi/2,\ldots,\pi ] \end{array } \right.\ ] ] this is a one - to - one mapping , covering the entire range of both @xmath126 and @xmath127 and which bijectively maps @xmath128 . we can now carry out the transformation according to _ 0^d^fp(^f)&=&_0^d^fp(^f ) + _ ^d^fp(^f ) + & = & _ 0^ p(^a)d^a . we obtain for the adjoint phase @xmath118 p(^a)&=&p^(t_c)(^a ) ( -a(t)-c(t)^2 ( ) ) + & & ( b(t ) ( ) ) , [ eq : phia_distrib1 ] with p^(t_c)(^a)=(^a)=^2 ( ) . likewise , for the adjoint loop @xmath127 we obtain p(tr*l*^a)&=&p^(t_c)(tr*l*^a)(-a(t)-(1+tr*l*^a ) ) + & & ( ) , with p^(t_c)(tr*l*^a)= . the simulation results ( figs . [ fig : su2adj1 ] and [ fig : su2evadj1 ] ) are again reproduced to such accuracy that the curves are virtually indistinguishable . we carry out the legendre transformations ( [ eq : legendre2 ] ) to obtain @xmath129 and @xmath130 . the results are shown in figs . [ fig : su2adj2eff ] and [ fig : su2evadj2eff ] . unfortunately , approximating these results via the saddle - point method is not possible due to the lack of a proper expansion point . ( top ) and @xmath59 ( bottom ) . the different temperatures correspond to @xmath63 respectively . [ fig : su2adj2eff],title="fig : " ] ( top ) and @xmath59 ( bottom ) . the different temperatures correspond to @xmath63 respectively . [ fig : su2adj2eff],title="fig : " ] ( top ) and @xmath59 ( bottom ) . the different temperatures correspond to @xmath63 respectively . [ fig : su2evadj2eff],title="fig : " ] ( top ) and @xmath59 ( bottom ) . the different temperatures correspond to @xmath63 respectively . [ fig : su2evadj2eff],title="fig : " ] we again obtain a taylor expansion of the classical potential @xmath131 . expanding ( [ eq : phia_distrib1 ] ) around @xmath132 to @xmath110 yields v_0(^a)&&-()+a(t ) + & & + ( 1+c(t)-- ) ( ) ^2 + & & + ( + + ) ( ) ^4 [ eq : phia_distrib2 ] in fig . [ fig : su2evadj_taylor ] a comparison of ( [ eq : phia_distrib2 ] ) and the simulation for @xmath102 is shown . a fourth order expansion reproduces the data with high accuracy . an expansion around @xmath106 is not possible due to the logarithmic term in the potential . ( top ) and @xmath59 ( bottom ) . @xmath102 results compared to taylor expansions . [ fig : su2evadj_taylor],title="fig : " ] ( top ) and @xmath59 ( bottom ) . @xmath102 results compared to taylor expansions . [ fig : su2evadj_taylor],title="fig : " ] in this work we simulated pure su(2 ) gauge theory at finite temperature , for two different fixed values of @xmath64 and hence of the lattice spacing . we changed the temperature by changing the time - like extent @xmath45 of the lattice . for each temperature we computed the classical potential of the bare polyakov loop and of the phase of the eigenvalues of the thermal wilson loop ( wrapped around the periodic boundary of the time direction ) in the fundamental and adjoint representation , using a histogram method . from these potentials , we obtained the effective potential via legendre transformation . we obtained model functions for the classical potentials for each case ( three parameters are required ) and investigated how the parameters depend on temperature and lattice spacing . we discussed how the models can be approximated by taylor expansions and how they can be converted into one another . we showed how the effective potential of the fundamental loop can be approximated to high precision by analytical expressions , when a next - to - leading order saddle expansion is used to compute the legendre transformation . several conclusions may be drawn from our work : the most obvious conclusion is that both the classical and effective potentials of the bare loop and its eigenvalues are well described by simple models , reminiscent of landau - ginzburg theories but with a symmetry breaking term , where the whole dependence on temperature and lattice - spacing is absorbed into three parameters . we have thus confirmed that the ansatz used in refs . @xcite to parametrize the classical potential in a su(2 ) matrix model of wilson lines is valid also in the full gauge theory , with the same qualitative behavior of the parameters . also , we have shown that a next - to - leading order approximation of the effective potential of the fundamental loop is reasonable . we stress that what we have computed in this work is the effective potential for the _ bare _ loop , which vanishes in the continuum limit . in a given irreducible representation of the gauge group , a renormalized loop is extracted from the bare loop by dividing by a renormalization constant , which depends upon the representation @xcite . one of the motivations of this work was the idea that the eigenvalues of the wilson line might be insensitive to renormalization . our results contradict this . instead , the eigenvalues of the bare loop are dominated by the effects of the vandermonde determinant . on the lattice , the vandermonde determinant arises naturally at each point in spacetime . in the action , its contribution is proportional to @xmath133 , where @xmath134 is the lattice spacing , and so it diverges in the continuum limit . in retrospect , it is natural that the lattice effective potential for a site potential is dominated by a vandermonde term . for the future , it will be interesting to investigate how the effective potential of renormalized loops can be obtained from our results . also , one should extract the non - perturbative contribution to the effective potential from our data and compare it to model calculations . furthermore , our investigation should be extended to higher representations of su(2 ) and also to other groups , such as su(3 ) or exceptional groups with a trivial center , such as g(2 ) . investigating the effects of dynamical fermions on the results discussed here is currently under way , using standard staggered fermions . it would be interesting to extend this investigation to non - zero chemical potential , which is possible for two colors , where there is no sign problem . we thank david scheffler for helpful discussions , for proof - reading and for a script to create histograms . this work was supported by the deutsche forschungsgemeinschaft within sfb 634 , by the helmholtz international center for fair within the loewe initiative of the state of hesse , by the u.s . department of energy under contracts # de - fg02 - 09er41620 and # de - ac02 - 98ch10886 , by the city university of new york through the psc - cuny research award program , grant 66514 - 00 44 , and by the european commission , fp7-people-2009-rg , no . 249203 . all results presented here were obtained using nvidia geforce gtx 580 graphics cards . to fix the physical units for the results discussed in this work we use a standard method which determines the lattice spacing @xmath134 through measurements of the string tension in lattice units at zero temperature ( which in practice means lattices of large volume which are either hypercubic or have @xmath135 ) . in this appendix we give a short summary of the method ( which is derived in detail in several textbooks ) and present numerical results for @xmath136 and related quantities . of course the running of the coupling with the cut - off scale has been investigated both analytically and numerically for both two- and three - color gauge theory , in many previous works ( see e.g. @xcite and references therein ) . our main motivation for repeating such an investigation is to make use of the advanced computing power of the gpus to measure at larger @xmath137 and @xmath45 and with larger statistics as was previously possible and thus reduce systematic and statistical errors . all physical units quoted in this work implicitly refer to the results presented in this appendix . in pure gauge theory the static quark - antiquark potential is known @xcite to be well described by v(r)=a+(1/r)+r + o(1/ r^2).[eq : quarkpot3 ] the linear coefficient @xmath138 is the string tension . on the lattice we can measure the potential in lattice units , _ i.e. _ @xmath139 , through the use of rectangular wilson loops @xmath140 in the @xmath141 plane ( our definition of @xmath142 implies that the trace is taken ) . these are related to the potential by w(n_x , n_t ) = e^-n_t a v(a n_x ) ( 1+o ( e^-n_t a e ) ) , [ eq : quarkpot1 ] where @xmath143 is the energy gap between the ground state and the first excited state of a quark - antiquark pair . since the second term is exponentially suppressed with @xmath144 , in the limit of large @xmath144 the potential can be directly obtained through the use of _ creutz ratios _ a v(a n_x ) .[eq : quarkpot2 ] since eq . ( [ eq : quarkpot3 ] ) is written in lattice units as a v(a n_x)=a a+(1/n_x)+ a^2 n_x + o(1/ ( a n_x ) ^2 ) , [ eq : quarkpot4 ] it is clear that what we actually obtain from a @xmath90 fit of the data is the dimensionless product @xmath145 . fixing the string tension @xmath138 to its physical @xmath146 value ( or some other definite value @xmath147 ) then determines @xmath134 for a given @xmath64 . the lattice spacing @xmath134 depends on the choice of @xmath138 of course , and for theories where this is not directly related to observable physics ( as in su(2 ) gauge theory ) one in fact often leaves @xmath138 undefined , which implies that @xmath134 and all derived quantities are expressed in units of an external parameter . the crucial point here is that the @xmath64 dependence of the lattice spacing is in fact temperature independent . this implies that once @xmath148 is known at @xmath146 , this fixes the temperature ( through @xmath149 ) and the physical units of all measured quantities for any given @xmath45 and @xmath64 . to improve the signal - to - noise ratio for our string tension measurements we employ a method known as ape smearing @xcite . this method consists of repeatedly replacing all gauge links @xmath150 with averages of the form u_(n)=(1-)u_(n)+ _ c_(n ) , [ eq : apesmear1 ] where @xmath151 represents the staple matrix c_(n)&=&u_(n ) u_(n+ ) u_^(n+ ) + & + & u_^(n- ) u_(n- ) u_(n-+ ) , [ eq : apesmear2 ] and re - unitarizing the link ( projecting back to @xmath152 ) after each iteration . the crucial effect of this procedure is to enhance the ground state overlap . ape smearing is applied to a copy of the lattice ( the monte - carlo thus remains unaffected ) on which the wilson loops are subsequently measured . a good choice for the parameter @xmath153 is known @xcite to be = 1 - ; w = 0.2 = 0.5454 . we adopt this choice as well and apply the smearing @xmath154 times before measuring wilson loops on a given lattice configuration . we measure @xmath155 on lattices of sizes @xmath156 , @xmath157 and @xmath158 , for @xmath64 values ranging from @xmath159 to @xmath160 using ( [ eq : quarkpot2 ] ) and ( [ eq : quarkpot4 ] ) . sample sizes range from measurements on each site of several hundreds ( for the larger @xmath64 values ) to several tens of thousands ( for the smaller @xmath64 values ) of independent ( smeared ) gauge field configurations . from these results we obtain @xmath161 , which is shown in fig . [ fig : runcoupl1 ] . wilson loops are measured up to spatial and timelike extents of @xmath162 and @xmath163 respectively . we find consistency with the data sets quoted in ref . @xcite . one must consider potential sources of systematic errors . firstly there are finite volume and discretization errors , for which we check by comparing the results obtained from lattices of different dimensions . we find that our different choices give consistent results in @xmath64 ranges where they overlap ( at least two different lattices overlap in any given region ) , therefore we consider these errors to be under control . second , one must take care to choose the timelike extent @xmath144 of the wilson loops large enough , such that the contribution from the exited states to the creutz ratios ( [ eq : quarkpot2 ] ) can be neglected . this is tricky since the signal diminishes with rising @xmath144 . we find that we can obtain , with reasonably small statistical errors , consistency between @xmath166 and @xmath167 on the two smaller lattices for each value of @xmath64 . on the largest lattice consistent results are obtained on @xmath168 . [ fig : runcoupl1 ] plots both respective choices for each lattice as the same dataset . we consider this source of error to be under control as well . the last , and potentially most severe , source of systematic errors are the higher order contributions to the potential ( [ eq : quarkpot3 ] ) , which one must account for . we find that an excellent fit of the data can be obtained by discarding higher order terms and restricting the fit to @xmath169 . alternatively one could introduce an additional term @xmath170 , or throw away more of the short - distance results . although we find that the resulting string - tension measurements are much noisier if either of the latter two options is chosen , we find evidence that using @xmath169 introduces a systematic error such that the string tension is over - estimated ( and thus the temperature is under - estimated ) for the region @xmath171 if higher - order terms are excluded . we choose to combine two sets of string tension data as our final results . one with @xmath169 where the next term of ( [ eq : quarkpot3 ] ) is included and one without such a term but fitting only to @xmath172 . we find that they are consistent . we take this as evidence that the systematic errors from higher - order terms are small for our final results . the data points shown in the figures in this section represent the measurements with the smallest errorbars . to avoid cluttering of the figures , we refrain from showing additional points which were obtained but come with huge errorbars . all points were used however to determine the parameters of the model functions discussed in the following . our main goal is to obtain the @xmath64 dependence of @xmath8 and its inverse for different @xmath45 . this can be achieved if both @xmath175 for a given @xmath45 and @xmath176 are known . furthermore , it is desirable to obtain not only a finite set of values but model - functions for each of these , valid over a large as possible range . a reasonable approach is to attempt to fit the data with known perturbative formulae , possibly with additional correction terms . from @xmath161 we directly obtain @xmath177 . the results are plotted in fig . [ fig : runcoupl2 ] for different @xmath45 . we find that it is possible to model our results by taking the inverse of the leading logarithmic term of the perturbative running - coupling formula ( see e.g. @xcite ) . defining r()= ( ) , [ eq : runcouplfit2 ] and using a = , n_t = r ( ) .[eq : runcouplfit1 ] we find that our data shown in figs . [ fig : runcoupl1 ] and [ fig : runcoupl2 ] are well described for the entire range of @xmath64 considered with the parameters d= 1.98(1 ) , b= 0.305(6 ) .[eq : runcouplfit3 ] this is similar to what was done in ref . @xcite although the @xmath64 range considered here is larger . it is in fact somewhat amazing that a leading - order formula works so well over such a large range . it is also very convenient , since eq . ( [ eq : runcouplfit2 ] ) can be easily inverted . if we restrict our fit - window to the same as was used in @xcite , we obtain a consistent value for the parameter @xmath178 . if we know a @xmath179 for a given @xmath45 we are done , for we can then obtain = =( ) , [ eq : totc ] and likewise ( ) b+_c= . [ eq : runcouplfit4 ] our numerical results for @xmath180 can analogously be converted to @xmath8 . the parameter @xmath138 drops out entirely . we show @xmath64 as a function of @xmath8 in fig . [ fig : runcoupl3 ] , both as raw data and model - function . we collected @xmath76 values for @xmath182 from refs . @xcite and fit them to interpolate for other @xmath45 values , using _ c(n_t)= a_0 + b_0 ( n_t ) - ( ( n_t))^c_0 , [ eq : nlo_bc ] we find that the next - to - leading logarithm is necessary to appropriately describe such a range of @xmath45 . our fit - parameters converge to a_0 = 1.1579(6 ) , b_0=0.9398(1 ) , c_0=1.627(2 ) , with @xmath183 . ( [ eq : nlo_bc ] ) is used in eq . ( [ eq : totc ] ) to determine @xmath76 for all @xmath45 instead of the raw data . one last thing should be mentioned : we realize that the results presented in this appendix go beyond what is necessary for the study presented in the main sections . in particular , to obtain @xmath8 , when simulating on a lattice of time - like extent @xmath45 at a given @xmath64 which is known to be @xmath76 for a different @xmath184 , one can simply use the relation = .[eq : totc_ntontc ] we conducted the investigation discussed here hoping that it will be useful in a more general setting and we have chosen to present @xmath8 values throughout this paper as our general method provides them . equation ( [ eq : totc_ntontc ] ) however is very useful to us to estimate errors . it turns out that the @xmath8 values quoted throughout this paper differ by no more than @xmath185 from what ( [ eq : totc_ntontc ] ) predicts , which is quite satisfactory . g. t hooft , nucl . b * 138 * , 1 ( 1978 ) . a. m. polyakov , phys . b * 72 * , 477 ( 1978 ) . l. susskind , phys . d * 20 * , 2610 ( 1979 ) . l. d. mclerran and b. svetitsky , phys . d * 24 * , 450 ( 1981 ) . a. dumitru , j. lenaghan and r. d. pisarski , phys . rev . d * 71 * , 074004 ( 2005 ) [ hep - ph/0410294 ] . a. dumitru , y. hatta , j. lenaghan , k. orginos and r. d. pisarski , phys . d * 70 * , 034511 ( 2004 ) [ hep - th/0311223 ] . n. meisinger , t. r. miller and m. c. ogilvie , phys . d * 65 * , 034009 ( 2002 ) [ hep - ph/0108009 ] . p. n. meisinger and m. c. ogilvie , phys . d * 65 * , 056013 ( 2002 ) [ hep - ph/0108026 ] . r. d. pisarski , phys . d * 74 * , 121703 ( 2006 ) [ hep - ph/0608242 ] . a. dumitru , y. guo , y. hidaka , c. p. k. altes and r. d. pisarski , phys . rev . d * 83 * , 034022 ( 2011 ) [ arxiv:1011.3820 [ hep - ph ] ] . a. dumitru , y. guo , y. hidaka , c. p. k. altes and r. d. pisarski , phys . rev . d * 86 * , 105017 ( 2012 ) [ arxiv:1205.0137 [ hep - ph ] ] . t. umeda , s. ejiri , s. aoki , t. hatsuda , k. kanaya , y. maezawa and h. ohno , phys . d * 79 * , 051501 ( 2009 ) [ arxiv:0809.2842 [ hep - lat ] ] . m. panero , phys . lett . * 103 * , 232001 ( 2009 ) [ arxiv:0907.3719 [ hep - lat ] ] . s. datta and s. gupta , phys . d * 82 * , 114505 ( 2010 ) [ arxiv:1006.0938 [ hep - lat ] ] . s. borsanyi , g. endrodi , z. fodor , s. d. katz and k. k. szabo , jhep * 1207 * , 056 ( 2012 ) [ arxiv:1204.6184 [ hep - lat ] ] . m. caselle , l. castagnini , a. feo , f. gliozzi , u. gursoy , m. panero and a. schafer , jhep * 1205 * , 135 ( 2012 ) [ arxiv:1111.0580 [ hep - th ] ] . b. lucini and m. panero , phys . rept . * 526 * , 93 ( 2013 ) [ arxiv:1210.4997 [ hep - th ] ] . l. fister and j. m. pawlowski , arxiv:1301.4163 [ hep - ph ] . j. braun , h. gies and j. m. pawlowski , phys . b * 684 * , 262 ( 2010 ) [ arxiv:0708.2413 [ hep - th ] ] . d. smith , phys . d * 82 * , 034503 ( 2010 ) [ arxiv:0911.4037 [ hep - lat ] ] . d. smith ( 2010 ) , _ lattice simulation of a center symmetric three dimensional effective theory for su(2 ) yang - mills _ thesis , goethe university of frankfurt , germany . a. d. kennedy and b. j. pendleton , phys . b * 156 * , 393 ( 1985 ) . s. gupta , k. huebner and o. kaczmarek , phys . d * 77 * , 034503 ( 2008 ) [ arxiv:0711.2251 [ hep - lat ] ] . c. s. fischer , l. fister , j. luecker and j. m. pawlowski , arxiv:1306.6022 [ hep - ph ] . k. langfeld and j. m. pawlowski , arxiv:1307.0455 [ hep - lat ] . d. diakonov , c. gattringer and h. -p . schadler , jhep * 1208 * , 128 ( 2012 ) [ arxiv:1205.4768 [ hep - lat ] ] . b. a. berg , _ markov chain monte carlo simulations and their statistical analysis _ , world scientific publishing , singapore , 2004 j. greensite , phys . d * 86 * , 114507 ( 2012 ) [ arxiv:1209.5697 [ hep - lat ] ] . j. greensite and k. langfeld , arxiv:1301.4977 [ hep - lat ] . a. dumitru and d. smith , phys . d * 77 * , 094022 ( 2008 ) [ arxiv:0711.0868 [ hep - lat ] ] . d. smith , nucl . a * 820 * , 227 ( 2009 ) [ arxiv:0810.1129 [ hep - lat ] ] . g. s. bali and k. schilling , phys . d * 47 * , 661 ( 1993 ) [ hep - lat/9208028 ] . j. c. r. bloch , a. cucchieri , k. langfeld and t. mendes , nucl . b * 687 * , 76 ( 2004 ) [ hep - lat/0312036 ] . c. michael , phys . b * 283 * , 103 ( 1992 ) [ hep - lat/9205010 ] . a. cucchieri , hep - lat/0209076 . m. luscher , r. sommer , p. weisz and u. wolff , nucl . b * 413 * , 481 ( 1994 ) [ hep - lat/9309005 ] . s. necco and r. sommer , phys . b * 523 * , 135 ( 2001 ) [ hep - ph/0109093 ] . m. luscher and p. weisz , jhep * 0207 * , 049 ( 2002 ) [ hep - lat/0207003 ] . m. albanese _ et al . _ [ ape collaboration ] , phys . b * 192 * , 163 ( 1987 ) . n. cardoso and p. bicudo , j. comput . phys . * 230 * , 3998 ( 2011 ) [ arxiv:1010.4834 [ hep - lat ] ] . f. geles and c. b. lang , arxiv:1103.5368 [ hep - lat ] . c. morningstar and m. j. peardon , phys . d * 69 * , 054501 ( 2004 ) [ hep - lat/0311018 ] . j. fingberg , u. m. heller and f. karsch , nucl . b * 392 * , 493 ( 1993 ) [ hep - lat/9208012 ] . p. de forcrand and l. von smekal , phys . d * 66 * , 011504 ( 2002 ) [ hep - lat/0107018 ] . a. velytsky , int . j. mod . c * 19 * , 1079 ( 2008 ) [ arxiv:0711.0748 [ hep - lat ] ] . x. cheng and e. t. tomboulis , phys . d * 86 * , 074507 ( 2012 ) [ arxiv:1206.3816 [ hep - lat ] ] .

let @xmath9 be a zero mean process having second moments @xmath10 < \infty$ ] . consider the partial sum @xmath11 and its normalized variance @xmath12 $ ] . a very important issue in probability theory and statistics is whether or not the central limit theorem holds , that is , if we have @xmath13 where @xmath14 denotes the standard normal distribution function . going one step further , we can ask ourselves about the possible rate of convergence in ( [ eqclt ] ) , more precisely , if it holds that @xmath15 where @xmath16 is a probability metric , @xmath17 follows a standard normal distribution and @xmath18 denotes the probability measure induced by the random variable @xmath19 . the rate @xmath20 can be considered as a measure of reliability for statistical inference based on @xmath21 , and large rates are naturally preferred . the question of rate of convergence has been addressed under numerous different setups with respect to the metric and underlying structure of the sequence @xmath9 in the literature . perhaps one of the most important metrics is the kolmogorov ( uniform ) metric , given as @xmath22 the latter has been studied extensively in the literature under many different notions of ( weak ) dependence for @xmath23 . one general way to measure dependence is in terms of various mixing conditions . in the case of the uniform metric , bolthausen @xcite and rio @xcite showed that it is possible to obtain the rate @xmath24 in ( [ eqcltrateks ] ) , given certain mixing assumptions and a bounded support of the underlying sequence @xmath9 ; see also @xcite , among others , for related results and extensions . under the notion of @xmath25-mixing , tikhomirov @xcite obtained @xmath26 , provided that @xmath27 < \infty$ ] and the mixing coefficient decays exponentially fast ; see also @xcite . martingales constitute another important class for the study of ( [ eqcltrateks ] ) . some relevant contributions in this context are , for instance , brown and heyde @xcite , bolthausen @xcite and more recently dedecker et al . @xcite . in the special case of functionals of gaussian or poissonian sequences , deep results have been obtained by noudin and peccati et al . ; see , for instance , @xcite and @xcite . another stream of significant works focuses on stationary ( causal ) bernoulli - shift processes , given as @xmath28 the study of ( [ eqcltrateks ] ) given the structure in ( [ eqstructurecondition ] ) has a long history , and dates back to kac @xcite and postnikov @xcite . ibragimov @xcite established a rate of convergence , @xmath29 , subject to an exponentially fast decaying weak dependence coefficient . using the technique of tikhomirov @xcite , gtze and hipp obtained edgeworth expansions for processes of type ( [ eqstructurecondition ] ) in a series of works ; cf . @xcite ; see also heinrich @xcite and lahiri @xcite . this approach , however , requires the validity of a number of technical conditions . this includes in particular a conditional crmer - like condition subject to an exponential decay , which is somewhat difficult to verify . in contrast , it turns out that a berry esseen theorem only requires a simple , yet fairly general dependence condition where no exponential decay is required . indeed , we will see that many popular examples from the literature are within our framework . unlike previous results in the literature , we also obtain optimal rates for @xmath30 given ( infinite ) weak dependence , which to the best of our knowledge is new ( excluding special cases as linear processes ) . the proofs are based on an @xmath31-dependent approximation ( @xmath32 ) , which is quite common in the literature . the substantial difference here is the subsequent treatment of the @xmath31-dependent sequence . to motivate one of the main ideas of the proofs , let us assume @xmath33 for a moment . given a weakly , @xmath31-dependent sequence @xmath34 , one may show via classic arguments that @xmath35,\end{aligned}\ ] ] provided that @xmath36 < \infty$ ] and @xmath37 . note , however , since @xmath38 is weakly dependent , one finds that @xmath39 \bigr| \leq\frac { c}{\sqrt{m}}.\end{aligned}\ ] ] hence if one succeeds in replacing @xmath36 $ ] in ( [ eqboundmom2.9ex ] ) with ( [ eqboundmom3ex ] ) , one obtains the optimal rate @xmath24 . a similar reasoning applies to @xmath40 . unfortunately though , setting this idea to work leads to rather intricate problems , and a technique like that of tikhomirov @xcite is not fruitful , inevitably leading to a suboptimal rate . our approach is based on coupling and conditioning arguments and ideal ( zolotarev ) metrics . interestingly , there is a connection to more recent results of dedecker et al . @xcite , who consider different ( smoother ) probability metrics . we will see that at least some of the problems we encounter may be redirected to these results after some preparation . throughout this paper , we will use the following notation : for a random variable @xmath19 and @xmath41 , we denote with @xmath42^{1/p}$ ] the @xmath43 norm . let @xmath44 be a sequence of independent and identically distributed random variables with values in a measurable space @xmath45 . denote the corresponding @xmath46-algebra with @xmath47 . given a real - valued stationary sequence @xmath48 , we always assume that @xmath38 is adapted to @xmath49 for each @xmath50 . hence we implicitly assume that @xmath51 can be written as in ( [ eqstructurecondition ] ) . for convenience , we write @xmath52 with @xmath53 . the class of processes that fits into this framework is large and contains a variety of functionals of linear and nonlinear processes including arma , garch and related processes ( see , e.g. , @xcite ) , but also examples from dynamic system theory . some popular examples are given below in section [ secex ] . a nice feature of the representation given in ( [ eqstructurecondition ] ) is that it allows us to give simple , yet very efficient and general dependence conditions . following wu @xcite , let @xmath54 be an independent copy of @xmath55 on the same probability space , and define the `` filter '' @xmath56 as @xmath57 we put @xmath58 and @xmath59 , and in particular we set @xmath60 . as a dependence measure , we then consider the quantity @xmath61 , @xmath41 . dependence conditions of this type are quite general and easy to verify in many cases ; cf . @xcite and the examples below . observe that if the function @xmath62 does not depend on @xmath63 , we obtain the simpler version @xmath64 note that it is actually not trivial to construct a stationary process @xmath9 that can only be represented as @xmath65 ; that is , a function @xmath66 independent of @xmath63 such that @xmath67 for all @xmath68 does not exist . we refer to corollary 2.3 in feldman and rudolph @xcite for such an example . we will derive all of our results under the following assumptions . [ assmaindependence ] let @xmath69 be stationary such that for some @xmath70 : @xmath71 , @xmath72 = 0 $ ] , @xmath73 , @xmath74 , where @xmath75 $ ] . in the sequel , @xmath76 denotes a varying absolute constant , depending only on @xmath77 , @xmath78 and @xmath79 . the following theorem is one of the main results of this paper . [ thmberry ] grant assumption [ assmaindependence ] for some @xmath1 $ ] , and let @xmath80 . then @xmath81 and hence we may select @xmath82 . theorem [ thmberry ] provides optimal convergence rates under mild conditions . in particular , it seems that this is the first time optimal rates are shown to hold under general infinite weak dependence conditions if @xmath83 . examples to demonstrate the versatility of the result are given in section [ secex ] . in particular , we consider functions of the dynamical system @xmath84 in example [ exnumber ] , a problem which has been studied in the literature for decades . combining theorem [ thmberry ] with results of dedecker and rio @xcite , we also obtain optimal results for the @xmath85-norm for martingale differences . [ thmlp ] grant assumption [ assmaindependence ] for some @xmath3 , and let @xmath80 . if @xmath69 is a martingale difference sequence , then for any @xmath6 we have @xmath86 note that in the case @xmath87 , the results of dedecker and rio @xcite are more general . the nonuniform analogue to theorems [ thmberry ] and [ thmlp ] is given below . here , we obtain optimality up to logarithmic factors . [ thmnonunif ] grant assumption [ assmaindependence ] for some @xmath8 . then for any @xmath88 , @xmath89 where @xmath90 . as a particular application of theorem [ thmnonunif ] , consider @xmath91 where the function @xmath92 satisfies @xmath93 for some @xmath94 , and the derivative @xmath95 exists for @xmath96 . if @xmath97 , property ( [ eqgconditforapprox ] ) implies the identity @xmath98 = \int_{0}^{\infty } f'(x ) p \bigl(|s_n|\big/\sqrt{n s_n^2 } \geq x \bigr ) \,dx,\end{aligned}\ ] ] and we thus obtain the following corollary . [ corgconvergence ] grant assumption [ assmaindependence ] for some @xmath8 . if ( [ eqgconditforapprox ] ) holds , then @xmath99 - \int_{\mathbb{r}}f \bigl(|x| \bigr)\,d \phi(x ) \biggr| \leq { b}n^{-(p \wedge 3)/2 + 1 } ( \log n ) ^{p/2}.\end{aligned}\ ] ] as a special case , consider @xmath100 , @xmath101 . we may then use corollary [ corgconvergence ] to obtain rates of convergence for moments . [ cormomconvergence ] grant assumption [ assmaindependence ] for some @xmath8 . then for any @xmath102 , we have @xmath103 in the special case of i.i.d . sequences and @xmath104 , sharp results in this context have been obtained in hall @xcite . it seems that related results for dependent sequences are unknown . all examples considered here are time - homogenous bernoulli - shift processes ; that is , @xmath62 does not depend on @xmath63 , and hence equality ( [ eqhomodepcondi ] ) holds . let @xmath105 , and suppose that the sequence @xmath106 satisfies @xmath107 . if @xmath108 , then one may show that the linear process @xmath109 let @xmath110 be a measurable function such that @xmath111 = 0 $ ] , where @xmath112 . if @xmath110 is hlder continuous with regularity @xmath113 , that is , @xmath114 hence if @xmath115 and @xmath74 , then assumption [ assmaindependence ] holds . [ exnumber ] consider the measure preserving transformation @xmath84 on the probability space @xmath116,\mathcal{b},\bolds{\lambda } ) $ ] , with borel @xmath46-algebra @xmath117 and lebesgue measure @xmath118 . let @xmath119 $ ] . then @xmath120 , where @xmath121 are bernoulli random variables . the flow @xmath122 can then be written as @xmath123 ; see @xcite . the study about the behavior of @xmath124 for appropriate functions @xmath110 has a very long history and dates back to kac @xcite . since then , numerous contributions have been made ; see , for instance , @xcite , to name a few . here , we consider the following class of functions . let @xmath110 be a function defined on the unit interval @xmath125 $ ] , such that @xmath126 \label{eqibracondi } \\[-8pt ] \nonumber & & \int_0 ^ 1 t^{-1 } \bigl| \log(t)\bigr|^2 w_p ( f , t ) \ , dt < \infty,\end{aligned}\ ] ] where @xmath127 denotes the @xmath128 , \bolds{\lambda } ) $ ] modulos of continuity of @xmath129,\break\bolds{\lambda } ) $ ] . this setup is a little more general than in @xcite . for @xmath130 , let @xmath131 ; that is , @xmath132 is the one - periodic extension to the positive real line . one then often finds the equivalent formulation @xmath133 in the literature . consider now the partial sum @xmath134 . ibragimov @xcite showed that @xmath135 by alternative methods , le borgne and pne @xcite , extending rio @xcite , managed to remove the logarithmic factor if @xmath136 . a priori , the sequence @xmath137 does not directly fit into our framework , which , however , can be achieved by a simple time flip . define the function @xmath138 for @xmath139 , and let @xmath140 . then we may write @xmath141 note that we have to perform this time flip for every @xmath142 , which , however , has no impact on the applicability of our results . using the same arguments as in @xcite , we find that ( [ eqibracondi ] ) implies that for @xmath1 $ ] @xmath143 if @xmath74 , we see that assumption [ assmaindependence ] holds . in particular , an application of theorem [ thmberry ] gives the rate @xmath82 , thereby removing the unnecessary @xmath144 factor in ( [ eqexnumber1 ] ) for the whole range @xmath1 $ ] . [ exmdep ] @xmath145consider the zero mean @xmath31-dependent process @xmath146 , where @xmath147 and @xmath110 is a measurable function and @xmath148 is i.i.d . and takes values in @xmath149 . @xmath31 may depend on @xmath150 such that @xmath151 , but we demand in addition that @xmath152\big/ ( n m ) > 0.\end{aligned}\ ] ] in this context , it is useful to work with the transformed block - variables @xmath153 and write @xmath154 where @xmath155 ; hence @xmath9 is a two - dependent sequence . this representation ensures that assumption [ assmaindependence](i ) and ( ii ) hold for @xmath156 , independently of the value of @xmath31 . the drawback of this block - structure is that we loose a factor @xmath31 , since we have @xmath157 where we assume that @xmath158 for simplicity . however , this loss is known in the literature : theorem [ thmberry ] now yields the commonly observed rate @xmath159 in the context of @xmath31-dependent sequences satisfying ( [ eqmdepexvarcondi ] ) ; see , for instance , theorem 2.6 in @xcite . in the latter , the rate @xmath159 is not immediately obvious , but follows from elementary computations using ( [ eqmdepexvarcondi ] ) . iterated random functions ( cf . @xcite ) are an important class of processes . many nonlinear models like arch , bilinear and threshold autoregressive models fit into this framework . let @xmath105 and @xmath160 be defined via the recursion @xmath161 commonly referred to as _ iterated random functions _ ; see , for instance , @xcite . let @xmath162 be the lipschitz coefficient . if @xmath163 and @xmath164 for some @xmath165 , then @xmath38 can be represented as @xmath166 for some measurable function @xmath66 . in addition , we have @xmath167 see @xcite . hence if @xmath168 = 0 $ ] and @xmath169 , assumption [ assmaindependence ] holds . as an example , consider the _ stochastic recursion _ @xmath170 where @xmath171 is an i.i.d . sequence . let @xmath172 . if we have , for some @xmath70 , @xmath173 = 0,\end{aligned}\ ] ] then @xmath174 , and assumption [ assmaindependence ] holds if @xmath74 . in particular , if @xmath175 are independent , then one readily verifies that @xmath176}{1 - \mathbb{e}[a_0 ] } \biggr),\end{aligned}\ ] ] which is strictly positive since @xmath177 | < 1 $ ] by jensen s inequality . hence if ( [ stocheqcondi ] ) holds for @xmath178 , then assumption [ assmaindependence ] holds for @xmath77 . analogue conditions can be derived for higher order recursions . [ exgarch ] let @xmath105 . another very prominent stochastic recursion is the @xmath179 sequence , given through the relations @xmath180 with @xmath181 . we assume that @xmath182 for some @xmath70 . an important quantity is @xmath183 where we replace possible undefined @xmath184 with zero . if @xmath185 , then @xmath9 is stationary ; cf . in particular , it was shown in @xcite that @xmath9 may be represented as @xmath186 using this representation and the fact that @xmath187 for @xmath188 , @xmath41 , one can follow the proof of theorem 4.2 in @xcite to show that @xmath189 since @xmath190 = \mathbb{e } [ \epsilon _ k ] = 0 $ ] , assumption [ assmaindependence ] holds if @xmath74 . we remark that previous results on @xmath191 , in the case of garch(@xmath192 sequences , either require heavy additional assumptions or have suboptimal rates ; cf . @xcite . in the study of nonlinear processes , volterra processes are of fundamental importance . following berkes et al . @xcite , we consider @xmath193 where @xmath105 and @xmath194 for @xmath70 , and @xmath195 are called the @xmath63th volterra kernel . let @xmath196 then there exists a constant @xmath197 such that @xmath198 thus if @xmath199 and @xmath200 , then assumption [ assmaindependence ] holds . the main approach consists of an @xmath31-dependent approximation where @xmath201 , followed by characteristic functions and esseen s inequality . however , here the trouble starts , since we can not factor the characteristic function as in the classic proof , due to the @xmath31-dependence . tikhomirov @xcite uses a chaining - type argument , which is also fruitful for edgeworth expansions ; cf . however , since this approach inevitably leads to a loss in the rate , this is not an option for berry esseen - type results . in order to circumvent this problem , we first work under an appropriately chosen conditional probability measure @xmath202 . unfortunately though , this leads to rather intricate problems , since all involved quantities of interest are then random . we first consider the case of a weakly @xmath31-dependent sequence @xmath9 , where @xmath32 as @xmath150 increases . note that this is different from example [ exmdep ] . for the general case , we then construct a suitable @xmath31-dependent approximating sequence such that the error of approximation is negligible , which is carried out in section [ secproofofmaintheorem ] . the overall proof of theorem [ thmberry ] is lengthy . important technical auxiliary results are therefore established separately in section [ secmainlemmas ] . minor additionally required results are collected in section [ secauxresults ] . the proofs of theorems [ thmlp ] and [ thmnonunif ] are given in sections [ secprooflp ] and [ secthmproofnonunif ] . to simplify the notation in the proofs , we restrict ourselves to the case of homogeneous bernoulli shifts , that is , where @xmath203 , and the function @xmath66 does not depend on @xmath63 . this requires substantially fewer indices and notation throughout the proofs , and , in particular , ( [ eqhomodepcondi ] ) holds . the more general nonhomogenous ( but still stationary ) case follows from straightforward ( notational ) adaptations . this is because the key ingredient we require for the proof is the bernoulli - shift structure ( [ eqstructurecondition ] ) in connection with the summability condition , assumption [ assmaindependence](ii ) . whether or not @xmath66 depends on @xmath63 is of no relevance in this context . in order to deal with @xmath31-dependent sequences , we require some additional notation and definitions . throughout the remainder of this section , we let @xmath204 and measurable functions @xmath205 , where @xmath206 as @xmath150 increases . we work under the following conditions : [ assdependence ] let @xmath207 be such that for some @xmath208 , uniformly in @xmath31 : @xmath71 , @xmath72 = 0 $ ] , @xmath209 , @xmath210 , where @xmath211 = \sum_{k = -m}^{m}\mathbb{e } [ x_0x_k ] $ ] . observe that this setup is fundamentally different from that considered in example [ exmdep ] . in particular , here we have that @xmath212 \thicksim n $ ] . define the following @xmath46-algebra : @xmath213 where we recall that @xmath214 and @xmath215 are mutually independent , identically distributed random sequences . we write @xmath216 for the conditional law and @xmath217 $ ] ( or @xmath218 $ ] ) for the conditional expectation with respect to @xmath219 ( or some other filtration @xmath220 ) . we introduce @xmath221 \quad \mbox{and}\quad s_{|m}^{(2 ) } = \sum_{k = 1}^n \mathbb{e } [ x_k|\mathbb{f } _ m ] , \end{aligned}\ ] ] hence @xmath222 to avoid any notational problems , we put @xmath223 for @xmath224 . let @xmath225 , where @xmath226 are chosen such that @xmath227 and @xmath228 is an absolute constant , independent of @xmath229 . for @xmath230 , we construct the block random variables @xmath231 \quad \mbox{and}\quad r_j = \sum_{k = ( 2j-1)m + 1}^{2jm } x_k - \mathbb{e } [ x_k|\mathbb{f}_m ] , \end{aligned}\ ] ] and put @xmath232 , hence @xmath233 . note that by construction of the blocks , @xmath234 , @xmath235 are independent random variables under the conditional probability measure @xmath236 , and are identically distributed at least for @xmath237 under @xmath238 . we also put @xmath239 $ ] and @xmath240 $ ] for @xmath241 . note that @xmath242 , @xmath243 is a sequence of independent random variables . the following partial and conditional variances are relevant for the proofs : @xmath244\quad \mbox{and}\quad \sigma_j^2 = \mathbb{e } \bigl[\sigma_{j|m}^2 \bigr ] , \\ \sigma_{|m}^2 & = & \frac{1}{n}\mathbb{e } \bigl[\bigl(s_{|m}^{(1)}\bigr)^2 | \mathbb{f}_m \bigr ] = \frac{1}{n + m'/2 m } \sum _ { j = 1}^n \sigma_{j|m}^2 , \\ \overline{\sigma}_m^2 & = & \mathbb{e } \bigl [ \sigma_{|m}^2 \bigr ] = \frac { 1}{n + m'/2 m } \sum _ { j = 1}^n \sigma_j^2 , \\ \widehat{\sigma}_m^2 & = & \frac{1}{2m}\sum _ { k = 1}^m\sum_{l = 1}^m \mathbb{e } [ x_kx_l ] .\end{aligned}\ ] ] as we shall see below , these quantities are all closely connected . note that @xmath245 for @xmath246 , but @xmath247 in general . moreover , we have the equation @xmath248 yields that under assumption [ assdependence ] we have @xmath249 . moreover , lemma [ lemsigexpansion ] gives @xmath250 for @xmath251 . we conclude that @xmath252 the same is true for @xmath253 , since @xmath254 . summarizing , we see that we do not have any degeneracy problems for the partial variances @xmath255 , @xmath230 under assumption [ assdependence ] . for the second part @xmath256 , we introduce @xmath257 . one then readily derives via conditioning arguments that @xmath258 we are now ready to give the main result of this section . [ thmmdependent ] grant assumption [ assdependence ] , and let @xmath1 $ ] . assume in addition that @xmath259 for @xmath260 . then @xmath261 where @xmath262 depends on @xmath263 , @xmath77 , @xmath264 and @xmath265 . the proof of theorem [ thmmdependent ] is based on the following decomposition . let @xmath266 be independent unit gaussian random variables . then @xmath267 where @xmath268 are defined as @xmath269 \bigr| , \\ \mathbf{b } & = & \sup_{x \in\mathbb{r } } \bigl|\mathbb{e } \bigl[p_{|\mathbb{f}_m } \bigl(z_1 \sigma_{|m } \leq x - s_{|m}^{(2)}/ \sqrt{n } \bigr ) - p_{|\mathbb{f } _ m } \bigl(z_1 \overline{\sigma}_{m } \leq x - s_{|m}^{(2)}/\sqrt { n } \bigr ) \bigr ] \bigr| , \\ \mathbf{c } & = & \sup_{x \in\mathbb{r } } \bigl|p \bigl(s_{|m}^{(2)}/ \sqrt{n } \leq x - z_1 \overline{\sigma}_{m } \bigr ) - p ( z_2 { \overline { \varsigma}}_{m } \leq x - z_1 \overline{\sigma}_{m } ) \bigr|.\end{aligned}\ ] ] we will treat the three parts separately , and show that @xmath270 , which proves theorem [ thmmdependent ] . as a brief overview , the proof consists of the following steps : apply esseen s smoothing inequality , and factor the resulting characteristic function into a ( conditional ) product of characteristic functions @xmath271 under the conditional probability measure @xmath202 ; use ideal metrics to control the distance between @xmath272 and corresponding gaussian versions under @xmath202 ; based on renyi s representation , control the ( conditional ) characteristic functions @xmath271 under @xmath238 ; replace conditional variances under the overall probability measure @xmath238 . one of the main difficulties arises from working under the conditional measure @xmath202 . for the proof , we require some additional notation . in analogy to the filter @xmath56 , we denote with @xmath273 , @xmath274 we put @xmath275 and @xmath276 , and in particular , we have @xmath277 . similarly , let @xmath278 be independent copies of @xmath214 and @xmath215 . for @xmath279 , we then introduce the quantities @xmath280 in analogy to @xmath281 . this means that we replace every @xmath282 with @xmath283 at all corresponding places . for @xmath284 , we also introduce the @xmath46-algebras @xmath285 \\[-8pt ] \nonumber \mathcal { e}_k^ * & = & \sigma \bigl(\epsilon_j , 1 \leq j \leq k \mbox { and } \epsilon_i ' , i \leq0 \bigr).\end{aligned}\ ] ] similarly , we introduce filtrations @xmath286 and @xmath287 . throughout the proofs , we make the following conventions : we do not distinguish between @xmath288 and @xmath289 since the difference @xmath290 is not of any particular relevance for the proofs . we use @xmath288 for both expressions . the abbreviations @xmath291 for expressions ( possible with some additional indices ) vary from proof to proof . we use @xmath292 , @xmath293 , ( @xmath294 ) to denote ( two - sided ) inequalities involving a multiplicative constant . if there is no confusion , we put @xmath295 for @xmath235 to lighten the notation , particularly in part @xmath296 . we write [ as in ( [ defnsnm ] ) ] @xmath297 if we make definitions on the fly . the proof of part @xmath298 is divided into four major steps . some more technical arguments are deferred to sections [ seczolo ] and [ secboundcondichar ] . for @xmath299 , put @xmath300 , and denote with @xmath301 its complement . since @xmath302 , we obtain that @xmath303 \bigr| \\ & \leq & \mathbb{e } \bigl[\sup_{y \in \mathbb{r } } \bigl|p_{|\mathbb{f}_m } \bigl(s_{|m}^{(1)}/\sqrt{n } \leq y \bigr ) - p_{|\mathbb{f } _ m } ( z_1 \sigma_{|m } \leq y ) \bigr|\mathbh{1}(\mathcal { b}_{n } ) \bigr ] + 2p \bigl(\mathcal{b}_{n}^{c } \bigr).\end{aligned}\ ] ] corollary [ corsigreplaceissmall ] yields that @xmath304 since @xmath305 , and it thus suffices to treat @xmath306 berry esseen inequality . denote with @xmath307 the smoothed version of @xmath308 ( cf . @xcite ) as in the classical approach . since @xmath309 on the set @xmath310 by construction , the smoothing inequality ( cf . @xcite , lemma 1 , xvi.3 ) is applicable , and it thus suffices to treat @xmath307 . let @xmath311 $ ] , and put @xmath312 , where @xmath313 will be specified later . due to the independence of @xmath314 under @xmath315 and since @xmath316 , it follows that @xmath317 \leq\int_{-t}^{t}\mathbb{e } \biggl [ \biggl|\prod _ { j = 1}^n \varphi_j ( \xi/ \sqrt{n } ) - \prod_{j = 1}^n e^{-\sigma_{j|m}^2 \xi^2/2n } \biggr| \biggr ] \big/|\xi| \,d \xi.\end{aligned}\ ] ] put @xmath318 . then @xmath319 , where we use the convention that @xmath320 if @xmath321 or . hence we have @xmath322 note that both @xmath323 and @xmath324 are two - dependent sequences . since @xmath325 , @xmath326 , it then follows by the triangle inequality , stationarity and `` leave one out '' that @xmath327 we proceed by obtaining upper bounds for @xmath328 and @xmath329 . bounding @xmath330 , @xmath331 . let @xmath332 , @xmath331 be two zero mean standard gaussian random variables . then @xmath333 \bigr\|_1 \\ & & { } + \bigl\|\mathbb{e}_{\mathbb{f}_m } \bigl[\sin(ty_i ) - \sin(t\sigma _ { i|m}z_i ) \bigr ] \bigr\|_1.\end{aligned}\ ] ] due to the very nice analytical properties of @xmath334 , one may reformulate the above in terms of ideal - metrics ; cf . @xcite and section [ seczolo ] . this indeed leads to the desired bound @xmath335 the precise derivation is carried out in section [ seczolo ] via lemmas [ lemtrunccharcosandsin ] and [ lemapproxchar ] , and corollary [ corzolo ] . whether @xmath336 or @xmath337 makes no difference . bounding @xmath338 , we require good enough estimates for @xmath339 where @xmath340 . as already mentioned , we can not directly follow the classical approach . instead , we use a refined version based on a conditioning argument . to this end , let us first deal with @xmath341 . put @xmath342 we first consider the case @xmath343 . introduce @xmath344 \bigr ) + r_1\quad\mbox{and } \nonumber \\[-8pt ] \label{eqthmauxcharequationbound2.89 } \\[-8pt ] \nonumber v_{1}^{(l)}(m ) & = & \mathit{iv}_1^{(l)}(m ) - \mathbb{e}_{\mathcal { g}_1^{(l ) } } \bigl[\mathit{iv}_1^{(l)}(m ) \bigr].\end{aligned}\ ] ] then since @xmath345 , we have @xmath346 \bigr| \bigr].\end{aligned}\ ] ] clearly , this is also valid for @xmath341 , @xmath241 , with corresponding @xmath347 and @xmath348 , @xmath349 , defined analogously to ( [ eqthmauxcharequationbound2.89 ] ) . let @xmath350,\end{aligned}\ ] ] and @xmath351 , and hence @xmath352 denotes the set of all even numbers between @xmath353 and @xmath354 . then @xmath355 \bigr| \bigr ] = \prod _ { j \in\mathcal{j } } \mathbb{e}_{\mathbb{f}_m } \bigl [ \bigl|\varphi _ j^{(l)}(x ) \bigr| \bigr],\end{aligned}\ ] ] where @xmath356 . note that @xmath357 is a sequence of i.i.d . random variables , particularly with respect to @xmath202 . hence by independence and jensen s inequality , it follows from the above that @xmath358 \bigr\|_1 \nonumber \\[-8pt ] \label{eqthmauxcharequationbound2.905 } \\[-8pt ] \nonumber & \leq & \prod _ { j \in \mathcal{j } } \bigl\|\varphi_j^{(l)}(x ) \bigr\|_1 = \biggl\|\prod_{j \in\mathcal{j } } \bigl| \varphi_j^{(l)}(x ) \bigr| \biggr\|_1.\end{aligned}\ ] ] we thus see that it suffices to deal with @xmath359 . the classical argument uses the estimate @xmath360 for the characteristic function @xmath361 . since in our case @xmath362 is random , we can not use this estimate . instead , we will use lemma [ lemgeneralvarphibound ] , which provides a similar result . in order to apply it , set @xmath363 , @xmath364 for the applicability of lemma [ lemgeneralvarphibound ] , we need to verify that : @xmath365 = 0 $ ] ; there exists a @xmath366 such that @xmath367 \leq u^- ) < 1/7 $ ] , uniformly for @xmath368 ; @xmath369 uniformly for @xmath368 and some @xmath370 . now ( i ) is true by construction . claim ( ii ) is dealt with via lemma [ lemauxhsigcomp ] , which yields that @xmath371 \leq\widehat { \sigma } _ { m - l}^2 \bigr ) \lesssim \frac{1}{\sqrt{m - l}}.\end{aligned}\ ] ] since @xmath372 for large enough @xmath373 ( say @xmath374 ) by lemma [ lemsigexpressionsrelations ] , we may set @xmath375 . for showing ( iii ) , it suffices to treat the case @xmath343 . note that ( for @xmath376 ) @xmath377 = \mathbb { e}_{\mathcal{e}_l } \bigl[x_k - x_k^{(k - l , * ) } \bigr]\quad \mbox{and } \quad \mathbb{e}_{\mathbb { f}_m } [ x_k ] = \mathbb{e}_{\mathbb{f}_m } \bigl[x_k - x_{k}^ * \bigr].\end{aligned}\ ] ] by stationarity and the triangle and jensen inequalities , we then have that @xmath378 \biggr\|_p \nonumber \\[-8pt ] \label{eqthmauxcharequationbound4 } \\[-8pt ] \nonumber & & { } + 2 \biggl\|\sum_{k = l+1}^m \mathbb{e}_{\mathbb{f}_m } \bigl[x_k - x_k^ * \bigr ] \biggr\|_p + 2 \|r_1 \|_p.\end{aligned}\ ] ] using jensen s inequality and arguing similar to lemma [ lemboundr1 ] , it follows that @xmath379 \biggr\|_p & \leq & \sum_{k = l+1}^m \bigl\|x_k - x_k^{(k - l , * ) } \bigr\|_p \\ & \leq & \sum_{k = 1}^\infty k \bigl\|x_k - x_k ' \bigr\|_p < \infty.\end{aligned}\ ] ] similarly , using also lemma [ lemboundr1 ] to control @xmath380 , we obtain that @xmath381 \biggr\|_p + \|r_1 \|_p < \infty.\end{aligned}\ ] ] by lemma [ lemwuoriginal ] , we have @xmath382 , and hence ( iii ) follows . we can thus apply lemma [ lemgeneralvarphibound ] with @xmath383 and @xmath384 , which yields @xmath385 where @xmath356 . it is important to emphasize that both @xmath386 do not depend on @xmath387 and are strictly positive . moreover , we find from ( [ eqthmauxcharequationbound2.92 ] ) that @xmath388 can be chosen freely , as long as @xmath373 is larger than @xmath389 , which will be important in the next step . bounding and integrating @xmath390 . we first treat @xmath391 . recall that @xmath318 , hence @xmath392 by ( [ eqthmauxcharequationbound1 ] ) , ( [ eqthmauxcharequationbound2.905 ] ) and ( [ eqthmauxcharequationbound6 ] ) , it then follows for @xmath393 that @xmath394 to make use of this bound , we need to appropriately select @xmath395 . recall that @xmath396 , @xmath397 by assumption . choosing @xmath398 and @xmath399 , we obtain from the above that @xmath400 in order to treat @xmath401 , let @xmath402 , and @xmath403 . denote with @xmath404 its complement . then by corollary [ corsigreplaceissmall ] ( straightforward adaption is necessary ) and ( [ eqthmauxcharequationbound1 ] ) , it follows that @xmath405 similarly , using @xmath406 one obtains @xmath407 hence employing ( [ eqthmauxcharequationbound9 ] ) and ( [ eqthmauxcharequationbound10 ] ) yields @xmath408 \\[-8pt ] \nonumber & \lesssim & n^{-p/2 + 1 } + n^{-p/2 + 1 } n^{-p/2 } n^{p+1 } + n^{-p/2 } n^2 \log t \\ \nonumber & \lesssim & n^{-p/2 + 1},\end{aligned}\ ] ] since @xmath396 , @xmath397 by assumption . similarly , one obtains the same bound for @xmath329 . this completes the proof of part @xmath296 . let @xmath410.\end{aligned}\ ] ] recall that @xmath411 and @xmath412 by corollary [ corsigreplaceissmall ] . using properties of the gaussian distribution , it follows that @xmath413 \bigr| + \sup _ { x\in\mathbb{r } } \bigl|\mathbb{e } \bigl[\delta^{(2)}(x ) \mathbh{1}\bigl ( \mathcal{b}_{n}^c\bigr ) \bigr ] \bigr| \\ & \lesssim & \mathbb{e } \bigl [ \bigl|1/\sigma_{|m } - 1/\overline { \sigma}_{m } \bigr|\mathbh{1 } ( \mathcal{b}_n ) \bigr ] + n^{-p/2 + 1}.\end{aligned}\ ] ] using @xmath414 , hlders inequality and lemma [ lemsigexpansion ] , we obtain that @xmath415 \lesssim \bigl\| \sigma_{|m}^2 - \overline { \sigma}_{m}^2 \bigr\|_{p/2 } \lesssim n^{-p/2 + 1}.\end{aligned}\ ] ] hence we conclude that @xmath416 . due to the independence of @xmath418 , we may rewrite @xmath419 as @xmath420 where @xmath421 denotes the c.d.f . of a standard normal distribution . this induces a `` natural '' smoothing . the claim now follows by repeating the same arguments as in part @xmath296 . note however , that the present situation is much easier to handle , due to the already smoothed version , and since @xmath422 , @xmath423 is a sequence of independent random variables . alternatively , one may also directly appeal to the results in @xcite . the proof of theorem [ thmberry ] mainly consists of constructing a good @xmath31-dependent approximation and then verifying the conditions of theorem [ thmmdependent ] . to this end , set @xmath424 for some @xmath425 , and note that @xmath426 for @xmath1 $ ] . let @xmath427 , and define the approximating sequence as @xmath428\quad \mbox{and } \nonumber \\[-8pt ] \\[-8pt ] \nonumber x_k^{(>m ) } & = & x_k - x_k^{(\leq m ) } = x_k - \mathbb{e } \bigl[x_k | \mathcal{e}_k^{m } \bigr].\end{aligned}\ ] ] we also introduce the corresponding partial sums as @xmath429 further , let @xmath430 and @xmath431 . we require the following auxiliary result ( lemma 5.1 in @xcite ) . [ lemsiegidecomposition ] for every @xmath432 , every @xmath433 and every @xmath434 , the following estimate holds : @xmath435 where : @xmath436 proof of theorem [ thmberry ] as a preparatory result , note that @xmath437.\end{aligned}\ ] ] using the same arguments as in lemma [ lemsigexpressionsrelations ] , it follows that @xmath438 . by the properties of gaussian distribution , @xmath439 and we may thus safely interchange @xmath440 and @xmath79 . we first deal with @xmath441 . for @xmath442 , denote with @xmath443 the projection operator @xmath444 - \mathbb{e } \bigl[x_k^{(>m ) } |\mathcal { e}_{j-1 } \bigr].\end{aligned}\ ] ] proceeding as in the proof of lemma 3.1 in @xcite , it follows that for @xmath284 , @xmath445 an application of theorem 1 in @xcite now yields that @xmath446 for some absolute constant @xmath447 that only depends on @xmath77 . by ( [ eqthmgen4 ] ) , it follows that the above is of magnitude @xmath448 setting @xmath449 , we obtain the bound @xmath450 . we thus conclude from the markov inequality that @xmath451 hence @xmath452 note that a much sharper bound can be obtained via moderate deviation arguments ( cf . @xcite ) , but the current one is sufficient for our needs , and its deviation requires fewer computations . next , we deal with @xmath453 . the aim is to apply theorem [ thmmdependent ] to obtain the result . in order to do so , we need to verify assumption [ assdependence](i)(iii ) for @xmath454 . note first that @xmath455 = \mathbb{e } [ x_k ] = 0 $ ] . moreover , jensen s inequality gives @xmath456 \bigr\|_p \leq \|x_k \|_p < \infty.\end{aligned}\ ] ] hence assumption [ assdependence](i ) is valid . note that we may assume @xmath457 , since otherwise @xmath458 , and assumption [ assdependence](ii ) is trivially true . put @xmath459 since @xmath460 ' = \mathbb { e } [ x_k ' |\mathcal{e}_k^{(m,\prime ) } ] $ ] , it follows that @xmath461 - \mathbb{e}_{\mathcal { e}_k^{m } } [ x_k ] \\ & = & \mathbb{e}_{\mathcal{e}_k^{(m,\prime ) } } \bigl[x_k ' - x_k \bigr ] + \mathbb{e}_{\mathcal{e}_k^{(m,\prime ) } } [ x_k ] - \mathbb { e}_{\mathcal { e}_k^{m } } [ x_k ] \\ & = & \nonumber \mathbb{e}_{\mathcal { e}_k^{(m,\prime ) } } \bigl[x_k ' - x_k \bigr ] + \mathbb{e}_{\mathcal { e}_k^{m } } \bigl[x_k ' \bigr ] - \mathbb{e}_{\mathcal{e}_k^{m } } [ x_k ] \\ & = & \mathbb{e } _ { \mathcal{e}_k^{(m,\prime ) } } \bigl[x_k ' - x_k \bigr ] + \mathbb { e}_{\mathcal { e}_k^{m } } \bigl[x_k ' - x_k \bigr].\end{aligned}\ ] ] hence by jensen s inequality @xmath462 , which gives the claim . we have @xmath463 \bigr\|_p \leq \bigl\|x_k - x_k^{(m , * ) } \bigr\|_p \nonumber \\[-8pt ] \label{eqthmgen7 } \\[-8pt ] \nonumber & \leq & m^{-2}\sum_{l = m}^{\infty}l^2 \bigl\| x_l - x_l ' \bigr\|_p \lesssim m^{-2}.\end{aligned}\ ] ] by the cauchy schwarz , triangle and jensen inequalities , we have @xmath464 - \mathbb{e } \bigl[x_k^{(\leq m ) } x_0^{(\leq m ) } \bigr ] \bigr| \\ & & \qquad\leq \|x_0 \|_2 \bigl\|x_k^ { ( > m ) } \bigr\|_2 + \|x_k \|_2 \bigl\|x_0^ { ( > m ) } \bigr\|_2 + \bigl\|x_0^ { ( > m ) } \bigr\|_2 \bigl\|x_k^ { ( > m ) } \bigr\|_2.\end{aligned}\ ] ] by ( [ eqthmgen7 ] ) , this is of the magnitude @xmath465 . we thus conclude that @xmath466 - \sum_{k = 0}^m \mathbb{e } \bigl[x_k^{(\leq m ) } x_0^{(\leq m ) } \bigr ] \biggr| \lesssim m^{-1}.\end{aligned}\ ] ] on the other hand , we have @xmath467 \biggr| \leq\sum_{k > m } \|x_0 \|_2 \bigl\|x_k^ * - x_k \bigr\|_2 \leq\frac { 1}{m}\sum_{k > m}k^2 \bigl\|x_k - x_k ' \bigr\|_2 \|x_0 \|_2 \lesssim\frac{1}{m}.\end{aligned}\ ] ] this yields @xmath468 - s^2 \biggr| \lesssim \frac{1}{m},\end{aligned}\ ] ] which gives ( iii ) for large enough @xmath31 . since @xmath469 , we see that we may apply theorem [ thmmdependent ] which yields @xmath470 next , we deal with @xmath471 . properties of the gaussian distribution function give @xmath472 however , by the cauchy schwarz inequality and ( [ eqthmgen6 ] ) , it follows that @xmath473 and we thus conclude that @xmath474 finally , setting @xmath475 , standard arguments involving the gaussian distribution function yield that @xmath476 piecing together ( [ eqthmgena1 ] ) , ( [ eqthmgena2 ] ) , ( [ eqthmgena3 ] ) and ( [ eqthmgena0 ] ) , lemma [ lemsiegidecomposition ] yields @xmath477 which completes the proof . recall that @xmath478 we first consider the case @xmath479 . using theorem [ thmberry ] , we have @xmath480 in order to bound @xmath481 , we apply @xcite , theorem 3.2 , which will give us the bound @xmath482 to this end , we need to verify that @xmath483 |\mathcal{e}_0 \bigr ] \bigr ) \bigr\|_1 \\ \label{eqriol1 } & & \qquad{}+ \frac{1}{k}\sum _ { i = 1}^k \bigl\|x_{-i } x_0 \mathbb{e } \bigl[x_k^2 - \mathbb { e}\bigl[x_k^2 \bigr ] | \mathcal{e}_0 \bigr ] \bigr\|_1 \biggr)<\infty\quad\mbox{and } \\ \nonumber & & \sum_{k > 0 } \frac{1}{k } \sum _ { i = \lfloor k/2 \rfloor}^k \bigl\||x_0| \vee1 \mathbb{e } \bigl[x_i x_k^2 - \mathbb{e } \bigl[x_i x_k^2\bigr ] | \mathcal{e}_0 \bigr ] \bigr\|_1 < \infty.\end{aligned}\ ] ] applying the hlder , jensen and triangle inequalities , we get @xmath484 |\mathcal { e}_0 \bigr ] \bigr ) \bigr\|_1 & \leq & \bigl\|x_0 ^ 2 \vee1 \bigr\|_2 \bigl\|x_k - x_k^ * \bigr\|_4 \bigl\|x_k + x_k^ * \bigr\|_4 \\ & \lesssim & k \bigl\|x_k - x_k ' \bigr\|_4.\end{aligned}\ ] ] similarly , with @xmath485 , we obtain that @xmath486 | \mathcal { e}_0 \bigr ] \bigr\|_1 & \lesssim & \|x_{-i } \|_4 \bigl\| \mathbb{e } [ x_{0}|\mathcal{e}_{-i } ] \bigr\|_4 k \bigl\|x_k - x_k ' \bigr\|_4 \\ & \lesssim & \bigl\|x_i - x_i^ * \bigr\|_4 k \bigl\|x_k - x_k ' \bigr\|_4 \\ & \lesssim & i \bigl\|x_i - x_i ' \bigr\|_4 k \bigl\|x_k - x_k ' \bigr\|_4.\end{aligned}\ ] ] in the same manner , we get that @xmath487 | \mathcal{e}_0 \bigr ] \bigr\|_1 & \lesssim & \bigl\|x_i - x_i^ * \bigr\| _ 4 + \bigl\|x_k - x_k^ * \bigr\|_4 \\ & \lesssim & i \bigl\|x_i - x_i ' \bigr\|_4 + k \bigl\|x_k - x_k ' \bigr\|_4.\end{aligned}\ ] ] combining all three bounds , the validity of ( [ eqriol1 ] ) follows , and hence ( [ eqthmlpboundrio3 ] ) . for ( [ eqthmlpboundrio2 ] ) , we thus obtain @xmath488 which completes the proof for @xmath489 . for @xmath87 , we may directly refer to @xcite , theorem 3.2 , using the above bounds . for the proof , we require the following result ; cf . @xcite , lemma 5.4 . [ lemosipov ] let @xmath490 be a real - valued random variable . put @xmath491 and assume that @xmath492 for @xmath101 and @xmath493 . then @xmath494 for all @xmath495 , where @xmath496 is a positive constant depending only on @xmath497 , and @xmath498 \biggr|.\end{aligned}\ ] ] consider first the case where @xmath499 , for @xmath500 large enough ( see below ) . then by the markov inequality and lemma [ lemwuoriginal ] , it follows that @xmath501 combining theorem [ thmberry ] with ( [ eq_truncated_be ] ) and lemma [ lemosipov ] , we see that it suffices to consider @xmath502 with @xmath503 . using again theorem [ thmberry ] together with ( [ eq_truncated_be ] ) , standard tail bounds for the gaussian distribution and elementary computations give @xmath504 according to a fuk nagaev - type inequality for dependent sequences in @xcite , theorem 2 , if it holds that @xmath505 then for large enough @xmath228 and @xmath506 we get @xmath507 and hence , @xmath508 however , setting @xmath509 , an application of the cauchy schwarz inequality yields @xmath510 & & \qquad\leq \sum_{k = 1}^{\infty } a_k^2 \sum_{k = 1}^{\infty } a_k^{-2 } k^{({p - 2})/({p+1 } ) } \bigl\|x_k - x_k'\bigr\| _ p^{{2p}/({p+1 } ) } \\[-2pt ] & & \qquad\lesssim \sum_{k = 1}^{\infty } k^{2 } \bigl\|x_k - x_k'\bigr\|_p < \infty\end{aligned}\ ] ] by assumption [ assmaindependence ] . hence ( [ eqthmnonunif3 ] ) holds , and thus ( [ eqthmnonunif4 ] ) and ( [ eqthmnonunif5 ] ) . to complete the proof , it remains to treat the case @xmath511 . but in this case , we may directly appeal to ( [ eqthmnonunif4 ] ) which gives the result . suppose we have a sequence of random variables @xmath512 and a sequence of filtrations @xmath513 , such that both @xmath514 \}_{1 \leq j \leq j}$ ] and @xmath515 \}_{1 \leq j \leq j}$ ] are independent sequences . note that this does not necessarily mean that @xmath516 is independent , and indeed this is not the case when we apply lemma [ lemgeneralvarphibound ] in step 4 of the proof of part @xmath296 . introduce the conditional characteristic function @xmath517.\end{aligned}\ ] ] given the above conditions , we have the following result . [ lemgeneralvarphibound ] let @xmath8 , and assume that : @xmath365 = 0 $ ] uniformly for @xmath518 , there exists a @xmath366 such that @xmath519 \leq u^- ) < 1/7 $ ] uniformly for @xmath520 , @xmath521 \leq c_1 < \infty$ ] uniformly for @xmath518 . then there exist constants @xmath522 , only depending on @xmath523 and @xmath77 , such that @xmath524 \lesssim e^{-c_{\varphi,1}x^2 j } + e^{-\sqrt{j/4}\log 8/7}\qquad \mbox{for $ x^2 \leq c_{\varphi,2}$.}\end{aligned}\ ] ] let @xmath525.\end{aligned}\ ] ] using a taylor expansion and writing @xmath526 , we obtain that @xmath527 = 1 - \mathbb{e}_{\mathcal { h}_j } \bigl[h_j^2 \bigr ] x^2/2 + x^2/2\int_0 ^ 1(1-s)i(s , x)\,ds.\end{aligned}\ ] ] using the lipschitz property of @xmath528 and @xmath529 , it follows that @xmath530,\qquad h > 0.\end{aligned}\ ] ] for @xmath531 we have from the markov inequality @xmath532 & \leq & 2 \int_h^{\infty } x p_{\mathcal{h}_j } \bigl(|h_j| \geq x \bigr)\,dx + h^2 p_{\mathcal{h}_j}\bigl(|h_j| \geq h\bigr ) \\ & \leq & 2 h^{-p + 2}\int_0^{\infty } x^{p-1 } p_{\mathcal{h}_j } \bigl(|h_j| \geq x \bigr)\,dx+h^2 p_{\mathcal{h}_j}\bigl(|h_j| \geq h\bigr ) \\ & \leq & \frac{2+p}{p } h^{-p+2 } \mathbb{e}_{\mathcal{h}_j } \bigl[|h_j|^p \bigr ] < h^{-p+2 } \mathbb{e}_{\mathcal{h}_j } \bigl[|h_j|^p \bigr].\end{aligned}\ ] ] we thus conclude from ( [ eqlemgeneralvarphibound1 ] ) that @xmath533|xh| + 4 h^{-p+2 } \mathbb{e}_{\mathcal{h}_j } \bigl[|h_j|^p \bigr].\end{aligned}\ ] ] this gives us @xmath534 - 1 + \mathbb{e } _ { \mathcal{h}_j } \bigl[h_j^2 \bigr ] x^2/2 \bigr| \nonumber \\[-8pt ] \label{eqlemgeneralvarphibound2 } \\[-8pt ] \nonumber & & \qquad\leq\mathbb { e}_{\mathcal { h}_j } \bigl[h_j^2 \bigr]h|x|^3 + 2h^{-p+2}x^2 \mathbb{e}_{\mathcal { h}_j } \bigl[|h_j|^p \bigr].\end{aligned}\ ] ] let @xmath535 , and put @xmath536 $ ] and @xmath537 $ ] . consider @xmath538 where @xmath539 denotes the @xmath540th largest random variable for @xmath541 . let @xmath542 , @xmath518 denote i.i.d . unit exponential random variables , and denote with @xmath543 the @xmath540th largest . further , denote with @xmath544 the c.d.f . of @xmath545 , @xmath518 , and with @xmath546 . using the transformation @xmath547 , we thus obtain @xmath548\\[-8pt]\nonumber & & \qquad \stackrel{d}{= } \bigl\{e_j \leq -\log\bigl(1 - f_{\rho_j}(x_j)\bigr ) : 1 \leq j \leq j \bigr\ } , \qquad x_j \in \mathbb{r},\end{aligned}\ ] ] which is the well - known renyi representation ; cf . @xcite . in particular , by the construction of @xmath549 it follows that @xmath550 for @xmath551 . let @xmath552 and @xmath553 . we wish to find a @xmath554 such that @xmath555 . this is implied by @xmath556 . we will now construct such an @xmath554 . since @xmath557 = \int_0^{\infty } \bigl(1 - f_{\rho_j}(x ) \bigr)\,dx \geq c_2 p \bigl(\rho _ j^{\mathcal{h } } \geq c_2\bigr)\qquad \mbox{for $ c_2 > 0$},\end{aligned}\ ] ] it follows that @xmath558 . hence choosing @xmath559 , we obtain @xmath560 and hence @xmath561 , which leads to @xmath562 . thus by known properties of exponential order statistics ( cf . @xcite ) , we have @xmath563 e^{-u_{\mathcal{h}}^+(j ) } \geq1 - \mathcal { o } \bigl(e^{-\sqrt{j/2}\log2 } \bigr),\end{aligned}\ ] ] for sufficiently large @xmath564 . we thus conclude that @xmath565 let us denote this set with @xmath566 , and put @xmath567 . note that the index set @xmath568 has at least cardinality @xmath569 given event @xmath570 . for the sake of simplicity , let us assume that @xmath571 , which , as is clear from the arguments below , has no impact on our results . let us introduce @xmath572 the order statistics of @xmath573 within the index set @xmath568 . this means that @xmath574 is not necessarily the smallest value of @xmath575 , @xmath541 . more generally , it holds that @xmath576 now , similar to as before , let @xmath577 be the c.d.f . of @xmath578 , @xmath518 , and put @xmath579 , @xmath580 and @xmath581 for some @xmath582 . we search for a @xmath583 such that @xmath584 , which is true if @xmath585 . however , this is precisely what we demanded in the assumptions . then proceeding as before , we have @xmath586 e^{-u_{\mathcal{h}}^-(j ) } \\ & \geq & 1 - \mathcal{o } \bigl(e^{-\sqrt{j/4}\log8/7 } \bigr).\end{aligned}\ ] ] we thus conclude from ( [ eqlemboundchar6.5 ] ) and the construction of @xmath587 that @xmath588 \label{eqlemboundchar7 } \\[-8pt ] \nonumber & \geq & 1 - \mathcal { o } \bigl(e^{-\sqrt{j/4}\log8/7 } \bigr).\end{aligned}\ ] ] put @xmath589 . combining ( [ eqlemboundchar6 ] ) and ( [ eqlemboundchar7 ] ) we obtain @xmath590 we denote this set with @xmath591 . also note that by the ( conditional ) lyapunov inequality , we have @xmath592 note that @xmath593 on the event @xmath594 , and , by the above , we get @xmath595 using ( [ eqlemgeneralvarphibound2 ] ) , this implies that for every @xmath596 , we have @xmath597 - 1 + \mathbb{e } _ { |\mathcal{h}_j } \bigl[h_j^2 \bigr ] x^2/2 \bigr| \leq\bigl(u^+\bigr)^{p/2 } h|x|^3 + 2h^{-p+2 } u^+ x^2.\end{aligned}\ ] ] hence , if @xmath598 and @xmath599 , we conclude from the above and the triangle inequality that for @xmath600 , @xmath601 \label{eqlemboundchar9 } \\[-8pt ] \eqntext{\mbox{for $ \bigl(u^+\bigr)^{p/2 } x^2/2 < 1$,}}\end{aligned}\ ] ] where @xmath602 . since @xmath603 and @xmath604 , there exist absolute constants @xmath605 , chosen sufficiently small , such that @xmath606 next , observe that @xmath607 & = & \mathbb{e } \biggl [ \biggl|\prod _ { j = 1}^{j}\varphi _ j^{\mathcal{h } } ( x ) \biggr| \bigl(\mathbh{1}(\mathcal{a } ) + \mathbh{1}\bigl(\mathcal { a}^c \bigr ) \bigr ) \biggr ] \\[-3pt ] \label{eqlemboundchar9.1 } & \leq & p \bigl(\mathcal{a}^c \bigr ) + \mathbb{e } \biggl [ \biggl|\prod_{j \in\mathcal{i}}\varphi_j^{\mathcal{h}}(x ) \biggr|\mathbh { 1}(\mathcal { a } ) \biggr ] \\[-3pt ] \nonumber & \leq & p \bigl(\mathcal{a}^c \bigr ) + \mathbb{e } \biggl [ \biggl|\prod _ { j \in\mathcal{i}_{\mathcal{a}}}\varphi_j^{\mathcal { h}}(x ) \biggr| \mathbh{1}(\mathcal{a } ) \biggr].\end{aligned}\ ] ] moreover , using ( [ eqlemboundchar9 ] ) and ( [ eqlemboundchar9.2 ] ) and since @xmath608 on @xmath609 , it follows that @xmath610 & \leq & \mathbb{e } \biggl [ \biggl|\prod_{j \in\mathcal{i}_{\mathcal{a } } } \bigl(1 - u(x ) \bigr ) \biggr|\mathbh{1}(\mathcal{a } ) \biggr ] \\ & \leq & \mathbb{e } \biggl[\prod_{j \in\mathcal { i}_{\mathcal{a}}}e^{- u(x ) } \mathbh{1}(\mathcal{a } ) \biggr]\leq e^{-u(x ) j/8 } \leq e^{-c_{\varphi,1}jx^2}.\end{aligned}\ ] ] hence we conclude from the above and ( [ eqlemboundchar8 ] ) that @xmath611 \lesssim e^{-c_{\varphi,1}x^2 j } + e^{-\sqrt { j/4}\log8/7}\qquad \mbox{for $ x^2 \leq c_{\varphi,2}$,}\end{aligned}\ ] ] which yields the claim . [ lemsigexpressionsrelations ] grant assumption [ assdependence ] . then @xmath612 | < \infty$ ] and @xmath613 . moreover , we have @xmath614 as @xmath615 . since @xmath616 = \mathbb { e } [ x_k - x_k^ * |\mathcal{e}_0 ] $ ] , the cauchy schwarz and jensen inequalities imply @xmath617 \bigr| & \leq & \|x_0 \|_2 \sum_{k = 0}^{\infty } \bigl\|\mathbb{e } [ x_k |\mathcal{e}_0 ] \bigr\|_2 \leq \|x_0 \|_2\sum_{k = 0}^{\infty } \bigl\|x_k - x_k^ * \bigr\|_2 \\ & \leq & \|x_0 \| _ 2\sum_{k = 1}^{\infty}k^2 \bigl\|x_k - x_k ' \bigr\|_2 < \infty.\end{aligned}\ ] ] the decomposition @xmath618 now follows from ( [ decompsigma ] ) . claim @xmath619 as @xmath615 readily follows from the previous computations . [ lemsigexpansion ] grant assumption [ assdependence ] . then : @xmath620 for @xmath230 , @xmath621 for @xmath230 , @xmath622 . we first show ( i ) . without loss of generality , we may assume @xmath343 , since @xmath623 . to lighten the notation , we use @xmath624 . we will first establish that @xmath625 . we have that @xmath626 \bigr ) + r_1 \biggr)^2 \biggr ] - 2 m \widehat{\sigma}_m^2.\end{aligned}\ ] ] by squaring out the first expression , we obtain a sum of square terms and a sum of mixed terms . let us first treat the mixed terms , which are @xmath627 + { \mathbb{e}}_{{\mathbb{f}}_m}[x_k]\bigl(x_l - x_l^{(**)}\bigr)\bigr ] \\[-2pt ] & & \quad{}+ 2 \sum_{k = 1}^m \mathbb{e}_{\mathbb{f}_m } \bigl[r_1 x_k^ { ( * * ) } + r_1 \bigl(x_k - x_k^{(**)}\bigr ) + r_1 \mathbb{e}_{\mathbb{f}_m}[x_k ] \bigr ] \\[-2pt ] & & \qquad= i_m + \mathit{ii}_m + \mathit{iii}_m + \mathit{iv}_m + v_m + \mathit{vi}_m.\end{aligned}\ ] ] we will handle all these terms separately . we have @xmath628 & = & \sum_{l = 1}^m \sum _ { k = l}^m \mathbb { e}_{\mathbb{f } _ m } \bigl [ \bigl(x_l - x_l^{(**)}\bigr)\mathbb{e } \bigl[x_k^ { ( * * ) } |\sigma \bigl(\mathbb{f}_m , \mathcal{e}_l,\mathcal{e}_l^ { ( * * ) } \bigr ) \bigr ] \bigr ] \\[-2pt ] & & { } + \sum_{l = 1}^m \sum _ { k = 1}^{l-1}\mathbb{e}_{\mathbb{f}_m } \bigl[x_k^{(**)}\bigl(x_l - x_l^{(**)}\bigr ) \bigr].\end{aligned}\ ] ] since @xmath629 \stackrel{d } { = } \mathbb{e } [ x_k |\mathcal{e}_l ] = \mathbb{e } \bigl[x_k - x_k^{(k - l , * ) } |\mathcal { e}_l \bigr],\end{aligned}\ ] ] the cauchy schwarz ( with respect to @xmath630 ) and jensen inequalities thus yield @xmath631 since @xmath632 = \mathbb { e}[x_k ] = 0 $ ] ( @xmath457 ) it follows that @xmath633 . it follows via the jensen and triangle inequalities that @xmath634 \bigr\|_p & = & \bigl\|\mathbb{e } \bigl[x_l - x_l^ { ( * * ) } \nonumber \\[-10pt ] \label{eqlemsigexpansion3}\\[-10pt ] \nonumber & \leq & \bigl\|x_l - x_l^{(**)}\bigr\|_p \leq\sum _ { j = l}^{\infty } \bigl\|x_j - x_j ' \bigr\|_p.\end{aligned}\ ] ] the cauchy schwarz ( with respect to @xmath630 ) and jensen inequalities then give @xmath635 note that @xmath636 and @xmath637 are independent for @xmath638 and @xmath639 . hence since @xmath640 = 0 $ ] , we have @xmath641 & = & \sum _ { k = 1}^m\sum_{l = m+1}^{2 m } \mathbb{e}_{\mathbb{f}_m } \bigl[x_k^{(**)}x_l \bigr ] \\[-3pt ] & = & \sum_{k = 1}^m\sum _ { l = m+1}^{2 m } \mathbb{e}_{\mathbb{f}_m } \bigl[x_k^{(**)}\bigl(x_l - x_l^{(l - k , * ) } + x_l^{(l - k,*)}\bigr ) \bigr ] \\[-3pt ] & = & \sum_{k = 1}^m\sum _ { l = m+1}^{2 m } \mathbb{e}_{\mathbb{f}_m } \bigl[x_k^{(**)}\bigl(x_l - x_l^{(l - k,*)}\bigr ) \bigr].\end{aligned}\ ] ] the cauchy schwarz ( with respect to @xmath630 ) and jensen inequalitities then yield @xmath642 & \leq & 2\sum_{k = 1}^m\sum _ { l = m+1}^{2 m } \sum_{j = l - k}^{\infty } \bigl\|x_j - x_j ' \bigr\|_{p } \|x_1 \|_{p } \\[-3pt ] & \leq & 2\sum_{k = 1}^m\sum _ { j = m - k}^{\infty } ( j - m+k ) \bigl\|x_j - x_j ' \bigr\| _ { p } \|x_1 \|_{p } \\[-3pt ] & \leq & 2 \sum_{j = 1}^{\infty } j^2 \bigl\|x_j - x_j ' \bigr\|_{p } \|x_1 \|_{p } < \infty.\end{aligned}\ ] ] the cauchy schwarz ( with respect to @xmath643 ) and jensen inequalities yield @xmath644 & \leq & \sum_{l = 1}^{\infty } l^2 \bigl\|x_l - x_l ' \bigr\|_p \|r_1 \|_p < \infty,\end{aligned}\ ] ] since @xmath645 by lemma [ lemboundr1 ] . proceeding as above and using ( [ eqlemsigexpansion3 ] ) , we get @xmath646 . it thus remains to deal with the squared terms , which are @xmath647\mathbb{e}_{\mathbb{f } _ m}[x_l ] \bigr ] + \mathbb{e}_{\mathbb{f}_m } \bigl[r_1 ^ 2 \bigr ] \\ & & \qquad= 2m\widehat { \sigma}_m^2 + \mathit{vii}_m + \mathit{viii}_m + \mathit{ix}_m.\end{aligned}\ ] ] however , using the results from the previous computations and lemma [ lemboundr1 ] , one readily deduces that @xmath648 piecing everything together , we have established that @xmath649 . however , from the above arguments one readily deduces that @xmath650 , and hence ( i ) and ( ii ) follow . we now treat ( iii ) . since @xmath651 is an independent sequence under @xmath202 , we have @xmath652 let @xmath653 and @xmath654 such that @xmath655 . then @xmath656 note that @xmath657 is a sequence of independent random variables , and the same is true for @xmath658 . then by lemma [ lemwuoriginal ] , it follows that @xmath659$,}\end{aligned}\ ] ] which by ( i ) is of the magnitude @xmath660 . hence we conclude from ( [ eqlemsigexpansion6 ] ) that @xmath661 [ corsigreplaceissmall ] grant assumption [ assdependence ] . let @xmath662 . then @xmath663 by markov s inequality , lemma [ lemsigexpressionsrelations ] and lemma [ lemsigexpansion ] , it follows that for large enough @xmath31 @xmath664 since @xmath210 by assumption [ assdependence](iii ) . the aim of this section is to give a proof for the inequality @xmath665 in corollary [ corzolo ] . we will achieve this by employing ideal metrics . let @xmath666 . then we can represent @xmath667 as @xmath668 , where @xmath669 = \mathfrak{m}$ ] denotes the integer part , and @xmath670 . let @xmath671 be the class of all real - valued functions @xmath110 , such that the @xmath672th derivative exists and satisfies @xmath673 note that since @xmath674 are bounded in absolute value and are lipschitz continuous , it follows that up to some finite constant @xmath675 we have @xmath676 for any @xmath666 . as already mentioned in step 2 of the proof of part @xmath296 , we will make use of some special ideal - metrics @xmath677 ( zolotarev metric ) . for two probability measures @xmath678 , the metric @xmath677 is defined as @xmath679 the metric @xmath680 has the nice property of homogeneity . for random variables @xmath681 , induced probability measures @xmath682 and constant @xmath425 , this means that @xmath683 . we require some further notation . for @xmath684 , put @xmath685 note that @xmath686 is independent of @xmath219 , and hence @xmath687 $ ] . let @xmath688 be a sequence of zero mean , standard i.i.d . gaussian random variables . in addition , let @xmath689/\widehat { \sigma}_m^2 = \sigma_{j|m}^2/ \widehat { \sigma}_m^2 \qquad \mbox{for $ 1 \leq j \leq n-1$},\end{aligned}\ ] ] and @xmath690 for @xmath691 . as first step toward ( [ eqphiboundcor ] ) , we have the following result . [ lemtrunccharcosandsin ] grant assumption [ assdependence ] . then for @xmath692 , it holds that @xmath693 \bigr\|_1 \lesssim m^{-p/2 + 1}|x|^p \qquad\mbox{if $ p \in(2,3]$},\end{aligned}\ ] ] where @xmath235 . to lighten the notation , we use @xmath694 and @xmath695 in the following . using taylor expansion , we have @xmath696 note @xmath697 = 0 $ ] and @xmath698 = \eta_{j|m}\mathbb{e } [ s_{j } ] = 0 $ ] . moreover , since @xmath699 $ ] by construction , we obtain from ( [ eqlemtrunccharcosandsin2 ] ) that @xmath700 \\ & & \qquad= x^2\int_0 ^ 1(1-t ) \mathbb{e}_{\mathbb{f}_m } \bigl[y_j^2 \bigl(f''(txy_j)-f''(0 ) \bigr)\\ & & \qquad\qquad{}\hspace*{68pt}- ( s_{j}\eta_{j|m})^2 \bigl(f''(txs_{j}\eta_{j|m})-f''(0 ) \bigr ) \bigr]/2 \,dt \\ & & \qquad\stackrel{\mathrm{def}}{= } x^2i_m(x).\end{aligned}\ ] ] we have @xmath701 where we recall that @xmath702 and @xmath703 are independent . using the jensen , triangle and hlder inequalities and @xmath704 , it follows that @xmath705 \bigr\|_1 \\ & & \qquad\leq 2 \bigl\| \bigl(y_j^2 - ( s_{j}\eta_{j|m})^2 \bigr ) |txy_j \\ & & \qquad\leq 2 \bigl\|y_j^2 - ( s_{j } \eta_{j|m})^2 \bigr\| _ { p/2 } \bigl\||txy_j|^{p-2 } \bigr\|_{p/(p-2)}\\ & & \qquad \lesssim2 \bigl\| |txy_j|^{p-2 } \bigr\|_{p/(p-2 ) } \\ & & \qquad\quad{}\times\bigl ( \|y_j - s_{j } \|_p \|y_j + s_{j } \|_{p } + \bigl\|s_j^2 \bigr\|_{p/2 } \bigl\|\sigma_{j|m}^2 - \widehat{\sigma } _ m^{2 } \bigr\|_{p/2 } \bigr),\end{aligned}\ ] ] where we used that @xmath706 for large enough @xmath31 . by lemmas [ lemboundr1 ] and [ lemsigexpansion ] , this is of magnitude @xmath707 . hence by adding and subtracting @xmath708 and using similar arguments as before , we obtain from the above @xmath709 \bigr\|_1 \ , dt \nonumber \\[-8pt ] \label{eqlemtrunccharcosandsin3 } \\[-8pt ] \nonumber & & \qquad \lesssim m^{-1/2}|x|^p + |x|^p \bigl\|s_{j}^2 \bigr\|_{p/2 } \bigl\|\eta _ { j|m}^2 \bigr\|_{p/2 } \bigl\||s_{j } - y_j|^{p-2 } \bigr\|_{p/(p-2 ) } \\ \nonumber & & \qquad\lesssim m^{-1/2}|x|^p + m^{-(p-2)/2}|x|^p \lesssim m^{-p/2 + 1}|x|^p,\end{aligned}\ ] ] where we use that @xmath702 and @xmath710 are independent . this gives the desired result . as next step toward ( [ eqphiboundcor ] ) , we have the following . [ lemapproxchar ] grant assumption [ assdependence ] . then for @xmath692 , it holds that @xmath711 \bigr\|_1 \lesssim|x|^p m^{-p/2 + 1}\qquad \mbox{if $ p \in(2,3]$},\end{aligned}\ ] ] where @xmath235 . to increase the readability , we use the abbreviations @xmath712 and @xmath713 in the following . the main objective is to transfer the problem to the setup in @xcite and apply the corresponding results . to this end , we first perform some necessary preparatory computations . we have that @xmath714 \bigr\|_p \leq\sum _ { k > l } \bigl\|x_k - x_k^ * \bigr\|_p \leq\sum_{k > l}k \bigl\| x_k - x_k ' \bigr\|_p \to 0\end{aligned}\ ] ] as @xmath715 , hence it follows that @xmath716\qquad \mbox{converges in $ \|\cdot \|_p$}.\end{aligned}\ ] ] next , note that @xmath717.\end{aligned}\ ] ] using exactly the same arguments as in the proof of lemma [ lemsigexpansion ] ( the present situation is much simpler ) , we get that @xmath718 - 2m\widehat{\sigma}^2 \biggr\|_{p/2 } < \infty.\end{aligned}\ ] ] we thus obtain that @xmath719 - 2m\widehat{\sigma}^2 \biggr\|_{p/2 } \lesssim\sum _ { m = 1}^{\infty } m^{-3/2 } < \infty.\end{aligned}\ ] ] we will now treat the cases @xmath83 and @xmath33 separately . since @xmath720 , the homogeneity of order @xmath77 implies that @xmath721 \bigr\|_1 & \leq & 2 \bigl\|\zeta_{p } ( p_{xs_{j}\eta_{j|m}|\mathbb{f}_m } , p_{xz_{j}\widehat{\sigma}\eta_{j|m}|\mathbb{f}_m } ) \bigr\| _ 1 \\ & = & 2 \bigl\||x|^p|\eta_{j|m}|^p\zeta_{p } ( p_{s_{j}|\mathbb{f}_m } , p_{z_{j}\widehat{\sigma}|\mathbb{f}_m } ) \bigr\|_1.\end{aligned}\ ] ] since @xmath722 are independent of @xmath219 , we have @xmath723 . hence @xmath724 \bigr\|_1 \leq2 \bigl\||x|^p|\eta_{j|m}|^p \bigr\|_1 \zeta_{p } ( p_{s_{j } } , p_{z_{j}\widehat{\sigma } } ) .\end{aligned}\ ] ] by lemma [ lemsigexpansion ] , we have @xmath725 . note that @xmath726 = \mathbb { e } [ z_j ] = 0 $ ] and @xmath727 . hence due to ( [ eqlemapproxchar1 ] ) and ( [ eqlemapproxchar2 ] ) , we may apply theorem 3.1(1 ) in @xcite , which gives us @xmath728 . hence @xmath729 \bigr\|_1 \lesssim |x|^p m^{-p/2 + 1}.\end{aligned}\ ] ] we may proceed as in the previous case , with the exception that here we need to apply theorem 3.2 in @xcite . we may do so due to ( [ eqlemapproxchar1 ] ) and ( [ eqlemapproxchar2 ] ) . hence we obtain @xmath730 \bigr\|_1 \lesssim |x|^3 m^{-1/2}.\end{aligned}\ ] ] using lemmas [ lemtrunccharcosandsin ] and [ lemapproxchar ] , the triangle inequality gives the following corollary , which proves ( [ eqphiboundcor ] ) . we will frequently use the following lemma , which is essentially a restatement of theorem 1 in @xcite , adapted to our setting . [ lemwuoriginal ] put @xmath732 , @xmath733 . if @xmath734 , then @xmath735 for the sake of completeness , we sate this result in the general , nontime - homogenous but stationary bernoulli - shift context . recall that @xmath736 s_{j|m}^ { ( * * ) } & = & \frac{1}{\sqrt{2m}}\sum _ { k = ( 2j-2)m + 1}^{(2j-1)m } x_k^{((2j - 2)m , * * ) } .\end{aligned}\ ] ] [ lemboundr1 ] grant assumption [ assdependence ] . then : @xmath737 for @xmath235 , @xmath738 for @xmath235 . without loss of generally , we assume that @xmath343 since @xmath623 . ( i ) we have the decomposition @xmath739 \bigr\|_p + \|r_j \|_p.\end{aligned}\ ] ] we will deal with all three terms separately . the triangle inequality gives @xmath740 next , note that @xmath741 = \mathbb{e } [ x_k ] = 0 $ ] for @xmath638 . hence it follows via the jensen and triangle inequalities that @xmath742 \biggr\|_p & = & \biggl\|\sum _ { k = 1}^m\mathbb{e } \bigl[x_k - x_k^ { ( * * ) } |\mathbb { f}_m \bigr ] \biggr\|_p \\ & \leq & \sum_{k = 1}^m \bigl\|x_k - x_k^ { ( * * ) } \bigr\|_p \leq\sum _ { k = 1}^{\infty}k \bigl\|x_k - x_k ' \bigr\|_p < \infty.\end{aligned}\ ] ] similarly , since @xmath743 \stackrel{d}{= } \mathbb{e}_{\mathbb{f}_m } [ x_k^{(k - m , * ) } - x_k ] $ ] for @xmath744 , we have @xmath745 \biggr\|_p & = & \biggl\|\sum _ { k = m+1}^{2 m } \mathbb{e}_{\mathbb { f}_m } \bigl[x_k^{(k - m , * ) } - x_k \bigr ] \biggr\|_p \\ & \leq & \sum_{k = m+1}^{2 m } \bigl\| x_k^{(k - m , * ) } - x_k \bigr\|_p \leq\sum _ { k = 1}^{\infty}k \bigl\|x_k ' - x_k \bigr\|_p < \infty.\end{aligned}\ ] ] combining all three bounds gives ( i ) . this implies that for ( ii ) , it suffices to show that @xmath746 . using the above bounds and lemma [ lemwuoriginal ] , we get @xmath747 \bigr\|_p \lesssim\sqrt{m}.\end{aligned}\ ] ] [ lemauxhsigcomp ] grant assumption [ assdependence ] , and let @xmath748 and @xmath749 , @xmath750 be as in ( [ eqthmauxcharequationbound2.91 ] ) . then @xmath751 \leq\widehat { \sigma } _ { m - l}^2 \bigr ) < 1/7.\end{aligned}\ ] ] since @xmath623 , it suffices to treat the case @xmath343 . recall that @xmath752 and @xmath753 - r_1 + \mathbb{e}_{\mathcal{g}_1^{(l ) } } [ r_1 ] , \\ 2(m - l ) \widehat{\sigma}_{m - l}^2 & = & \mathbb{e}_{\mathcal{e}_l } \biggl [ \biggl(\sum_{k = l+1}^m x_k^{(k - l , * ) } \biggr)^2 \biggr].\end{aligned}\ ] ] let @xmath754 + ( r_1 - \mathbb{e}_{\mathcal{e}_l}[r_1])\mathbh{1}(k = l+1)$ ] . using @xmath755 and applying the cauchy schwarz and jensen inequalities then yields @xmath756 - 2\widehat { \sigma}_{m - l}^2 \bigr\|_1 \\ & & \qquad = \biggl\|\mathbb{e}_{\mathcal{e}_l } \biggl [ \biggl(\sum_{k = l+1}^m x_k - x_k^{(k - l , * ) } - i_k^{(l ) } \biggr ) \biggl(\sum_{k = l+1}^m x_k + x_k^{(k - l , * ) } - i_k^{(l ) } \biggr ) \biggr ] \biggr\|_1 \\ & & \qquad\leq \biggl\|\sum_{k = l+1}^m x_k - x_k^{(k - l , * ) } - i_k^{(l ) } \biggr\|_2 \biggl\|\sum _ { k = l+1}^m x_k + x_k^{(k - l , * ) } - i_k^{(l ) } \biggr\|_2 \\ & & \qquad\stackrel{\mathrm{def } } { = } i_1(l , m ) \mathit{ii}_2(l , m).\end{aligned}\ ] ] by lemma [ lemboundr1 ] and the arguments therein , it follows that @xmath757 , uniformly for @xmath758 . similarly , one obtains that @xmath759 . hence @xmath760 - 2 \widehat { \sigma } _ { m - l}^2 \bigr\|_1 \lesssim \frac{1}{\sqrt{m - l}}.\end{aligned}\ ] ] we then have that @xmath761 \leq\widehat { \sigma } _ { m - l}^2 \bigr ) & \leq & p \bigl ( \bigl| \mathbb{e}_{\mathcal { h}_j } \bigl[h_j^2 \bigr ] - 2 \widehat{\sigma}_{m - l}^2 \bigr| > \widehat{\sigma } _ { m - l}^2 \bigr ) \\ & \leq & \widehat{\sigma}_{m - l}^{-2 } \bigl\|\mathbb{e } _ { \mathcal{h}_j } \bigl[h_j^2 \bigr ] - 2\widehat { \sigma}_{m - l}^2 \bigr\|_1 \lesssim\frac{1}{\sqrt{m - l}}\end{aligned}\ ] ] by markov s inequality . hence the claim follows if @xmath762 is large enough . note that more detailed computations , as in lemma [ lemsigexpansion ] , would give a more precise result . however , the current version is sufficient for our needs . i would like to thank the associate editor and the anonymous reviewer for a careful reading of the manuscript and the comments and remarks that helped to improve and clarify the presentation . i also thank istvan berkes and wei biao wu for stimulating discussions .

complex grassmann manifold has found its extensive applications in multiple - input multiple - output ( mimo ) wireless communication [ 1]-[5 ] in the last ten years . many of these applications focused on solving grassmann packing or quantization problems for better performance in codebook design [ 3]-[5 ] , where the eigenspace of the wireless channel matrix is modeled as a point in complex grassmann manifold . there are also published works on theoretic analyses of the quantization bound on the complex grassmann manifold [ 6 ] , [ 7 ] . for the latter , the normalized volume of hyperball needs to be calculated . various of approximated volume formulas were obtained in [ 6]-[11 ] and used extensively , where the estimation results in [ 8 ] were extensively applied and accurate enough for these applications . however , they are not suitable for some other special applications , e.g. , when the normalized hyperball volume is used as the probability of the distance between any two points in complex grassmann manifold [ 5 ] , [ 12 ] . hence , a more precise normalized volume formula of hyperball in complex grassmann manifold is needed . large - scale mimo or massive mimo system was firstly introduced in [ 13 ] where the base station ( bs ) is equipped with dozens to several hundreds transmit antennas . it has received enormous attention due to its ability of providing linear capacity growth without the need of increased power or bandwidth [ 13 ] , [ 14 ] . this advantage is realized by employing multi - user mimo ( mu - mmo ) which simultaneously beam - forms to many users . in this system , the bs selects users at each scheduling slot and transmits data to these users on the same time and frequency resource . in order to remove the mutual interference among these users and maximize the multi - user channel capacity , the bs needs to know the channel state information ( csi ) and channel quality information ( cqi ) of each user . however , in practical systems , e.g. , 3rd generation partnership project ( 3gpp ) long term evolution - advanced ( lte - a ) [ 15 ] , there always exists unavoidable errors in csi estimation because of the limited feedback bandwidth in frequency - division duplexing ( fdd ) systems or the noise and interference in the measured results of time - division duplexing ( tdd ) systems . on the other hand , the bs generally only knows the cqi of each user in single - user mimo ( su - mimo ) mode but not in mu - mimo mode because it is impractical for the system to actually measure the cqi in mu - mimo mode . therefore , even if zero - forcing ( zf ) precoding is used at the bs side for mu - mimo , there is still residual interference due to inaccurate csi in this situation . as a result , it affects mu - mimo systems in two aspects . firstly , the real sinr at each receiver side is greatly changed compared to the sinr of su - mimo mode , hence a reasonable estimation of each user s real sinr is needed so that it could be used to properly select each user s data transmission rate ( e.g. , the modulation scheme and channel coding rate ) in mu - mimo mode . secondly , as the channel capacity of each user in mu - mimo mode is smaller than su - mimo mode because of not only the decrease of transmit power but also the residual interference , it requires us to forecast the system throughput gain achieved by mu - mimo cautiously . especially , when the error of csi is large , mu - mimo may have no advantage compared to su - mimo but significantly increase system complexity . therefore , how to choose the proper modulation order and channel coding rate and to predict the capacity of mu - mimo with non - ideal csi is a fundamental question in actual mu - mimo communication systems . however , there are only a few papers analyzing the real sinr and capacity for mu - mimo in practical systems . in [ 16 ] , a cursory sinr prediction scheme was presented for some specific scenarios . in [ 17 ] , the cumulative density function ( cdf ) of sinr with ideal csi for mu - mimo was derived but the errors in practical systems were not considered . in this paper , precise normalized volume formulas of hyperball based on two general distance definitions ( projective - frobenius norm and projective-2 norm ) are obtained by applying the probability density function of canonical angle between any two points in complex grassmann manifold introduced in [ 18 ] . simulation results show that they have very high accuracy thus verified their reasonableness . one of the formulas is applied to estimate the sinrs in large - scale mu - mimo communication systems based on su - mimo sinrs . in various simulated cases , our analytical expressions of sinrs match the real values in the actual system almost perfectly . it indicates that these expressions could be used to forecast the mu - mimo sinrs and the capacity gain over su - mimo . the rest of this paper is organized as follows . firstly , the foundation of complex grassmann manifold is introduced in section i. the hyperball volume formula for various cases based on different definitions of norm distances is proposed in section ii , and simulation results are given to demonstrate accuracy . in section iii , we apply the formula to estimate the sinrs in large - scale mu - mimo systems and approximation expressions are derived . simulation results are provided for verification . finally , in section iv , conclusions are drawn . in this section , we first provide the definitions of complex grassmann manifold and the distance on it . then , the normalized volume of hyperball in grassmann manifold is introduced , and the equivalent relation between cdf of distance in the manifold and normalized volume of hyperball is constructed . finally , the normalized volume formula is obtained for various cases . in complex vector space @xmath0 , the set of all @xmath1 matrices with mutually orthogonal columns where each column has unit norm are defined as complex stiefel manifold , i.e. , @xmath2 an equivalent relation between @xmath3 and @xmath4 is constructed on @xmath5 as @xmath6 for any @xmath7 if @xmath8 , where @xmath9 and @xmath10 is the unitary group with order @xmath11 . based on the equivalent relation , a quotient space of @xmath5 is expressed as @xmath12 we call the quotient space @xmath13 complex grassmann manifold . in this paper , a point is used to represent a matrix in @xmath5 or @xmath14 , then all the points which are equivalent in @xmath5 are denoted by anyone of them . if we call this point the effective point , then @xmath13 is made of all the effective points in @xmath5 . according to the definition , a point in @xmath13 is a @xmath15-dimensional subspace of @xmath16 , while all the equivalent points in @xmath5 represent different bases for the same @xmath15-dimensional subspace . in order to study the volume on @xmath13 , we have to define the distance on it . other than the general properties of distance , e.g. , non - negativity and triangle inequality , an additional property that the distance between any two equivalent points is zero has to be satisfied for the defined distance . next , we provide two common definitions . _ definition 1 : _ ( projective - frobenius norm distance ) for any two points @xmath17 , the projective - frobenius norm distance between them is @xmath18 where @xmath19 are the canonical angles of @xmath20 and @xmath21 , which are defined by the singular value decomposition ( svd ) of @xmath22 @xmath23 where @xmath24 , and @xmath25 . obviously , @xmath26 satisfy the inequality @xmath27 . _ definition 2 : _ ( projective-2 norm distance ) for any two points @xmath28 , the projective-2 norm distance between them is @xmath29 where @xmath30 , are the same as in definition 1 . obviously @xmath26 satisfies inequality @xmath31 . based on the above definitions , if we know the canonical angles of any two points in @xmath13 , it is easy to calculate the distances between them . in @xmath13 , the hyperball @xmath32 with center @xmath33 and radius @xmath34 is defined as @xmath35 where @xmath36 is one of the two defined distances in ( [ f4 ] ) and ( [ f5 ] ) . let the volumes of @xmath32 and @xmath14 be @xmath37 and @xmath38 respectively , then we define the normalized volume of hyperball @xmath39 as @xmath40 according to the haar measure theory of compact manifold [ 19 ] , this normalized volume defines the probability that the distance between any point and the center @xmath33 is no more than @xmath41 , i.e. , @xmath42 because of the unique right unitary transformation invariant measure [ 19 ] on @xmath14 , equation ( [ f8 ] ) implies that @xmath43 is exactly the probability of the distance between any two points which is no more than @xmath34 . as a result , we could calculate the normalized volume of hyperball by the probability @xmath44 . when @xmath45 , the probability density function of the canonical angles between any two points in @xmath13 are provided in reference [ 18 ] as @xmath46 @xmath47 where @xmath48 and @xmath49 . notice that @xmath50 are ordered with @xmath51 . according to ( 4 ) , two sets of different ordered angles correspond to two equivalent points in @xmath13 . with formula ( [ f9 ] ) , combining the distances defined in ( [ f3 ] ) and ( [ f5 ] ) , we write the volume of @xmath32 in @xmath52 as @xmath53 @xmath54 where @xmath55 . for the projective - f norm distance , the domain of integration is @xmath56 @xmath57 for the projective-2 norm distance , the domain of integration is @xmath58 @xmath59 next , we provide the results for various cases of @xmath15 and @xmath41 based on the two distances . when @xmath60 , a point in @xmath61 is a complex vector with unit norm in @xmath62 , then the canonical angle @xmath63 between two points is the usual euclidean angle of two vectors . both of the projective - f norm distance and projective-2 norm distance degenerate to the same as @xmath64 . substituting it into ( [ f9 ] ) results in @xmath65 hence , the volume of @xmath66 in @xmath67 is integrated as @xmath68 @xmath69 moreover , as the correlation coefficient of two unit - norm vectors in @xmath70 is @xmath71 , the cdf of the correlation coefficient could be acquired from ( [ f14 ] ) as @xmath72 @xmath73 when @xmath75 , the distances defined in ( [ f3 ] ) and ( [ f5 ] ) are not the same anymore . we next calculate the normalized volume of the hyperball respectively . due to the fact that @xmath76 , combining ( [ f10 ] ) and ( [ f11 ] ) , the volume of hyperball @xmath77 in @xmath78 is @xmath79 @xmath80 where the integration is taken over @xmath81 @xmath82 since this is a complicated problem , and the result depends on the value of @xmath83 , we proide the result as a theorem below . the details are presented in appendix a. the normalized volume of hyperball @xmath84 in complex grassmann manifold @xmath85 is @xmath86 where @xmath87 @xmath88 and @xmath89 @xmath90 in ( [ f19 ] ) , @xmath91 is the beta difference function , where @xmath92 is the beta function defined as @xmath93 and @xmath94 denotes the incomplete beta function , which relates to the beta function as @xmath95 where @xmath96 in ( [ f20 ] ) , @xmath97 is the incomplete beta difference function . the constant @xmath98 in ( [ f18 ] ) is @xmath99 for the projective-2 norm distance , we have @xmath100 . combining ( [ f10 ] ) and ( [ f12 ] ) , the normalized volume of @xmath77 in @xmath78 is @xmath101 @xmath102 where the integration is taken over @xmath103 @xmath104 substituting the variables @xmath105 and @xmath106 , after derivation , the result becomes @xmath107 this means @xmath108 . when @xmath109 , the situation becomes more complicated if we use the same method as @xmath75 , because it is very hard to determine the domain of integration as @xmath83 changes . before further analysis , we introduce the selberg beta integration and generalized selberg beta integration [ 20 ] as two lemmas which are the keys to solve the problem . _ lemma 1 : _ ( selberg beta integration [ 20 ] ) let the set @xmath110 be the integration domain , and the parameters @xmath111 , @xmath112 , and @xmath113 satisfy @xmath114 , @xmath115 , and @xmath116 respectively , then the selberg beta integration is @xmath117 @xmath118 the generalization of the above integration is the generalized selberg integration stated in lemma 2 . _ lemma 2 : _ ( generalized selberg beta integration [ 20 ] ) let @xmath119 be the domain of integration , and the parameters @xmath111 , @xmath112 , and @xmath120 satisfy @xmath114 , @xmath114 , @xmath121 respectively , then the generalized selberg beta integral is @xmath122 based on the two lemmas , we could solve the volume calculation problem with the two defined distances . first , we consider the case of @xmath123 , then the volume of the ball @xmath124 in @xmath67 is written as @xmath125 @xmath126 the integration is taken over the domain in ( [ f11 ] ) . notice that the polynomial to be integrated is is symmetric and there are @xmath127 permutaions of @xmath128 , which means that ( [ f30 ] ) could be transferred to the following equivalent integration problem @xmath129 @xmath130 where the new domain @xmath131 is @xmath132 . substituting the variables @xmath133 , @xmath134 , with @xmath135 , ( [ f31 ] ) is further transferred to @xmath136 @xmath137 where the integration is taken over @xmath138 . according to lemma 2 , we get the final formula for the case of @xmath139 as @xmath140 @xmath141 which is further simplified to @xmath142 for the case of @xmath143 , the domain of integration is a complicated polyhedron . in order to obtain the precise result , the integration domain has to be divided into @xmath144 sections and calculated in each section respectively . even so , it is still hard to get an analytic solution . fortunately , the requirement of the volume formula of hyperball @xmath145 for @xmath146 when @xmath147 and @xmath148 seldom occures , and if needed , the formula in reference [ 8 ] is accurate enough for practical applications . therefore , we cease to study this problem further in this paper . according to the projective-2 norm distance , the volume of @xmath83-ball in @xmath67 is calculated as the following integration @xmath149 @xmath150 where @xmath151 is defined in ( [ f12 ] ) . again , using the property of symmetric polynomial and substituting the variable @xmath133 with @xmath152 , @xmath134 , the above integration is transformed to @xmath153 @xmath154 where the integral is taken over @xmath155 . using lemma 1 and simplifying the results , we get @xmath156 this means @xmath157 . when @xmath158 , for any two points @xmath159 , let @xmath160,{{\mathbf{v}}_2 } = \left [ { \begin{array}{*{20}{c } } { { \mathbf{v}}_2 ^ 1}&{{\mathbf{v}}_2 ^ 2 } \end{array } } \right]\ ] ] where @xmath161 and @xmath162 are the first @xmath163 columns of @xmath164 and @xmath165 respectively , while @xmath166 and @xmath167 are the last @xmath168 columns of @xmath164 and @xmath165 respectively . as any two @xmath11-dimensional subspaces in a @xmath169-dimensional space have an overlapped subspace with dimension of @xmath168 , let @xmath166 and @xmath167 be the overlapped subspace , which means @xmath170 , where @xmath171 is a @xmath172 unitary matrix . using @xmath173 , we get @xmath174 = \left [ { \begin{array}{*{20}c } { { \mathbf{v}}_1^{1,h } { \mathbf{v}}_2 ^ 1 } & { \mathbf{0 } } \\ { \mathbf{0 } } & { \mathbf{u } } \\ \end{array } } \right].\ ] ] let the svd of @xmath175 be @xmath176 , then @xmath177 could be expressed as @xmath178\left [ { \begin{array}{*{20}c } { \cos { \mathbf{\theta _ a } } } & { \mathbf{0 } } \\ { \mathbf{0 } } & { \cos { \mathbf{0 } } } \\ \end{array } } \right]\left [ { \begin{array}{*{20}c } { { \mathbf{v}}_a^h } & { \mathbf{0 } } \\ { \mathbf{0 } } & { \mathbf{i } } \\ \end{array } } \right]\ ] ] where @xmath179 is a @xmath180 diagonal matrix and @xmath181 is the canonical angle matrix in @xmath182 . according to the distance definitions , the volume of hyperball in @xmath183 is equal to that of @xmath182 . it also means @xmath184 in this section , we present some illustrative numerical results to show that the formulas provided in section iii yield very accurate results in various cases . in simulation , we generate @xmath185 couples of uniformly distributed points in @xmath183 , and calculate the projective - f norm distance and projective-2 norm distance of each couple of points . according to ( [ f8 ] ) , the cdf of these distance values denotes the probability @xmath186 . in order to generate uniformly distributed points in @xmath183 , a random matrix @xmath187 with @xmath188 elements is generated first . next , let the thin qr decomposition of @xmath189 be @xmath190 where the diagonal elements of the matrix @xmath191 is positive ( e.g. , implemented by the gram - schmidt process ) , then the matrix @xmath192 is the desired matrix . here we provide a simple explanation of this method . for more details please refer to [ 21 ] . the ginibre ensemble @xmath193 consists of matrices @xmath194 , whose elements @xmath195 are independent and identically distributed ( i.i.d . ) standard normal complex random variables . then , the probability density function of @xmath196 is @xmath197 . let the probability density function @xmath198 define a measure of @xmath193 as @xmath199 . given any @xmath200 , we define the equivalent class @xmath201 = \left\ { { \left . { { \mathbf{uqr } } } \right|{\mathbf{u } } \in \mathbb{u}\left ( { n } \right ) } \right\}$ ] . note that the set @xmath202 forms the stiefel manifold @xmath203 . since the measure @xmath204 is invariant under left - multiplication by @xmath205 , the restriction of @xmath206 to equivalent @xmath201 $ ] is also left - invariant for every @xmath207 . according to the harr measure theory , it means that the matrix @xmath208 is uniformly distributed on @xmath203 given any matrix @xmath207 . therefore , the method generates uniformly distributed points in grassmann manifold considering the equivalent relationship in stiefel manifold . the comparisons between the simulated and calculated results are shown in fig . 1 , fig . 2 , fig . 3 , and fig . 4 for various cases . in fig . 1 , the cdf of correlation coefficient in @xmath209 is obtained by monte carlo simulation ( solid line ) and it is compared with the corresponding values ( circle ) from closed form ( [ f15 ] ) . obviously , we can say that the simulation results match the calculated results perfectly . for @xmath75 with projective - f norm distance , the simulated and calculated volume of hyperball with radius @xmath83 are compared in fig . 2 , while for @xmath210 and @xmath123 , similar comparisons are shown in fig . these two figures show the accuracy of ( [ f18 ] ) , ( [ f34 ] ) , and ( [ f41 ] ) . in fig . 4 , the calculated values of formulas ( [ f37 ] ) and ( [ f41 ] ) are verified by simulation results for various @xmath11 and @xmath169 . again , they match each other almost perfectly . all the numerical results indicate that the closed - form formulas in this paper are accurate . furthermore , as @xmath211 is an exponential function of the dimension of complex vector space @xmath169 for given @xmath83 and @xmath11 , the volume of the hyperball with radius @xmath83 decreases very fast as @xmath169 grows , which means that the distance of any two @xmath11-dimensional subspaces in @xmath70 approaches to the maximum value rapidly as @xmath169 grows . in other words , the probability of any two @xmath11-dimensional subspaces being mutually orthogonal approaches to 1 very fast . this provdes a foundation for large - scale mu - mimo communication systems where the eigenspace of the wireless channel matrix is a point in @xmath183 . in this section , the formulas in section iv are applied to analyze the sinrs of mu - mmo with non - ideal csi in large - scale mimo systems . firstly , an error model of csi is constructed , then an approximation expression of sinr of each user is derived . after that , the expectation of sinr is obtained based on the probability density function of correlation coefficient between any two points in complex grassmann manifold . this expectation can be used to predict each user s data transmission rate in mu - mimo systems with non - ideal csi , and to predict the total data transmission rate in mu - mimo systems . finally , simulation results are provided to verify the validity of the approximation expression . considering a wireless communication system where a bs serves @xmath212 users . suppose that the bs has @xmath213 transmit antennas and each user has one receive antenna . in this system , the bs chooses the transmission mode adaptively according to the information of the @xmath212 users when transmitting data . the adaptive transmission mode includes switching between su - mimo and mu - mimo , and changing the the selection of users to serve in mu - mimo mode . in this paper , we assume that the bs only has cqi of each user in su - mimo mode as @xmath214 , which is the case in practical systems , e.g. , 3gpp lte / lte - a . let @xmath215 $ ] , @xmath216 , be the channel vector of the @xmath217th user with @xmath188 elements , then the eigenvector space of @xmath218 is @xmath219 . because of system errors , e.g. , limited feedback bandwidth or measurement errors , the bs obtains the csi as @xmath220 instead of @xmath221 , and the relation between them could be modeled as @xmath222 where @xmath223 is the correlation coefficient between @xmath221 and @xmath220 defined as @xmath224 . since only the @xmath217th user is studied in this paper , the @xmath111 is used to replace @xmath225 for simplification . although @xmath226 satisfies the constraint @xmath227 , it should be much closer to 1 than to 0 so that the mu - mimo system can work in practice . in a tdd system , by making use of the channel reciprocity , the bs obtains the downlink csi through the pilot sent by users in the uplink , hence @xmath111 is mainly determined by the sinr of each user in the uplink , and it is different for each user . the symbol @xmath228 represents an unit vector in the null space of @xmath220 . obviously , @xmath221 , @xmath220 , and @xmath228 are elements in @xmath229 . at the bs side , the data of @xmath212 users transmitted on the same time - frequency resource in mu - mimo systems are mapped onto the @xmath213 antennas by the precoding matrix @xmath230 $ ] , where @xmath231 is the @xmath213-dimensional precoding vector belonging to the @xmath217th user . then , the sinr of the @xmath217th user with non - ideal csi could be written as @xmath232 where @xmath233 denotes the power of noise plus interference from neighboring cells . next , we analyze the sinr based on two general precoding methods : conjugate beamforming ( cb ) and zf . for cb , the precoding vector of the @xmath217th user is @xmath234 where @xmath235 is a scale factor which denotes the power allocated to the @xmath217th user . in this paper , we assume that the total transmit power is @xmath236 , and it is equally distributed among users . hence , @xmath237 , and @xmath238 . then , the sinr of the @xmath217th user is @xmath239 @xmath240 where @xmath241 . substituting ( [ f42 ] ) into ( [ f45 ] ) and making use of @xmath242 , it becomes @xmath243 @xmath244 in ( 46 ) , as @xmath213 is large , the correlation coefficients @xmath245 and @xmath246 are almost the same and close to zero , hence ignoring the term @xmath247 in approaximation 1 and replacing @xmath246 with @xmath245 in approaximation 2 would cause little error to the estimated sinr . actually , when the csi is ideal , the approximation ( [ f46 ] ) is equal to the accurate sinr . when @xmath212 is large , the term @xmath248 is close to @xmath249 $ ] . let @xmath250 , then it is a random variable denoting the correlation coefficient between any two points in @xmath251 . hence , the probability function of @xmath252 could be derived from ( [ f15 ] ) as @xmath253 using the definition of beta function , @xmath254 $ ] is calculated as @xmath255 = \int_0 ^ 1 { x^2 f\left ( x \right)dx } = \left ( { n - 1 } \right)b\left ( { 2,n - 1 } \right ) = \frac{1}{n}.\ ] ] therefore , when @xmath212 is large , the sinr could be further simplified to @xmath256 with ( [ f49 ] ) , we can predict the capacity of large - scale mu - mimo systems when cb is used . the estimation error of ( [ f46 ] ) is mainly caused by ignoring the cross term @xmath257 . as @xmath258 and @xmath259 are independent variables , the cross term approaches to the fixed value @xmath260 $ ] gradually as @xmath212 grows , which means that the standard derivation ( std . ) of the estimation error decreases when @xmath212 grows . on the other hand , the values of @xmath258 and @xmath259 are closer and closer to 0 when @xmath213 increases , thus the std . of estimation error decreases correspondingly . furthermore , as @xmath111 is close to 1 , @xmath261 decreases rapidly when @xmath111 increases , hence as the accuracy of cqi estimation increases , the std . of estimation error decreases . for ( 49 ) , it is mostly affected by the value of @xmath212 according to its derivation process , hence increasing @xmath212 will decrease the estimation error of ( [ f49 ] ) . in order to verify the validity of ( 46 ) , we provide some numerical results in fig . 5 and fig . 6 , where the mean and std . of estimation errors are used to measure the accuracy of ( [ f46 ] ) . as @xmath111 is mainly determined by the uplink sinrs of users , we consider three typical sinr values of 0 , 3 , and 6 db , which correspond to the @xmath111 values of 0.7 , 0.8 , and 0.9 respectively according to ( [ f42 ] ) . if the uplink sinr of a user is less than 0 db , we consider it not suitable for mu - mimo . in fig . 5 , we can see that the mean of estimation error is almost 0 regardless of the parameters @xmath213 , @xmath212 , and @xmath111 , which means that ( [ f46 ] ) provides an unbiased estimation of sinr . the std . of estimation error , which is affected by @xmath213 , @xmath212 , and @xmath111 , is shown in fig . 6 , and it is consistant with the analysis in 2 ) above . the accuracy of ( [ f49 ] ) is shown in fig . 7 , where the estimated values ( est . in fig . 7 ) are compared with the real values ( real in fig . obviously , ( [ f49 ] ) provides an approximation of the expectation of sinr with very high accuracy , hence it is helpful for us to carry out capacity estimation . zf precoding is used to remove the multi - user interference completely , where the multi - user interference channel of the @xmath217th user is defined as @xmath262^t.\ ] ] applying the null space projection , the precoding vector of the @xmath217th user could be written as @xmath263 although zf could remove the multi - user interference , it would reduce the sum of channel capacity if the grouped users for mu - mimo have high mutual channel correlation . therefore , the bs should group the users whose mutual correlation coefficients @xmath264 , are lower than a predefined threshold . since @xmath265 , @xmath266 approaches to @xmath267 when @xmath213 becomes large according to the conclusion in section iii . hence , it is reasonable to replace @xmath268 with @xmath269 to simplify the analysis in this paper , and ( [ f51 ] ) becomes @xmath270 without losing generality , considering the @xmath217th user , substituting ( [ f52 ] ) into ( [ f43 ] ) and after derivation , it becomes @xmath271 @xmath272 in order to simplify the denominator in ( [ f53 ] ) , we approximate it as @xmath273 when @xmath274 , the left side is strictly equal to the right side in ( [ f54 ] ) . when @xmath275 , as @xmath276 , are independent and uniformly distributed in @xmath277 , the left side and right side of ( [ f54 ] ) approach to the same value as @xmath212 grows . it means that ( [ f54 ] ) is a reasonable approximation for a large value of @xmath212 . with ( [ f54 ] ) , ( [ f53 ] ) could be further reduced to ( [ f55 ] ) as @xmath278 where @xmath279 . substituting ( [ f42 ] ) into ( [ f55 ] ) , we get @xmath280 we get ( [ f56 ] ) at the top of the next page . @xmath281 since the above formula is still too complicated , we continue to simplify the numerator and denominator respectively by ignoring the secondary factors . for the numerator , as the probability of @xmath282 approaches to 1 when @xmath283 , the term @xmath284 can be dropped and the result becomes @xmath285 . similarly for the denominator , we reduce it to @xmath286 . in addition , we have @xmath287 . therefore , the sinr of @xmath217th user could be rewritten as @xmath288 with ( [ f57 ] ) , the bs can choose the suitable combination of modulation scheme and channel coding rate for each user in a mu - mimo user group . as a result , the bs can properly group mu - mimo users and adaptively switch between mu - mimo and su - mimo . in some cases , the expectation of sinr provides a more reliable prediction of the mu - mimo spectral efficiency at a given error level of csi . let @xmath289 , then @xmath290 could be written as a function of @xmath291 : @xmath292 . since @xmath293 has a form of @xmath294 , it is a convex function . thus by making using of jensen inequality ( @xmath295 when @xmath296 is a convex function ) , the expectation of ( [ f57 ] ) has a lower bound @xmath297 \geq \frac{{\alpha ^2 \left ( { 1 - { \mathbb e}\left [ z \right ] } \right)^2 } } { { \left ( { 1 - \alpha ^2 - k\gamma } \right){\mathbb e}\left [ z \right ] + k\gamma } } .\ ] ] as @xmath298 , are i.i.d . random variables , according to ( [ f48 ] ) , we have @xmath299 = ( k - 1)(n - 1 ) b ( 2,n - 1)$ ] , and we obtain theorem 2 as summarized below . for an @xmath300 mu - mimo communication system , where @xmath213 is the number of transmit antenna at the bs side , and @xmath212 is the number of grouped users . if the @xmath301 of the @xmath217th user is @xmath302 which is defined as @xmath303 , then the expected sinr at the receiver has a lower bound @xmath304 \geq \frac{{\alpha ^2 \left ( { n - k + 1 } \right ) } } { { \left ( { 1 - \alpha ^2 - k\gamma } \right)\left ( { k - 1 } \right ) + nk\gamma } } .\ ] ] @xmath305 \nonumber \\ & \geq \frac{{\alpha ^2 \left ( { n - k + 1 } \right ) } } { { \left ( { 1 - \alpha ^2 - k\gamma } \right)\left ( { k - 1 } \right ) + nk\gamma } } .\end{aligned}\ ] ] with ( [ f59 ] ) , we could estimate the expectation of channel capacity of mu - mimo when zf precoding is used for large - scale mu - mimo systems . at this point , we provide some discussion about the approximations ( [ f57 ] ) and ( [ f59 ] ) . for ( [ f57 ] ) , there are three factors affecting its accuracy . firstly , given the values of @xmath212 and @xmath111 , the probability of @xmath306 increases as @xmath213 grows in ( [ f51 ] ) , hence the approximation error of the numerator in ( [ f56 ] ) is almost ignorable when @xmath213 is large , e.g. , @xmath307 . on the other hand , the values of @xmath308 and @xmath309 are closer and closer to zero as @xmath213 grows . as a result , the approximations of the numerator and denominator in ( [ f56 ] ) are reasonable for large @xmath213 . therefore , we conclude that the gap between the approximated value in ( [ f57 ] ) and the real sinr becomes smaller as @xmath213 grows . secondly , given the values of @xmath213 and @xmath111 , the probability of @xmath306 decreases as @xmath212 grows , while the values of @xmath308 and @xmath309 show inverse tendency . this means that the approximation of @xmath310 in ( [ f51 ] ) and the omission of @xmath308 and @xmath309 in ( [ f56 ] ) would cause larger error as @xmath212 grows . thirdly , given the values of @xmath213 and @xmath212 , the value of @xmath311 decreases as @xmath111 increases . hence , ignoring this term causes little error in ( [ f56 ] ) for @xmath111 close to 1 . therefore , as the csi is more accurate , the approximation in ( [ f57 ] ) resluts in less error . in conclusion , larger values of @xmath213 , @xmath111 , or smaller value of @xmath212 result in a more accurate approximation in ( [ f57 ] ) . as ( [ f59 ] ) provides a lower bound of expectation of sinr , we can analyze the factor which affects the gap between the lower bound and the real value . firstly , the term @xmath312 approaches to @xmath313 $ ] as @xmath212 grows , which means that for a large value of @xmath212 , the inequality in ( [ f59 ] ) approaches to equality . secondly , if @xmath312 is negligible relative to the value of @xmath314 , e.g. , @xmath213 or @xmath113 is very large , the inequality also approaches to equality . as a result , the lower bound of ( [ f59 ] ) is asymptotically tight as @xmath212 or @xmath213 grows . with the approximation of sinr in ( [ f57 ] ) and @xmath315 $ ] in ( [ f59 ] ) , it is easy for the bs to predict the capacity of each user in mu - mimo mode , the sum capacity gain over su - mimo , and the total capacity of the grouped users . we present numerical simulation results in this section to verify the validity of ( [ f57 ] ) and ( [ f59 ] ) . similarly to the conjugate beamforming case , the mean and std . of the estimation error are used to measure the accuracy of ( [ f57 ] ) , and the results with different values of @xmath212 , @xmath213 , and @xmath111 are shown in fig . 8 and fig . 9 respectively . for ( [ f59 ] ) , the estimation value and the real expectation of sinr are compared in fig . 10 . in fig . 8 , we can see that the mean of estimation error increases as the su - mimo sinr increases , which is because when the value of su - mimo sinr is small , the noise and inter - cell interference are the major factors affecting the mu - mimo sinr . as the value of su - mimo sinr increases , the residual multi - user interference becomes the major factor , and the estimation error of ( [ f57 ] ) becomes dominant . in addition , the mean of estimation error increases as the value of @xmath212 increases , and decrease as the value of @xmath213 or @xmath111 decreases , which verifies the analysis in 2 ) above . for the same reasons , fig . 9 shows similar variation tendency to fig . 8 . considering that the number of grouped users is about 1/10 to 1/5 of the number of antennas in large - scale mu - mimo systems ( e.g. , @xmath316 , @xmath317 ) , ( [ f57 ] ) provides sinr estimation with relatively high accuracy . in fig . 10 , the difference between the estimation and real values is close to 0 for most cases , and it is about 1 db in the high sinr region . hence , it provides a reasonable estimation of the expectation of sinr . in this section , we compare the sinrs of cb and zf in ( 44 ) and ( 59 ) . the sinr gain of zf over cb at a given csi error level is @xmath318 where @xmath291 is the same as in ( [ f58 ] ) which denotes the multi - user interference power among the users in mu - mimo . according to ( [ f60 ] ) , @xmath319 when @xmath212 is large , then ( [ f60 ] ) is reduced to a simple function of @xmath212 , @xmath213 , and @xmath111 as @xmath320 here , we offer a discussion about ( [ f60 ] ) and ( [ f61 ] ) . in the high @xmath301 region , as @xmath321 and @xmath322 , we have @xmath323 , and @xmath324 \to \frac{{{n^2 } + \left ( { k - 1 } \right)\left ( { k - 1 - 2n } \right)}}{{{n^2}\left ( { 1 - { \alpha ^2 } } \right)}}$ ] . therefore , the sinr gain of zf over cb increases from @xmath325 to positive infinity when @xmath111 varies from 0 to 1 . it indicates that zf has an overwhelming advantage compared to cb when the csi is ideal , which coincides with our conventional understanding . however , as the error of csi increases , the sinr gain of zf begins to decrease , and both zf and cb has the same sinr value when @xmath326 . if @xmath111 decreases further , the sinr of zf becomes smaller than that of cb . in the low @xmath301 region , as @xmath327 , @xmath328 , and @xmath324 \to \frac{{n - k + 1}}{n}$ ] . it means that the sinr gain of zf over cb is definitely smaller than 1 , regardless of the considered parameters . given the value of @xmath111 , increasing the value of @xmath213 or decreasing the value of @xmath212 could result in smaller @xmath291 according to the definition of @xmath291 , thus enlarge the sinr gain of zf over cb . in conclusion , if the csi is not ideal , the sinr gain of zf over cb is determined by the parameter @xmath111 given the values of @xmath213 and @xmath212 in the high @xmath301 region , while the sinr gain of zf is always smaller than 1 in the low @xmath301 region and independent of @xmath111 . we provide some numerical results to verify the rationality of formula ( [ f60 ] ) and ( [ f61 ] ) in fig . 12 , and fig . 13 . in fig . 11 , the formulas ( [ f60 ] ) ( estimated gain ) and ( [ f61 ] ) ( asmp gain ) are compared with the real gain , where we can see that both the formulas ( [ f60 ] ) and ( [ f61 ] ) provide underestimation of the real gain . nevertheless , the estimation errors are relatively small , and these two formulas could be used to predict the gain of zf over cb for large - scale mu - mimo systems . the std . of ( [ f60 ] ) and ( [ f61 ] ) are shown in fig . 12 and fig . 13 respectively . for the same reason , the variation of std . with parameters @xmath212 , @xmath213 , and @xmath111 in fig . 12 is similar to that of fig . 13 shows a relatively different variation of std . , where the std . increases rapidly as @xmath212 decreases because ( [ f61 ] ) can be considered accurate only when @xmath212 is large . in summery , these figures indicate that formulas ( [ f60 ] ) and ( [ f61 ] ) can be used to predict the gain of zf over cb , and switch between these two precoding methods adaptively . when the csi is ideal , the multi - user interference is removed completely when zf precoding is employed , then the sinr of the @xmath217th user for mu - mimo is @xmath329 where @xmath330 with the matrix @xmath331 defined as @xmath332^t $ ] . note that the factor @xmath333 is ignored in @xmath334 here for the same reason as in ( [ f52 ] ) . substituting @xmath334 into ( [ f62 ] ) and after simplification , we have @xmath335 using the same method as in section iv.a , the expectation of @xmath336 is calculated as @xmath337 = \frac{{1 - { \mathbb e}\left [ { { \bf{u}}_i { \mathbf{\tilde h}}_i { \mathbf{\tilde h}}_i^h { \bf{u}}_i } \right ] } } { { k\gamma } } = \frac{{1 - \left ( { k - 1 } \right){\mathbb e}\left [ { x^2 } \right ] } } { { k\gamma } } = \frac{{n - k + 1 } } { { nk\gamma } } .\ ] ] @xmath338 & = \frac{{1 - { \mathbb e}\left [ { { \bf{u}}_i { \mathbf{\tilde h}}_i { \mathbf{\tilde h}}_i^h { \bf{u}}_i } \right ] } } { { k\gamma } } \nonumber \\ & = \frac{{1 - \left ( { k - 1 } \right){\mathbb e}\left [ { x^2 } \right ] } } { { k\gamma } } \nonumber \\ & = \frac{{n - k + 1 } } { { nk\gamma } } .\end{aligned}\ ] ] it is not difficult to find that the value of ( [ f64 ] ) is equal to that of ( [ f59 ] ) when @xmath339 . 14 and fig . 15 show the accuracy of formula ( [ f63 ] ) , where both the mean and std . of estimation error is very close to zero . hence , formula ( [ f63 ] ) provides a very accurate upper bound of the sinr of mu - mimo given @xmath340 . in fig . 16 , the estimation matches the real value almost perfectly . therefore , formula ( [ f64 ] ) could be used to estimate ergodic capacity given @xmath341 . we studied the volume of a hyperball in a complex grassmann manifold based on the probability of canonical angles between any two points and obtained closed - form formulas for various @xmath342 values and radii of the hyperball . in addition , we provided an application of the formulas in large - scale mu - mimo communication systems and derived closed - form approximations of mu - mimo sinrs given the number of bs antennas @xmath213 , mu - mimo grouped user number @xmath212 , csi error @xmath111 , and the user channel quality @xmath301 in su - mimo mode . simulation results verified the accuracy of the formulas and the closed - form approximations . our results solve a fundamental problem whose solution has been missing but is necessary for a practical deployment of massive mu - mimo systems . in the future , the cases of more than one receive antennas and correlated channel matrix will be considered , where the volume formulas of hyperball in grassmann manifold when @xmath343 will be applied . let @xmath344 and @xmath345 , then the probability could be written as @xmath346 when @xmath347 , the domain of integration is the set @xmath348 which is shown in fig . 17 , where @xmath151 could be divided into two parts @xmath349 and @xmath350 . the sum integration on these two parts is @xmath351 @xmath352 where @xmath353 @xmath354 let @xmath355 , ( [ f67 ] ) could be further integrated as @xmath356 @xmath357 when @xmath358 , the domain of integration is the set @xmath359 as shown in fig . similar to the process of @xmath347 , the integration is written as @xmath360 @xmath361 the first term on the right hand of the second equation is the same as the case of @xmath362 , so it could be written as @xmath363 @xmath364 where @xmath365 @xmath366 let @xmath91 and @xmath97 , combining the constant @xmath98 , ( [ f66 ] ) , and ( [ f70 ] ) , ( 65 ) could be written as the final formula @xmath367 where @xmath368 @xmath369 and @xmath370 @xmath371 1 a. narula , m. j. lopez , m. d. trott , and g. w. wornell , efficient use of side information in mulitiple - antenna data transmission over fading channels,_ieee j. select . areas in commun . 16 , no . 8 , pp . 1423 - 1436 , oct . k. k. mukkavilli , a. sabharwal , e. erkip , and b. aazhang , on beamforming with finite rate feedback in multiple antenna systems,_ieee trans . inf . theory _ , 10 , pp . 2562 - 2579 , oct . j. c. roh and b. d. rao , an efficient feedback method for mimo systems with slowly time - varying channels , `` _ in proc.of wireless comm . and networking _ , vol . 2 , pp . 760 - 764 , march 2004 . d. j. love , r. w. heath , and t. strohmer , grassmannian beamforming for multiple - input multiple - output wireless systems , '' _ ieee trans . inf . theory _ , 2735 - 2747 , oct . d. j. love and r. w. heath , limited feedback unitary precoding for spatial multiplexing , `` _ ieee trans . inf . theory _ , vol.51 , no.8 , pp . 2967 - 2976 , aug . b. mondal , dutta , and r. w. heath , quantization on the grassmann manifold , '' _ ieee transaction on signal processing _ , vol . 4208 - 4216 aug . w. dai , y. liu , and b. rider , quantization bounds on grassmann manifolds and applications to mimo communications . `` _ ieee trans . inf . theory _ , vol.54 , no . 3 , pp 1108 - 1123 , oct . 2008 . a. barg and d. y. nogin , bounds on packings of spheres in the grassmann manifold , '' _ ieee trans . info . theory _ , 2450 - 2454 , aug . 2002 . henkel and oliver , sphere - packing bounds in the grassmann and stiefel manifolds , `` _ ieee trans . inf . theory _ , 4208 - 4216 oct . r. t. krishnamachari , and m. k. varanasi , volume of geodesic balls in the complex stiefel manifold , '' _ 46th annual allerton conference on communication , control , and computing _ , urbana - champaign , il , 2008 . r. a. pitaval and o. tirkkonen , volume of ball and hamming - type bounds for stiefel manifold with euclidean distance , `` _ signals , systems and computers ( asilomar),2012 _ , pp . 483 - 487 , nov . 2012 . e. g. larsson , very large mimo systems part i : theory and analysis , '' _ icassp 2012 tutorial _ , march 26 , 2012 . t. l. marzetta , noncooperative cellular wireless with unlimited numbers of base station antennas , `` _ ieee trans . wireless . _ , vol . 9 , no . 11 , pp 3590 - 3600 , nov . v. mancuso and s. alouf , reducing costs and pollution in cellular networks , '' _ ieee communications mag _ , pp . 63 - 71 , august 2011 . 3gpp technical specication group radio access network ; evolved universal terrestrial radio access ( e - utra ) ; further advancement for e - utra physical layer aspects ( rel-11 ) , 3gpp tr36.814 v10.0.0 , july 2012 . r. balraj , scholand , t. bai , and s.iwelski , analysis of cqi prediction for mu - mimo in lte systems , `` _ vehicular technology conference ( vtc spring ) _ , 2012 ieee 75th date of conference : 6 - 9 may 2012 . z. h. lin , p. xiao , and y. wu , performance analysis for linearly precoded lte downlink multiuser mimo , '' _ eurasip journal on wireless communications and networking 2011_. b. m. hochwald , t. l. marzetta , and b. hassibi , space time autocoding , `` _ ieee trans . inf . 47 , no . 7 , nov . 2001 . c. chevalley , the theory of lie groups , 1th ed , 1em plus 0.5em minus 0.4emprinceton university press , 1946 . m. l. mehta , random matrices . 1th ed , 1em plus 0.5em minus 0.4em elsevier academic press , 2004 . p. diaconis , what is a random matrix ? '' _ notices of the ams _ , vol . 1348 - 1349 , dec . 2005 . a. fehske , g. fettweis , j. malmodin and g. biczok , the global footprint of mobile communications : the ecological and economic perspective , " _ ieee communications magazine _ , pp . 55 - 62 , aug 2011 .

it has been known that the naive factorization assumption ( fa ) @xcite does not apply to exclusive @xmath6 meson decays into charmonia , such as @xmath7 @xcite . these modes belong to the color - suppressed category @xcite , for which predictions from fa are always small due to the vanishing wilson coefficient @xmath8 . however , the branching ratios measured by babar recently @xcite @xmath9 imply a larger parameter @xmath10 @xcite . the same difficulty has appeared in other similar decays @xmath11 : the observed branching ratios are usually many times larger than the expectations from fa . for example , the @xmath12 decays do not receive factorizable contributions , but the data of their branching ratios are comparable to those of @xmath13 in eq . ( [ eq : brexp ] ) . many attempts to resolve this puzzle have been made in more sophisticated approaches ( for a review , see @xcite ) . the large @xmath14 leads to a natural conjecture that nonfactorizable contributions , such as the vertex and spectator corrections from figs . [ fvs](a)-[fvs](d ) and from figs . [ fvs](e ) and [ fvs](f ) , respectively , may play an important role . it has been found @xcite in the qcd - improved factorization ( qcdf ) @xcite that the above nonfactorizable contributions , resulting in the branching ratio @xmath15 , are too small to explain the data . since only the leading - twist ( twist-2 ) kaon distribution amplitude was included in @xcite , the authors of @xcite added the twist-3 contribution , which is chirally enhanced , though being power - suppressed in the heavy quark limit . unfortunately , [ fvs](e ) and [ fvs](f ) generate logarithmical divergences from the end - point region , where the parton momentum fraction of the kaon is small . to make an estimation , arbitrary cutoffs for parameterizing the divergences , i.e. , large theoretical uncertainties , have been introduced . the end - point singularities become more serious in the qcdf analysis of the @xmath16 decays @xcite . it has been claimed that the data of those modes involving @xmath17 , @xmath18 , and @xmath19 are not understandable in qcdf , and only those involving @xmath20 are @xcite . the @xmath13 decays have been also studied in light - cone sum rules ( lcsr ) @xcite , and the small branching ratio from the factorizable contribution @xcite , @xmath21 was confirmed . the nonfactorizable contributions considered in @xcite refer , however , to those arising from the three - parton kaon distribution amplitudes . in the corresponding diagrams an additional valence gluon from the kaon attaches one of the charm quarks in the @xmath17 meson . adding these pieces , one arrived at the parameter @xmath22 , which is still insufficient to account for the data . the calculation of fig . [ fvs ] in lcsr , involving two - loop integrals , has not yet been performed . hence , the conclusion in @xcite simply indicates that the nonfactorizable contributions to the @xmath13 decays from higher fock states of the involved mesons are negligible . we doubt that the conclusion from qcdf is solid for two reasons . first , the end - point singularities render the estimation of the nonfactorizable contributions out of control . second , the simple asymptotic models were adopted for the charmonium distribution amplitudes , which , without any theoretical and experimental bases , may underestimate the nonfactorizable contributions . motivated by the above observations , we shall investigate the nonfactorizable corrections to fa from fig . [ fvs ] using the following method . we take into account the vertex corrections through the variation of the renormalization scale for the factorizable contributions , because their evaluation requires more theoretical inputs , which have not yet been available . the spectator amplitudes will be computed in the perturbative qcd ( pqcd ) approach based on @xmath1 factorization theorem @xcite . the formalism , similar to that of the color - suppressed mode @xmath23 @xcite , is free from the end - point singularities . we infer the charmonium distribution amplitudes from the non - relativistic heavy quarkonium bound - state wave functions , which have been shown to well explain the cross section of charmonium production in @xmath24 collisions @xcite . based on the universality of hadron distribution amplitudes , these more sophisticated models should be able to account for the exclusive @xmath6 meson decays into charmonia , if qcd factorization theorem works . the @xmath25 decays have been studied in pqcd @xcite , compared to which the new ingredients of this work are : 1 ) we clarify several controversial statements on the end - point singularities in the @xmath26 analysis in the literature ; 2 ) the @xmath27 meson distribution amplitudes derived from qcd sum rules @xcite are included up to two - parton and twist-3 level ; 3 ) the models for the charmonium distribution amplitudes have both theoretical and experimental bases ; 4 ) not only the @xmath25 decays , but @xmath28 are investigated . it will be shown that the obtained branching ratios , polarization fractions , and relative phases among various helicity amplitudes , except those of @xmath2 , are all in consistency with the existing data . the observed @xmath2 branching ratios , significantly larger than our results , are the only puzzle . therefore , our conclusion differs from that drawn in qcdf @xcite . the predictions for the @xmath29 modes can be compared with future measurements . we present our formalism for the @xmath25 decays as an example in sec . the factorizable contributions are treated in fa , since the @xmath30 transition form factors in the present case are characterized by a low scale , and may not be calculable . the vertex and spectator corrections are handled in the way stated above . the same formalism is then applied to the other exclusive @xmath6 meson decays into charmonia in sec . section iv is the conclusion . the appendix collects the expressions of the involved meson distribution amplitudes and of the factorization formulas for all the spectator amplitudes . for a review of the studies of semi - inclusive @xmath6 meson decays into charmonia , refer to @xcite . we write the @xmath6 ( @xmath17 ) meson momentum @xmath31 ( @xmath32 ) in the light - cone coordinates as @xmath33 with the mass ratio @xmath34 . the @xmath27 meson momentum is then given by @xmath35 . the polarization vectors of the @xmath17 meson are parameterized as @xmath36 the relevant effective hamiltonian for the @xmath25 decays is @xmath37-v_{tb}v_{ts}^{*}\sum_{k=3}^{10}c_k(\mu)o_k\right\}\ ; , \label{effj}\end{aligned}\ ] ] with the cabibbo - kobayashi - maskawa ( ckm ) matrix elements @xmath38 and the four - fermion operators , @xmath39 @xmath40 being the color indices . below we shall neglect the tiny product @xmath41 , and assume @xmath42 for convenience . the @xmath13 decay rate has the expression , @xmath43 in pqcd , we found that its characteristic scale is around 1 gev . with such a small hard scale , the perturbation theory may not be reliable . therefore , we do not attempt to calculate the factorizable contribution , but parameterize it in fa . the amplitude @xmath44 is then written as @xmath45 with the @xmath17 meson decay constant @xmath46 and the spectator amplitude @xmath47 . the form factor @xmath48 is defined via the matrix element , @xmath49 + f_0(q^2)\frac{m_b^2-m_k^2}{q^2}q_\mu\;,\label{fp}\end{aligned}\ ] ] @xmath50 being the momentum transfer , and @xmath51 the kaon mass . the effective wilson coefficient @xmath52 sums the contributions from both the tree and penguin operators in eq . ( [ effj ] ) : @xmath53 with @xmath54 and the charm quark charge @xmath55 . after parameterizing the factorizable contribution in fa , the soft dynamics is absorbed into the form factor @xmath56 . the remaining piece , i.e. , the wilson coefficient , is dominated by hard dynamics of @xmath57 , @xmath58 being the @xmath59 quark mass . we shall set the renormalization scale to @xmath60 gev , which corresponds to @xmath61 . the spectator amplitude @xmath62 is characterized by a scale of @xmath63 with @xmath64 being a hadronic scale @xcite : the hard gluons in figs . [ fvs](e ) and [ fvs](f ) , carrying the difference between the momenta of the soft spectator in the @xmath6 meson and of the energetic spectator in the kaon , are off - shell by @xmath63 . strictly speaking , once a @xmath6 meson transition form factor is treated as a soft object , the vertex corrections in figs . [ fvs](a)-[fvs](d ) should be added , which become of the same order as the spectator amplitude . this is the counting rules of qcdf . the pqcd formalism for the vertex corrections requires the inclusion of the transverse momenta @xmath1 of the charm quarks , because of the end - point singularities in some modes . the @xmath1-dependent charmonium wave functions are then the necessary inputs , which , however , have been determined neither theoretically nor experimentally . therefore , the vertex corrections appear as a theoretical uncertainty eventually , which can be covered by a variation of the scale @xmath65 between @xmath66 and @xmath67 as shown below . in terms of the notation in @xcite , we decompose the nonlocal matrix elements associated with longitudinally and transversely polarized @xmath17 mesons into @xmath68_{lj}\psi^l(x)+[{\not{\!\epsilon}}_l{\not{\!p}}]_{lj } \psi^{t}(x ) \bigg\}\ ; , \label{lpf}\\ \langle j/\psi(p,\epsilon_t)|\bar c(z)_jc(0)_l|0\rangle & = & \frac{1}{\sqrt{2n_c}}\int_0 ^ 1 dx e^{ixp\cdot z } \bigg\{m_{j/\psi}[{\not{\!\epsilon}}_t]_{lj}\psi^v(x)+ [ { \not{\!\epsilon}}_t{\not{\!p}}]_{lj}\psi^t(x ) \bigg\}\ ; , \label{spf}\end{aligned}\ ] ] respectively , which define the twist-2 distribution amplitudes @xmath69 and @xmath70 , and the twist-3 distribution amplitudes @xmath71 and @xmath72 with the @xmath73 quark carrying the fractional momentum @xmath74 . it is confusing that both the structures @xmath75 and @xmath76 were regarded as being leading - twist in the analysis of the @xmath13 decays @xcite . in fact , it is @xmath77 that is leading - twist , which contributes only to the @xmath78 decays . the vertex corrections to the @xmath13 decays , denoted as @xmath79 in qcdf , have been calculated in the ndr scheme @xcite . their effect can be combined into the wilson coefficients associated with the factorizable contributions : @xmath80 with the function , @xmath81\;,\end{aligned}\ ] ] where those terms proportional to @xmath82 have been neglected . we have also neglected the contributions from the electroweak penguin operators for simplicity . as shown in fig . [ rg ] , including the vertex corrections leads to a larger @xmath83 , which reproduces eq . ( [ eq : brnf ] ) roughly . we have confirmed that the variation of @xmath65 between @xmath84 and @xmath85 is wide enough for taking into account the effect of the vertex corrections . the definitions of the @xmath6 meson wave function @xmath86 and of the kaon distribution amplitudes @xmath87 are referred to @xcite . it is then straightforward to derive the spectator amplitude , @xmath88 where the amplitudes @xmath89 and @xmath90 result from the @xmath91 and @xmath92 operators in eq . ( [ effj ] ) , respectively . their factorization formulas are given by @xmath93\int_{0}^{\infty } b_{1}db_{1 } \phi _ { b}(x_{1},b_{1 } ) \nonumber \\ & & \times \big\ { \big [ ( 1 - 2r^{2}_{2})(1-x_{2 } ) \phi _ { k } ( x_{3 } ) \psi^{l}(x_{2 } ) + \frac{1}{2 } r^{2}_{2 } \phi_{k}(x_{3})\psi^{t } ( x_{2 } ) \nonumber \\ & & - r_{k } ( 1-r^2_2)x_3 \phi^{p}_{k}(x_3 ) \psi _ { l}(x_{2})+ r_{k } \left ( 2r^{2}_{2}(1-x_2)+(1-r^2_2)x_3 \right ) \phi^{\sigma}_{k}(x_3 ) \psi^{l}(x_{2 } ) \big]\nonumber \\ & & \times e_{4}(t_d^{(1)})h_d^{(1)}(x_1,x_2,x_3,b_1 ) \nonumber \\ & & - \big [ ( x_2+(1 - 2r^2_{2})x_{3})\phi _ { k } ( x_{3 } ) \psi^{l}(x_{2})+r^{2}_{2}(2r_{k}\phi^{\sigma}_{k}(x_{3 } ) -\frac{1}{2}\phi_{k}(x_{3}))\psi^{t}(x_{2 } ) \nonumber \\ & & -r_{k } ( 1-r^2_2)x_3 \phi^{p}_{k}(x_3 ) \psi _ { l}(x_{2})- r_{k } \left ( 2r^{2}_{2}x_2+(1-r^2_2)x_3 \right ) \phi^{\sigma}_{k}(x_3 ) \psi^{l}(x_{2})\big]\nonumber \\ & & \times e_{4}(t^{(2)}_d ) h_d^{(2)}(x_1,x_2,x_3,b_1)\;,\label{psi4}\\ { \cal m}_{6}^{(j/\psi k ) } & = & 16\pi c_{f}\sqrt{2n_{c}}\int_{0}^{1}[dx]\int_{0}^{\infty } b_{1}db_{1 } \phi _ { b}(x_{1},b_{1 } ) \nonumber \\ & & \times \big\ { \big [ ( 1-x_2+(1 - 2r^2_{2})x_{3})\phi _ { k } ( x_{3 } ) \psi^{l}(x_{2})+ r^{2}_{2}(2r_{k}\phi^{\sigma}_{k}(x_{3})-\frac{1}{2}\phi_{k}(x_{3}))\psi^{t}(x_{2 } ) \nonumber \\ & & -r_{k } ( 1-r^2_2)x_3 \phi^{p}_{k}(x_3 ) \psi^{l}(x_{2})- r_{k } \left ( 2r^{2}_{2}(1-x_2)+(1-r^2_2)x_3 \right ) \phi^{\sigma}_{k}(x_3 ) \psi^{l}(x_{2})\big]\nonumber \\ & & \times e_{6}(t_d^{(1)})h_d^{(1)}(x_1,x_2,x_3,b_1 ) \nonumber \\ & & - \big [ ( 1 - 2r^{2}_{2})x_{2 } \phi _ { k } ( x_{3 } ) \psi^{l}(x_{2 } ) + \frac{1}{2 } r^{2}_{2 } \phi_{k}(x_{3})\psi^{t } ( x_{2 } ) \nonumber \\ & & - r_{k } ( 1-r^2_2)x_3 \phi^{p}_{k}(x_3 ) \psi^{l}(x_{2})+ r_{k } \left ( 2r^{2}_{2}x_2+(1-r^2_2)x_3 \right ) \phi^{\sigma}_{k}(x_3 ) \psi^{l}(x_{2 } ) \big]\nonumber \\ & & \times e_{6}(t^{(2)}_d ) h_d^{(2)}(x_1,x_2,x_3,b_1 ) \big\}\;,\label{psi6}\end{aligned}\ ] ] with the color factor @xmath94 , the number of colors @xmath95 , the symbol @xmath96\equiv dx_1 dx_2 dx_3 $ ] and the mass ratio @xmath97 , @xmath98 being the chiral scale associated with the kaon . the evolution factors are written as @xmath99 with the wilson coefficients , @xmath100 the sudakov exponent is given by @xmath101\ ; , \label{sb } \\ s_{k}(t)&=&\exp\left[-s(x_{3}p_{3}^{-},b_{3 } ) -s((1-x_{3})p_{3}^{-},b_{3 } ) -2\int_{1/b_{3}}^{t}\frac{d{\bar{\mu}}}{\bar{\mu } } \gamma ( \alpha_{s}({\bar{\mu}}))\right]\ ; , \label{sbk}\end{aligned}\ ] ] with the quark anomalous dimension @xmath102 . note that the coefficient @xmath103 of the quark anomalous dimension in eq . ( [ sb ] ) differs from 2 in eq . ( [ sbk ] ) . the reason is that the rescaled heavy - quark field adopted in the definition of the heavy - meson wave function has a self - energy correction different from that of the full heavy - quark field @xcite . the variables @xmath104 and @xmath105 , conjugate to the parton transverse momenta @xmath106 and @xmath107 , represent the transverse extents of the @xmath6 and @xmath108 mesons , respectively . the explicit expression of the exponent @xmath109 can be found in @xcite . the above sudakov exponentials decrease fast in the large @xmath59 region , such that the @xmath13 spectator amplitudes remain sufficiently perturbative in the end - point region of the momentum fractions . the hard functions @xmath110 , @xmath111 and 2 , are @xmath112 with the arguments , @xmath113r_2 ^ 2\ ; , \\ d_2 & = & x_1x_2r_2^{2}+(x_1-x_2)x_3(1-r_2^{2 } ) + \left(\frac{1}{4}-x_2 ^ 2\right)r_2 ^ 2\;.\end{aligned}\ ] ] the hard scales @xmath114 are chosen as @xmath115 without the transverse momenta @xmath1 , the terms containing @xmath116 in eqs . ( [ psi4 ] ) and ( [ psi6 ] ) are logarithimically divergent due to the end - point singularities from @xmath117 @xcite . since the end - point singularities appear only at the power of @xmath118 , instead of at leading power , the spectator contributions are expected to be evaluated more reliably than the factorizable contributions in pqcd . for the @xmath78 modes , the parametrization of the kinematic variables in eqs . ( [ mom ] ) and ( [ pol ] ) still apply . we further define the polarization vectors @xmath119 for the @xmath120 meson , which is orthogonal to the @xmath120 meson momentum @xmath121 . in this case the @xmath120 meson mass should be kept in the intermediate stage of the calculation , and then dropped at the end according to the power counting rules in @xcite . the decay rate is given by @xmath122 the amplitudes for different final helicity states are expressed as @xmath123 , \nonumber\\ & & i\ , \epsilon^{\alpha \beta\gamma \rho } \epsilon^{*}_{2\alpha}\epsilon^{*}_{3\beta } \frac{p_{2\gamma } p_{3\rho } } { m_b^2}\ , \left[r_2a_{\rm eff}(\mu)f_{j/\psi}v(m_{j/\psi}^2)+{\cal m}^{(j/\psi k^*)}_{t}\right]\bigg\}\ ; , \label{m1}\end{aligned}\ ] ] with the combination of the form factors , @xmath124 and the mass ratio @xmath125 . the first term in eq . ( [ m1 ] ) corresponds to the configuration with both the vector mesons being longitudinally polarized , and the second ( third ) term to the two configurations with both the vector mesons being transversely polarized in the parallel ( perpendicular ) directions . the effective wilson coefficient @xmath52 is the same as in eq . ( [ a ] ) . the @xmath126 transition form factors @xmath127 , @xmath128 , and @xmath38 are defined via the matrix elements , @xmath129 the spectator amplitudes @xmath130 from figs . [ fvs](e ) and [ fvs](f ) are written as @xmath131 where the explicit factorization formulas for the amplitudes @xmath132 and for @xmath133 are presented in the appendix . the helicity amplitudes are then defined as , @xmath134\ ; , \nonumber\\ a_{\parallel}&=&-g \sqrt{2}\left[r_2a_{\rm eff}(\mu)f_{j/\psi}a_1(m_{j/\psi}^2)+{\cal m}^{(j/\psi k^*)}_{n}\right]\ ; , \nonumber \\ r_2r_{k^ * } \sqrt{2[(v_2\cdot v_3)^{2}-1 ] } \left[r_2a_{\rm eff}(\mu)f_{j/\psi}v(m_{j/\psi}^2)+{\cal m}^{(j/\psi k^*)}_{t}\right]\ ; , \label{ase}\end{aligned}\ ] ] for the longitudinal , parallel , and perpendicular polarizations , respectively , with the velocities @xmath135 and @xmath136 . the normalization factor @xmath137 has been chosen such that the following relation is satisfied : @xmath138 ( @xmath139 ) between @xmath140 ( @xmath141 ) and @xmath142 takes the value of @xmath143 in the heavy - quark limit , ie . , as @xmath144 is real and @xmath130 vanish . we define the polarization fractions @xmath145 , @xmath146 , and @xmath147 . the @xmath78 polarizations , significantly affected by the nonfactorizable contributions , were not discussed in @xcite . to perform the numerical analysis , we need the information of the involved meson distribution amplitudes . the @xmath6 meson wave function @xmath86 and the @xmath27 meson distribution amplitudes have been studied intensively , whose expressions are collected in the appendix . the charmonium distribution amplitudes attract less attention in the literature . in @xcite they were assumed to be identical to the corresponding light - meson distribution amplitudes : those of the @xmath17 ( @xmath19 ) meson are the same as of the @xmath148 ( @xmath143 ) meson . then the twist-3 charmonium distribution amplitudes do not vanish at the end points @xmath149 , 1 , where the valence charm quarks become highly off - shell . such a functional shape seems to contradict to the intuition . more realistic asymptotic models have been proposed recently in @xcite , which were inferred from the non - relativistic heavy quarkonium bound - state wave functions with the same quantum numbers . we shall adopt these @xmath17 and @xmath19 meson distribution amplitudes , and derive the @xmath20 and @xmath18 meson distribution amplitudes following the similar procedure in the next section . the @xmath17 meson asymptotic distribution amplitudes are given by @xcite @xmath150^{0.7}\;,\nonumber\\ \psi^t(x)&=&10.94\frac{f_{j/\psi}}{2\sqrt{2n_c}}(1 - 2x)^2 \left[\frac{x(1-x)}{1 - 2.8x(1-x)}\right]^{0.7}\;,\nonumber\\ \psi^v(x)&=&1.67\frac{f_{j/\psi}}{2\sqrt{2n_c}}\left[1+(2x-1)^2\right ] \left[\frac{x(1-x)}{1 - 2.8x(1-x)}\right]^{0.7}\;,\label{jda}\end{aligned}\ ] ] in which the twist-3 ones @xmath151 vanish at the end points due to the additional factor @xmath152^{0.7}$ ] . compared to @xcite , we have distinguished the distribution amplitudes associated with the longitudinally and transversely polarized @xmath17 mesons , which exhibit the different asymptotic behaviors . we do not distinguish the decay constant @xmath46 for the normalization of @xmath69 and @xmath153 for the normalization of @xmath71 . the distribution amplitudes associated with the structures @xmath154 ( the identity ) and @xmath155 ( the pseudo - vector ) diminish like @xmath156 , @xmath157 being the @xmath73 quark mass . the above models have been shown to yield the observed cross section of charmonium production in @xmath24 collisions @xcite . if factorization theorem works , the same distribution amplitudes , due to their universality , should be able to explain the exclusive @xmath6 meson decays into charmonia . c|ccc ' '' '' & @xmath158 & @xmath159 & @xmath59 + @xmath48 & 0.35 & 1.58 & 0.68 + @xmath160 & 0.35 & 0.71 & 0.04 + @xmath161 & 0.31 & 1.79 & 1.18 + @xmath162 & 0.31 & 1.68 & 1.08 + @xmath163 & 0.26 & 0.93 & 0.19 + @xmath164 & 0.24 & 1.63 & 0.98 + c|cc ' '' '' @xmath165 & @xmath166[gev ] & @xmath167[mev ] + @xmath168 & 2.980 & @xmath169 + @xmath170 & 3.097 & 405 @xmath171 14 + @xmath172 & 3.415 & 360 + @xmath173 & 3.511 & 335 + for the @xmath30 transition form factors , we employ the models derived from the light - front qcd @xcite , which have been parameterized as @xmath174 with the constants @xmath158 , @xmath159 , and @xmath59 being listed in table [ tab:1 ] . using the inputs in table [ tab : proper ] @xcite for the various charmonium states , the ckm matrix elements @xmath175 and @xmath176 , the quark mass @xmath177 for the wilson coefficients , the lifetimes @xmath178 and @xmath179 , and the fermi constant @xmath180 , we derive the branching ratios , @xmath181 we have also derived the the polarization fractions and the relative phases among the helicity amplitudes for the @xmath182 decays , @xmath183 both the @xmath184 and @xmath185 modes have the same polarization fractions , since they differ only in the lifetimes in our formalism . the theoretical uncertainties come from the variation of the renormalization scale @xmath65 . we have checked the sensitivity of our results to the choice of the @xmath59 quark mass : @xmath186 gev increases the branching ratio @xmath187 only by 10% . the recent babar measurement gave @xcite @xmath188 and the belle measurement gave @xcite @xmath189 it is found that the consistency between the theoretical and experimental values for the @xmath25 branching ratios is reasonable . the predicted longitudinal polarization fraction @xmath190 for the @xmath78 is a bit larger , and the predicted relative phases are a bit smaller than the data . however , we point out that the polarization fractions and the relative phases could be modified by adding the higher gegenbauer terms to the @xmath17 meson distribution amplitudes , and that the consistency will be improved . this fine tuning will not be performed here . roughly speaking , it is very promising to understand the @xmath191 data in the pqcd framework . another remark is as follows . most of the @xmath17 meson distribution amplitudes , except @xmath71 , exhibit maxima at @xmath192 , such that the valence charm quark , carrying the invariant mass @xmath193 , is almost on - shell . however , @xmath71 has two humps with a dip at @xmath192 . we argue that the models in eq . ( [ jda ] ) are the asymptotic ones . including the higher gegenbauer terms , such as those for a light vector meson , @xmath194\right . \nonumber \\ & & \left.+a_2^{j/\psi}[3 - 30(2x-1)^2 + 35(2x-1)^4]\right\ } \left[\frac{x(1-x)}{1 - 2.8x(1-x)}\right]^{0.70}\;,\end{aligned}\ ] ] @xmath71 possess three humps with the major one located at @xmath192 . adopting this parametrization , the gegenbauer coefficients @xmath195 and @xmath196 should be determined from the data of charmonium production in @xmath24 annihilation , before it is applied to the exclusive @xmath6 meson decays into charmonia . before calculating the spectator contributions to other similar modes , we clarify a confusing statement in the literature . it has been claimed @xcite that the end - point singularities from @xmath117 cancel between the spectator diagrams figs . [ fvs](e ) and [ fvs](f ) in the qcdf formalism for the @xmath13 decays , but do not for the @xmath198 decays . our opinion is that the end - point singularities should appear in both modes at the power of @xmath199 , if no approximation is made . the singularities cancel in @xcite because of the relation , @xmath200 where the decay constant @xmath153 is associated with the normalization of the twist-3 distribution amplitude @xmath71 . this relation arises from the exact on - shell condition of the valence charm quarks . the singularities were present in @xcite , since the authors , considering only the leading - twist @xmath18 meson distribution amplitude , got no chance to apply eq . ( [ onsh ] ) . we agree on the comment made in @xcite : it is not clear why the singularities cancel in @xcite for the @xmath13 decays , viewing that they treated the ratio on the left - hand side of eq . ( [ onsh ] ) as a constant . the exact on - shell condition is indeed not necessary , because a deviation from mass shell , as long as being power - suppressed , is allowed . furthermore , the momentum fraction @xmath201 , running between 0 and 1 , is hardly related to the left - hand side . once eq . ( [ onsh ] ) is not postulated , the end - point singularities exist in the terms proportional to @xmath199 for the @xmath13 decays @xcite , and also for the @xmath202 decays . note that these singularities disappear in the pqcd approach based on @xmath1 factorization theorem . the pqcd analysis of the @xmath203 decays is similar to that of @xmath25 . to calculate the nonfactorizable spectator amplitudes , we consider the @xmath18 meson distribution amplitudes defined via the nonlocal matrix elements associated with the longitudinal and transverse polarizations , @xmath204_{lj}\chi_{1}^{l}(x)+[\gamma_5{\not{\!\epsilon}}_l{\not{\!p}}]_{lj } \chi_{1}^{t}(x ) \bigg\ } , \label{lpf1}\\ \langle \chi_{c1}(p,\epsilon_t)|\bar c(z)_jc(0)_l|0\rangle & = & \frac{1}{\sqrt{2n_c}}\int_0 ^ 1 dx e^{ixp\cdot z } \bigg\{m_{\chi_{c1}}[\gamma_5{\not{\!\epsilon}}_t]_{lj}\chi_1^v(x)+ [ \gamma_5{\not{\!\epsilon}}_t{\not{\!p}}]_{lj}\chi_1^t(x ) \bigg\ } , \label{spf1}\end{aligned}\ ] ] respectively , with the @xmath18 meson mass @xmath205 . the asymptotic models for the twist-2 distribution amplitudes @xmath206 and @xmath207 , and for the twist-3 distribution amplitudes @xmath208 and @xmath209 will be derived following the prescription in @xcite : they can be extracted from the @xmath210 , @xmath211 schrodinger states for a coulomb potential , whose radial dependence is given by @xmath212 @xmath213 being the bohr momentum . fourier transformation of the above solution leads to @xmath214 with @xmath215 . employing the substitution @xcite , @xmath216 @xmath217 ( @xmath218 ) being the @xmath73 ( @xmath219 ) quark momentum fraction , we obtain the heavy quarkonium distribution amplitude , @xmath220}{[1 - 4x\bar x(1-v^2)]^2}\right\}\;,\label{x1}\end{aligned}\ ] ] with @xmath221 . to impose a fractional power @xmath222 to the factor in the curved brackets @xcite , we neglect the @xmath223 term in the numerator . otherwise , the numerator may become negative for @xmath224 and the analyticity of the distribution amplitude will be lost . we then propose the @xmath18 meson distribution amplitudes inferred from eq . ( [ x1 ] ) , @xmath225 ^ 2}\right\}^{1-v^2}\;,\label{int}\end{aligned}\ ] ] with @xmath226 being set to the asymptotic models of the corresponding twists for light vector mesons . it is now clear that after imposing the fractional power @xmath222 , eq . ( [ int ] ) approaches the light meson distribution amplitudes as @xmath227 , and the heavy quarkonium distribution amplitudes the same as eq . ( [ x1 ] ) as @xmath228 @xcite . fixing the normalization , we derive @xmath229}{[1 - 2.8x(1-x)]^2}\right\}^{0.7}\;,\nonumber\\ \chi_1^t(x)&=&15.17\frac{f_{\chi_{c1}}}{2\sqrt{2n_c}}(1 - 2x)^2 \left\{\frac{x(1-x)[1 - 4x(1-x)]}{[1 - 2.8x(1-x)]^2}\right\}^{0.7}\;,\nonumber\\ \chi_1^v(x)&=&3.60\frac{f_{\chi_{c1}}}{2\sqrt{2n_c}}\left[1+(2x-1)^2\right ] \left\{\frac{x(1-x)[1 - 4x(1-x)]}{[1 - 2.8x(1-x)]^2}\right\}^{0.7}\;,\label{chda}\end{aligned}\ ] ] where the same decay constant @xmath230 has been assumed for the longitudinally and transversely polarized @xmath18 meson . note that the dip at @xmath192 is a consequence of the @xmath231-wave schrodinger wave functions . the @xmath203 decay amplitudes are written as @xmath232 , \nonumber\\ & & i\ , \epsilon^{\alpha \beta\gamma \rho } \epsilon^{*}_{2\alpha}\epsilon^{*}_{3\beta } \frac{p_{2\gamma } p_{3\rho } } { m_b^2}\ , \left[r_2f_{\chi_{c1}}v(m_{\chi_{c1}}^2)+{\cal m}^{(\chi_{c1 } k^*)}_{t}\right]\bigg\}\ ; , \label{m7}\end{aligned}\ ] ] with the mass ratio @xmath233 , and the effective wilson coefficient , @xmath234 similarly , we shall vary the renormalization scale @xmath65 in the range between @xmath84 and @xmath67 , which covers the effect of the vertex corrections . as in the case of @xmath13 , the infrared divergences in the vertex corrections cancel in the @xmath235 decays : the hard kernel of the vertex corrections is odd for the structure @xmath236 under the exchange of @xmath217 and @xmath237 , while the corresponding distribution amplitude is even . the hard kernel vanishes for the structure @xmath238 . express the spectator amplitudes @xmath239 and @xmath240 as @xmath241 where the factorization formulas of the amplitudes @xmath242 , @xmath243 , @xmath244 and @xmath245 are shown in the appendix . their expressions are similar to those of @xmath25 with some terms flipping signs . using the inputs listed in table [ tab : proper ] @xcite , we get @xmath246 the errors of the above predictions arise from the variation of the renormalization scale @xmath65 for the factorizable contribution . for the @xmath247 branching ratios , the imaginary part of the spectator amplitude @xmath248 dominates , such that the influence of the @xmath65 dependence becomes mild . this is also the reason we obtain a large @xmath249 . the @xmath235 branching ratios in eq . ( [ chi01 ] ) are consistent with the recent babar measurement @xcite , @xmath250 note that these branching ratios , with a parametrization for the logarithmical end - point singularities , were estimated to be only about @xmath251 in qcdf @xcite . our predictions for the @xmath247 decays , including the branching ratios , the polarization fractions ( especially the large @xmath249 ) , and the relative phases , can be compared with data in the future . we then discuss the @xmath252 decays , for which the nonfactorizable contributions dominate due to the absence of the factorizable ones . the nonlocal matrix element associated with the @xmath20 meson is decomposed into , @xmath253_{lj } \chi_0^{v}(x)+m_{\chi_{c0}}[i]_{lj}\chi_0^s(x ) \bigg\}\ ; , \label{lpf0}\end{aligned}\ ] ] which defines the twist-2 and twist-3 distribution amplitudes , @xmath254 and @xmath255 , respectively , @xmath256 being the @xmath20 meson mass . to satisfy the identity @xmath257 , @xmath254 must be anti - symmetric under the exchange of @xmath217 and @xmath237 . following the similar ansatz of constructing the @xmath18 meson distribution amplitudes , we propose the following asymptotic models , @xmath258}{[1 - 2.8x(1-x)]^2}\right\}^{0.7}\ ; , \nonumber\\ \chi_0^s(x)&=&4.73\frac{f_{\chi_{c0}}}{2\sqrt{2n_c } } \left\{\frac{x(1-x)[1 - 4x(1-x)]}{[1 - 2.8x(1-x)]^2}\right\}^{0.7}\ ; , \label{cda}\end{aligned}\ ] ] where the decay constant @xmath259 for the normalization of @xmath260 and @xmath261 has been assumed to be equal . in the qcdf analysis of the @xmath262 decays , the vertex corrections from figs . [ fvs](a)-[fvs](d ) are infrared safe . however , for the @xmath263 decays , the hard kernels of the vertex corrections are even for the structure @xmath154 , and odd for @xmath264 under the exchange of @xmath217 and @xmath237 . the corresponding @xmath20 meson distribution amplitudes are also even for the structure @xmath154 , and odd for @xmath264 . therefore , the infrared divergences do not cancel @xcite . to regularize these divergences , a binding energy @xmath265 and a gluon mass @xmath266 have been introduced in @xcite and @xcite , respectively . the vertex corrections to the @xmath267 decays can also be handled in the pqcd approach . for the reasons stated in the introduction , we do not attempt such a calculation here . instead , we shall demonstrate that the spectator contributions are sufficient to account for the observed @xmath252 branching ratios . the decay amplitudes are written as @xmath268 with the explicit factorization formulas of the amplitudes @xmath269 being referred to the appendix . there are also logarithmic end - point singularities in these amplitudes , if using qcdf , indicating that the pqcd approach based on @xmath1 factorization theorem is more appropriate . adopting the inputs in table [ tab : proper ] , we obtain the branching ratios , @xmath270 there is no theoretical uncertainty from the variation of @xmath65 , since the factorizable contributions vanish in this case . babar gave the upper bounds ( the values ) @xcite , @xmath271 and belle gave @xcite @xmath272 where the third error comes from a model uncertainty . hence , we conclude that the above data are understandable . the @xmath273 branching ratio was found to be as small as @xmath274 in @xcite ( with a large theoretical uncertainty from parameterizing the end - point singularities in qcdf ) and @xmath275 for the gluon mass @xmath276 gev in @xcite . despite of several discrepancies in their factorization formulas for the vertex corrections and for the spectator amplitudes , the difference in their numerical results is not really essential due to the tunable parameters . the analysis of the @xmath277 decays is similar to that of @xmath278 . the nonlocal matrix element associated with the @xmath20 meson is decomposed into @xmath279_{lj } \eta^{v}(x)+m_{\eta_{c}}[\gamma_5]_{lj}\eta^s(x ) \bigg\}\ ; , \label{epf0}\end{aligned}\ ] ] which defines the twist-2 and twist-3 @xmath280 meson distribution amplitudes , @xmath281 and @xmath282 , respectively , @xmath283 being the @xmath280 meson mass . the asymptotic models for the @xmath280 meson distribution amplitudes have been determined in @xcite : @xmath284^{0.7}\ ; , \nonumber\\ \eta^s(x)&=&1.97\frac{f_{\eta_{c}}}{2\sqrt{2n_c } } \left[\frac{x(1-x)}{1 - 2.8x(1-x)}\right]^{0.7}\ ; , \label{eda}\end{aligned}\ ] ] where we have assumed the same the decay constant @xmath285 for the normalization of @xmath286 and @xmath287 . the decay amplitudes are written as @xmath288 for which the factorization formulas of the spectator amplitudes , @xmath289 can be found in the appendix . the @xmath290 transition form factor @xmath291 and the @xmath126 transition form factor @xmath292 have been defined in eqs . ( [ fp ] ) and ( [ a012 ] ) , respectively . the effective wilson coefficient @xmath293 for the @xmath294 decays is also given by eq . ( [ b ] ) . similarly , we shall vary the renormalization scale @xmath65 in the range between @xmath84 and @xmath67 , which covers the effect of the vertex corrections . the hard kernels of the vertex corrections are even for the structure @xmath295 , and odd for @xmath296 under the exchange of @xmath217 and @xmath237 . the corresponding @xmath19 meson distribution amplitudes are both even for @xmath295 and @xmath296 . therefore , the infrared divergences in the vertex corrections to the @xmath297 decay , cancelling only for @xmath296 , still exist . using the inputs in table [ tab : proper ] @xcite and the form factor parametrization in eq . ( [ fpa ] ) , we obtain the branching ratios , @xmath298 since the @xmath19 meson distribution amplitudes have no two humps , the @xmath299 branching ratios are expected to be smaller than the @xmath13 ones . however , the belle measurement @xcite , @xmath300 and the babar measurement @xcite , @xmath301 are significantly larger than our predictions . the qcdf approach gave the @xmath2 branching ratios about @xmath302 for a rough estimate , which are also too small @xcite . because the @xmath19 meson distribution amplitudes in eq . ( [ eda ] ) have been shown to produce the observed cross section for charmonium production in @xmath24 annihilation @xcite , they are supposed to explain the branching ratios of the exclusive @xmath6 meson decays into charmonia according to the universality , even though these two processes involve dramatically different dynamics . therefore , we conclude that the @xmath2 decays are the only puzzle , and demand more studies . the exclusive @xmath6 meson decays into charmonia , @xmath303 , for @xmath304 , and @xmath19 have been analyzed in the qcdf approach in the literature . among these modes , those with @xmath305 were claimed to be explainable , and those with @xmath306 , and @xmath19 were not . however , this conclusion is not solid due to the serious end - point singularities in many qcdf decay amplitudes for the vertex corrections and for the spectator contributions , and due to the ad - hoc models of the charmonium distribution amplitudes . in this paper we have investigated these decays in an improved framework : the factorizable contributions are still treated in fa , since the involved @xmath30 transition form factors , evaluated at the charmonium mass , may not be calculated perturbatively ; the effect of the vertex corrections was taken into account by varying the renormalization scale @xmath65 , because their estimation in the pqcd approach needs additional nonperturbative information , i.e. , the distribution of the charm quark in its transverse degrees of freedom inside a charmonium ; the spectator contributions were computed in pqcd , in which the end - point singularities are smeared by the sudakov factor associated with the @xmath27 meson . the @xmath17 and @xmath19 meson distribution amplitudes have been inferred from the @xmath307 , @xmath308 schrodinger state for a coulomb potential @xcite , which produce the measured cross section of charmonium production in @xmath24 collisions . to analyze the @xmath309 decays , we have obtained the @xmath310 meson distribution amplitudes from the @xmath210 , @xmath211 schrodinger states following the similar procedure . that is , our models for the charmonium distribution amplitudes have been constrained theoretically and experimentally . our investigation has indicated that only the @xmath299 decays exhibit a puzzle , whose branching ratios are significantly smaller than the observed values . the data of the @xmath311 , @xmath312 decays , including the branching ratios , the polarization fractions , and the relative phases among the various helicity amplitudes , are all understandable within theoretical uncertainty . if qcd factorization theorem works , the @xmath19 meson distribution amplitudes , explaining the charmonium production data , are supposed to explain the exclusive @xmath6 meson decays into charmonia . hence , the @xmath313 modes require a more thorough study . we have predicted the branching ratios , the polarization fractions , and the relative phases associated with the @xmath314 decays , which can be compared with future measurements . 1.0 cm we thank k.t . chao and h.y . cheng for useful discussions . this work was supported by the national science council of r.o.c . under grant nsc-93 - 2112-m-001 - 014 . for the @xmath6 meson wave function , we adopt the model @xcite , @xmath315 \label{bw } \;,\end{aligned}\ ] ] with the shape parameter @xmath316 gev . the normalization constant @xmath317 gev is related to the decay constant @xmath318 mev ( in the convention @xmath319 mev ) . the @xmath108 and @xmath120 meson distribution amplitudes have been derived from qcd sum rules @xcite , @xmath320\right\}\ ; , \label{kk}\\ \phi _ { k}^{p}(x ) & = & \frac{f_k}{2\sqrt{2n_{c}}}\left [ 1 + 0.24c_{2}^{1/2}(1 - 2x)-0.11c_{4}^{1/2}(1 - 2x)\right ] \ ; , \label{kp}\\ \phi _ { k}^{\sigma } ( x ) & = & \frac{f_k}{2\sqrt{2n_{c}}}(1 - 2x)\left [ 1 + 0.35(10x^{2}-10x+1)\right ] \ ; , \label{ks } \\ \phi_{k^ * } ( x ) & = & \frac{3f_{k^*}}{\sqrt{2n_{c}}}x(1-x ) \big[1 + 0.57(1 - 2x)+0.07c^{3/2}_2(1 - 2x)\big]\ ; , \label{pk2}\\ \phi_{k^*}^{t } ( x ) & = & \frac{f_{k^*}^t}{2\sqrt{2n_{c } } } \bigg\ { 0.3(1 - 2x)\left[3(1 - 2x)^2 + 10(1 - 2x)-1\right]+1.68c^{1/2}_4(1 - 2x ) \nonumber \\ & & + 0.06(1 - 2x)^2\left[5(1 - 2x)^2 - 3\right ] + 0.36\left [ 1 - 2(1 - 2x)(1+\ln(1-x))\right ] \bigg\ } \ ; , \label{pk3t}\\ \phi _ { k^*}^s ( x ) & = & \frac{f_{k^*}^t}{2\sqrt{2n_{c } } } \bigg\ { 3(1 - 2x)\left[1 + 0.2(1 - 2x)+0.6(10x^2 - 10x+1)\right ] \nonumber \\ & & -0.12x(1-x)+0.36[1 - 6x-2\ln(1-x)]\bigg\ } \ ; , \label{pk3s}\\ \phi_{k^*}^{t}(x)&=&\frac{3f_{k^*}^t}{\sqrt{2n_c } } x(1-x ) \big[1 + 0.6(1 - 2x)+0.04c^{3/2}_2(1 - 2x)\big]\ ; , \label{pkt}\\ \phi_{k^*}^v(x)&=&\frac{f_{k^*}}{2\sqrt{2n_c } } \bigg\{\frac{3}{4}\big[1+(1 - 2x)^2 + 0.44(1 - 2x)^3\big ] + 0.4c^{1/2}_2(1 - 2x ) \nonumber \\ & & + 0.88c^{1/2}_4(1 - 2x)+0.48[2x+\ln(1-x ) ] \bigg\}\ ; , \label{pkv}\\ \phi_{k^*}^a(x ) & = & \frac{f_{k^*}}{4\sqrt{2n_{c } } } \big\{3(1 - 2x)\big[1 + 0.19(1 - 2x)+0.81(10x^2 - 10x+1)\big]\nonumber \\ & & -1.14x(1-x)+0.48[1 - 6x-2\ln(1-x)]\big\}\ ; , \label{pka}\end{aligned}\ ] ] with the gegenbauer polynomials @xmath321 and the decay constants @xmath322 mev , @xmath323 mev , and @xmath324 mev . the coefficients of the gegenbauer polynomials correspond to the masses @xmath325 gev and @xmath326 gev . we adopt the @xmath120 meson mass @xmath327 gev . we present the spectator amplitudes for the @xmath329 decays below , where the symbol @xmath330 represents the charmonium @xmath17 or @xmath18 : @xmath331\int_{0}^{\infty } b_{1}db_{1 } \phi _ { b}(x_{1},b_{1 } ) \nonumber \\ & & \times \big\ { \big [ ( 1 - 2r^{2}_{2})(1-x_{2 } ) \phi _ { k } ( x_{3 } ) x^{l}(x_{2 } ) + \zeta_{y } \frac{1}{2 } r^{2}_{2 } \phi_{k}(x_{3})x^{t } ( x_{2 } ) \nonumber \\ & & - r_{k } ( 1-r^2_2)x_3 \phi^{p}_{k}(x_3 ) x^{l}(x_{2})+ r_{k } \left ( 2r^{2}_{2}(1-x_2)+(1-r^2_2)x_3 \right ) \phi^{\sigma}_{k}(x_3 ) x^{l}(x_{2 } ) \big]\nonumber \\ & & \times e_{4}(t_d^{(1)})h_d^{(1)}(x_1,x_2,x_3,b_1 ) \nonumber \\ & & - \big [ ( x_2+(1 - 2r^2_{2})x_{3})\phi _ { k } ( x_{3 } ) x^{l}(x_{2})+r^{2}_{2}(2r_{k}\phi^{\sigma}_{k}(x_{3})-\frac{1}{2}\phi_{k}(x_{3}))x^{t}(x_{2 } ) \nonumber \\ & & -r_{k } ( 1-r^2_2)x_3 \phi^{p}_{k}(x_3 ) x^{l}(x_{2})- r_{k } \left ( 2r^{2}_{2}x_2+(1-r^2_2)x_3 \right ) \phi^{\sigma}_{k}(x_3 ) x^{l}(x_{2})\big]\nonumber \\ & & \times e_{4}(t^{(2)}_d ) h_d^{(2)}(x_1,x_2,x_3,b_1 ) \big\}\;,\end{aligned}\ ] ] @xmath332\int_{0}^{\infty } b_{1}db_{1 } \phi _ { b}(x_{1},b_{1 } ) \nonumber \\ & & \times \big\ { \big [ ( 1-x_2+(1 - 2r^2_{2})x_{3})\phi _ { k } ( x_{3 } ) x^{l}(x_{2})+ \zeta_{x } r^{2}_{2}(2r_{k}\phi^{\sigma}_{k}(x_{3})-\frac{1}{2}\phi_{k}(x_{3}))x^{t}(x_{2 } ) \nonumber \\ & & -r_{k } ( 1-r^2_2)x_3 \phi^{p}_{k}(x_3 ) x^{l}(x_{2})- r_{k } \left ( 2r^{2}_{2}(1-x_2)+(1-r^2_2)x_3 \right ) \phi^{\sigma}_{k}(x_3 ) x^{l}(x_{2})\big]\nonumber \\ & & \times e_{6}(t_d^{(1)})h_d^{(1)}(x_1,x_2,x_3,b_1 ) \nonumber \\ & & - \big [ ( 1 - 2r^{2}_{2})x_{2 } \phi _ { k } ( x_{3 } ) x^{l}(x_{2 } ) + \frac{1}{2 } r^{2}_{2 } \phi_{k}(x_{3})x^{t } ( x_{2 } ) \nonumber \\ & & - r_{k } ( 1-r^2_2)x_3 \phi^{p}_{k}(x_3 ) x^{l}(x_{2})+ r_{k } \left ( 2r^{2}_{2}x_2+(1-r^2_2)x_3 \right ) \phi^{\sigma}_{k}(x_3 ) x^{l}(x_{2 } ) \big]\nonumber \\ & & \times e_{6}(t^{(2)}_d ) h_d^{(2)}(x_1,x_2,x_3,b_1 ) \big\}\;,\end{aligned}\ ] ] @xmath333\int_{0}^{\infty } b_{1}db_{1 } \phi _ { b}(x_{1},b_{1 } ) \nonumber \\ & & \times \big\ { \big [ ( 1 - 2r^{2}_{2})(1-x_{2 } ) \phi _ { k^ * } ( x_{3 } ) x^{l}(x_{2 } ) + \zeta_{x } \frac{1}{2 } r^{2}_{2 } \phi_{k^*}(x_{3})x^{t } ( x_{2 } ) \nonumber \\ & & - r_{k^ * } ( 1-r^2_2)x_3 \phi^{s}_{k^*}(x_3 ) x^{l}(x_{2})+ r_{k^ * } \left ( 2r^{2}_{2}(1-x_2)+(1 - 2r^2_2)x_3 \right ) \phi^{t}_{k^*}(x_3 ) x^{l}(x_{2 } ) \big]\nonumber \\ & & \times e_{4}(t_d^{(1)})h_d^{(1)}(x_1,x_2,x_3,b_1 ) \nonumber \\ & & - \big [ ( x_2+(1 - 2r^2_{2})x_{3})\phi _ { k^ * } ( x_{3 } ) x^{l}(x_{2})+r^{2}_{2}(2r_{k^*}\phi^{t}_{k^*}(x_{3})-\frac{1}{2}\phi_{k^*}(x_{3}))x^{t}(x_{2 } ) \nonumber \\ & & -r_{k^ * } ( 1-r^2_2)x_3 \phi^{s}_{k^*}(x_3 ) x^{l}(x_{2})- r_{k^ * } \left ( 2r^{2}_{2}x_2+(1 - 2r^2_2)x_3 \right ) \phi^{t}_{k^*}(x_3 ) x^{l}(x_{2})\big]\nonumber \\ & & \times e_{4}(t^{(2)}_d ) h_d^{(2)}(x_1,x_2,x_3,b_1 ) \big\}\;,\end{aligned}\ ] ] @xmath334\int_{0}^{\infty } b_{1}db_{1 } \phi _ { b}(x_{1},b_{1 } ) \nonumber \\ & & \times \big\ { \big [ ( 1-x_2+(1 - 2r^2_{2})x_{3})\phi _ { k^ * } ( x_{3})x^{l}(x_{2})+ \zeta_{x } r^{2}_{2}(2r_{k^*}\phi^{t}_{k^*}(x_{3})-\frac{1}{2}\phi_{k^*}(x_{3}))x^{t}(x_{2 } ) \nonumber \\ & & - r_{k^ * } ( 1-r^2_2)x_3 \phi^{s}_{k^*}(x_3 ) x^{l}(x_{2})- r_{k^ * } \left ( 2r^{2}_{2}(1-x_2)+(1 - 2r^2_2)x_3 \right ) \phi^{t}_{k^*}(x_3 ) x^{l}(x_{2})\big ] \nonumber \\ & & \times e_{6}(t_d^{(1)})h_d^{(1)}(x_1,x_2,x_3,b_1 ) \nonumber \\ & & - \big [ ( 1 - 2r^{2}_{2})x_{2 } \phi _ { k^ * } ( x_{3 } ) x^{l}(x_{2 } ) + \frac{1}{2 } r^{2}_{2 } \phi_{k^*}(x_{3})x^{t } ( x_{2 } ) \nonumber \\ & & - r_{k^ * } ( 1-r^2_2)x_3 \phi^{s}_{k^*}(x_3 ) x^{l}(x_{2})+ r_{k^ * } \left ( 2r^{2}_{2}x_2+(1 - 2r^2_2)x_3 \right ) \phi^{t}_{k^*}(x_3 ) x^{l}(x_{2 } ) \big]\nonumber \\ & & \times e_{6}(t^{(2)}_d ) h_d^{(2)}(x_1,x_2,x_3,b_1 ) \big\}\;,\end{aligned}\ ] ] @xmath335\int_{0}^{\infty } b_{1}db_{1 } \phi _ { b}(x_{1},b_{1 } ) \nonumber \\ & & \times r_{2}\big\ { \big[(1-r^2_{2})(1-x_{2})\psi^{t}_{k^*}(x_3)x^{v}(x_2 ) \nonumber \\ & & + \zeta_{x } \frac{1}{2}r_{k^ * } ( ( 1+r^{2}_{2 } ) \phi^{v}_{k^*}(x_{3 } ) -(1-r^{2}_{2})\psi^{a}_{k^*}(x_{3 } ) ) x^{t}(x_{2})\big ] e_{4}(t^{(1)}_d)h_d^{(1)}(x_1,x_2,x_3,b_1 ) \nonumber \\ & & + \big[(1-r^2_{2})x_{2}\psi^{t}_{k^*}(x_3)x^{v}(x_2 ) \nonumber \\ & & -\left((1-r^2_{2})\phi^{t}_{k^*}(x_3)- \frac{1}{2}r_{k^ * } ( ( 1+r^{2}_{2 } ) \phi^{v}_{k^*}(x_{3})+(1-r^{2}_{2})\psi^{a}_{k^*}(x_{3 } ) \right ) x^{t}(x_{2 } ) \nonumber \\ & & -2r_{k^*}((1+r^{2}_{2})x_{2}+(1-r^{2}_{2})x_{3 } ) \phi^{v}_{k^*}(x_{3 } ) x^{v}(x_2 ) \big ] e_{4}(t^{(2)}_d ) h_d^{(2)}(x_1,x_2,x_3,b_1 ) \big\}\;,\end{aligned}\ ] ] @xmath336\int_{0}^{\infty } b_{1}db_{1 } \phi _ { b}(x_{1},b_{1 } ) \nonumber \\ & & \times r_{2}\big\ { \big[(1-r^2_{2})(1-x_{2})\psi^{t}_{k^*}(x_3)x^{v}(x_2 ) \nonumber \\ & & -\zeta_{x } \left((1-r^2_{2})\phi^{t}_{k^*}(x_3)- \frac{1}{2}r_{k^ * } ( ( 1+r^{2}_{2 } ) \phi^{v}_{k^*}(x_{3})+(1-r^{2}_{2})\psi^{a}_{k^*}(x_{3 } ) \right ) x^{t}(x_{2 } ) \nonumber \\ & & -2r_{k^*}((1+r^{2}_{2})(1-x_{2})+(1-r^{2}_{2})x_{3 } ) \phi^{v}_{k^*}(x_{3 } ) x^{v}(x_2 ) \big ] e_{6}(t^{(1)}_d)h_d^{(1)}(x_1,x_2,x_3,b_1 ) \nonumber \\ & & + \big[(1-r^2_{2})x_{2}\psi^{t}_{k^*}(x_3)x^{v}(x_2 ) \nonumber \\ & & + \frac{1}{2}r_{k^ * } ( ( 1+r^{2}_{2 } ) \phi^{v}_{k^*}(x_{3})-(1-r^{2}_{2})\psi^{a}_{k^*}(x_{3 } ) ) x^{t}(x_{2})\big ] e_{6}(t^{(2)}_d ) h_d^{(2)}(x_1,x_2,x_3,b_1 ) \big\}\;,\end{aligned}\ ] ] @xmath337\int_{0}^{\infty } b_{1}db_{1 } \phi _ { b}(x_{1},b_{1 } ) \nonumber \\ & & \times r_{2}\big\ { \big [ ( 1-x_2)\phi^{t}_{k^*}(x_3)x^{v}(x_2 ) -\zeta_{x } \frac{1}{2 } r_{k^ * } \left(\phi^{v}_{k^*}(x_3)-(1 + 2r^{2}_{2})\psi^{a}_{k^*}(x_3 ) \right ) x^{t}(x_{2 } ) \big ] \nonumber \\ & & \times e_{4}(t^{(1)}_d)h_d^{(1)}(x_1,x_2,x_3,b_1 ) \nonumber \\ & & + \big [ x_2 \phi^{t}_{k^*}(x_3)x^{v}(x_2 ) - \left(\phi^{t}_{k^*}(x_3)-\frac{1}{2 } r_{k^ * } ( \phi^{v}_{k^*}(x_3)+(1 + 2r^{2}_{2})\psi^{a}_{k^*}(x_3 ) ) \right ) x^{t}(x_{2 } ) \nonumber \\ & & -2r_{k^ * } ( ( 1 + 2r^{2}_{2})x_2 + x_{3})\phi^{a}_{k^*}(x_3 ) x^{v}(x_2 ) \big]e_{4}(t^{(2)}_d ) h_d^{(2)}(x_1,x_2,x_3,b_1 ) \big\}\;,\end{aligned}\ ] ] @xmath338\int_{0}^{\infty } b_{1}db_{1 } \phi _ { b}(x_{1},b_{1 } ) \nonumber \\ & & \times r_{2}\big\ { \big [ ( 1-x_2 ) \phi^{t}_{k^*}(x_3)x^{v}(x_2 ) \nonumber \\ & & -\zeta_{x } \left(\phi^{t}_{k^*}(x_3)-\frac{1}{2 } r_{k^ * } ( \phi^{v}_{k^*}(x_3)+(1 + 2r^{2}_{2})\psi^{a}_{k^*}(x_3 ) ) \right ) x^{t}(x_{2 } ) \nonumber \\ & & -2r_{k^ * } ( ( 1 + 2r^{2}_{2})(1-x_2 ) + x_{3})\phi^{a}_{k^*}(x_3 ) x^{v}(x_2 ) \big ] e_{6}(t^{(1)}_d)h_d^{(1)}(x_1,x_2,x_3,b_1 ) \nonumber \\ & & + \big [ x_2\phi^{t}_{k^*}(x_3)x^{v}(x_2 ) -\frac{1}{2 } r_{k^ * } \left(\phi^{v}_{k^*}(x_3)-(1 + 2r^{2}_{2})\psi^{a}_{k^*}(x_3 ) \right ) x^{t}(x_{2 } ) \big ] \nonumber\\ & & \times e_{6}(t^{(2)}_d ) h_d^{(2)}(x_1,x_2,x_3,b_1 ) \big\}\;,\end{aligned}\ ] ] we present the spectator amplitudes for the @xmath342 decays below , where the symbol @xmath330 represents the charmonium @xmath20 or @xmath19 : @xmath343\int_{0}^{\infty } b_{1}db_{1 } \phi _ { b}(x_{1},b_{1 } ) \nonumber \\ & & \times \big\ { \big [ ( 1-x_{2 } ) \phi _ { k } ( x_{3 } ) x^v(x_{2 } ) -\zeta_{x } r^{2}_{2}(2r_{k}\phi^{p}_{k}(x_3)-\frac{1}{2 } \phi_{k}(x_{3}))x^{s}(x_{2 } ) \nonumber \\ & & + r_{k } ( 1-r^2_2)x_3 \phi^{\sigma}_{k}(x_3 ) x^{v}(x_{2})- r_{k } \left ( 2r^{2}_{2}(1-x_2)+(1-r^2_2)x_3 \right ) \phi^{p}_{k}(x_3 ) x^{v}(x_{2 } ) \big]\nonumber \\ & & \times e_{4}(t_d^{(1)})h_d^{(1)}(x_1,x_2,x_3,b_1 ) \nonumber \\ & & - \big [ ( x_2+(1 - 2r^2_{2})x_{3})\phi _ { k } ( x_{3 } ) x^{v}(x_{2})+r^{2}_{2}(2r_{k}\phi^{p}_{k}(x_{3})-\frac{1}{2}\phi_{k}(x_{3}))x^{s}(x_{2 } ) \nonumber \\ & & -r_{k } ( 1-r^2_2)x_3 \phi^{\sigma}_{k}(x_3 ) x^{v}(x_{2})- r_{k } \left ( 2r^{2}_{2}x_2+(1-r^2_2)x_3 \right ) \phi^{p}_{k}(x_3 ) x^{v}(x_{2})\big]\nonumber \\ & & \times e_{4}(t^{(2)}_d ) h_d^{(2)}(x_1,x_2,x_3,b_1 ) \big\}\;,\end{aligned}\ ] ] @xmath344\int_{0}^{\infty } b_{1}db_{1 } \phi _ { b}(x_{1},b_{1 } ) \nonumber \\ & & \times \big\ { \big [ ( 1-x_2+(1 - 2r^2_{2})x_{3})\phi _ { k } ( x_{3 } ) x^{v}(x_{2})- \zeta_{x } r^{2}_{2}(2r_{k}\phi^{p}_{k}(x_{3})-\frac{1}{2}\phi_{k}(x_{3}))x^{s}(x_{2 } ) \nonumber \\ & & -r_{k } ( 1-r^2_2)x_3 \phi^{\sigma}_{k}(x_3 ) x^{v}(x_{2})- r_{k } \left ( 2r^{2}_{2}(1-x_2)+(1-r^2_2)x_3 \right ) \phi^{p}_{k}(x_3 ) x^{v}(x_{2})\big]\nonumber \\ & & \times e_{6}(t_d^{(1)})h_d^{(1)}(x_1,x_2,x_3,b_1 ) \nonumber \\ & & -\big [ x_{2 } \phi _ { k } ( x_{3 } ) x^{v}(x_{2 } ) + r^{2}_{2}(2r_{k}\phi^{p}_{k}(x_3)-\frac{1}{2 } \phi_{k}(x_{3}))x^{s}(x_{2 } ) \nonumber \\ & & + r_{k } ( 1-r^2_2)x_3 \phi^{\sigma}_{k}(x_3 ) x^{v}(x_{2})- r_{k } \left ( 2r^{2}_{2}x_2+(1-r^2_2)x_3 \right ) \phi^{p}_{k}(x_3 ) x^{v}(x_{2 } ) \big]\nonumber \\ & & \times e_{6}(t^{(2)}_d ) h_d^{(2)}(x_1,x_2,x_3,b_1 ) \big\}\;,\end{aligned}\ ] ] @xmath345\int_{0}^{\infty } b_{1}db_{1 } \phi _ { b}(x_{1},b_{1 } ) \nonumber \\ & & \times \big\ { \big [ ( 1-x_{2 } ) \phi _ { k^ * } ( x_{3 } ) x^{v}(x_{2 } ) -\zeta_{x } r^{2}_{2}(2r_{k^*}\phi^{s}_{k^*}(x_3)-\frac{1}{2 } \phi_{k^*}(x_{3}))x^{s}(x_{2 } ) \nonumber \\ & & + r_{k^ * } ( 1-r^2_2)x_3 \phi^{t}_{k^*}(x_3 ) x^{v}(x_{2})- r_{k^ * } \left ( 2r^{2}_{2}(1-x_2)+(1-r^2_2)x_3 \right ) \phi^{s}_{k^*}(x_3 ) x^{v}(x_{2 } ) \big]\nonumber \\ & & \times e_{4}(t_d^{(1)})h_d^{(1)}(x_1,x_2,x_3,b_1 ) \nonumber \\ & & - \big [ ( x_2+(1 - 2r^2_{2})x_{3})\phi _ { k^ * } ( x_{3 } ) x^{v}(x_{2})+r^{2}_{2}(2r_{k^*}\phi^{s}_{k}(x_{3 } ) -\frac{1}{2}\phi_{k^*}(x_{3}))x^{s}(x_{2 } ) \nonumber \\ & & -r_{k^ * } ( 1-r^2_2)x_3 \phi^{t}_{k^*}(x_3 ) x^{v}(x_{2})- r_{k^ * } \left ( 2r^{2}_{2}x_2+(1-r^2_2)x_3 \right ) \phi^{s}_{k^*}(x_3 ) x^{v}(x_{2})\big]\nonumber \\ & & \times e_{4}(t^{(2)}_d ) h_d^{(2)}(x_1,x_2,x_3,b_1 ) \big\}\;,\end{aligned}\ ] ] @xmath346\int_{0}^{\infty } b_{1}db_{1 } \phi _ { b}(x_{1},b_{1 } ) \nonumber \\ & & \times \big\ { \big [ ( 1-x_2+(1 - 2r^2_{2})x_{3})\phi _ { k^ * } ( x_{3 } ) x^{v}(x_{2})-\zeta_{x}r^{2}_{2}(2r_{k}\phi^{s}_{k^*}(x_{3})-\frac{1}{2}\phi_{k^*}(x_{3}))x^{s}(x_{2 } ) \nonumber \\ & & - r_{k^ * } ( 1-r^2_2)x_3 \phi^{t}_{k^*}(x_3 ) x^{v}(x_{2})- r_{k^ * } \left ( 2r^{2}_{2}(1-x_2)+(1-r^2_2)x_3 \right ) \phi^{s}_{k^*}(x_3 ) x^{v}(x_{2})\big]\nonumber \\ & & \times e_{6}(t_d^{(1)})h_d^{(1)}(x_1,x_2,x_3,b_1 ) \nonumber \\ & & -\big [ x_{2 } \phi _ { k^ * } ( x_{3 } ) x^{v}(x_{2 } ) + r^{2}_{2}(2r_{k^*}\phi^{s}_{k^*}(x_3)-\frac{1}{2 } \phi_{k^*}(x_{3}))x^{s}(x_{2 } ) \nonumber \\ & & + r_{k^ * } ( 1-r^2_2)x_3 \phi^{t}_{k^*}(x_3 ) x^{v}(x_{2})- r_{k^ * } \left ( 2r^{2}_{2}x_2+(1-r^2_2)x_3 \right ) \phi^{s}_{k^*}(x_3 ) x^{v}(x_{2 } ) \big]\nonumber \\ & & \times e_{6}(t^{(2)}_d ) h_d^{(2)}(x_1,x_2,x_3,b_1 ) \big\}\;,\end{aligned}\ ] ] m. bauer , b. stech , m. wirbel , z. phys . c * 29 * , 637 ( 1985 ) ; _ ibid . _ * 34 * , 103 ( 1987 ) . a. ali , g. kramer and c.d . l , phys . d * 58 * , 094009 ( 1998 ) ; y.h . chen , h.y . cheng , b. tseng , and k.c . yang , phys . d * 60 * , 094014 ( 1999 ) . cheng and k.c . yang , phys . d * 59 * , 092004 ( 1999 ) . m. neubert and a.a . petrov , phys . b * 519 * , 50 ( 2001 ) . babar collaboration , b. aubert _ et al . _ , hep - ex/0412062 . li , prog . part . phys . * 51 * , 85 ( 2003 ) ; czech . * 53 * , 657 ( 2003 ) . j. chay and c. kim , hep - ph/0009244 . m. beneke , g. buchalla , m. neubert , and c.t . sachrajda , phys . lett . * 83 * , 1914 ( 1999 ) ; nucl . * b591 * , 313 ( 2000 ) . cheng and k.c . yang , phys . d * 63 * , 074011 ( 2001 ) . z.z . song , c. meng , and k.t . chao , phys . b * 568 * , 127 ( 2003 ) ; z.z . song , c. meng , y.j . gao , and k.t . chao , phys . d * 69 * , 054009 ( 2004 ) . z.z . song , c. meng , and k.t . chao , eur . j. c * 36 * , 365 ( 2004 ) . b. melic , lect . notes phys . * 647 * , 287 ( 2004 ) ; phys . d * 68 * , 034004 ( 2003 ) ; b. melic and r. rckl , hep - ph/0210353 . wang , l. li , and t. huang , phys . d * 70 * , 074006 ( 2004 ) . li and h.l . yu , phys . lett . * 74 * , 4388 ( 1995 ) ; phys . b * 353 * , 301 ( 1995 ) ; phys . d * 53 * , 2480 ( 1996 ) . c.h . chang and h - n . li , phys . d * 55 * , 5577 ( 1997 ) . keum , h - n . li , and a.i . sanda , phys . b * 504 * , 6 ( 2001 ) ; phys . d * 63 * , 054008 ( 2001 ) ; y.y . keum and h - n . li , phys . rev . * d63 * , 074006 ( 2001 ) . c. d. l , k. ukai , and m. z. yang , phys . d * 63 * , 074009 ( 2001 ) . keum , t. kurimoto , h .- lu , and a.i . sanda , phys . d * 69 * , 094018 ( 2004 ) . a.e . bondar and v.l . chernyak , hep - ph/0412335 . t.w . yeh and h - n . li , phys . d * 56 * , 1615 ( 1997 ) . sanda , n. sinha , r. sinha , and k. ukai , in _ proceedings for the 4th international workshop on @xmath6 physics and cp violation _ , ed . t. ohshima and a.i . sanda ( world scientific , singapore , 2001 ) . chen , phys . d * 67 * , 094011 ( 2003 ) . p. ball , jhep * 01 * , 010 ( 1999 ) . p. ball , v.m . braun , y. koike , and k. tanaka , nucl . b529 * , 323 ( 1998 ) . quarkonium working group , n. brambilla _ et al . _ , hep - ph/0412158 . m. nagashima and h - n . li , phys . d * 67 * , 034001 ( 2003 ) . chen , y.y . keum , and h - n . li , phys . d * 64 * , 112002 ( 2001 ) ; phys . d * 66 * , 054013 ( 2002 ) . t. kurimoto , h - n . li , and a.i . sanda , phys . d * 65 * , 014007 , ( 2002 ) ; phys . d * 67 * , 054028 ( 2003 ) . li and h.s . liao , phys . d * 70 * , 074030 ( 2004 ) . collins and d.e . soper , nucl . phys . * b193 * , 381 ( 1981 ) . j. botts and g. sterman , nucl . * b325 * , 62 ( 1989 ) . li and g. sterman , nucl . phys . * b381 * , 129 ( 1992 ) . li and s. mishima , hep - ph/0411146 , to appear in phys . ma and z.g . si , phys . d * 70 * , 074007 ( 2004 ) . particle data group collaboration , s. eidelman _ et al . _ , b * 592 * , 1 ( 2004 ) . d.s . hwang and g.h . kim , z. phys . c * 76 * , 107 ( 1997 ) . novikov , l.b . okun , m.a . shifman , a.i . vainshtein , m.b . voloshin and v.i . zakharov , phys . rep . * 41 * , 1 ( 1978 ) . cheng , c.k . chua , and c.w . hwang , phys . d * 69 * , 074025 ( 2004 ) and references therein ; h.y . cheng , hep - ph/0410316 . b. melic , phys . b * 591 * , 91 ( 2004 ) . belle collaboration , k. abe _ et al . _ , hep - ex/0408104 . pham and g. zhu , hep - ph/0412428 . c. meng , y.j . gao , and k.t . chao , hep - ph/0502240 . babar collaboration , b. aubert _ et al . _ , hep - ex/0501061 . belle collaboration , a. garmash _ et al . _ , hep - ex/0412066 . belle collaboration , f. fang _ et al . _ , lett . * 90 * , 071801 ( 2003 ) . babar collaboration , b. aubert _ et al . d * 70 * , 011101 ( 2004 ) .

modal logics have proven to be a valuable tool in mathematics and computer science . the traditional uni - modal logic enriches the propositional language with the operator @xmath2 where @xmath3 is interpreted as _ @xmath4 possibly holds_. the usual semantics interpret modal formulas over graphs , where @xmath3 means `` there is a successor world where @xmath4 is true . '' in addition to their mathematical interest , modal logics are widely used in practical applications : in artificial intelligence , modal logic is used to model the knowledge and beliefs of an agent , see e.g. @xcite . modal logics also can be applied in cryptographic and other protocols @xcite . for many specific applications , there exist tailor - made variants of modal logics @xcite . due to the vast number of applications , complexity issues for modal logics are very relevant , and have been examined since ladner s seminal work @xcite . depending on the application , modal logics with different properties are studied . for example , one might want the formula @xmath5 to be an axiom if something is true , then it should be considered possible . or @xmath6if it is possible that @xmath4 is possible , then @xmath4 itself should be possible . classical results @xcite show that there is a close correspondence between modal logics defined by axioms and logics obtained by restricting the class of considered graphs . requiring the axioms mentioned above corresponds to restricting the classes of graphs to those which are reflexive or transitive , respectively . determining the complexity of a given modal logic , defined either by the class of considered graphs or via a modal axiom system , has been an active line of research since ladner s results . in particular , the complexity classes @xmath0 and @xmath1 have been at the center of attention . most complexity results have been on a case - by - case basis , proving results for individual logics both for standard modal logics and variations like temporal or hybrid logics @xcite . examples of more general results include halpern and rgo s proof that logics including the _ negative introspection _ axiom , which corresponds to the euclidean graph property , have an @xmath0-complete satisfiability problem @xcite . in @xcite , schrder and pattinson show a way to prove @xmath1 upper bounds for modal logics defined by modal axioms of modal depth 1 . in @xcite , ladner proved @xmath1-hardness for all logics for which reflexive and transitive graphs are admissible models . in @xcite , hemaspaandra showed that all normal logics extending s4.3 have an @xmath0-complete satisfiability problem , and work on the _ guarded fragment _ has shown that some classes of modal logics can be seen as a decidable fragment of first - order logic @xcite . while these results give hardness or upper bounds for classes of logics , they do not provide a full case distinction identifying _ all _ `` easy '' or `` hard '' cases in the considered class . we achieve such a result : for a large class of modal logics containing many important representatives , we identify _ all _ cases which have an @xmath0-complete satisfiability problem , and show that the satisfiability problem for _ all _ other non - trivial logics in that class is @xmath1-hard . hence these problems avoid the infinitely many complexity classes between @xmath0 and @xmath1 , many of which have natural complete problems arising from logical questions . to our knowledge , such a general result has not been achieved before . to describe the considered class of modal logics , note that many relevant properties of modal models can be expressed by first - order formulas : a graph is transitive if its edge - relation @xmath7 satisfies the clause @xmath8 and symmetric if it satisfies @xmath9 . many other graph properties can be defined using similar formulas , where the presence of a certain pattern of edges in the graph forces the existence of another . analogously to propositional logic , we call conjunctions of such clauses _ universal horn formulas_. many relevant logics can be defined in this way : all examples form @xcite fall into this category , as well as logics over euclidean graphs . we study the following problem : given a universal horn formula @xmath10 what is the complexity of the modal satisfiability problem over the class of graphs defined by @xmath11 ? the main results of this paper are the following : first , we identify all cases which give a satisfiability problem solvable in @xmath0 ( which then for every nontrivial logic is @xmath0-complete ) , and show that all other cases are @xmath1-hard . second , we prove a generalization of a `` tree - like model property , '' and use it to obtain @xmath1 upper bounds for a large class of logics . as a corollary , we prove that ladner s classic hardness result is `` optimal '' in the class of logics defined by universal horn formulas . a further corollary is that in the universal horn class , all logics whose satisfiability problem is not @xmath1-hard already have the `` polynomial - size model property , '' which is only one of several known ways to prove @xmath0 upper bounds for modal logics . various work was done on restricting the syntax of the modal formulas by restricting the propositional operators @xcite , the nesting degree and number of variables @xcite or considering modal formulas in horn form @xcite . while these results are about restricting the _ syntax _ of the modal formulas , the current work studies different _ semantics _ of modal logics , where the semantics are specified by horn formulas . the organization of the paper is as follows : in section [ sect : preliminaries ] , we introduce terminology and generalize classic complexity results . section [ sect : modal models ] then establishes techniques to restrict the size of models for modal formulas , which are important tools for the @xmath0-membership later . section [ sect : universal horn clauses ] contains the main results of the paper about universal horn formulas . after introducing them in section [ sect : horn definitions ] and proving their relationship to homomorphisms in section [ sect : horn homomorphism ] , we prove @xmath0-results for special cases of horn formulas in sections [ section : varphi k to l section ] and [ section : np results ] . using these results , section [ sect : main dichotomy result ] then proves our main dichotomy result , which is corollary [ corollary : horn cunjunction without algorithm ] . the remainder of the paper establishes @xmath1 upper bounds for many of the @xmath1-hard logics . an important tool for these proofs is introduced in section [ sect : trees for horn ] , where we show a tree - like model property for all @xmath1-hard logics defined by universal horn formulas . section [ sect : pspace for non - transitive ] contains our @xmath1-algorithm , which generalizes many previously known algorithms for modal logics . finally , section [ sect : applications ] obtains a series of corollaries , proving the above - mentioned optimality result for ladner s hardness condition , and exhibiting a number of cases to which our @xmath1-algorithm can be applied . the paper closes with a summary and open questions in section [ sect : conclusion ] . modal logic is an extension of propositional logic . a modal formula is a propositional formula using variables , the usual logical symbols @xmath12 and a unary operator @xmath13 . ( a dual operator @xmath14 is often considered as well , this can be regarded as abbreviation for @xmath15 . ) a model for a modal formula is a set of connected `` worlds '' with individual propositional assignments . to be precise , a _ frame _ is a directed graph @xmath16 where the vertices in @xmath17 are called `` worlds , '' and an edge @xmath18 is interpreted as @xmath19 is `` considered possible '' from @xmath20 . a _ model _ @xmath21 consists of a frame @xmath16 a set @xmath22 of propositional variables and a function @xmath23 assigning each variable @xmath24 a subset of @xmath25 the set of worlds in which @xmath26 is true . we say the model @xmath27 is _ based _ on the frame @xmath28 . if @xmath29 is a class of frames , then a model is an @xmath29-model if it is based on a frame in @xmath29 . with @xmath30 we denote the number of worlds in the model @xmath27 . for a world @xmath31 we define when a formula @xmath32 is _ satisfied _ at @xmath33 in @xmath27 ( written @xmath34 . ) if @xmath32 is a variable @xmath35 then @xmath34 if and only if @xmath36 . as usual , @xmath37 if and only if @xmath38 and @xmath39 , and @xmath40 iff @xmath41 . for the modal operator , @xmath42 if and only if there is a world @xmath43 such that @xmath44 and @xmath45 . .common modal logics [ cols="<,^,<,<,<",options="header " , ] the construction works as follows : with formula @xmath46 we give the start of the model and define the world @xmath33 in which @xmath4 is satisfied to have the level @xmath47 . formula @xmath48 requires each node which is reachable in at most @xmath49 steps from @xmath33 to have a well - defined level ( or none ) . formulas @xmath50 and @xmath51 require each level @xmath52 to be followed by level @xmath53 and in the case that the @xmath52th quantifier is @xmath54 formula @xmath51 requires the corresponding branching of the quantifier tree . formula @xmath55 then forces the truth values of the @xmath56 to be `` sent down '' the tree , as described earlier . formula @xmath57 finally requires that the propositional formula @xmath58 is true for all possible truth assignments to @xmath59 generated by the quantifier tree . ladner s construction for the @xmath1 upper bound for the logic @xmath60 also reveals the following . the main idea behind this corollary is that any model for some modal formula can be transformed into a strict tree by `` unrolling . '' a much more general version of this result will be proven later as theorem [ theorem : tree - like models for horn logics ] . [ corollary : every k - satisfiable formula satisfiable in strict tree ] a modal formula is @xmath61-satisfiable if and only if it can be satisfied in a strict tree . corollary [ corollary : every k - satisfiable formula satisfiable in strict tree ] shows that we only need to apply our formulas to trees as long as our first order formula does not `` say anything '' about trees , the generated logic is the same as @xmath61 , although these logics do not necessarily have the same set of models . [ corollary : formulas satisfiable in trees give pspace complete logics ] let @xmath62 be a first - order formula over the frame language such that @xmath62 is satisfied in every strict tree . then the satisfiability problem for @xmath63 is @xmath1-complete . this follows immediately from corollary [ corollary : every k - satisfiable formula satisfiable in strict tree ] and theorem [ theorem : ladner ] , since every modal formula is @xmath60-satisfiable if and only it is @xmath63-satisfiable . for our complete classification in theorem [ theorem : horn conjunction classification ] , we need a hardness result which is a slight variation of ladner s hardness result for all logics between @xmath60 and @xmath64 . however , the proof is merely a closer inspection of ladner s construction . in addition to the already mentioned graph properties , we define a generalization of transitivity . for a natural number @xmath65 we say that a graph @xmath66 is @xmath67-transitive , if every pair of vertices @xmath68 in @xmath66 such that there is a @xmath67-step path from @xmath20 to @xmath19 in @xmath66 is connected with an edge . it is easy to see that a graph is transitive if and only if it is @xmath69-transitive , and a transitive graph is also @xmath67-transitive for every @xmath70 . for a set @xmath71 we say that a graph is @xmath72-transitive if it is @xmath67-transitive for every @xmath73 . [ theorem : ladner hardness cases for k - transitivity ] let @xmath11 be a first - order formula over the frame language such that one of the following cases applies : * @xmath11 is satisfied in every strict tree , * @xmath11 is satisfied in every reflexive tree , * there is a set @xmath74 such that @xmath11 is satisfied in every @xmath72-transitive tree , * @xmath11 is satisfied in every symmetric tree , * @xmath11 is satisfied in every tree which is both reflexive and symmetric , * there is a set @xmath74 such that @xmath11 is satisfied in every tree which is both reflexive and @xmath72-transitive . then @xmath75 is @xmath76-hard . note that for the cases not including symmetry , the result `` almost '' follows directly from ladner s theorem [ theorem : ladner ] . it does not follow directly , since we only require that our first - order formulas are satisfied in @xmath72-transitive _ trees _ , but nothing about arbitrary @xmath72-transitive graphs . nevertheless , the result for this case follows directly from ladner s proof . first consider the cases in which symmetry does not occur . now the result follows directly from ladner s proof : ladner proves @xmath1-hardness with a reduction from @xmath77 . from a quantified boolean formula @xmath78 he constructs a modal formula @xmath32 such that the following holds : 1 . if @xmath79 is true , then there is a modal model @xmath27 and a world @xmath80 such that @xmath27 is a strict tree , and @xmath34 . moreover , adding any number of reflexive or transitive edges in @xmath27 preserves the fact that @xmath34 . 2 . if @xmath79 is false , then @xmath32 is not @xmath60-satisfiable . it is immediate that this reduction also proves the desired hardness result for @xmath81 first assume that @xmath79 is true . then , by the above , the model @xmath82 satisfies @xmath83 and is a tree . if we close this model under the reflexive and/or @xmath72-transitive closures , then by the above this still is a model for @xmath32 . since @xmath11 is satisfied in every reflexive and/or @xmath72-transitive tree , this is a @xmath84-model for @xmath83 and hence @xmath32 is @xmath84-satisfiable . on the other hand , if @xmath79 is false , then @xmath32 is not @xmath60-satisfiable , and therefore not @xmath84-satisfiable . now the cases involving symmetry follow from corollary [ corollary : ufos satisfied in symmetric tree are pspace hard ] . as this section indicated , trees are an important subclass of modal models . and in fact , a good intuition to read this paper is to always think of the graphs we deal with as `` near - trees , '' i.e. , trees with additional edges . @xmath0-results in this paper are shown with the explicit construction of small models for a given formula . we therefore introduce some notation on graphs . for a graph @xmath85 the set of vertices of @xmath66 is denoted with @xmath86 and @xmath87 is the set of its edges . as usual , a _ homomorphism _ from a graph @xmath88 to a graph @xmath89 is a function preserving the edge relation . a _ strict tree _ is a tree in the usual sense , i.e. , a directed , acyclic , connected graph which has a root @xmath33 from which all other vertices can be reached . we now define notation to describe paths in graphs . note that we often identify modal models and their frames , when the propositional assignments are clear from the context or not important for our arguments . let @xmath66 be a graph , @xmath90 vertices , and @xmath91 . we write @xmath92 if in @xmath85 there is a path of length @xmath52 from @xmath33 to @xmath19 . additionally , @xmath93 for all @xmath94 . we also say that @xmath33 is a @xmath52-step predecessor of @xmath95 and @xmath19 is a @xmath52-step successor of @xmath33 in @xmath85 if @xmath96 . the _ maximal depth _ of @xmath94 is defined as @xmath97 . similarly , the _ maximal height _ of @xmath94 is @xmath98 . note that the maximal depth and maximal height of nodes can be ( countably ) infinite , even in finite graphs . the next definition is a restriction on graphs which is very natural for modal logics : for deciding whether @xmath34 holds for some modal @xmath99 a world @xmath100 and a modal formula @xmath83 it is obvious that only the worlds which are reachable from @xmath33 are important . let @xmath66 be a graph , and let @xmath94 . the graph @xmath101 is obtained from @xmath66 by restricting @xmath66 to the worlds which can be reached from @xmath33 . proving the polynomial - size model property for some logic is usually done starting with an arbitrary model for a given modal formula and building a smaller model out of it , which still satisfies the modal formula . however , we also need to ensure that the new model still satisfies the conditions of the logic under consideration . therefore , we need results that allow us to perform modifications on our models and leave the modal and the first - order properties invariant . the first result in this way concerns the first - order aspect of the frames for universal formulas : [ theorem : ufo invariants ] let @xmath66 and @xmath102 be graphs , and let @xmath103 be a natural number . the following conditions are equivalent : 1 . every universal first - order formula ( with at most @xmath103 variables ) satisfied by @xmath102 is also satisfied by @xmath85 2 . every existential first - order formula ( with at most @xmath103 variables ) satisfied by @xmath66 is also satisfied by @xmath104 3 . for each finite @xmath105 ( such that @xmath106 ) , there exists a functions @xmath107 such that for @xmath108 it holds that @xmath109 iff @xmath110 . @xmath111 : : let @xmath112 be an existential first - order formula over the frame language which holds in @xmath85 and assume that it does not hold in @xmath102 . in this case , the negation of the formula holds in @xmath104 i.e. , @xmath113 . this is equivalent to @xmath114 which is a universal first - order formula over the frame language . hence by the prerequisites , we know that @xmath115 which is a contradiction . @xmath116 : : analogously to the above . @xmath117 : : we construct a function @xmath107 with the desired properties . let @xmath118 and let @xmath119 . we construct an existential first - order formula + @xmath120 + obviously , @xmath11 has @xmath103 variables , and @xmath66 obviously is a model for @xmath10 it follows that @xmath121 . therefore , there are @xmath122 such that @xmath123 if and only if @xmath124 . define @xmath125 . this function obviously meets the criteria : let @xmath123 . then @xmath126 is a clause in @xmath11 . therefore for the values @xmath127 and @xmath128 chosen by the existential quantifiers , @xmath129 must hold . since @xmath130 and @xmath131 the claim follows . for the condition @xmath132 the proof is the same . @xmath133 : : let such a function @xmath134 exist for every @xmath105 with @xmath106 . let @xmath135 be a first - order existential formula , let @xmath103 be the number of variables in @xmath11 , and let @xmath136 . we show that @xmath121 . since @xmath137 there are @xmath138 such that @xmath139 holds in @xmath66 . this implies that @xmath140 holds in @xmath102 . therefore , @xmath121 . note that while the function @xmath134 required to exist in the conditions of the above theorem shares some properties with an isomorphism , it is not required to be injective . the following is an important special case , which immediately follows from this observation and theorem [ theorem : ufo invariants ] : [ prop : trivial subgraph property ] let @xmath66 be a graph , @xmath62 a universal first - order formula over the frame language such that @xmath141 . then for every subgraph @xmath102 of @xmath85 it holds that @xmath142 . for a modal model , a restriction of the model is a restriction of the graph , where the propositional assignment for the remaining worlds is unchanged . we now consider restrictions which are `` compatible '' with the modal properties of the formulas in question . the following lemma describes a standard way to reduce the number `` relevant '' of successors to worlds in models . this is an application of the more general idea of bounded morphisms , which we will encounter in section [ sect : trees for horn ] . for a modal formula @xmath83 @xmath143 denotes the set of its subformulas . with @xmath144 we denote the modal depth of a formula @xmath83 i.e. , the maximal nesting degree of the modal operator @xmath13 in @xmath32 . [ lemma : modal restriction invariance ] let @xmath32 be a modal formula , and let @xmath34 . let @xmath145 be a restriction of @xmath27 such that the following holds : 1 . @xmath146 2 . for all @xmath147 and all @xmath148 such that @xmath149 there is some @xmath150 such that @xmath68 is an edge in @xmath99 and @xmath151 . then @xmath152 . lemma [ lemma : modal restriction invariance ] immediately follows from the following lemma . the version stated in lemma [ lemma : modal restriction invariance ] is the one we almost exclusively use , hence we stated this simpler version explicitly . we now prove a slightly more general result , which also takes into account that for a modal formula , worlds which are not reachable on a path with at most the length of the modal depth of the formula , are irrelevant . [ lemma : modal restriction invariance with levels ] let @xmath32 be a modal formula , and let @xmath34 . let @xmath145 be a restriction of @xmath27 such that the following holds : 1 . @xmath146 2 . for all @xmath147 and all @xmath153 such that @xmath154 and there exists an @xmath91 such that @xmath155 and @xmath156 there is some @xmath150 such that @xmath68 is an edge in @xmath99 and @xmath151 . then @xmath152 . we show the following claim : let @xmath157 and @xmath158 such that @xmath159 and @xmath160 then @xmath161 if and only if @xmath162 . for @xmath163 @xmath164 and @xmath165 this implies the lemma , since @xmath33 is an element of @xmath145 by definition . we show the claim by induction on @xmath79 . if @xmath79 is a variable , then this holds trivially , since @xmath145 is a restriction of @xmath27 and therefore , propositional assignments are not changed . the induction step for propositional operators is trivial . therefore , assume that @xmath166 for some @xmath153 such that the claim holds for @xmath167 . now let @xmath168 meet the prerequisites of the claim , i.e. , let @xmath160 and let @xmath155 . first assume that @xmath161 . since @xmath169 it follows that @xmath170 and hence @xmath171 . since @xmath149 the prerequisites of the lemma therefore imply that there is a world @xmath150 such that @xmath68 is an edge in @xmath99 and @xmath151 . since @xmath159 it follows that @xmath172 . by the induction hypothesis , we know that @xmath173 . since @xmath145 is a restriction of @xmath99 @xmath68 is an edge in @xmath145 as well , and therefore we conclude that @xmath174 i.e. , @xmath162 . for the other direction , assume that @xmath162 . therefore , there is a node @xmath150 such that @xmath175 and @xmath68 is an edge in @xmath145 . since @xmath159 we know that @xmath172 holds as well , and since @xmath176 from the induction hypothesis we conclude that @xmath151 . since @xmath68 is an edge in @xmath27 as well , it therefore follows that @xmath149 i.e. , @xmath177 concluding the proof . the following easy proposition shows how lemma [ lemma : modal restriction invariance with levels ] can be applied : [ prop : rooted models exist ] let @xmath178 be a universal elementary logic , let @xmath32 be a modal formula , and let @xmath179 where @xmath27 is a @xmath180-model . then there is a @xmath180-model @xmath145 which is a restriction of @xmath99 which is rooted at @xmath181 and where every world can be reached from @xmath33 in at most @xmath182 steps , and @xmath152 . the model @xmath145 is obtained by simply removing all worlds from @xmath27 which can not be reached from @xmath33 in at most @xmath182 steps . due to proposition [ prop : trivial subgraph property ] , @xmath145 is still a @xmath180-model . we now show that @xmath152 holds , by proving that @xmath145 satisfies the conditions of lemma [ lemma : modal restriction invariance with levels ] . by definition , since @xmath33 can be reached from @xmath33 in @xmath47 steps , we know that @xmath183 . hence let @xmath147 and let @xmath153 and let @xmath149 such that @xmath184 for some @xmath52 such that @xmath185 . since @xmath149 we know that there is a world @xmath186 such that @xmath187 and @xmath68 is an edge in @xmath27 . it follows that @xmath188 . since @xmath156 and @xmath189 we know that @xmath190 and hence we know that @xmath19 is an element of @xmath145 . therefore , @xmath145 satisfies the conditions of lemma [ lemma : modal restriction invariance with levels ] , and therefore the lemma implies that @xmath191 as claimed . as mentioned , our @xmath0-containment results are obtained by proving the polysize model property and applying proposition [ proposition : elementary plus polysize model property is in np ] . the polynomial models are obtained by restricting arbitrary models to polynomial size . we will now show some restrictions which we can make in any model , showing that we can assume certain parts of the model to be only polynomial in size . [ lemma : few nodes with few predecessors ] let @xmath192 . then for any modal formula @xmath32 and any @xmath179 there is a submodel @xmath145 of @xmath27 such that @xmath152 and the following holds : @xmath193 for each world @xmath20 in @xmath99 let @xmath194 . for each @xmath195 let @xmath196 be a subset of the @xmath197-step successors of @xmath20 in @xmath27 such that for every @xmath198 there is a world @xmath199 such that @xmath68 is an edge in @xmath99 and @xmath187 and @xmath200 . now define @xmath201 and for each @xmath202 let @xmath203 . finally , define @xmath145 to be the restriction of @xmath27 to @xmath204 . to show that @xmath191 we prove that @xmath145 satisfies the conditions of lemma [ lemma : modal restriction invariance ] . obviously , @xmath145 is a restriction of @xmath27 and @xmath205 . therefore , let @xmath20 be a world from @xmath206 and let @xmath167 be a subformula of @xmath32 such that @xmath154 . since @xmath147 there is some @xmath52 such that @xmath207 . since @xmath149 we know that @xmath198 and hence there is a world @xmath199 such that @xmath187 and @xmath68 is an edge in @xmath27 . it follows that @xmath208 and hence @xmath145 satisfies the conditions of lemma [ lemma : modal restriction invariance ] as claimed . now let @xmath209 it remains to show the cardinality bound for @xmath210 . it is obvious that @xmath211 since for every @xmath212 every vertex in @xmath213 has a predecessor in @xmath214 and hence inductively , every vertex in @xmath215 has an @xmath52-step predecessor in @xmath145 . obviously , @xmath216 and @xmath217 . therefore , @xmath218 for all @xmath91 . now , due to the above , @xmath219 . since @xmath220 the claim follows . by construction , the model @xmath145 given in the proof of the above lemma is countable . hence we obtain the following corollary : [ corrollary : universal elementary logics have countable model property ] let @xmath178 be a universal elementary logic . then every @xmath180-satisfiable formula @xmath32 is satisfiable in a countable @xmath180-model . moreover , every @xmath180-model satisfying @xmath32 at a node @xmath33 contains a countable @xmath180-submodel satisfying @xmath32 at the node @xmath33 . the following lemma shows that it is sufficient to restrict the number of those vertices in the model which have a minimal height in the graph . in combination with lemma [ lemma : few nodes with few predecessors ] , this shows that we only need to be concerned about vertices which have both a certain number of predecessors , and a certain number of successors . we already saw in proposition [ prop : rooted models exist ] that we are only interested in rooted graphs . in such graphs , lemmas [ lemma : few nodes with few predecessors ] and [ lemma : few nodes with few successors ] can be seen as limiting the number of vertices near the `` top '' or the `` bottom '' of the model . this is useful , because in graphs satisfying some universal formula , often special cases can occur in these regions of the graph . these lemmas show that we do not need to look too closely at these exceptions . [ lemma : few nodes with few successors ] let @xmath192 be a constant . then for any modal formula @xmath32 and any @xmath179 there is a submodel @xmath145 of @xmath27 such that @xmath191 and for which the following holds : @xmath221 where @xmath222 . let @xmath223 i.e. , the set of nodes in @xmath27 which have a @xmath224-step successor . we now define a sequence of submodels of @xmath225 let @xmath226 and for @xmath227 let @xmath213 be defined as follows : * @xmath228 * for every @xmath229 and each @xmath148 such that @xmath149 add one world @xmath19 from @xmath27 to @xmath213 such that @xmath68 is an edge in @xmath27 and @xmath151 . now let @xmath145 be the restriction of @xmath27 to the worlds in @xmath230 . we show that @xmath191 by showing that it satisfies the conditions of lemma [ lemma : modal restriction invariance ] . by definition , hence let @xmath147 @xmath153 and let @xmath154 . since @xmath147 there exists a minimal @xmath52 such that @xmath207 . first assume that @xmath232 . by construction , for every relevant @xmath233 every node in @xmath234 has a @xmath197-step predecessor in @xmath235 and hence , inductively , the node @xmath236 is a @xmath237-step successor of a node @xmath26 in @xmath238 . since @xmath149 we know that @xmath20 has a successor in @xmath27 . this implies that @xmath26 has a @xmath224-step successor in @xmath99 and hence @xmath239 which is a contradiction , since @xmath240 and @xmath241 . therefore , we know that @xmath242 . by construction , there is a world @xmath19 in @xmath243 such that @xmath68 is an edge in @xmath99 and @xmath244 . therefore , the model @xmath145 satisfies the conditions of lemma [ lemma : modal restriction invariance ] , and therefore we conclude that @xmath152 . by definition , it holds that @xmath245 and for @xmath227 @xmath246 . since @xmath247 this implies that @xmath248 as claimed . the main purpose of lemmas [ lemma : few nodes with few predecessors ] and [ lemma : few nodes with few successors ] is the following : if for a modal logic @xmath249 there is a constant @xmath250 such that every @xmath180-satisfiable formula has a model in which we can restrict the number of nodes which have both a @xmath224-step predecessor and a @xmath224-step successor , then lemmas [ lemma : few nodes with few predecessors ] and [ lemma : few nodes with few successors ] can be used to show the polynomial model property for @xmath180 . this idea plays a crucial role in the proof of our main @xmath0-containment result , theorem [ theorem : above k to k+1 leads to np ] , and is formalized in the following corollary : [ corollary : suffices to restrict middle vertices for np ] let @xmath178 be a universal elementary modal logic such that there exists a constant @xmath192 such that there is a polynomial @xmath251 such that for all @xmath180-satisfiable formulas @xmath83 there is a @xmath180-model @xmath27 and a world @xmath80 such that @xmath179 and @xmath252 then @xmath178 has the polynomial - size model property , and @xmath253 . note that the function @xmath251 is only required to be polynomial in its argument @xmath254 and not in the value @xmath250 which is a constant depending only on the logic , and not on the formula . let @xmath32 be a @xmath180-satisfiable formula , and let @xmath27 be a @xmath180-model and @xmath80 meeting the prerequisites of the corollary . by lemma [ lemma : few nodes with few predecessors ] , there is a submodel @xmath145 of @xmath27 such that @xmath191 and @xmath255 therefore , since the conditions required in the prerequisites of the corollary are invariant under further restrictions of the model , assume without loss of generality that @xmath27 already satisfies this condition . now let @xmath256 where @xmath257 note that @xmath258 and @xmath259 are not necessarily disjoint ( @xmath258 contains the nodes with only small depth at the `` * * t**op '' of the model , @xmath260 represents the `` * * c**enter '' of the model , and @xmath259 is the `` * * b**ottom . '' ) . by the prerequisites of the corollary , we can assume that @xmath261 and by the above we know that @xmath262 . now let @xmath263 . it follows that @xmath264 and hence @xmath265 by lemma [ lemma : few nodes with few successors ] , there is a submodel @xmath145 of @xmath27 such that @xmath152 and @xmath266 and hence by the above we have that @xmath267 ( note that the cardinality of the size @xmath210 defined with respect to the submodel @xmath145 is bounded by the cardinality of the original set @xmath210 ) . since @xmath251 is a polynomial and @xmath224 is a constant only depending on the logic @xmath180 , this is a polynomial size bound in @xmath254 and therefore we have proven the polynomial - size model property . by proposition [ prop : trivial subgraph property ] , @xmath145 is a @xmath180-model . the complexity result now follows from proposition [ proposition : elementary plus polysize model property is in np ] . we now consider a syntactically restricted case of universal first order formulas , namely horn formulas . many well - known logics can be expressed in this way . usually , a horn clause is defined as a disjunction of literals of which at most one is positive . if a positive literal occurs , then the clause can be written as an implication , since @xmath268 is equivalent to @xmath269 . if no positive literal occurs , then the clause @xmath270 can be written as @xmath271 . since in the context of the frame language , an atomic proposition is of the form @xmath272 the following is the natural version of horn clauses for our purposes : a _ universal horn clause _ is a formula of the form @xmath273 or of the form @xmath274 where all ( not necessarily distinct ) variables are implicitly universally quantified . a _ universal horn formula _ is a conjunction of universal horn clauses . with universal horn formulas , many of the usually considered graph properties can be expressed , like transitivity , symmetry , euclidicity , etc . in the following definition , we show how universal horn clauses can be represented as graphs . let @xmath62 be a universal horn clause . * the _ prerequisite graph of @xmath62 _ , denoted with @xmath275 , consists of the variables appearing on the left - hand side of the implication @xmath276 where @xmath277 is an edge if the clause @xmath278 appears . * if @xmath62 is a universal horn clause where the right - hand side of the implication in @xmath62 is @xmath279 then the _ conclusion edge of @xmath62 _ , denoted with @xmath280 , is the edge @xmath281 . if the right - hand side of the implication is @xmath282 then @xmath283 is the empty set . the definition of the prerequisite graph and the conclusion edge of a universal horn formula establishes a one - to - one correspondence between universal horn clauses and their representation as graphs . these definitions allow us to relate truth of a horn clause to homomorphic images of the involved graphs : [ prop : homomorphism and horn clauses ] 1 . let @xmath62 be a universal horn clause with @xmath284 . a graph @xmath66 satisfies @xmath62 if and only the following holds : for every homomorphism @xmath285 , @xmath286 is an edge in @xmath287 2 . let @xmath62 be a universal horn clause such that @xmath288 . then a graph @xmath66 satisfies @xmath62 is and only if there is no homomorphism @xmath289 1 . let @xmath290 where all variables are implicitly universally quantified . first assume that @xmath291 and let @xmath285 be a homomorphism . due to the definition of @xmath292 there are edges @xmath293 in @xmath294 . since @xmath295 is a homomorphism , this implies that @xmath296 are edges in @xmath66 . hence the nodes @xmath297 satisfy the formula @xmath298 . therefore , the nodes @xmath299 satisfy the prerequisites of the clause @xmath62 . since @xmath291 this implies that @xmath286 is an edge in @xmath287 + now for the other direction , assume that @xmath66 fulfills the homomorphism property , and let @xmath300 . let @xmath301 be nodes in @xmath66 satisfying the prerequisite clause of @xmath302 i.e. , if @xmath303 is a clause in @xmath292 then @xmath304 is an edge in @xmath66 . then obviously the function @xmath295 mapping the variable @xmath305 to the node @xmath306 is a homomorphism from @xmath294 to @xmath66 . by the prerequisites , we know that @xmath286 is an edge in @xmath66 . hence , @xmath66 satisfies the formula @xmath62 . 2 . analogous . there is a natural correspondence between implications of these formulas and graph homomorphisms . [ prop : homomorphism and horn clause implication ] 1 . let @xmath307 and @xmath308 be universal horn clauses such that there exists a homomorphism @xmath309 which maps the conclusion edge of @xmath307 to the conclusion edge of @xmath308 . then @xmath307 implies @xmath310 2 . let @xmath307 and @xmath308 be universal horn clauses such that @xmath311 and let @xmath312 be a homomorphism . then @xmath307 implies @xmath310 1 . let @xmath313 then by the prerequisites it follows that @xmath314 . now let @xmath66 be a graph such that @xmath315 and let @xmath316 be a homomorphism . by proposition [ prop : homomorphism and horn clauses ] , it suffices to show that @xmath317 is an edge in @xmath66 . since @xmath295 and @xmath318 are homomorphisms , @xmath319 is a homomorphism as well . since @xmath315 proposition [ prop : homomorphism and horn clauses ] implies that @xmath317 is an edge in @xmath85 as claimed . 2 . let @xmath66 be a graph such that @xmath320 . due to proposition [ prop : homomorphism and horn clauses ] , to show that @xmath321 it suffices that there is no homomorphism @xmath322 . hence assume that such a homomorphism exists . then @xmath323 is a homomorphism from @xmath324 into @xmath85 which is a contradiction to proposition [ prop : homomorphism and horn clauses ] , since @xmath325 we now consider special cases of horn clauses , which will be central for the logics having satisfiability problems in @xmath0 . the following definition captures the case where the variables in the conclusion edge have a common predecessor in the prerequisite graph , but there is not necessarily a direct path between them . using results about graphs satisfying generalizations of formulas of this type , we will be able to show all of the @xmath0-containment results that the proof of the later classification theorem , theorem [ theorem : horn conjunction classification ] , depends on . let @xmath326 be the formula @xmath327 where all variables are universally quantified ( and in the case that @xmath328 or @xmath329 appear in the formula , we replace them with @xmath33 ) . l3.3 cm in figure [ fig : varphi 2 to 4 example ] , we present the graph representation of the formula @xmath330 . the graph property described by these formulas is easy to see : [ prop : varphi k to l property ] let @xmath66 be a graph , and let @xmath331 . then @xmath332 if and only if for any nodes @xmath333 if @xmath334 and @xmath335 then @xmath336 is an edge in @xmath287 this definition generalizes several well - known examples in particular , a graph is reflexive if and only if it satisfies @xmath337 symmetry is expressed with @xmath338 and transitivity with @xmath339 . finally , a graph is euclidean iff it satisfies @xmath340 . therefore , this notation allows us to capture many interesting graph properties , and it is not surprising that generalizations of this idea are the main ingredients for our polynomial size model proofs . we start with looking at some properties and implications of formulas of the form @xmath341 [ lemma : varphi k to l implications ] let @xmath342 and let @xmath66 be a graph such that @xmath343 1 . @xmath344 2 . if @xmath345 then for any @xmath346 , @xmath347 3 . there is some @xmath348 such that @xmath349 for all @xmath350 _ proof _ 1 . let @xmath33 be some node in @xmath85 such that @xmath351 and @xmath352 and let the ( not necessarily distinct ) intermediate vertices be denoted with @xmath353 . since @xmath326 holds in @xmath66 , this implies that there is an edge @xmath354 . by choice of nodes , @xmath355 . combining these , we obtain a path of length @xmath67 from @xmath356 to @xmath357 . on the other hand , @xmath358 . since @xmath326 holds in @xmath66 , it follows that there is an edge from @xmath357 to @xmath359 proving that @xmath360 holds in @xmath66 . clearly it suffices to show @xmath361 the claim for arbitrary @xmath52 follows inductively . let @xmath362 be chosen such that @xmath363 and there are edges @xmath364 and @xmath365 . we need to show that there is an edge @xmath366 + since @xmath367 it follows that @xmath368 is an edge in @xmath66 . since there obviously is a path of length @xmath369 from @xmath370 to @xmath356 , it follows that there is a path of length @xmath67 from @xmath370 to @xmath371 . since there also is a path of length @xmath372 from @xmath370 to @xmath373 it follows that there is an edge @xmath374 in @xmath66 , which concludes the proof . 3 . this follows immediately from the above : from points @xmath197 and @xmath375 it follows that the claim holds for @xmath376 @xmath14 l4.5 cm the formula @xmath326 is supposed to capture the case where the variables in the conclusion edge of a universal horn clause have a common predecessor . but not all of these cases are covered with this formula . the above figure [ fig : varphi 2 to 4 example ] is a graphical representation of what the implication @xmath330 does . but what if this is only a subgraph of the prerequisite graph ? in a more general case , the node @xmath33 and the nodes @xmath377 will have more predecessors and successors . figure [ fig : transformation original formula ] gives an example of a more general formula . we will now see that this formula can be `` simplified . '' this simplification is not an equivalent transformation of the formula , but we construct a new formula which is implied by the original one . the one - sided implication suffices to show many of the results we need . the simpler formula is presented in figure [ fig : transformation resulting formula ] . it is easy to see that every graph which satisfies the formula displayed in figure [ fig : transformation original formula ] also satisfies the formula from figure [ fig : transformation resulting formula ] . this follows directly from proposition [ prop : homomorphism and horn clause implication ] , since the prerequisite graph from figure [ fig : transformation original formula ] can obviously be mapped homomorphically to the prerequisite graph from figure [ fig : transformation resulting formula ] ( the homomorphism @xmath295 is defined as @xmath378 and maps the other nodes to the ones with the same labels ) . hence , if we can show @xmath0-containment for all universal elementary modal logics extending the one defined by the later formula , this puts the logic defined by the original formula into @xmath0 as well . l2.4 cm any universal horn clause which can be mapped onto a tree can be embedded in a graph with certain properties , namely the properties of the formula we now define . due to corollary [ corollary : formulas satisfiable in trees give pspace complete logics ] , it is natural that tree - like homomorphic images of our universal horn formulas are of interest to us . these formulas capture the generalizations of @xmath326 mentioned above , where we demand that the nodes @xmath379 have a sufficient number of predecessors or successors . we again use the representation of horn clauses as graphs . l3.3 cm for @xmath380 the formula @xmath381 is defined as the universal horn clause displayed in figure [ fig : general xyz stu formula ] . it should be noted that the notation @xmath381 suggests that @xmath382 can be compared to natural numbers , but what is meant in that notation is simply that the number of predecessors ( successors , resp . ) of @xmath379 can be compared to @xmath383 and @xmath384 respectively . hence , if we use `` natural names '' for the vertices , i.e. we have vertices @xmath385 and @xmath386 then this ensures that the vertices up to @xmath387 @xmath388 and @xmath389 exist . when proving that this formula holds in a graph , we will usually rely on the notation provided in figure [ fig : general xyz stu formula ] , and assume that there are nodes @xmath390 with edges as seen in figure [ fig : general xyz stu formula ] , i.e. , most of the time we do not mention the homomorphism explicitly . the formula @xmath381 can be seen to be only a slight generalization of the formulas @xmath326 we already considered , as exhibited by the following proposition : [ prop : genvarphi property ] let @xmath66 be a graph , and let @xmath391 . then @xmath392 if and only if the following condition holds : for any nodes @xmath333 such that @xmath33 has a @xmath251-step predecessor , @xmath26 has a @xmath393-step successor and @xmath394 has an @xmath395-step successor , @xmath334 and @xmath335 it follows that @xmath336 is an edge in @xmath287 the above proposition immediately implies the following : [ prop : cut to many pred many suc part gives non - gen varphi ] let @xmath396 and let @xmath102 be the restriction of @xmath66 to the set @xmath397 then @xmath398 this proposition is one of the reasons why corollary [ corollary : suffices to restrict middle vertices for np ] is important : we can use it together with proposition [ prop : cut to many pred many suc part gives non - gen varphi ] to restrict our attention to the vertices in the `` middle '' of the graph , and then talk about the formula @xmath326 instead of @xmath381 . this is the general approach for our @xmath0-containment proofs , although some technical difficulties remain , as we will see in the proof of theorem [ theorem : above k to k+1 leads to np ] . there obviously is a close relationship between formulas of the form @xmath326 and formulas of the form @xmath381 . in particular , this relationship allows the `` lifting '' of implications , as is shown in the following easy lemma . [ lemma : varphi k to l implication gives genvarphi implications ] let @xmath399 and let @xmath326 imply @xmath400 . then @xmath381 implies @xmath401 where @xmath402 let @xmath66 be a graph satisfying @xmath381 . let @xmath403 be the nodes in the graph connected as the nodes in the prerequisite graph of @xmath404 . let @xmath102 be the graph @xmath66 restricted to the set of vertices which have a @xmath251-step predecessor , and @xmath405-step successor in @xmath66 . then , by choice of nodes , and propositions [ prop : genvarphi property ] and [ prop : varphi k to l property ] , it follows that @xmath406 . hence , due to the prerequisites , we know that @xmath407 . in particular , since the nodes @xmath403 satisfy the prerequisite graph of @xmath401 we know that @xmath408 . hence , it follows that @xmath336 is an edge in @xmath102 , and therefore it is an edge in @xmath85 as claimed . we need three more implications between formulas of this kind for our later @xmath0-results : [ lemma : genvarphi k to 0 implications][lemma:(prop):varphi 0 to k gives varphi 0 to k^2 ] let @xmath409 and let @xmath66 be a graph . 1 . if @xmath410 then @xmath411 2 . if @xmath412 then @xmath413 where @xmath414 3 . let @xmath70 . then @xmath415 implies @xmath416 let @xmath417 be nodes in the graph fulfilling the prerequisite graph of the respective formulas . 1 . by the @xmath418-property , it is clear that @xmath419 are edges in @xmath66 . hence , @xmath420 . again due to the property , it follows that @xmath421 is an edge in @xmath85 as required . 2 . consider the subgraph @xmath422 . by choice of @xmath423 every node in @xmath102 has a @xmath251-step predecessor , a @xmath393-step successor , and an @xmath424-step successor in @xmath66 . hence , @xmath425 . due to part @xmath426 this implies that @xmath427 . hence , in @xmath102 there are edges @xmath428 . therefore , due to the @xmath418-property , it follows that @xmath429 is an edge in @xmath102 . hence , it follows that @xmath430 . the @xmath418-property implies that @xmath431 is an edge , as required . 3 . this follows from lemma [ lemma : test case 1 ] , since @xmath67-transitivity implies @xmath432-transitivity . in this section , we prove @xmath0-containment results for logics defined by universal horn clauses . we show that `` most '' of the logics of the form @xmath433 give rise to a satisfiability problem in @xmath0 . it is comparably easy to show that a logic of the form @xmath434 leads to a satisfiability problem in @xmath0 , by carefully `` copying '' vertices and adding the correct neighbors . however , while this process leads to a model which still satisfies @xmath435 it does not give the desired result that the problem can be solved in @xmath436 for all universal horn logics which are extensions of @xmath434 . in order to prove this , our model - manipulations must be consistent with the conditions of theorem [ theorem : ufo invariants ] . in the proof of the following theorem , we construct a small model using only restriction . note that the proof for this theorem is the only occasion where we actually prove the polynomial - size model property by explicitly constructing the model . all further @xmath0-results make use of theorem [ theorem : above k to k+1 leads to np ] , by showing that the logic in question is an extension of one of the logics that this theorem deals with . for example , the logic @xmath437 satisfies the conditions of this theorem . it is therefore a central theorem for our complexity classification . the proof relies on one later result , namely the second case of theorem [ theorem : tree - like models for horn logics ] . however , the proof for this case of theorem [ theorem : tree - like models for horn logics ] does not rely on any facts about the classification algorithm studied later . in the following , a graph @xmath66 with root @xmath33 is @xmath33-canonical if for every node @xmath438 if @xmath439 and @xmath440 then @xmath441 . note that this does not mean that every node @xmath26 is reachable on only one path from the root , but that all paths from the root @xmath33 to @xmath26 have the same length . we say that a graph @xmath66 is _ canonical _ if it is @xmath33-canonical for a root @xmath33 of @xmath66 . it is easy to see that a graph is canonical if and only if it can be homomorphically mapped onto a strict line . [ theorem : above k to k+1 leads to np ] let @xmath62 be a universal horn formula implying @xmath442 for some @xmath443 . then @xmath63 has the polynomial - size model property , and @xmath444 due to proposition [ proposition : elementary plus polysize model property is in np ] , it suffices to prove the polynomial - size model property for @xmath445 hence , let @xmath32 be a @xmath63-satisfiable modal formula , and , following proposition [ prop : rooted models exist ] , let @xmath27 be a @xmath63-model with root @xmath33 such that @xmath34 . define @xmath446 to be the maximal length of a strict line on which @xmath62 is satisfied , and @xmath446 to be zero in the case that @xmath62 is satisfied on every strict line ( note that due to proposition [ prop : trivial subgraph property ] , if @xmath62 is satisfied on the strict line of length @xmath212 then it is also satisfied on the strict line of length @xmath447 ) . this number is obviously a constant depending only on @xmath62 . if @xmath62 is satisfied on every strict line , we can , due to theorem [ theorem : tree - like models for horn logics ] , assume that @xmath27 is @xmath33-canonical . note that we will work with restrictions of the model during the course of the proof since any restriction of a canonical graph having the same root is still canonical , the submodels we consider later are canonical as well . now , define @xmath448 the set @xmath260 contains all the nodes in @xmath27 which have `` enough '' predecessors and successors to ensure that the formula @xmath442 gives us all the necessary edges that we are interested in . to be more precise , we show the following fact : [ prooffact : c models varphi i to i+1 ] for all @xmath449 it holds that @xmath450 it suffices to show the claim for @xmath451 . the result for arbitrary @xmath52 then follows from lemma [ lemma : varphi k to l implications ] , and the observation that @xmath452 is true in any graph . hence , we show @xmath453 let @xmath454 be vertices from @xmath455 such that they satisfy the prerequisites of @xmath456 i.e. , let @xmath457 @xmath458 and @xmath459 be edges in @xmath27 . since @xmath460 we know that @xmath461 has a @xmath369-step predecessor @xmath462 which in turn has a @xmath251-step predecessor . due to the edges mentioned above , it obviously holds that @xmath463 and @xmath464 . since both @xmath465 and @xmath466 are elements of @xmath455 they have the required number of successors , and therefore , since @xmath467 proposition [ prop : genvarphi property ] implies that @xmath468 is an edge in @xmath99 as claimed . due to lemma [ lemma : few nodes with few predecessors ] , we can assume , without loss of generality , that @xmath469 is polynomially bounded ( in the length of @xmath32 ) . we can additionally assume , again due to lemma [ lemma : few nodes with few predecessors ] , that the number of nodes @xmath19 with @xmath470 is also restricted by a polynomial . due to proposition [ prop : rooted models exist ] , we can also assume that every world in @xmath27 can be reached from @xmath33 in @xmath182 steps . therefore , we can choose a set @xmath471 with the following properties : * @xmath66 is polynomially bounded , * @xmath472 * for every node @xmath473 and every @xmath474 such that @xmath68 is an edge in @xmath99 and every subformula @xmath167 of @xmath32 such that @xmath187 there is a node @xmath475 such that @xmath476 is an edge in @xmath99 and @xmath477 * every node in @xmath260 can be reached on a path from some node in @xmath66 . additionally , if @xmath4 is satisfied on every strict line , every node in @xmath260 can be reached from some node in @xmath66 on a path of length at most @xmath478 such a set can be chosen with an application of the technique used to prove lemma [ lemma : few nodes with few predecessors ] : a set @xmath66 can be chosen with the properties above . first consider the case that @xmath62 is satisfied on every strict line . in this case , since @xmath27 is @xmath33-canonical , we know that @xmath210 is exactly the set of vertices on the first @xmath479 levels of @xmath27 ( the @xmath52-th level is the set of vertices @xmath19 in @xmath27 such that @xmath480 . note that a canonical graph can only have one root ) . since @xmath27 is @xmath33-canonical , we can simply choose @xmath66 to be the set of those nodes at level @xmath481 in the model @xmath27 which are also elements of @xmath260 . since in a canonical model @xmath99 for each node @xmath19 @xmath482 is exactly the level of @xmath19 in @xmath99 this is a set of polynomial size , and every path from @xmath33 to a node in @xmath260 passes through @xmath66 . further , every successor @xmath474 of a node @xmath473 is trivially a member of @xmath66 . since every node in @xmath27 can be reached from the root @xmath33 in at most @xmath182 steps , the claim follows . now consider the case where @xmath62 is not satisfied on every strict line . for each world @xmath483 let @xmath484 and let @xmath196 be a subset of successors of @xmath20 in @xmath27 such that for every @xmath198 there is a world @xmath199 such that @xmath68 is an edge in @xmath99 @xmath187 and @xmath485 . if it is possible to choose @xmath486 then do so . define @xmath201 and for each @xmath487 let @xmath488 now define @xmath145 to be the restriction of @xmath27 to the worlds in @xmath489 . we show that @xmath152 by proving that @xmath145 satisfies the conditions of lemma [ lemma : modal restriction invariance ] . by construction , we know that @xmath183 . now let @xmath147 and let @xmath167 be a subformula of @xmath32 with @xmath490 . then @xmath491 and therefore there is a world @xmath199 with @xmath492 . since @xmath147 there is some @xmath52 with @xmath207 . it follows that @xmath493 and hence @xmath150 as claimed . since @xmath145 is a restriction of @xmath99 we know that @xmath145 also is a @xmath63-model , and inherits all size restrictions on submodels that we have already established . therefore , we can , without loss of generality , assume that @xmath494 now define @xmath495 . by the proof of lemma [ lemma : few nodes with few predecessors ] , since @xmath481 is a constant , we conclude that @xmath66 is polynomial in @xmath496 . by construction , @xmath497 . we prove that every node @xmath498 can be reached on a path from a node in @xmath260 . for every @xmath483 there exists a sequence of nodes @xmath499 such that for all relevant @xmath212 @xmath500 @xmath501 is an edge in @xmath99 and @xmath502 . let @xmath503 be minimal such that @xmath504 . since @xmath505 it follows that @xmath506 and we know that such a @xmath503 exists , since @xmath507 . it suffices to prove that @xmath508 . since @xmath509 it remains to prove that @xmath510 . assume that this is not the case , from the above it then follows that @xmath511 . by choice of nodes , we know that @xmath512 has a @xmath481-step predecessor , and since @xmath512 is a predecessor of @xmath195 @xmath512 also has a @xmath513-step successor . since @xmath514 we know that @xmath515 therefore it follows that @xmath516 a contradiction to the minimality of @xmath517 now , let @xmath518 and let there be some @xmath474 such that @xmath68 is an edge in @xmath99 and @xmath151 for some @xmath148 . since @xmath518 we know that @xmath519 or @xmath20 does not have a @xmath481-step predecessor , and hence @xmath207 for some @xmath520 . by construction of @xmath521 there is some world @xmath522 which satisfies the requirements , since the successors are chosen to be from @xmath260 if possible . since @xmath523 this proves that @xmath66 indeed satisfies the conditions . for every @xmath524 we define @xmath525 to be the set @xmath526 . obviously , @xmath525 is a graph with root @xmath527 and is a restriction of @xmath260 ( recall that @xmath528 is the restriction of @xmath27 to all vertices reachable from @xmath529 on a directed path ) . hence , proposition [ prop : trivial subgraph property ] and fact [ prooffact : c models varphi i to i+1 ] immediately imply the following : [ prooffact : m_c_g models varphi i to i+1 ] let @xmath530 . then @xmath525 is a graph with root @xmath529 such that @xmath531 for all @xmath532 for any @xmath530 and @xmath202 we define @xmath533 we observe the following fact : [ prooffact : x in l(g)(i ) y in l(g)(i+1 ) gives edge ( x , y ) ] let @xmath524 @xmath202 @xmath534 and @xmath535 . then @xmath281 is an edge in @xmath536 this immediately follows from fact [ prooffact : m_c_g models varphi i to i+1 ] and the definition of @xmath537 by definition , it follows that @xmath538 and @xmath539 . from fact [ prooffact : c models varphi i to i+1 ] , we know that @xmath540 and hence there is an edge @xmath281 in @xmath260 as claimed . graphs fulfilling the formula @xmath541 are `` layered : '' whenever there are nodes @xmath26 and @xmath394 such that there is a path of length @xmath52 from the root to @xmath212 and of length @xmath542 to the node @xmath543 then there is an edge between @xmath26 and @xmath394 . from the definition , it is obvious that for any subgraph @xmath22 of @xmath260 with root @xmath527 that @xmath22 is canonical , if for every pair of natural numbers @xmath544 it follows that @xmath545 . we make a distinction between those elements in @xmath66 which lead to a canonical graph , and those which do not . in light of lemma [ lemma : modal restriction invariance with levels ] , it is natural that we do not need to look at the entire graph , but can ignore nodes which do not have `` short '' paths from the root of the graph leading to it . for a natural number @xmath546 we say that @xmath22 is _ @xmath547-canonical _ , if for all @xmath548 it holds that @xmath549 we define the following : @xmath550 it is obvious that @xmath551 . for each @xmath552 let @xmath553 denote natural numbers , and let @xmath554 denote some node such that @xmath555 with @xmath556 and @xmath557 is minimal with these properties . in particular , observe that @xmath557 and @xmath558 are polynomial in @xmath559 we now show that for @xmath552 the graph @xmath560 still has root @xmath527 unless it is empty . it particular , this means that it makes sense to ask if these graphs are @xmath182-canonical . [ prooffact : reduces models still rooted ] let @xmath552 such that @xmath560 is not empty . then @xmath560 has root @xmath561 assume that this is not the case , i.e. , that there is some @xmath562 and in this graph , there is no path from @xmath529 to @xmath19 . since @xmath19 is an element of @xmath563 a path from @xmath529 to @xmath19 exists in the original model @xmath27 . let @xmath564 be this path . since the path does not exist in the graph @xmath565 it follows that one of the @xmath566 must be an element of @xmath567 . since there is a path from this @xmath566 to @xmath95 it follows that @xmath19 is an element of @xmath567 as well , which is a contradiction . the nodes @xmath554 can be seen as `` minimal non - canonical points '' in @xmath525 . the following makes this more precise : let @xmath568 . then @xmath560 is @xmath182-canonical . due to fact [ prooffact : reduces models still rooted ] , we know that @xmath560 has root @xmath527 and by definition this graph is a subset of @xmath260 . assume that it is not @xmath182-canonical . then there exist natural numbers @xmath569 and some node @xmath570 . due to the minimality of @xmath571 it follows that @xmath572 and hence @xmath573 . in particular , it follows by induction on fact [ prooffact : x in l(g)(i ) y in l(g)(i+1 ) gives edge ( x , y ) ] that there is a path from @xmath574 to @xmath575 i.e. , @xmath576 . this is a contradiction . as our next connectivity result , we show the following : [ prooffact : forward and back edges ] let @xmath552 and let @xmath577 @xmath578 where @xmath579 and @xmath580 are natural numbers such that @xmath581 . then @xmath281 is an edge in @xmath536 since @xmath582 and @xmath583 by induction on fact [ prooffact : x in l(g)(i ) y in l(g)(i+1 ) gives edge ( x , y ) ] we know that @xmath584 and obviously this implies that for every multiple of @xmath585 a path of that length exists in @xmath27 from @xmath554 to @xmath554 itself . now assume that @xmath586 for some integer @xmath587 . we make a case distinction : case 1 : @xmath588 : : the above implies that @xmath589 and by choice of @xmath233 we also know that @xmath590 . since all of the involved nodes are elements of @xmath455 and from fact [ prooffact : c models varphi i to i+1 ] we know that @xmath591 for all @xmath592 this implies that there is an edge from @xmath26 to @xmath394 in @xmath99 as claimed . case 2 : @xmath593 : : by choice of @xmath594 we know that @xmath595 . by choice of @xmath233 the above and since @xmath596 is positive , we also know that @xmath597 . hence the existence of the edge @xmath281 again follows from fact [ prooffact : c models varphi i to i+1 ] . we therefore have the following structure of the model @xmath225 by definition , @xmath598 and by choice of @xmath85 every node in @xmath260 can be reached on a path from a node in @xmath66 . therefore , @xmath599 and hence @xmath27 can be written as the union of sub - graphs as follows : @xmath600 recall that the set @xmath66 is polynomial in @xmath496 . due to corollary [ corollary : suffices to restrict middle vertices for np ] , it suffices to restrict the `` middle part '' of this equation , i.e. , the components except @xmath210 and @xmath601 to polynomial size in order to obtain the desired polynomial model . in order to do this , we will now prove further connectivity results for the sub - models @xmath567 . it is important to note that due to proposition [ prop : trivial subgraph property ] , all of these submodels inherit all of the properties of their respective super - models which can be expressed by a universal first order formula . the idea behind the construction is the following : for each node in the original model , we add enough successors in our new model to ensure that the new model satisfies the conditions of lemma [ lemma : modal restriction invariance with levels ] . for the @xmath182-canonical submodels , we know that we can stop adding nodes at depth @xmath144 since other nodes can not be reached on a shorter path from the root . for the non-@xmath182-canonical submodels , we can not do this , but we also do not need to : due to fact [ prooffact : forward and back edges ] , we have `` circular '' edges , hence we know that nodes in a `` low '' level also are in a `` high '' level . for the construction , let @xmath602 and let @xmath603 for @xmath604 define @xmath605 ( note that in this case , by definition @xmath606 is a member of @xmath607 ) . for @xmath608 define @xmath609 note that by definition and fact [ prooffact : reduces models still rooted ] it follows that @xmath610 is a graph which is either empty or has root @xmath611 . it also follows from the above that for @xmath612 the graph @xmath610 is @xmath182-canonical . since the union over all @xmath610 is the same as the union over all @xmath525 ( for all @xmath613 ) , it follows that @xmath614 note that @xmath615 is polynomial in @xmath254 since @xmath66 is . we need one helpful fact about the non-@xmath182-canonical @xmath610-models : [ prooffact : non canonical models are union over restricted levels ] let @xmath616 i.e. , let @xmath610 be not @xmath182-canonical . then @xmath617 the inclusion @xmath618 is obvious , since @xmath619 . for the inclusion @xmath620 let @xmath621 . since , by definition , @xmath622 and hence @xmath623 there is a minimal @xmath624 such that @xmath625 . if @xmath626 then the claim holds . therefore , assume that @xmath627 and thus @xmath628 . since @xmath629 we also know that @xmath630 therefore let @xmath631 . by definition , it holds that @xmath632 and in particular , @xmath633 . by fact [ prooffact : forward and back edges ] , we know that there is an edge from @xmath631 to @xmath625 . this implies , by definition , that @xmath634 and by minimality of @xmath233 we know that @xmath635 and hence @xmath636 i.e. , @xmath637 a contradiction . we now define a series of models @xmath215 for @xmath638 which , step by step , integrate `` enough '' vertices of the original model @xmath27 to ensure that the formula @xmath32 still holds , but restrict the size to a polynomial . as the induction start , we define @xmath639 . for @xmath227 the construction is as follows : * add every world from @xmath640 to @xmath214 * add @xmath611 to @xmath214 * if @xmath641 i.e. , if @xmath610 is @xmath182-canonical , then for each @xmath642 and each formula @xmath643 if there is a world @xmath644 such that @xmath187 then add _ one _ of these worlds into @xmath645 * if @xmath646 i.e. , if @xmath610 is not @xmath182-canonical , then for each @xmath647 and each formula @xmath153 perform the following : * * if there is a world @xmath644 such that @xmath187 then add _ one _ of these worlds @xmath19 into @xmath215 . * * if there is a world @xmath20 which has been added into one of the @xmath182-canonical submodels in the step above , and there is a successor @xmath644 of @xmath20 such that @xmath648 then add _ one _ of these worlds @xmath19 into @xmath645 by construction , since @xmath558 and @xmath557 are polynomial in @xmath496 for each @xmath552 and @xmath220 there are only polynomially many worlds in @xmath649 . we now define @xmath145 to be the model @xmath650 . then the set of worlds in @xmath145 which have @xmath651-step successor and a @xmath481-step predecessor is polynomially bounded , since this is a subset of @xmath649 . hence , due to corollary [ corollary : suffices to restrict middle vertices for np ] , it suffices to show that @xmath152 in order to exhibit a model of @xmath32 which is polynomial in size . in order to prove this , we show that @xmath145 satisfies the conditions of lemma [ lemma : modal restriction invariance with levels ] . since @xmath652 by definition and @xmath653 @xmath33 is an element of @xmath145 by definition . therefore , let @xmath20 be an element of @xmath206 and let @xmath167 be a subformula of @xmath83 such that @xmath149 and let @xmath654 be a natural number such that @xmath655 and @xmath656 . it suffices to show that there is a world @xmath150 such that @xmath187 and @xmath68 is an edge in @xmath536 since @xmath149 there is a world @xmath657 such that @xmath476 is an edge in @xmath99 and @xmath658 . there are several cases to consider . if @xmath659 then by the construction of @xmath206 @xmath660 is an element of @xmath145 as well . hence we can choose @xmath19 to be @xmath660 . if @xmath518 and @xmath661 then it holds that either @xmath662 or @xmath663 . if @xmath664 then we know that @xmath665 and we can choose @xmath666 . hence assume that @xmath662 . in this case , by choice of @xmath85 there is a world @xmath667 such that @xmath668 and @xmath669 is an edge in @xmath27 . since @xmath670 we can choose @xmath671 . if @xmath672 then @xmath673 holds as well . therefore , it remains to consider the case @xmath674 . first , assume that there is no @xmath675 such that @xmath676 . since @xmath677 it follows that @xmath621 for some @xmath52 . since @xmath678 @xmath610 must be @xmath182-canonical , since all other @xmath610 contain all successors from @xmath260 to nodes in @xmath610 it follows that @xmath679 and therefore @xmath680 holds as well . since @xmath678 there is a non-@xmath182-canonical submodel @xmath681 such that @xmath682 and hence , due to fact [ prooffact : non canonical models are union over restricted levels ] , there is some @xmath683 such that @xmath684 . in this case , a node @xmath19 fulfilling the requirements has been added to the model @xmath145 due to the last condition in the construction . now assume that there is some @xmath685 such that @xmath676 . if it is possible to choose this @xmath52 in such a way that @xmath610 is not @xmath182-canonical , then we do so . in particular , since @xmath476 is an edge in @xmath99 there is some natural number @xmath587 such that @xmath686 and @xmath687 . if it is only possible to choose an @xmath52 leading to a @xmath182-canonical model @xmath688 then choose @xmath52 and @xmath587 in such a way that @xmath587 is minimal with this property . we make a case distinction . case 1 : @xmath616 i.e. , @xmath610 is not canonical . : : by fact [ prooffact : non canonical models are union over restricted levels ] , due to the minimality of @xmath233 we know that @xmath689 . now , since @xmath690 and @xmath691 it follows that @xmath692 . hence , by construction there is a @xmath693 such that @xmath187 and due to fact [ prooffact : x in l(g)(i ) y in l(g)(i+1 ) gives edge ( x , y ) ] , there is an edge @xmath68 in @xmath536 case 2 : @xmath694 i.e. , @xmath610 is @xmath182-canonical . : : if @xmath695 then , due to the construction of @xmath206 a world @xmath19 from @xmath696 satisfying @xmath167 was added in the construction , and due to fact [ prooffact : x in l(g)(i ) y in l(g)(i+1 ) gives edge ( x , y ) ] , @xmath68 is an edge in @xmath536 + therefore , assume that @xmath697 i.e. , @xmath698 . since @xmath610 is @xmath182-canonical , we know that @xmath20 can not be reached from @xmath611 with a path shorter than @xmath182 steps . due to the choice of @xmath52 and @xmath233 we also know that @xmath20 does not appear in a non-@xmath182-canonical submodel ( otherwise , since the non-@xmath182-canonical submodels are successor - closed there would be a non-@xmath182-canonical submodel @xmath610 containing both @xmath20 and @xmath699 and we would have chosen this ) , and that in each @xmath182-canonical submodel where @xmath20 appears , it has depth of at least @xmath700 . + if @xmath62 is satisfied on every strict line , we know that every node from @xmath260 can be reached from an element in @xmath66 with at most @xmath182 steps , which is a contradiction . + now consider the case that @xmath62 is not satisfied on every strict line . since in the @xmath182-canonical submodels , strict lines having the length of the depth of the submodel appear , and @xmath62 is satisfied in @xmath99 we know that these submodels can not have depth of more than @xmath446 . therefore , we know that @xmath701 a contradiction . hence , we know that @xmath191 concluding the proof of theorem [ theorem : above k to k+1 leads to np ] . we will now show that theorem [ theorem : above k to k+1 leads to np ] implies @xmath0-results for a number of related logics . the following theorem shows that modal logics for classes of frames which fulfill a natural generalization of the euclidean property have the polynomial - size model property , and hence can be solved in @xmath0 . note that the case @xmath702 of corollary [ corollary : main np result corollary ] follows from the main result of @xcite . our results and theirs are incomparable : they achieve the @xmath0-result for _ all _ normal modal logics extending what in our notation is @xmath703 where our results only hold for logics defined by universal horn clauses ( although the proof of theorem [ theorem : above k to k+1 leads to np ] in most cases gives the @xmath0-result for all extensions of the logic which are defined by universal formulas over the frame language the only exception is the case where the graph formula is satisfied on every strict line , note that this case needed special treatment in the proof of theorem [ theorem : above k to k+1 leads to np ] , and relies on the tree - like property for these logics proven later in theorem [ theorem : tree - like models for horn logics ] ) . we achieve @xmath0-results for many logics which are not extensions of @xmath703 but do not prove these results for all extensions of these logics . with the preceding theorem and the results on the implication for formulas of the form @xmath704 we obtain the following corollary : [ corollary : main np result corollary ] let @xmath62 be a universal horn formula implying @xmath381 for some @xmath705 such that one of the following conditions holds : * @xmath706 * @xmath707 and @xmath708 then @xmath63 has the polynomial - size model property , and @xmath444 first assume that @xmath709 . lemma [ lemma : varphi k to l implications ] shows that @xmath326 implies @xmath710 for some @xmath348 . from lemma [ lemma : varphi k to l implication gives genvarphi implications ] , we conclude that @xmath381 implies @xmath711 for some natural numbers @xmath712 and @xmath713 . therefore , @xmath62 also implies @xmath714 and due to theorem [ theorem : above k to k+1 leads to np ] , @xmath63 has the polynomial - size model property and @xmath444 for the case @xmath707 and @xmath715 observe that due to lemma [ lemma : genvarphi k to 0 implications ] , @xmath63 is also an extension of @xmath716 for some constants @xmath717 . due to the prerequisites , it holds that @xmath718 . hence , the result follows from the case above . corollary [ corollary : main np result corollary ] covers all cases obtained from formulas of the form @xmath381 where we possibly could expect the polynomial size property to hold : the requirement demanding that @xmath719 is crucial , since the formula @xmath720 is satisfied in any symmetric graph , and hence satisfiability problems for the corresponding logics are @xmath1-hard due to corollary [ corollary : ufos satisfied in symmetric tree are pspace hard ] . additionally , logics defined by formulas of the form @xmath415 are @xmath1-hard as well : this formula is satisfied in every reflexive , transitive graph , and therefore the complexity result follows from theorem [ theorem : ladner hardness cases for k - transitivity ] . we therefore have proven that for horn formulas defined by these single clauses , as soon as they are not satisfied in various combinations of strict , transitive , reflexive , and symmetric trees , they already imply the polynomial - size model property . up to now , we only considered the effect of a single universal horn clause . however , there are many cases where the logics defined by individual formulas have a @xmath1-hard satisfiability problem , but the complexity of the problem for logic defined by their conjunction drops to @xmath0 . a well - known example for such a case is the modal logic @xmath721 which is the logic over graphs which are both symmetric and transitive . on their own , both of these properties lead to @xmath1-hard logics , but their conjunction gives a logic which is in @xmath0 . in our notation , it can easily be seen that @xmath722 . for this concrete example and many generalizations , the following theorem gives this result . [ theorem : transitive and symmetric gives np generalization ] let @xmath723 and let @xmath724 . let @xmath62 be a universal horn formula such that @xmath62 implies @xmath725 . then @xmath63 has the polynomial size model property , and @xmath444 we first prove that @xmath726 implies @xmath727 if @xmath67 is odd , and @xmath728 if @xmath67 is even . let @xmath66 be a graph satisfying @xmath729 and let @xmath730 ( @xmath731 if @xmath67 is even ) and @xmath732 ( @xmath733 if @xmath67 is even ) be nodes in @xmath66 such that @xmath734 and @xmath735 are edges for all relevant @xmath52 . first , let @xmath67 be odd . since @xmath736 it follows that @xmath66 is symmetric . by applying the @xmath415-property @xmath369 times , we get a path of length @xmath369 from @xmath33 to @xmath737 . hence , since @xmath738 is an edge , it follows that @xmath739 and due to the @xmath415-property , this implies that @xmath740 is an edge as required . now let @xmath67 be even . since @xmath741 and @xmath715 the symmetry of @xmath66 ensures that there is a path of length @xmath67 from @xmath33 to @xmath742 . hence , there is an edge @xmath743 and due to the symmetry of @xmath85 an edge @xmath744 . by @xmath745 applications of the @xmath746-property , we know that @xmath747 . hence , we conclude that @xmath748 and another application of the @xmath415-property gives the edge @xmath749 as required . for the @xmath0-result , observe that @xmath725 implies @xmath750 where @xmath751 @xmath752 and @xmath753 . due to the above and lemma [ lemma : varphi k to l implication gives genvarphi implications ] , we know that this formula implies @xmath754 for some @xmath717 if @xmath67 is odd , and it implies @xmath755 for some @xmath756 if @xmath67 is even . further , if @xmath67 is odd , then it follows that @xmath757 . hence , @xmath758 . if @xmath67 is even , then , since @xmath715 we know that @xmath759 and in both cases the @xmath0 result follows from corollary [ corollary : main np result corollary ] . theorem [ theorem : transitive and symmetric gives np generalization ] concludes our results about logics defined by specific horn formulas . we now have collected all tools required to prove the main result of the paper , the complexity classification of satisfiability problems for logics defined by universal horn formulas . in this section , we show a dichotomy theorem , which classifies the complexity of the satisfiability problem for logics of the form @xmath760 where @xmath11 is a conjunction of universal horn clauses , into solvable in @xmath0 and @xmath1-hard . the classification is given in the form of the algorithm @xmath761 presented in figure [ figure : classification algorithm ] . in order to explain the algorithm , we need some more definitions . for a set @xmath762 we say that a graph @xmath66 satisfies the conditions of @xmath763 if it has the corresponding properties , i.e. , if @xmath764 then @xmath66 is required to be reflexive , if @xmath765 then @xmath66 is required to be symmetric , and if @xmath766 then @xmath66 is required to be @xmath67-transitive . a @xmath763-tree is a graph which can be obtained from a strict tree @xmath258 by adding exactly those edges required to make it satisfy the conditions of @xmath763 ( note that this is a natural closure operator ) . similarly , a @xmath763-line is the @xmath763-closure of a strict line . finally , for a universal horn clause @xmath276 let @xmath767 denote the pairs @xmath768 such that @xmath258 is a @xmath763-tree , and @xmath769 is a homomorphism . intuitively , due to proposition [ prop : homomorphism and horn clauses ] , this is the set of @xmath763-trees about which the clause @xmath62 makes a statement , along with the corresponding homomorphisms . we first define the cases of universal horn clauses leading to @xmath0-containment of the satisfiability problem : let @xmath62 be a universal horn clause , and @xmath770 . we say that @xmath771 satisfies the _ @xmath0-case _ , if one of the following occurs : 1 . @xmath284 for @xmath772 and there is @xmath773 such that there is no directed path connecting @xmath774 and @xmath775 in @xmath776 2 . @xmath777 and @xmath288 or the vertices from @xmath283 are different and not connected with an undirected path in @xmath294 ( this also applies if @xmath26 or @xmath394 do not appear in @xmath294 ) 3 . @xmath284 for @xmath778 and there is a homomorphism @xmath779 where @xmath780 is the @xmath763-line @xmath781 and there are @xmath782 such that @xmath783 such that @xmath784 and @xmath785 4 . @xmath284 for @xmath778 and there exist @xmath786-lines @xmath787 and @xmath788 and homomorphisms @xmath789 @xmath790 such that @xmath791 and @xmath792 @xmath793 and @xmath794 where @xmath708 we now define the properties of horn clauses which do not lead to @xmath0-containment of the satisfiability problems on their own . recalling section [ subsection : ladner ] , it is natural that clauses which are satisfied in every reflexive , transitive , or symmetric tree are among these . this is captured by the following definitions : if @xmath62 is a universal horn clause with @xmath284 for @xmath795 or @xmath778 we say that @xmath771 satisfies the _ reflexive case _ , if @xmath795 or for every @xmath796 such that @xmath286 is not an edge in @xmath776 it holds that @xmath797 . @xmath771 satisfies the _ transitive case for @xmath70 _ , if it does not satisfy the reflexive case , and for every @xmath798 such that @xmath286 is not an edge in @xmath776 there is a path from @xmath774 to @xmath775 in @xmath776 and there is some @xmath773 such that there is no edge @xmath799 and @xmath775 is exactly @xmath67 levels below @xmath774 in @xmath258 . finally , @xmath771 satisfies the _ symmetric case _ if it does not satisfy the reflexive of the transitive case , and for every @xmath773 such that @xmath286 is not an edge in @xmath776 there is an edge @xmath800 in @xmath258 . we can now state the classification theorem the proof will follow from the individual results in this section . note that the algorithm as stated can not be implemented directly , since it uses tests of the form if a given first - order formula is satisfied in certain infinite classes of graphs , and checks if certain elements are present in the infinite set @xmath767 . however , we believe that size - restrictions for the structures actually required to look at can be proven , and hence the algorithm hopefully can be implemented to give a deterministic decision procedure . however , the main usage of the algorithm is to show more general classification theorems , as we will see in corollary [ corollary : horn cunjunction without algorithm ] . [ theorem : horn conjunction classification ] let @xmath167 be a conjunction of universal horn clauses . then the complexity of @xmath801 is correctly determined by @xmath761 . a first look at the algorithm in figure [ figure : classification algorithm ] reveals that it is obviously necessary to prove that the choices that the algorithm has to make always can be made : in the relevant situations , at least one of the `` reflexive , '' `` transitive , '' or `` symmetric '' conditions occurs . however , before starting with the proof , we explain the general idea of the algorithm and give an example for a logic which @xmath761 proves to have a satisfiability problem in @xmath0 . in the variable @xmath802 the algorithm maintains a list of implications of the formula @xmath167 . for example , @xmath761 puts @xmath803 into @xmath763 if it detects the formula @xmath167 to require a graph satisfying it to be `` near - reflexive '' ( meaning , reflexive in all nodes with sufficient height and depth ) . similarly , @xmath804 means that @xmath11 requires a graph to be `` near - symmetric , '' and @xmath805 means that @xmath11 requires a graph to be `` near-@xmath67-transitive '' ( this will be made precise in lemma [ lemma : horn conjunction algorithm adds correctly to types - list ] ) . if at one point @xmath761 detects that a clause @xmath62 satisfies one of the @xmath0-conditions , then this means that the clause @xmath302 in addition with the requirements kept in @xmath802 implies a graph - property which leads to the polynomial - size model property . it is well - known that the modal logic over the class of frames which are both transitive and symmetric has a satisfiability problem in @xmath0 . generalizing this , when @xmath761 detects that @xmath11 requires `` near - symmetry '' and `` near-@xmath67-transitivity , '' this also leads to the polynomial - size model property and thus to @xmath0-membership of the satisfiability problem , applying theorem [ theorem : transitive and symmetric gives np generalization ] . l6.5 cm as an example , let @xmath62 be the universal horn clause with prerequisite graph as shown in figure [ fig : applications section example graph 1 ] , with @xmath806 and let @xmath11 be the horn formula having @xmath62 as its only conjunct . the algorithm @xmath761 starts with @xmath807 and hence in its first iteration , checks if @xmath62 is satisfied in every strict tree . this is not the case , as figure [ fig : applications section example graph 2 ] shows ( here , we simply marked each note with the names of the vertices which are preimages of the homomorphism ) : this is a homomorphic image of @xmath294 as a strict line ( which in particular is a strict tree ) , in which the images of @xmath26 and @xmath394 are not connected with an edge . therefore , @xmath62 is not satisfied in this line , and hence not in every strict tree . however , it is clear that if we want to map @xmath294 into a strict tree , then all the vertices between @xmath808 and @xmath503 must be `` pairwise identified , '' just like in figure [ fig : applications section example graph 2 ] . l6.5 cm therefore the transitive case is satisfied , and the figure shows that this is true for @xmath809 ( among possibly others ) . hence @xmath761 adds the element @xmath810 to @xmath763 . next it checks if @xmath62 is satisfied in every @xmath811-tree . this again is not the case , and the homomorphic image of @xmath294 as a @xmath811-line in figure [ fig : applications section example graph 3 ] shows that @xmath62 satisfies @xmath0-condition @xmath812 ( we only included those lines added by the @xmath810-closure that are required for our function to be a homomorphism ) . therefore , the logic @xmath63 has the polynomial - size model property , and its satisfiability problem is in @xmath0 . this example demonstrates that in the run of @xmath761 , a clause @xmath62 can meet different cases depending on the content of the variable @xmath813 in the situation that @xmath807 the clause @xmath62 satisfies the transitive case , but when @xmath814 this is no longer the case . -line ] as mentioned above , the first fact that we need to prove about @xmath761 is that it is well - defined , that is , the case distinction between the `` reflexive , '' `` transitive '' and `` symmetric '' cases is complete . [ lemma : big classification algorithm is well - defined ] @xmath761 is well - defined : in all relevant situations , at least one of the `` reflexive , '' `` transitive , '' and `` symmetric '' cases occurs . assume that the algorithm is not well - defined , and let @xmath11 be an instance for which the algorithm behavior is unspecified . let @xmath62 be a clause for which none of the @xmath0-conditions and none of the reflexive , transitive , and symmetric conditions hold . let @xmath763 be as determined by @xmath761 when encountering @xmath815 since the algorithm only chooses @xmath62 if @xmath62 is not satisfied on every @xmath763-tree , we know that due to proposition [ prop : homomorphism and horn clauses ] , there is some @xmath798 . now assume that @xmath288 or @xmath284 for @xmath816 and @xmath817 do not both appear in @xmath294 . then @xmath0-condition @xmath69 is satisfied , and we have a contradiction . therefore we know that @xmath62 is a universal horn clause such that @xmath284 for some variables @xmath817 such that @xmath795 or @xmath818 since the reflexive case does not apply , we know that @xmath816 and hence @xmath819 since @xmath436-condition @xmath197 does not apply , we know that for every @xmath820 there is a directed path connecting @xmath774 and @xmath821 since the transitive case does not apply , we know that there is some @xmath822 in which there is no path from @xmath774 to @xmath775 . we therefore know , by the above , that there is a path from @xmath775 to @xmath774 in @xmath258 . hence , @xmath258 is not symmetric . since @xmath258 is a @xmath763-tree , we know that @xmath823 since there is a path from @xmath775 to @xmath774 in @xmath258 but not vice versa , we know that @xmath824 and since there is a path from @xmath775 to @xmath774 in the ( possibly reflexive and/or @xmath72-transitive for some set @xmath72 ) tree @xmath776 this implies that @xmath774 is on a lower level in @xmath258 than @xmath775 . let @xmath67 denote the difference in levels between @xmath774 and @xmath825 then @xmath826 . by mapping @xmath258 into a @xmath763-line @xmath787 with the homomorphism @xmath318 assigning each vertex @xmath20 the element @xmath827 where @xmath52 is the level of @xmath20 in @xmath776 we can , using the homomorphism @xmath828 map @xmath294 homomorphically into @xmath829 in such a way that @xmath830 and @xmath831 where @xmath52 denotes the level of @xmath774 in @xmath832 if @xmath715 then the line @xmath829 satisfies @xmath0-condition @xmath833 a contradiction . therefore , we know that @xmath834 and thus @xmath835 and @xmath829 satisfy @xmath0-condition @xmath836 . since @xmath62 also does not satisfy the symmetric case , we know that there is some @xmath773 such that there is no edge @xmath286 and no edge @xmath800 in @xmath258 . by @xmath436-condition @xmath426 we know that there is a directed path connecting @xmath774 and @xmath775 in @xmath258 . since @xmath837 and @xmath258 is a @xmath786-tree , we know that @xmath258 is not symmetric . if @xmath764 and @xmath258 is a @xmath763-tree , this implies that @xmath258 is reflexive , and hence @xmath838 . if @xmath839 then we know that , since @xmath840 also is not an element of @xmath802 that no node in @xmath258 is connected to itself with a directed path , and therefore we also know that @xmath838 . now assume that there is a path from @xmath775 to @xmath774 in @xmath258 . then the shortest of these paths must have length @xmath719 ( since there is no edge @xmath800 and therefore we can map @xmath258 ( and therefore @xmath294 ) homomorphically into a line @xmath829 as in the case above , where the distance of the image of @xmath774 and @xmath775 is @xmath65 and since @xmath775 is mapped to a predecessor of @xmath841 this line satisfies @xmath0-condition @xmath833 a contradiction . therefore we know that in @xmath776 there is a directed path from @xmath774 to @xmath775 . since @xmath286 is no edge in @xmath776 we know that the shortest of these paths must have length @xmath842 . since @xmath258 is not symmetric , this implies that @xmath775 must be on a lower level in @xmath258 than @xmath841 and the difference in levels is @xmath842 . similarly , to the above , we can construct a @xmath786-line @xmath843 and a homomorphism @xmath844 such that for @xmath845 if @xmath52 is the level of @xmath19 in @xmath776 then @xmath846 . let @xmath847 denote the homomorphism @xmath323 . due to the choice of @xmath67 and the definition of @xmath848 we know that @xmath849 and @xmath850 for some @xmath212 and hence @xmath847 and @xmath851 satisfy @xmath0-condition @xmath836 . since above , we already constructed @xmath835 and @xmath829 satisfying @xmath0-condition @xmath852 this implies that @xmath62 satisfies @xmath0-condition @xmath852 a contradiction . also note that in the transitive case , a @xmath719 can always be chosen : since by construction of the algorithm , @xmath62 is not satisfied on every @xmath786-tree , there is some @xmath768 such that @xmath286 is not an edge in @xmath258 . since the reflexive case does not apply , we can choose @xmath768 in such a way that @xmath838 . due to the transitive case , there is a path from @xmath774 to @xmath775 in @xmath776 and hence the shortest of these paths must have length of at least @xmath375 hence the difference of levels is at least @xmath69 ( we know that even if @xmath765 @xmath775 must be below @xmath774 in @xmath776 otherwise @xmath0-condition @xmath812 would apply ) . now that we know that @xmath761 is well - defined , we show that it always comes to a halt and hence , generates an answer this is not immediate : note that in the example formula from figure [ fig : applications section example graph 1 ] discussed above , we already saw that it is possible for the algorithm to revisit a clause . therefore one might consider it possible for the algorithm to repeatedly add the same element to the variables @xmath763 without coming to a halt , or the list @xmath763 to grow infinitely . in order to show that this does not happen , we prove a bound on the number of the elements that can be added into this list during the run of the algorithm . [ lemma : path from x to y in transitive case ] let @xmath62 be a clause in a universal horn formula @xmath11 such that @xmath62 satisfies the transitive case for a list @xmath763 such that @xmath853 for some @xmath854 and @xmath855 . then there is a path from @xmath26 to @xmath394 in @xmath856 the main idea of the proof is to show that if there is no path from @xmath26 to @xmath394 in @xmath292 then these variables are `` independent '' enough in @xmath294 to satisfy @xmath0-condition @xmath833 which then is a contradiction , since the clause satisfies the transitive case . by the prerequisites , since @xmath62 satisfies the transitive case , @xmath62 is not satisfied in every @xmath763-tree , and hence in particular there is some @xmath773 such that @xmath286 is not an edge in @xmath258 . since @xmath4 satisfies the transitive case , there is a path from @xmath774 to @xmath775 in @xmath258 . by the prerequisites , we know that there is some @xmath857 . we then know that @xmath858 since @xmath859 is never adder to @xmath763 ( note that this is tautological , since any graph is @xmath197-transitive ) . if @xmath258 were reflexive , then by proposition [ prop : k - transitivity facts ] , we know that @xmath258 would also be @xmath69-transitive , i.e. , transitive , and hence there would be an edge @xmath286 in @xmath776 a contradiction . hence we know that @xmath258 is not reflexive , and therefore @xmath839 since @xmath258 is a @xmath763-tree . since by the prerequisites , we know that @xmath837 we know that @xmath258 is also not symmetric . since there is a path from @xmath774 to @xmath825 and @xmath258 is neither symmetric nor reflexive , we know that @xmath775 is on a lower level in @xmath258 that @xmath774 . we also know that there is no path from @xmath775 to @xmath774 in @xmath258 . let @xmath860 be the difference of levels between @xmath774 and @xmath775 in @xmath258 . since we know that there is no edge @xmath286 in @xmath776 we know that @xmath861 now define a partial order @xmath862 on the variables in @xmath302 by @xmath863 if @xmath864 or if there is a path from @xmath20 to @xmath19 in @xmath294 . since @xmath294 can be mapped into a @xmath763-tree which is acyclic , we know that @xmath862 is anti - symmetric . since @xmath862 is obviously reflexive and transitive , this implies that @xmath862 is a partial order . now map @xmath258 into a @xmath786-list @xmath865 with the homomorphism @xmath848 such that for all @xmath845 @xmath866 where @xmath52 denotes the level of @xmath19 in @xmath258 . from the above we know that @xmath867 and @xmath868 for some @xmath869 let @xmath870 such that @xmath871 . such am @xmath49 exists , since @xmath872 . let @xmath873 be the @xmath786-line of length @xmath874 and define a function @xmath875 by @xmath876 we show that @xmath877 is a homomorphism . for this , let @xmath68 be an edge in @xmath294 . since @xmath323 is a homomorphism , we know that @xmath878 is an edge in @xmath829 . we make a case distinction : @xmath879 and @xmath880 : : then , @xmath881 and @xmath882 have the same position in @xmath851 as @xmath883 and @xmath884 have in @xmath829 . since @xmath851 restricted to @xmath885 is isomorphic to @xmath886 this implies that @xmath887 is an edge in @xmath888 as required . @xmath889 and @xmath890 : : in this case , @xmath881 and @xmath882 have the same positions in @xmath851 restricted to @xmath891 as @xmath883 and @xmath884 have in @xmath829 . since the restricted line is isomorphic to @xmath886 we again conclude that @xmath887 is an edge in @xmath892 @xmath889 and @xmath880 : : since @xmath68 is an edge in @xmath292 this implies @xmath863 . since @xmath862 is transitive , this implies @xmath893 a contradiction . @xmath879 and @xmath894 : : let @xmath895 and let @xmath896 . then we know that @xmath897 is an edge in @xmath829 . therefore , since @xmath898 and @xmath885 are isomorphic , we also know that there is an edge @xmath899 in @xmath851 . in @xmath888 there obviously is a @xmath900-step path from @xmath901 to @xmath902 . therefore , in @xmath888 there is a @xmath903-step path from @xmath904 to @xmath902 . since @xmath853 and @xmath851 is a @xmath763-list , we know that @xmath851 is @xmath905-transitive , and by proposition [ prop : k - transitivity facts ] , this implies that @xmath851 is also @xmath906-transitive . therefore , there is an edge @xmath907 . by definition of @xmath908 since @xmath909 @xmath910 @xmath895 and @xmath911 we know that @xmath912 and @xmath913 . thus , from the above we know that @xmath887 is an edge in @xmath892 since @xmath887 is an edge in @xmath851 in all cases , we know that @xmath914 is indeed a homomorphism . now assume that there is no path from @xmath26 to @xmath394 in @xmath294 . since @xmath62 satisfies the transitive case , we know by the construction of @xmath761 that @xmath62 does not satisfy the reflexive case , and hence in particular , @xmath915 . therefore we know that @xmath916 . since @xmath917 obviously holds , and since @xmath867 and @xmath918 we know that @xmath919 since , by choice of @xmath920 @xmath921 this implies that @xmath877 and the @xmath763-line @xmath851 satisfy @xmath0-condition @xmath833 a contradiction . therefore we know that there is a path from @xmath26 to @xmath394 in @xmath292 as claimed . a first helpful tool to show this is the following proposition , which says that there are no `` redundant '' transitivity conditions which get added to @xmath786 . [ prop : algorithm does not add already implied transitivity ] let @xmath922 be added to the variable @xmath763 by @xmath761 . then @xmath763 did not imply @xmath67-transitivity before adding @xmath923 assume that @xmath763 already implied @xmath922 . since @xmath922 is added by @xmath761 , there is a clause @xmath62 with conclusion edge @xmath281 for @xmath818 which satisfies the transitive condition for @xmath67 at this point of the algorithm s run . therefore , there is some @xmath773 such that @xmath286 is not an edge in @xmath776 and @xmath775 is exactly @xmath67 levels below @xmath774 . since by the assumption the conditions of @xmath763 already imply @xmath67-transitivity , the @xmath67-step path from @xmath774 to @xmath775 implies that @xmath286 is an edge in @xmath776 a contradiction . now that we know that transitivity conditions which were already implied by the conditions present in @xmath786 do not get added to the list , there is a clear strategy how we can prove that @xmath786 does not grow infinitely : we first prove just how many transitivity conditions are implied by the conditions in @xmath786 , and then show that from a certain point on , everything that will be added to @xmath786 by @xmath761 already is implied . in order to prove this , we need some technical results about implications of various forms of @xmath72-transitivity . [ lemma : test case 1 ] let @xmath924 . then @xmath925-transitivity implies @xmath926-transitivity for all @xmath927 we show the claim by induction on @xmath928 . for @xmath929 this holds trivially , since @xmath925-transitivity by definition implies @xmath860-transitivity . now assume that it holds for @xmath930 let @xmath66 be some graph which is @xmath925-transitive , and let @xmath931 be nodes in @xmath66 such that @xmath501 is an edge for all relevant @xmath52 . we need to show that @xmath932 is an edge in @xmath66 . since by the induction hypothesis , @xmath925-transitivity implies @xmath926-transitivity , we know that @xmath66 is @xmath926-transitive . therefore there is an edge @xmath933 in @xmath66 . by choice of @xmath934 there is a @xmath935-step path from @xmath936 to @xmath937 . therefore there is a @xmath905-step path from @xmath938 to @xmath939 and since @xmath66 is @xmath905-transitive , this implies that @xmath940 is an edge in @xmath85 as required . [ lemma : test case 2 ] let @xmath941 such that @xmath942 . then @xmath943-transitivity implies @xmath944-transitivity for all @xmath927 let @xmath945 for some @xmath946 @xmath947 . then @xmath948 . since @xmath949 we know that @xmath943-transitivity implies @xmath950-transitivity . by lemma [ lemma : test case 1 ] , we know that @xmath951-transitivity implies @xmath952-transitivity , and since @xmath905-transitivity is obviously implied by the prerequisites , this proves the lemma . [ lemma : test case 4b ] let @xmath953 and let @xmath954 such that @xmath955 and for @xmath544 @xmath956 . then there exists some @xmath957 such that @xmath958 such that @xmath959-transitivity implies @xmath944-transitivity for all @xmath927 let @xmath960 and let @xmath961 where @xmath962 . due to lemma [ lemma : test case 1 ] , @xmath963-transitivity implies @xmath964-transitivity for all @xmath965 . therefore we can choose @xmath966 and we know that @xmath72-transitivity implies @xmath967-transitivity . since for all @xmath930 @xmath968 is equivalent to @xmath969 modulo @xmath970 and modulo @xmath971 ( since @xmath971 divides @xmath970 ) , we can assume without loss of generality that @xmath972 for all @xmath973 . let @xmath974 where @xmath975 . such numbers @xmath976 exist , since @xmath977 for all @xmath782 . since @xmath978 we know that @xmath979 and therefore @xmath980 for all @xmath981 define @xmath982 . note that since @xmath971 divides @xmath983 this implies that @xmath984 . for the @xmath969 it holds that @xmath985 and again , since @xmath971 divides @xmath983 we have that @xmath986 . now since @xmath974 it follows that @xmath987 is equivalent to @xmath424 modulo @xmath988 and hence @xmath49 and @xmath969 are equivalent modulo @xmath971 for all @xmath52 . now note that @xmath989 therefore , since @xmath72-transitivity implies @xmath969-transitivity for all @xmath212 we know that @xmath72-transitivity implies @xmath990-transitivity for all @xmath52 . since @xmath991 and for @xmath544 it also holds that @xmath992 ( otherwise it would follow that @xmath993 and hence @xmath994 a contradiction ) , it follows that @xmath995 . we therefore know that for each @xmath996 @xmath72-transitivity implies @xmath997-transitivity , and since @xmath998 the claim follows from lemma [ lemma : test case 2 ] . the following lemma is the final of our results about implied transitivity conditions . its technical formulation hides the fact that the statement of the lemma is actually rather natural . as an example , consider the case where @xmath999 and @xmath1000 . then @xmath1001 and then the lemma says that there is some odd number @xmath49 such that for all odd numbers @xmath1002 @xmath1003-transitivity implies @xmath1004-transitivity . the key idea is that then , once @xmath1005 and @xmath1006 are elements of @xmath802 we know that due to proposition [ prop : algorithm does not add already implied transitivity ] , @xmath761 does not add any more @xmath922-conditions to @xmath786 anymore for odd @xmath67s which are greater than or equal to @xmath49 . therefore there are only finitely many odd @xmath67s such that @xmath761 can add @xmath922 from this point on . if @xmath761 would add an infinite number of @xmath922-conditions , then one of them must therefore be one where @xmath67 is even . assume that this already happens with @xmath1007 i.e. , that @xmath1008 is an even number . then @xmath1009 is odd , and therefore the greatest common divisor of @xmath1010 and @xmath1009 is @xmath197 . by the lemma , we therefore know that there is some @xmath49 such that @xmath1011-transitivity implies @xmath1012-transitivity for all @xmath1013 and then @xmath761 can only add @xmath922-conditions for @xmath1014 and this is only a finite number of possibilities . thus we know that @xmath761 must stop adding transitivity conditions at some point . with the above proof - strategy in mind , it is clear that the statement of the following lemma is crucial in restricting the number of @xmath922-conditions added by @xmath761 . [ lemma : test case 5 ] let @xmath1015 and let @xmath1016 . then there exists some @xmath957 such that @xmath1017 and @xmath1018-transitivity implies @xmath944-transitivity for all @xmath927 let @xmath1019 . we show the claim inductively on @xmath103 . if @xmath1020 then @xmath1021 and by lemma [ lemma : test case 1 ] , @xmath1022-transitivity implies @xmath1023-transitivity for all @xmath592 hence the claim follows with @xmath1024 since we need the case @xmath1025 explicitly in the induction , we prove it individually . hence let @xmath1026 and let @xmath1027 where @xmath1028 . from lemma [ lemma : test case 1 ] , we know that @xmath925-transitivity implies @xmath1029-transitivity for all @xmath1030 . now note that for all @xmath1031 @xmath1032 due to lemma [ lemma : test case 4b ] , it suffices to show that all of these @xmath969 are equivalent to @xmath197 modulo @xmath988 and for @xmath1033 @xmath956 . it is obvious that all @xmath969 are equivalent to @xmath197 modulo @xmath971 . now assume that there are @xmath1034 such that @xmath1035 . then the following holds : @xmath1036 this is a contradiction , since due to the above , we know that @xmath1037 . this completes the proof for the case @xmath1038 now inductively assume that the lemma holds for @xmath1039 . let @xmath1040 . recall that @xmath1041 . we first show that @xmath1042 which is a standard fact from number theory . obviously , @xmath971 divides both @xmath1043 and @xmath1044 and hence @xmath1045 . on the other hand , let @xmath529 be a common divisor of @xmath1046 and @xmath1047 . since @xmath529 divides @xmath1044 @xmath529 is also a divisor of @xmath1048 and since @xmath529 divides @xmath1049 it is a common divisor of all @xmath1050 . therefore , @xmath1051 . in particular , this holds for @xmath1052 and therefore @xmath1053 . since the other direction holds due to the above , we have shown that @xmath1054 due to the induction hypothesis , we know that there is a natural number @xmath1012 such that @xmath1055 and for all @xmath1056 @xmath72-transitivity implies @xmath1057-transitivity . let @xmath1058 and let @xmath1059 be a natural number such that @xmath1060 . then , by choice of @xmath1061 we know that @xmath72-transitivity implies @xmath1057-transitivity . let @xmath1062 . since @xmath72-transitivity also implies @xmath1043-transitivity , we know from the case @xmath1025 that there exists a natural number @xmath49 such that @xmath1063 and for all @xmath592 @xmath72-transitivity implies @xmath1064-transitivity . in order to prove the lemma , it therefore suffices to show that @xmath1065 . it holds that @xmath1066 since @xmath971 is a divisor of @xmath1047 and of @xmath1049 it follows that @xmath1067 . on the other hand , @xmath1068 is a common divisor of @xmath1069 and @xmath1046 . since @xmath1070 we know that @xmath1068 divides @xmath1047 . since @xmath1068 also divides @xmath1049 this implies that @xmath1068 is a divisor of @xmath1071 . therefore @xmath1072 and @xmath1073 and hence we have proven that @xmath1074 as claimed . we can now prove that there are only finitely many additions of the form @xmath922 during the algorithm s run , using the arguments from the discussion before lemma [ lemma : test case 5 ] : [ lemma : algorithm adds only finitely many transitivity conditions ] let @xmath11 be a universal horn formula . then @xmath1075 on input @xmath10 only adds finitely many conditions of the form @xmath922 to @xmath1076 assume that this is not the case . due to proposition [ prop : algorithm does not add already implied transitivity ] , we know that each added @xmath922-element is not implied by the previous elements , in particular , no element is added twice . hence there is an infinite sequence @xmath1077 such that all @xmath1078 are added to @xmath802 they are added in this order , and for @xmath1079 we have @xmath1080 for @xmath1081 let @xmath1082 . then @xmath1083 is obviously decreasing , and bounded by @xmath197 . therefore , the sequence converges , i.e. , there is some @xmath1084 such that @xmath1085 for all @xmath1086 . due to lemma [ lemma : test case 5 ] , we know that there is some @xmath49 such that @xmath1087-transitivity implies @xmath944-transitivity for all @xmath592 and @xmath1017 . since the sequence @xmath1088 is infinite and no element is repeated , there is some @xmath1086 such that @xmath1089 . since @xmath1090 we know that @xmath971 divides @xmath1091 . therefore , let @xmath1092 . since @xmath1093 let @xmath1094 . since @xmath1095 we know that @xmath1096 . hence it follows that @xmath1097 due to the choice of @xmath49 and since @xmath1098 we therefore know that @xmath1099-transitivity implies @xmath1100-transitivity . this is a contradiction to proposition [ prop : algorithm does not add already implied transitivity ] , since @xmath1100 is added after @xmath1101 . therefore , there are only finitely many elements of the form @xmath922 added by the algorithm . since we have now restricted the number of elements @xmath922 added to @xmath802 we can prove that the algorithm halts on any input . @xmath761 always halts . assume that it does not halt for an instance @xmath11 . due to lemma [ lemma : algorithm adds only finitely many transitivity conditions ] , we know that there are only finitely many elements of the form @xmath922 which are added to @xmath763 by @xmath761 . since the only other elements which can be added to @xmath763 are @xmath1102 and @xmath1103 this implies that there are only finitely many elements of any type which are added to @xmath763 . now let @xmath763 be as determined by the algorithm . since the size of @xmath763 is bounded and the variable never shrinks , this is well - defined . since the algorithm does not halt , we know that there is a clause @xmath62 in @xmath11 which is not satisfied in every @xmath763-tree , and @xmath62 does not satisfy any of the @xmath836 @xmath0-conditions . thus there is some @xmath798 such that @xmath62 is not satisfied in @xmath258 . since we already showed that @xmath761 is well - defined , and none of the @xmath0-cases applies ( otherwise the algorithm would come to a halt ) , we know that one of the reflexive , transitive , or symmetric cases applies . in particular , @xmath806 and @xmath795 or @xmath818 and @xmath286 is not an edge in @xmath258 . if one of the cases corresponding to an element not in @xmath763 applies , then this leads to an enlargement of @xmath802 a contradiction to the choice of @xmath763 . hence one of the cases corresponding to one of the elements already in @xmath763 applies . we make a case distinction : case 1 : the reflexive case applies . : : in the case that @xmath1104 the clause is trivially satisfied in the reflexive graph @xmath258 . assume that @xmath915 . in this case , @xmath797 . but since @xmath764 and @xmath1105 we know that @xmath258 is a reflexive tree . hence , the edge @xmath286 exists in @xmath776 a contradiction . case 2 : the transitive case applies for some @xmath1106 : : in this case , there is a path of length @xmath67 from @xmath774 to @xmath775 in @xmath258 . but since @xmath766 and @xmath1105 we know that @xmath258 is a @xmath67-transitive tree . hence , the edge @xmath286 exists in @xmath776 a contradiction . case 3 : the symmetric case applies . : : in this case , there is an edge from @xmath775 to @xmath774 in @xmath258 . but since @xmath765 and @xmath1105 we know that @xmath258 is a symmetric tree . hence , the edge @xmath286 exists in @xmath776 a contradiction . since we have a contradiction in each case , this completes the proof . now that we know that the algorithm is both well - defined and comes to a halt , it remains to prove its correctness . the case in which the algorithm states @xmath1-hardness is easily seen to be correct : [ lemma : horn conjunction algorithm correct on pspace ] if for a universal horn formula @xmath10 @xmath761 states that @xmath75 is @xmath1-hard , then this is true . the only possibility for the algorithm to state that the problem is @xmath1-hard is when the while - loop does not discover any clauses @xmath62 anymore which are not satisfied in every @xmath763-tree , and @xmath763 does not contain both @xmath840 and @xmath1107 for any @xmath70 . since each clause @xmath62 in @xmath11 is satisfied in every @xmath763-tree , this implies that the conjunction @xmath11 is also satisfied in each @xmath763-tree . hence one of the following cases occurs : * @xmath11 is satisfied in every strict tree ( if @xmath1108 ) , * @xmath11 is satisfied in every reflexive tree ( if @xmath1109 ) , * there is a set @xmath74 such that @xmath11 is satisfied in every @xmath72-transitive tree ( if @xmath1110 ) , * @xmath11 is satisfied in every symmetric tree ( if @xmath1111 ) , * @xmath11 is satisfied in every tree which is both reflexive and symmetric ( if @xmath1112 ) , * there is a set @xmath74 such that @xmath11 is satisfied in every tree which is both reflexive and @xmath72-transitive ( if @xmath1113 ) . in each of these cases , the hardness result follows directly from theorem [ theorem : ladner hardness cases for k - transitivity ] . we are now interested in the @xmath0-cases . we first show that if @xmath761 adds one of the requirements @xmath1114 or @xmath1107 for some @xmath67 to the list @xmath802 then the formula @xmath11 requires any graph @xmath66 satisfying @xmath11 to have the corresponding property ( except for vertices with insufficient depth or height in the graph ) . more precisely , we show the following lemma : [ lemma : horn conjunction algorithm adds correctly to types - list ] let @xmath11 be a universal horn formula , and let @xmath1115 be added to @xmath763 by @xmath761 on input @xmath11 . then the following holds : * if @xmath1116 then @xmath11 implies @xmath1117 for some @xmath1118 * if @xmath1119 then @xmath11 implies @xmath1120 for some @xmath1118 * if @xmath1121 then @xmath11 implies @xmath1122 for some @xmath1123 inductively assume that all values added before @xmath1124 ( if any ) were `` correct '' in the sense that they satisfy the conditions of the lemma . let the content of the variable @xmath763 directly before adding @xmath1124 be denoted with @xmath1125 . let @xmath62 be the clause in @xmath11 for which @xmath1124 was added . then , by the construction of the algorithm , we know that @xmath62 satisfies the case in the algorithm corresponding to @xmath1126 it particular , @xmath284 for some variables @xmath795 or @xmath818 . first assume that @xmath819 we also know that @xmath62 is not satisfied in every @xmath1125-tree . hence , let @xmath1127 such that @xmath286 is not an edge in @xmath258 . this must exist due to proposition [ prop : homomorphism and horn clauses ] . by construction of @xmath761 , in the case that @xmath1128 we have that @xmath1129 and in the case that @xmath1130 we can choose @xmath768 in such a way that @xmath775 is is @xmath67 levels below @xmath774 in @xmath258 . finally , in the case that @xmath1131 we know that there is an edge @xmath800 in @xmath258 . since the reflexive condition does not apply , we can choose @xmath768 in such a way that @xmath824 and since @xmath1132 this implies that @xmath774 is in a lower level of @xmath258 than @xmath775 . since all additions before @xmath1124 were correct , we know that all the types in @xmath1125 have been added correctly , i.e. , we can assume that there is some @xmath251 such that * if @xmath1133 then @xmath11 implies @xmath1117 for some @xmath1118 * if @xmath1134 then @xmath11 implies @xmath1120 for some @xmath1118 * if @xmath1135 then @xmath11 implies @xmath1122 for some @xmath1123 we can use the same value @xmath251 for all formulas here by taking the maximum , since clearly , @xmath381 implies @xmath1136 if @xmath1137 @xmath1138 and @xmath1139 . now let @xmath1140 be the homomorphic image of @xmath258 under the homomorphism @xmath1141 as a @xmath1125-line , where @xmath1141 assigns each @xmath1142 the node @xmath827 with @xmath52 being the level of @xmath19 in @xmath258 . let @xmath774 be in the @xmath654th level of @xmath776 and let @xmath775 be in the @xmath1143th level of @xmath258 ( where @xmath103 is the height of @xmath258 ) . due to the construction of @xmath1144 we know that @xmath1145 and @xmath1146 . first note that : * if @xmath1128 then @xmath1129 and hence @xmath1147 . it follows that @xmath1148 * if @xmath1130 then by choice of @xmath776 @xmath775 is exactly @xmath67 levels below @xmath26 in @xmath776 and therefore we know that @xmath1149 . * if @xmath1131 then , since @xmath1132 we know that in @xmath1150 we only have edges between non - decreasing nodes . since the symmetric case applies , we know that @xmath800 is an edge in @xmath258 . since @xmath824 @xmath774 is at a lower level than @xmath775 in @xmath776 and it follows that @xmath1151 . if @xmath1152 then it follows that the line @xmath780 satisfies @xmath0-condition @xmath833 a contradiction . therefore we know that @xmath1153 we now show that @xmath1154 for this , let @xmath66 be a graph such that @xmath137 by induction we know that the nodes of @xmath66 which have both a @xmath251-step predecessor and a @xmath251-step successor satisfy the conditions of @xmath1125 . to prove the claim , let @xmath1155 be elements of @xmath66 which have both an @xmath1156-step predecessor and an @xmath1156-step successor , and @xmath1157 in order to prove the lemma , we need to prove that @xmath68 is an edge in @xmath66 . since @xmath137 and @xmath62 is a clause in @xmath10 it suffices to prove that there is a homomorphism @xmath1158 such that @xmath1159 and @xmath1160 . then there is an edge @xmath68 in @xmath66 due to proposition [ prop : homomorphism and horn clauses ] . since @xmath1161 is a homomorphism and @xmath1162 and @xmath1163 it suffices to show that there is a homomorphism @xmath1164 such that @xmath1165 and @xmath1166 . then the homomorphism @xmath1167 satisfies the necessary conditions . let @xmath33 be an @xmath654-step predecessor of @xmath20 in @xmath66 such that @xmath33 has a @xmath251-step predecessor in @xmath85 and let @xmath503 be a @xmath547-step successor of @xmath19 in @xmath66 which has a @xmath251-step successor in @xmath66 . both must exist due to the choice of @xmath1155 . additionally , if @xmath1168 and therefore @xmath1169 is an edge in @xmath85 choose @xmath33 to be an @xmath1170-step predecessor of @xmath19 ( in which case it is also a @xmath654-step predecessor of @xmath20 ) , and in this case also let @xmath503 to be a @xmath1171-step successor of @xmath20 ( in which case it is also a @xmath547-step successor of @xmath19 ) . now , define @xmath1172 and let @xmath1173 from @xmath66 be chosen in such a way that @xmath1174 and @xmath1175 and for all relevant @xmath212 @xmath735 is an edge in @xmath66 . this is possible since @xmath20 and @xmath19 satisfy the conditions corresponding to @xmath1176 * if @xmath1128 we know that @xmath1177 and @xmath864 . hence the nodes @xmath904 can be chosen satisfying the demanded conditions by choosing @xmath1178 to be the nodes on the @xmath654-step - path from @xmath33 to @xmath1179 and @xmath1180 to be the nodes on the @xmath547-step path from @xmath864 to @xmath517 * if @xmath1130 we know that @xmath1181 and we can choose @xmath1178 to be the nodes on the path from @xmath33 to @xmath195 @xmath1182 to denote the @xmath67-step path from @xmath20 to @xmath19 ( which exists due to the choice of @xmath20 and @xmath19 ) , and @xmath1180 be the nodes on the @xmath547-step path from @xmath19 to @xmath517 * if @xmath1131 then by the above we know that @xmath1183 and we know that there is an edge @xmath1169 in @xmath66 . we also know that in this case , @xmath33 is an @xmath1170-step predecessor of @xmath19 . hence we can choose the nodes in the following way : let @xmath1184 be chosen as the nodes on the @xmath1170-step path from @xmath33 to @xmath95 and let @xmath1185 be the nodes on the @xmath1171-step path from @xmath20 to @xmath503 . since @xmath1169 is an edge in @xmath85 this gives the edge @xmath1186 which is required since @xmath1187 we now construct the homomorphism @xmath1188 for each relevant @xmath212 let @xmath1189 . then by construction , @xmath1165 and @xmath1166 . hence it remains to prove that @xmath318 is a homomorphism . since @xmath780 is a _ prev - types - list_-line , let @xmath1190 be a strict line such that @xmath780 is the @xmath1125-closure of @xmath1190 . since @xmath735 is an edge in @xmath66 for all relevant @xmath212 it follows that @xmath1164 is a homomorphism . since every @xmath904 has a @xmath251-step successor and a @xmath251-step predecessor in @xmath85 we know that the subgraph @xmath1191 satisfies the conditions from @xmath1192 and since @xmath780 is the @xmath1193-closure of @xmath1194 this implies that for every edge present in @xmath1150 the images of the corresponding vertices are also connected with an edge in @xmath85 and hence @xmath318 is indeed a homomorphism , finishing the proof of the lemma for the case that @xmath819 now assume that @xmath1195 or @xmath1196 and therefore @xmath795 . since there obviously is a homomorphism @xmath769 for some @xmath1125-tree @xmath776 the clause @xmath62 forces every node in a graph containing a @xmath763-line of sufficient length to be reflexive . since due to the induction hypothesis , we know that every model of @xmath167 of sufficient depth contains arbitrary long @xmath1125-lines , this concludes the proof for the remaining case @xmath1197 due to lemma [ lemma : horn conjunction algorithm adds correctly to types - list ] , we know that the list @xmath763 maintained by @xmath761 is sensible , and we are now in a position to prove that the @xmath0-cases claimed by the algorithm are correct as well . let @xmath11 be a universal horn formula . if @xmath761 states that @xmath1198 has the polynomial - size model property and @xmath75 is in @xmath0 , then this is true . _ there are two possibilities for the algorithm to claim the polynomial - size model property , and hence @xmath0-membership . first let us assume that for some @xmath715 both @xmath840 and @xmath1107 were added to @xmath763 . in this case , due to lemma [ lemma : horn conjunction algorithm adds correctly to types - list ] , we know that there is some @xmath1199 such that @xmath11 implies both @xmath1122 and @xmath1117 . hence theorem [ theorem : transitive and symmetric gives np generalization ] implies both the polynomial - size model property of the logic @xmath84 and the @xmath0-membership of its satisfiability problem . the second case in which @xmath761 claims the @xmath0-result is if it detects a clause @xmath62 which satisfies one of the conditions @xmath1200 . we know by lemma [ lemma : horn conjunction algorithm adds correctly to types - list ] , that for each element from @xmath802 the formula @xmath11 implies a formula of the corresponding type . hence we can assume that there is a natural number @xmath1201 such that the set of vertices in @xmath66 which have both a @xmath251-step predecessor and a @xmath251-step successor satisfy the conditions from @xmath763 . we make a case distinction . * * @xmath62 satisfies @xmath0-condition @xmath1202 * * in this case , we know that @xmath284 for some @xmath772 and that there is a pair @xmath822 such that there is no directed path connecting @xmath774 and @xmath775 in @xmath258 . note that we can assume @xmath837 since otherwise every pair of vertices in @xmath258 would be connected with a directed path . first assume that it is possible to choose @xmath258 and @xmath295 in such a way that @xmath838 . in this case , since @xmath258 is a tree , we know that @xmath774 and @xmath775 have a common predecessor @xmath33 in @xmath258 . let @xmath33 be a common predecessor which is `` minimal , '' i.e. , no node in a lower level than @xmath33 is a common predecessor . we construct a @xmath1203tree @xmath1204 as a homomorphic image of @xmath258 via the homomorphism @xmath318 as follows : let @xmath103 be the height of the tree @xmath776 and let @xmath1205 be a @xmath763-line such that every element in @xmath258 is mapped on its corresponding level in @xmath1150 except the nodes on the path from @xmath33 to @xmath774 ( excluding @xmath33 ) and the successors of @xmath774 . let @xmath808 be the level of @xmath33 in @xmath776 i.e. , let @xmath1206 . let @xmath1207 and @xmath1208 be the levels of @xmath774 and @xmath775 in @xmath776 since there is no directed path connecting @xmath774 and @xmath825 it follows that @xmath1209 . now introduce nodes @xmath1210 such that @xmath1211 is an edge , and @xmath735 is an edge for every relevant @xmath212 and map the path from @xmath33 to @xmath774 and the successors to @xmath774 to the `` branch '' @xmath1210 ( add as many of these nodes as the `` branch '' of @xmath258 requires ) . now close the construction under the @xmath763-condition , and call the tree obtained in this way @xmath1204 . since @xmath1204 is the `` canonical homomorphic image '' of @xmath776 it is again a @xmath786-tree , and in @xmath1212 there is a path of length @xmath67 from @xmath1213 to @xmath1214 and a path of length @xmath928 from @xmath1213 to @xmath1215 . let @xmath1216 be the clause with prerequisite graph @xmath1204 and conclusion edge @xmath317 . since @xmath323 is a homomorphism , we know from proposition [ prop : homomorphism and horn clause implication ] that @xmath62 implies @xmath1217 intuitively , @xmath1216 is the @xmath763-closure of a clause of the form @xmath1218 with @xmath1219 chosen according to the height of the tree @xmath258 and the length of the branch containing @xmath774 ( without loss of generality , we assume that the branch containing @xmath774 does not contain the deepest node in the tree ) . we now show that @xmath1220 the complexity result then follows from corollary [ corollary : main np result corollary ] . hence , let @xmath66 be a graph such that @xmath137 and let @xmath20 and @xmath19 be vertices in @xmath66 such that @xmath20 and @xmath19 have a predecessor @xmath1221 and @xmath461 has a @xmath1222-step predecessor , @xmath20 has a @xmath1223-step successor , and @xmath19 has an @xmath1224-step successor , and there is a @xmath67-step path from @xmath461 to @xmath195 and an @xmath928-step path from @xmath461 to @xmath19 . then , since these vertices satisfy the conditions of @xmath763 by lemma [ lemma : horn conjunction algorithm adds correctly to types - list ] , we can homomorphically map the prerequisite graph of @xmath1204 into @xmath66 via the homomorphism @xmath1225 such that @xmath1226 @xmath1227 and @xmath1228 . the edges required in order for @xmath1225 to be a homomorphism exist because of the paths of the corresponding lengths connecting @xmath181 @xmath195 and @xmath95 and because all of the relevant nodes in @xmath66 satisfy the @xmath763-conditions . hence , by proposition [ prop : homomorphism and horn clauses ] , we know that @xmath68 is an edge in @xmath85 as required to show . therefore , assume that it is not possible to choose @xmath768 in such a way that @xmath824 i.e. , assume that in every @xmath820 @xmath774 and @xmath775 are connected with a directed path or are identical . since @xmath62 satisfies the first @xmath0-condition , we know that there is a pair @xmath822 such that there is no directed path connecting @xmath774 and @xmath775 in @xmath776 and hence @xmath1129 and this node is irreflexive in @xmath258 . in particular , we know that @xmath803 is not an element of @xmath763 . we also know that @xmath1229 since otherwise , every node in the connected graph @xmath258 would be connected to any other with a directed path ( @xmath258 obviously is not the irreflexive singleton , since @xmath838 ) . now define @xmath318 to be @xmath258 s canonical homomorphic mapping to a @xmath763-line @xmath1230 it then follows that @xmath1231 for some @xmath52 . since @xmath26 and @xmath394 are different nodes in @xmath292 we can modify this line as follows : we introduce a new node @xmath904 which is a `` neighbor '' to @xmath827 i.e. , a node which is connected to all predecessors and all successors of @xmath1232 call this line @xmath1233 . note that since @xmath763 only contains variations of transitivity , there is no condition requiring that @xmath305 is a reflexive node . since we connected @xmath904 to all successors and predecessors of @xmath827 the line @xmath1233 still satisfies all conditions from @xmath802 and @xmath305 and @xmath904 are not connected with an edge in @xmath1233 . we now construct a homomorphism @xmath1234 by defining @xmath1235 for all @xmath1236 and @xmath1237 . since the involved nodes are irreflexive and @xmath323 is a homomorphism , and @xmath904 has all edges that @xmath305 has , we know that @xmath1225 is a homomorphism . let @xmath1216 be the universal horn clause with prerequisite graph @xmath1238 and conclusion edge @xmath1239 . the homomorphism @xmath1225 and proposition [ prop : homomorphism and horn clause implication ] show that @xmath62 implies @xmath1216 . note that the clause @xmath1216 requires the following : for every pair of nodes @xmath1240 in a graph @xmath66 satisfying the conditions of @xmath763 which have a common predecessor @xmath461 and a common successor @xmath1241 with sufficient height and depth , there is an edge connecting @xmath1242 and @xmath1243 . since by lemma [ lemma : horn conjunction algorithm adds correctly to types - list ] , every graph satisfying @xmath11 also satisfies the requirements from @xmath763 for nodes with sufficient depth and height , we can apply @xmath1216 to nodes of sufficient depth and height in every such graph . l8 cm let @xmath66 be a graph satisfying @xmath1244 and let @xmath260 be the set of nodes in @xmath66 which have sufficient height and depth to be able to apply the conditions of @xmath1216 . we show that @xmath260 is reflexive , transitive , and symmetric . note that since @xmath66 satisfies @xmath1244 and @xmath62 is a clause in @xmath11 implying @xmath1245 we know that @xmath66 satisfies @xmath1216 as well . we first show that @xmath260 is reflexive . let @xmath19 be a node in @xmath260 . since @xmath19 satisfies the conditions of both @xmath305 and @xmath904 in the line @xmath1238 and since @xmath1246 and @xmath66 satisfies @xmath1245 we know that there is an edge @xmath1247 in @xmath85 and hence @xmath19 is reflexive . for symmetry , let there be nodes @xmath1248 such that @xmath1249 is an edge in @xmath66 . by the above , we know that both of these nodes are reflexive , and in particular have unbounded height and depth in @xmath66 . therefore , we can map the predecessor graph of @xmath1245 i.e. , the line @xmath1233 to these nodes in such a way that for @xmath1250 @xmath305 is mapped to @xmath1251 @xmath904 is mapped to @xmath1251 and all @xmath1252 for @xmath1253 are mapped to @xmath1254 . since in @xmath1238 all edges go from variables with lower indexes to variables with higher indexes , in this way we have constructed a homomorphism from @xmath1255 to @xmath455 such that the conclusion edge @xmath1239 of @xmath1216 is mapped to the pair @xmath1256 . hence , by proposition [ prop : homomorphism and horn clauses ] , and since @xmath66 satisfies @xmath1245 we know that there must be an edge @xmath1256 in @xmath455 as required . r8 cm finally , we show that @xmath260 is transitive . for this , assume that @xmath1257 and @xmath1258 are edges in @xmath260 . since we already proved symmetry for @xmath455 we know that in this case , @xmath1259 and @xmath1260 are also edges . we can again construct a homomorphism mapping all elements of @xmath1233 to @xmath1261 except mapping @xmath305 to @xmath808 and @xmath904 to @xmath20 . since the nodes @xmath1262 and @xmath20 are reflexive in @xmath260 by the above , and there is no edge @xmath1239 or @xmath1263 in @xmath1150 this is a homomorphism . since @xmath66 satisfies @xmath1264 proposition [ prop : homomorphism and horn clauses ] shows that @xmath1265 is an edge in @xmath455 as claimed . in particular , since @xmath260 is reflexive , symmetric , and transitive , this means that @xmath1266 . since @xmath260 is defined as the nodes which have some sufficient depth and height @xmath251 in the graph , and @xmath1267 it follows that @xmath1268 for some @xmath1201 and hence @xmath1269 . therefore , @xmath1198 is a logic extending @xmath1270 and therefore the complexity result as well as the polynomial - size model property follow from corollary [ corollary : main np result corollary ] . * @xmath62 satisfies @xmath0-condition @xmath1271 * by the prerequisites , since @xmath62 is not satisfied on every @xmath763-tree , we can homomorphically map @xmath294 onto a @xmath763-line . first assume that @xmath283 is empty . then any graph @xmath66 satisfying the conditions of @xmath763 which has a line of more than this length does not satisfy the clause @xmath62 . since every graph satisfying @xmath11 also satisfies @xmath62 and the conditions of @xmath763 for nodes of sufficient depth and height , every graph satisfying @xmath11 of sufficient depth and height does not satisfy @xmath302 and hence not @xmath11 . this implies that graphs satisfying @xmath11 can only be of depth limited by a constant , and hence lemma [ lemma : few nodes with few predecessors ] immediately implies the polynomial - size model property . therefore assume that @xmath284 for variables @xmath1272 and assume that @xmath817 are not connected with an undirected path in @xmath294 . since @xmath26 and @xmath394 lie in different connected components of the underlying directed graph consisting of @xmath1273 and @xmath294 can be homomorphically mapped into a @xmath763-line , the `` left side '' of the implication @xmath62 is satisfied by any pair of vertices @xmath68 in a graph satisfying the @xmath763-conditions of sufficient depth and height . therefore , the subgraph @xmath260 containing all nodes with sufficient height and depth forms a universal subgraph , and the @xmath0-result follows by the same reasoning as in case @xmath1202 * * @xmath62 satisfies @xmath0-condition @xmath1274 * * in this case , it follows by the same reasoning as in lemma [ lemma : horn conjunction algorithm adds correctly to types - list ] that any graph satisfying @xmath11 also needs to satisfy the formula @xmath1275 for some @xmath251 and some @xmath719 . hence the complexity result follows from corollary [ corollary : main np result corollary ] . * * @xmath62 satisfies @xmath0-condition @xmath1276 * * again , with the proof of lemma [ lemma : horn conjunction algorithm adds correctly to types - list ] it can easily be seen that @xmath11 implies formulas @xmath1117 and @xmath1122 for some @xmath719 and some natural number @xmath251 . therefore the @xmath0-result follows from theorem [ theorem : transitive and symmetric gives np generalization ] . @xmath14 the preceding lemmas have established that @xmath761 is well - defined , always comes to a halt , and in each case produces the correct result . hence , we have proven theorem [ theorem : horn conjunction classification ] . this theorem and its proof now yield an interesting corollary : [ corollary : horn cunjunction without algorithm ] let @xmath11 be a universal horn formula . if one of the following cases applies : * @xmath11 is satisfied in every strict tree , * @xmath11 is satisfied in every reflexive tree , * @xmath11 is satisfied in every @xmath72-transitive tree for some @xmath71 * @xmath11 is satisfied in every symmetric tree , * @xmath11 is satisfied in every symmetric and reflexive tree , * @xmath11 is satisfied in every @xmath72-transitive and reflexive tree for some @xmath71 then @xmath75 is @xmath1-hard and @xmath84 does not have the polynomial - size model property . in all other cases , @xmath84 has the polynomial - size model property and @xmath1277 we know by theorem [ theorem : horn conjunction classification ] that @xmath761 correctly determines the complexity of the problem @xmath75 . by the algorithm , it is obvious that the only arising cases are @xmath1-hard and @xmath0 . from the proof of lemma [ lemma : horn conjunction algorithm correct on pspace ] , we know that the @xmath1-cases all satisfy the statement of the corollary , and theorem [ theorem : ladner hardness cases for k - transitivity ] shows that in these cases , we always have @xmath1-hardness . finally note that all of our @xmath0-proofs also give the polynomial - size model property , and all @xmath1-hardness proofs also show that this property does not apply . in section [ sect : main dichotomy result ] , we have shown that logics defined by universal horn formulas have a satisfiability problem which is solvable in @xmath0 , or is @xmath1-hard . we now show a tree - like model property for these logics , which we will put to use in the next section by concluding @xmath1-membership for a broad class of these logics . recall that corollary [ corollary : every k - satisfiable formula satisfiable in strict tree ] stated the tree - like model property for the modal logic @xmath1278 every @xmath60-satisfiable formula has a tree - like model which also is a @xmath60-model ( which is easy , since every model is a @xmath60-model ) . what we want to show is that a logic @xmath760 where @xmath11 is a universal horn formula satisfied on every reflexive / symmetric/@xmath72-transitive tree , has the following property : every modal formula which has a @xmath84-model also has a @xmath84-model which is `` nearly '' the reflexive / symmetric/@xmath72-transitive closure of a strict tree . in order to prove this result , we first recall from the literature the concept of a bounded morphism , which allows us to prove modal equivalence of models . let @xmath258 and @xmath27 be modal models , and let @xmath1279 be a function . then @xmath134 is a _ bounded morphism _ if the following holds : 1 . for all @xmath1280 @xmath33 and @xmath1281 satisfy the same propositional variables , 2 . @xmath134 is a homomorphism , 3 . if @xmath1282 is an edge in @xmath99 then there is some @xmath1142 such that @xmath68 is an edge in @xmath258 and @xmath1283 the most important feature of bounded morphisms is that they leave the modal properties of the involved models invariant : [ prop : bounded morphism modal invariance ] let @xmath258 and @xmath27 be modal models , and let @xmath1279 be a bounded morphism . then for each modal formula @xmath32 and for each @xmath1280 it holds that @xmath1284 if and only if @xmath1285 with bounded morphisms , we can now prove that our logics have `` tree - like '' models . we first recall a standard result from the literature about the logic @xmath61 : [ prop : tree like model exists with bounded morphism ] let @xmath34 such that @xmath27 is rooted at @xmath33 . then there exists a tree - like model @xmath258 and a surjective bounded morphism @xmath1286 we now generalize this result to universal horn logics , and `` tree - like '' models . we already know from corollary [ corollary : horn cunjunction without algorithm ] that for each universal horn formula @xmath11 for which the logic @xmath84 has a satisfiability problem which can not be solved in @xmath0 , the formula @xmath11 is satisfied on every tree which additionally is closed under reflexivity , @xmath72-transitivity and/or symmetry . we now show that a version of the `` converse '' is also true : not only are all of these trees models for the corresponding logics , but for every modal formula satisfiable in such a logic , we can find a model which is `` almost '' such a tree . the reason for the `` almost '' is that our horn formulas usually will not imply a property like @xmath72-transitivity , but only @xmath72-transitivity for nodes at a certain depth in the graph , as shown in lemma [ lemma : horn conjunction algorithm adds correctly to types - list ] . we will see in sections [ sect : pspace for non - transitive ] and [ sect : applications ] , that the characterization of the involved models can be used to obtain @xmath1 upper bounds for a wide class of logics . [ theorem : tree - like models for horn logics ] 1 . let @xmath11 be a universal horn formula such that @xmath761 returns @xmath1-hard on input @xmath11 . let @xmath763 be as determined by @xmath761 on input @xmath11 . since the algorithm determines the logic to be @xmath1-hard , this is well - defined . then for every modal formula @xmath32 which is @xmath84-satisfiable , there exists a @xmath84-model @xmath258 and a world @xmath1287 such that @xmath1284 and there is a strict tree @xmath1288 such that @xmath258 and @xmath1288 have the same set of vertices , and + @xmath1289 + where @xmath1290 denotes the @xmath763-closure of @xmath1291 2 . let @xmath11 be a universal horn formula such that @xmath11 is satisfied on every strict line . then for every modal formula @xmath32 which is @xmath84-satisfiable , there exists a @xmath84-model @xmath258 and a world @xmath1287 such that @xmath1292 and @xmath258 is @xmath33-canonical . we prove both claims with nearly the same construction , indicating when differences are required . since @xmath32 is @xmath84-satisfiable , there is a @xmath84-model @xmath27 and a world @xmath80 such that @xmath34 . for the first claim , let @xmath763 be as determined by @xmath761 when started on input @xmath11 . since the algorithm does not return @xmath1293 we know that @xmath1294 or @xmath1295 . since the second @xmath0-condition does not apply , we know that for any clause @xmath62 in @xmath11 such that there is a homomorphism @xmath769 for some @xmath763-tree @xmath776 that @xmath284 for some variables @xmath817 with @xmath795 or @xmath819 for the second claim , we show that for every clause @xmath62 in @xmath1244 if @xmath62 is not satisfied in every @xmath33-canonical graph , then @xmath1296 . obviously , the prerequisite graph of such a clause @xmath62 can be mapped into some @xmath33-canonical graph , and therefore also into its image @xmath780 as a strict line . since @xmath11 is satisfied on every strict line , so is @xmath302 and hence @xmath1297 since otherwise , the clause would be unsatisfied on @xmath1150 a contradiction . for both claims , due to proposition [ prop : rooted models exist ] , we can assume that @xmath27 is rooted at @xmath33 . from proposition [ prop : tree like model exists with bounded morphism ] , we know that there is a model @xmath1298 which is a strict tree , and a surjective bounded morphism @xmath1299 . let @xmath1300 . let @xmath1301 denote the root of @xmath1298 . the strict tree @xmath1298 is trivially @xmath1301-canonical . we define a sequence of models @xmath1302 such that for each @xmath1081 it holds that when we are proving the first claim : 1 . @xmath1303 2 . @xmath1304 3 . @xmath1305 is a homomorphism . in the case of the second claim , we exchange the second point with `` @xmath1306 is @xmath1301-canonical . '' note that due to corollary [ corrollary : universal elementary logics have countable model property ] , we can assume that the model @xmath27 is countable . the proof of proposition [ prop : tree like model exists with bounded morphism ] from @xcite then constructs a model @xmath1298 which is also countable . hence assume that @xmath1298 is a countable model . now let @xmath1307 be a surjective enumeration of @xmath1308 i.e. , of all possible edges in the involved trees . the construction of our model is as follows : for @xmath1309 the model @xmath1298 from above obviously satisfies the conditions , since every bounded morphism is also a homomorphism . for @xmath1310 we make a case distinction : * if every clause in @xmath11 is satisfied in @xmath1311 then let @xmath1312 * otherwise , let @xmath1313 be a clause from @xmath11 which is not satisfied in @xmath1306 such that @xmath1314 for variables @xmath1315 and let @xmath1316 be a homomorphism such that @xmath1317 is not an edge in @xmath1306 . this must exist by proposition [ prop : homomorphism and horn clauses ] , since by the above , the case @xmath1318 can not occur . choose @xmath1319 in such a way that the pair @xmath1317 has a minimal index in the sequence @xmath1320 . now let @xmath1321 be defined with vertex set @xmath1322 edge set @xmath1323 and the same propositional assignments as @xmath1324 we show that the construction satisfies the requirements 1 - 3 . the first point , which is the same for both claims , holds by definition . for the first claim , we prove that for each @xmath1325 it holds that @xmath1326 . since by definition , @xmath1327 we know that for all @xmath1325 it holds that @xmath1328 and we also know that @xmath1329 hence assume that there is some minimal @xmath103 such that @xmath1330 . due to the minimality of @xmath1325 and since the claim holds for @xmath1309 we know that the edge which is not present in @xmath1331 is the edge @xmath1332 . by definition , @xmath1333 is a homomorphism . since due to minimality of @xmath1325 we know that @xmath1334 this implies that @xmath1335 is a homomorphism as well . since @xmath62 is a clause in @xmath10 and @xmath1336 we know that @xmath1337 and with proposition [ prop : homomorphism and horn clauses ] , we conclude that @xmath1332 is an edge in @xmath1338 a contradiction . therefore we know that @xmath1326 for all @xmath1325 and hence we have proven the second point in the case of the first claim . for the second claim , we need to show that @xmath1306 is a @xmath1301-canonical graph . for @xmath202 let @xmath1339 denote the nodes in the @xmath52-th level of @xmath1340 i.e. , the set @xmath1341 . in order to prove that @xmath1306 is @xmath1301-canonical , since @xmath1342 is , it suffices to prove that the edge @xmath1332 in the step from @xmath1342 to @xmath1343 does not destroy the property of being canonical , i.e. , we need to show that @xmath1344 and @xmath1345 for some @xmath52 . let @xmath780 be the homomorphical image of @xmath1342 as a strict line via the homomorphism @xmath318 ( since @xmath1342 is @xmath33-canonical , this exists and is unique ) . since @xmath1333 is a homomorphism , we know that @xmath1346 is a homomorphism as well . since @xmath11 is satisfied on every strict line , this is also true for @xmath302 and hence due to proposition [ prop : homomorphism and horn clauses ] , we know that @xmath1347 is an edge in @xmath780 . therefore , @xmath1348 is exactly one level above @xmath1349 in @xmath1340 as required . for both claims , we now show that @xmath1305 is a homomorphism for all @xmath103 . again , we prove the fact by induction , and for @xmath1309 this holds due to the choice of @xmath1350 since every bounded morphism is a homomorphism . therefore , let the claim hold for @xmath1325 and let @xmath68 be an edge in @xmath1321 . if @xmath68 is not the edge @xmath1351 then we know that @xmath68 is an edge in @xmath1306 as well , and since due to induction hypothesis , we know that @xmath1305 is a homomorphism , it follows that @xmath1352 is an edge in @xmath27 . therefore assume that @xmath1353 and @xmath1354 . we need to show that @xmath1355 is an edge in @xmath27 . by construction , we know that @xmath1316 is a homomorphism . since by induction hypothesis , we know that @xmath1305 is a homomorphism , it follows that @xmath1356 is a homomorphism as well . since @xmath27 is a @xmath84-model and @xmath1313 is a clause in @xmath10 we know that @xmath27 satisfies @xmath1357 and thus by proposition [ prop : homomorphism and horn clauses ] , we know that @xmath1355 is an edge in @xmath99 as required . we now construct the desired @xmath84-model as follows : define @xmath1358 as having @xmath1359 and @xmath1360 . we show that @xmath1358 is a @xmath84-model , and that @xmath1361 is a bounded morphism . assume that @xmath1358 is not a @xmath84-model . by construction , we know that in this case , there is no @xmath103 such that @xmath1362 and therefore each @xmath1321 has exactly one additional edge in comparison to @xmath1306 . hence we can define a sequence @xmath1363 such that @xmath1364 iff @xmath1365 . then @xmath1366 is a sequence of pairwise different natural numbers . since @xmath11 is not satisfied in @xmath1367 there is a clause @xmath62 from @xmath11 which is not satisfied in @xmath1358 . for both claims , we therefore know by the above that @xmath284 for variables @xmath817 with @xmath795 or @xmath818 . by proposition [ prop : homomorphism and horn clauses ] , we know that there is a homomorphism @xmath1368 such that @xmath286 is not an edge in @xmath1358 . let @xmath1369 such that @xmath1370 . since in the sequence @xmath1371 no number is repeated , there is some natural number @xmath1372 such that @xmath1373 for all @xmath1086 . since @xmath294 is a finite graph , and @xmath1327 and @xmath1374 we know that there is some @xmath1375 such that @xmath1376 is a homomorphism . then @xmath1377 is also a homomorphism for all @xmath1378 since every edge present in @xmath1379 is also present in every @xmath1306 for @xmath1380 . let @xmath1381 . then in the step from @xmath1306 to @xmath1382 the edge @xmath1383 was added , and by choice of @xmath103 we know that @xmath1373 . since @xmath1384 is not an edge in @xmath1367 we know that @xmath1384 is also not an edge in @xmath1306 . recall that @xmath1385 are chosen in such a way that the edge @xmath1386 has minimal index in the sequence @xmath1387 this is a contradiction , since the edge @xmath1384 is an edge with a smaller index than @xmath1388 and since @xmath1377 is a homomorphism , @xmath1384 satisfies the conditions of the edge @xmath1386 in the construction of @xmath1389 it remains to show that @xmath1361 is a bounded morphism . property @xmath1390 holds by construction , since we do not change propositional assignments , and @xmath1299 is a bounded morphism . we now show that @xmath134 is a homomorphism . hence let @xmath68 be an edge in @xmath1358 . since @xmath1358 is the union over all @xmath1311 there is some @xmath1391 such that @xmath68 is an edge in @xmath1306 . since by the above , @xmath1305 is a homomorphism , it follows that @xmath1352 is an edge in @xmath99 as claimed . for property @xmath1392 let @xmath1393 and @xmath660 in @xmath27 such that @xmath1282 is an edge in @xmath27 . since @xmath1299 is a bounded morphism , we know that there is some @xmath1394 such that @xmath68 is an edge in @xmath1350 and @xmath1395 . by construction , @xmath68 is also an edge in @xmath1367 and therefore @xmath1361 is a bounded morphism . by choice of @xmath1396 and since @xmath1397 @xmath134 is also surjective . hence there is some @xmath1398 such that @xmath1399 . since @xmath179 it follows from proposition [ prop : bounded morphism modal invariance ] that @xmath1400 . by construction , for the first claim it holds that @xmath1401 since this holds for the individual @xmath1306 and @xmath1358 is the union over the edges of all @xmath1311 and for the second claim , @xmath1358 is obviously @xmath1301-canonical : assume that it is not , then there is a node @xmath1402 and natural numbers @xmath1403 such that @xmath1404 and @xmath1405 . since only a finite number of edges is relevant for this path , there exists some @xmath103 such that @xmath1406 and @xmath1407 . this is a contradiction , since @xmath1306 is @xmath1301-canonical . note that we can assume that the node @xmath461 is the root of @xmath1367 since due to proposition [ prop : rooted models exist ] , we can assume that every node in @xmath1358 can be reached from @xmath1408 in the previous section , we showed that for logics defined by universal horn formulas , satisfiable formulas are always satisfiable in a tree - like model . tree - like models are the main argument in many proofs showing @xmath1-membership for satisfiability problems in modal logic . we now show that these models indeed allow us to construct @xmath1 algorithms for a wide class of modal logics . note that while the construction in the proof of the following theorem has similarities to the constructions by ladner in @xcite or by halpern and moses in @xcite , the focus of our result is different . many proofs of previous @xmath1-algorithms also gave proofs of a variant of some tree - like model property . our proof relies on this property ( which we already proved for our logics in theorem [ theorem : tree - like models for horn logics ] ) , and as a consequence , the verification that the algorithm works correctly with respect to the modal aspect of its task is very easy to verify . the main work of the proof is to prove that the algorithm handles the first - order part of the satisfiability problem correctly , i.e. , that the model it constructs is in fact a model satisfying the first - order formula @xmath11 defining the logic . whereas this is easy for standard classes of frames ( checking reflexivity , symmetry , transitivity etc is straightforward ) , in the general case that we cover here this requires most of the work . the main feature of the logics that we use here is that of _ locality _ : the proof makes extensive use of the fact that in order to verify that the first - order clauses are satisfied it is sufficient to consider local parts of the model constructed by the algorithm . this is the main reason why we believe that this proof does not easily generalize to cases where a variant of transitivity is among the conditions implied by the first - order formula @xmath1409 [ theorem : pspace membership for non - trans logics ] let @xmath11 be a universal horn formula such that @xmath761 does not add any element of the form @xmath922 to @xmath786 on input @xmath11 . then @xmath1410 _ proof . _ since @xmath1411 we can assume that @xmath1412 . let @xmath763 be as determined by @xmath761 on input @xmath11 . from the prerequisites , we know that @xmath1294 . from theorem [ theorem : tree - like models for horn logics ] , we know that for every @xmath84-satisfiable modal formula @xmath83 there is a model @xmath258 of @xmath32 such that @xmath258 is an edge - extension of a strict tree @xmath1288 , and every edge present in @xmath258 which is not an edge of @xmath1288 is a reflexive or a symmetric edge . by the proof of lemma [ lemma : few nodes with few predecessors ] , we can assume that in the tree @xmath1413 every node has at most @xmath1414 successors . the strategy of the @xmath1-algorithm is as follows : we nondeterministically guess the model @xmath258 and verify that it is a model of both @xmath32 and of @xmath11 by performing a depth - first - search . it is straightforward to see that in this way , we can verify that the modal formula @xmath32 holds in the model . to also check if @xmath11 is satisfied requires a bit more effort : even though we know that @xmath11 only adds reflexive or symmetric edges in addition to those present in @xmath1413 we need to be careful which edges are required and which are not ( there may very well be formulas which are not satisfiable on a symmetric and reflexive tree , but are @xmath84-satisfiable ) . the main reason why we can perform these tests is that the properties that we work with have a `` local character : '' to check if an edge is required between some nodes @xmath20 and @xmath19 in @xmath776 we need to know whether there is a clause @xmath62 in @xmath11 such that @xmath62 can be homomorphically mapped into @xmath258 in such a way that the conclusion edge is mapped to the pair @xmath68 . since we are working with a tree only extended with reflexive and symmetric edges , we know that a homomorphic image of a connected component of some @xmath294 including @xmath20 and @xmath19 contains only vertices which are `` near '' to both @xmath20 and @xmath19 . therefore we can verify that these clauses are satisfied by procedures looking only locally at the model @xmath258 . there are two main obstacles to this approach : for once , the clause @xmath62 that requires @xmath68 to be an edge might very well contain more than one connected component except that one containing the conclusion edge ( we can ignore the case where there is no conclusion edge , or the vertices from the conclusion edge are unconnected in @xmath292 since if such a clause can be applied , i.e. , homomorphically mapped to @xmath776 then it can also be homomorphically mapped to the @xmath763-closure of @xmath776 and hence to a @xmath763-tree , in which case @xmath761 reports @xmath0-membership , which is a contradiction to our assumption @xmath1412 ) . the other obstacle is that although we only need to look at vertices in the `` neighborhood '' of the current vertex to check that it has all the right edges coming in and out , we need to ensure that all the vertices that we looked at `` locally '' are consistent , when we revisit a part of the model which is close to a node that we already considered . the ways to deal with these obstacles is the following : for the first problem , we simply keep a list of connected components of @xmath294-graphs , and at the beginning of the algorithm , guess for each one if it will appear as a homomorphic image in the tree ( which of course later we need to verify ) . for the second problem , we keep more nodes in storage than just the ones in the neighborhood of the one we are currently visiting , but only a polynomial number . strictly speaking , the algorithm does not operate on a model , but on an `` annotated model . '' the annotation of a world is the set of subformulas and negated subformulas of the input formula @xmath32 which are true at this world , and are required to be true to ensure that the formula @xmath32 is true at the root - world . by proposition [ prop : rooted models exist ] , we can assume that the tree @xmath258 has height of at most @xmath182 . for a node @xmath19 in the @xmath52-th level of @xmath776 let @xmath1415 denote the set of subformulas and negated subformulas of @xmath32 which have a modal depth of at most @xmath1416 . these are exactly those formulas for which we need to know that they hold at @xmath19 in order to verify that the input formula @xmath32 holds at the root of @xmath832 we now describe the decision procedure , which is a nondeterministic @xmath1-algorithm . let @xmath72 be the cardinality of the largest connected component in any of the graphs @xmath294 for clauses @xmath62 of @xmath11 . note that this number only depends on @xmath10 and therefore can be regarded as constant . the algorithm as stated in figure [ fig : pspace algorithm ] does not work in polynomial space , since it guesses and stores the possibly exponentially - sized model @xmath258 . we will first show that the algorithm as stated is correct and then prove how it can be implemented using only polynomial space , by only storing a currently relevant subset of the model @xmath832 for the description of the algorithm , we will call a node @xmath1142 _ back - symmetric _ if there is an edge @xmath1417 where @xmath20 is the predecessor of @xmath19 in @xmath1288 . when the algorithm guesses the model @xmath776 it additionally guesses the set @xmath1415 for every node @xmath19 in @xmath776 and for each node it guesses if it is back - symmetric and if it is reflexive . the procedure @xmath1418 performs the following check : for a node @xmath19 on the @xmath52-th level of @xmath776 @xmath1419 is required to contain all subformulas and negated subformulas of @xmath32 which have modal depth of at most @xmath1420 and are true at @xmath19 . hence , @xmath1421 must contain exactly one of @xmath1422 or @xmath79 for each relevant @xmath78 and additionally , if @xmath1423 then both @xmath1424 and @xmath1425 need to be members as well . similarly , if @xmath1426 is a member , then at least one of them must be an element of @xmath1427 the procedure @xmath1428 works as follows : if there is a clause @xmath62 in @xmath11 with @xmath1429 for @xmath1195 such that all connected components of @xmath294 can be mapped homomorphically into @xmath776 then @xmath1428 ensures that @xmath19 is reflexive . note that all other horn clauses in @xmath11 satisfy that @xmath284 for some @xmath818 . for these clauses , the procedure considers the subgraph @xmath1430 consisting of all nodes of @xmath258 which can be reached from @xmath19 in at most @xmath72 undirected steps ( note that a node can be reached in at most @xmath72 steps in @xmath258 if and only if it can be reached in at most @xmath72 steps in @xmath1288 ) . for every connected component @xmath260 of @xmath292 @xmath1428 tests all functions @xmath1431 . if one of these @xmath295 is a homomorphism , then @xmath1428 rejects , if the algorithm guessed in the beginning that @xmath260 can not be mapped homomorphically into @xmath258 . if there is one clause @xmath62 in @xmath11 such that all connected components of @xmath294 can be mapped into @xmath258 ( according to the list of these possibilities maintained by the algorithm ) and @xmath1428 detected a homomorphism @xmath1432 ( where @xmath1433 is the connected component of @xmath294 containing the nodes from the conclusion edge of @xmath62 ) for some @xmath62 with @xmath284 such that @xmath1434 then @xmath1428 rejects if @xmath286 is not an edge in @xmath1435 we prove that the algorithm is correct . first note that for each connected component @xmath260 of some @xmath294 for a clause @xmath62 in @xmath10 if there is a homomorphism @xmath1436 then there is a node @xmath1142 such that @xmath1431 is a homomorphism . this holds because all edges in @xmath258 are already present in the strict tree @xmath1288 or are symmetric or reflexive edges , and the homomorphic image of @xmath260 under the homomorphism @xmath295 is a connected component of @xmath776 and the maximal distance of the nodes in this image is @xmath72 ( recall that this is the maximal cardinality of a connected component in any @xmath294 ) . therefore , we can assume that for each connected component @xmath260 of some @xmath292 if it can be homomorphically mapped into @xmath776 then @xmath1437 guessed this correctly in the beginning in every accepting run of the algorithm ( an incorrect guess would , due to the observation just made , be detected by @xmath1428 for some node @xmath19 ) . now assume that the algorithm accepts . we claim that the model obtained from the annotated model guessed by the algorithm where a variable @xmath26 is true at a world @xmath19 if and only if @xmath1438 is a model of both the modal formula @xmath32 ( at the root - world @xmath33 ) and of the horn formula @xmath11 . by the checks the algorithm performs , it can easily be verified by induction on the level of the nodes ( corresponding to the modal depth of the involved formulas ) that for every world @xmath19 in @xmath776 every formula in @xmath1415 is satisfied at @xmath19 . since @xmath1439 this implies that @xmath1284 . the base case for the induction is clear , since for worlds @xmath19 in the level @xmath144 @xmath1415 only contains literals , and the algorithm ensures that @xmath1415 is propositionally consistent . since for each subformula @xmath79 of @xmath32 of relevant modal depth , @xmath1415 contains exactly one of @xmath79 and @xmath1440 the induction hypothesis can be applied in the relevant cases . note that the algorithm also checks consistency for the cases in which we have reflexive and/or symmetric edges . it remains to show that @xmath258 is also a model of the first - order formula @xmath11 . assume that this is not the case . then there exists some clause @xmath62 in @xmath11 which is not satisfied in @xmath258 . in particular , this implies that @xmath294 can be homomorphically mapped into @xmath776 and since @xmath1441 this implies that @xmath294 can be homomorphically mapped into a @xmath786-tree . since we assumed that @xmath1442 does not return @xmath436 on input @xmath10 we know that none of the @xmath0-conditions from @xmath761 are satisfied . since none of the @xmath0-conditions from @xmath761 occur , we know that @xmath62 has conclusion edge @xmath284 for variables @xmath817 where @xmath795 or @xmath818 . first assume that @xmath795 . since @xmath294 can be mapped homomorphically into @xmath776 the procedure @xmath1443 required every node in @xmath258 to be reflexive , hence @xmath62 is satisfied in @xmath258 . not assume that @xmath818 . since none of the @xmath0-conditions apply , we know that @xmath26 and @xmath394 are connected with an undirected path in @xmath294 . in particular , they lie in the same connected component @xmath1433 of @xmath294 . since @xmath62 is not satisfied in @xmath776 this implies by proposition [ prop : homomorphism and horn clauses ] that there are nodes @xmath1444 such that @xmath68 is not an edge in @xmath776 there is no edge @xmath68 in @xmath776 and there is a homomorphism @xmath769 such that @xmath1445 and @xmath1446 . due to the above , we know that @xmath1432 is a homomorphism . therefore , the homomorphism @xmath295 was found by @xmath1447 and @xmath1428 . since @xmath294 can be homomorphically mapped into @xmath776 we know that every connected component @xmath260 of @xmath294 can be homomorphically mapped into @xmath776 and due to the above , we know that the @xmath1437 guessed this correctly in an accepting run of the algorithm . therefore , the procedure @xmath1443 ensured that @xmath68 is an edge in @xmath776 a contradiction . now assume that @xmath32 is @xmath84-satisfiable . due to the remarks at the beginning of the proof , we know that in this case , there exists a @xmath84-model @xmath258 such that @xmath258 is an edge - extension of a strict tree , and @xmath1448 . therefore the algorithm can guess this model and verify that it satisfies both @xmath32 and @xmath1409 it remains to prove that the algorithm can be implemented in nondeterministic polynomial space . the result then follows , since due to a classic result by savitch @xcite , @xmath1449 . in order to implement the algorithm using only polynomial space , the main change needed compared to the version stated in figure [ fig : pspace algorithm ] is how much of the guessed model @xmath258 is stored in memory at a given time . the @xmath1450-implementation does not guess the entire model @xmath258 at the start of the algorithm , but guesses each node the moment it is first accessed ( either by being created explicitly , or by being explored as an @xmath72-step neighbor of another node by the procedure @xmath1443 ) . it removes the node from memory at a time when it will not be accessed anymore in the remaining execution of the algorithm . to be precise , the algorithm at all times keeps in its memory the node @xmath1451 and all of its predecessors , and all nodes which can be reached from these in at most @xmath72 steps in the tree @xmath1288 . since in @xmath776 every node has at most @xmath1414 successors in the next level , and @xmath72 is a constant , this is a polynomial number of nodes . we now need to prove that no necessary information is removed from memory , i.e. , that no node is first created , then deleted and then accessed again . note that from the construction of the algorithm , it is obvious that new nodes are visited in a depth - first order . therefore assume that this happens for some node @xmath1452 . note that any node which gets deleted from memory is not reachable from the root world @xmath33 in at most @xmath72 steps , and therefore @xmath1452 is at some level @xmath1453 in the tree @xmath1454 . let @xmath1455 be the node for which @xmath1452 was visited for the first time , i.e. , the first node visited such that @xmath1456 ( recall that @xmath1457 is the set of nodes which can be reached from @xmath1455 in at most @xmath72 undirected steps in @xmath258 ) . since @xmath1452 is deleted from memory and required again later , there is some node @xmath1254 such that @xmath1452 can not be reached from any predecessor of @xmath1254 in at most @xmath72 steps , and a node @xmath1458 such that @xmath1452 can be reached from @xmath1458 in at most @xmath72 steps . let @xmath654 be the ( uniquely determined ) common predecessor of @xmath1452 and @xmath1254 with a maximal level in the tree . then , since @xmath654 is a predecessor of @xmath1459 we know that @xmath1460 . hence , @xmath1452 is at least @xmath72 levels below @xmath654 . since @xmath1288 is a tree , any node @xmath503 such that @xmath1461 must therefore be a successor of @xmath654 . in particular , @xmath1458 is a successor of @xmath654 . this is a contradiction , because @xmath1437 traverses the tree in depth - first - search , and hence does not leave the sub - tree with root @xmath654 and re - enters it later . therefore we have shown that it is sufficient to keep a polynomial number of nodes in storage , and which nodes to keep can be decided by an easy pattern . hence it follows that the algorithm can indeed by implemented in nondeterministic polynomial space as required , concluding the proof . @xmath14 theorem [ theorem : horn conjunction classification ] and [ theorem : pspace membership for non - trans logics ] can be used to classify the complexity of a lot of concrete logics , but they also imply more general results , for which we will give two examples . for once , recall that ladner proved that all normal modal logics @xmath180 such that @xmath64 ( the logic over all transitive and reflexive frames ) is an extension of @xmath180 give rise to a @xmath1-hard satisfiability problem . the following corollary shows that this result is optimal in the sense that every universal horn logic which is a `` proper extension '' of @xmath1462 in the way that they imply the conditions of @xmath1462 , already gives an @xmath0-solvable satisfiability problem . [ corollary : ladner hardness result optimal ] let @xmath11 be a universal horn formula such that @xmath11 implies @xmath1463 . then either @xmath1464 or @xmath84 has the polynomial - size model property and @xmath1277 by the prerequisites , we know that @xmath11 is equivalent to @xmath1465 . hence we can , without loss of generality , assume that @xmath1466 and @xmath1467 appear as clauses in @xmath1409 if every clause in @xmath11 is satisfied in every transitive and reflexive tree , then every modal formula @xmath32 which is satisfiable in a transitive and reflexive tree is @xmath84-satisfiable . note that a special case of theorem [ theorem : tree - like models for horn logics ] gives the result that every @xmath64-satisfiable formula also is satisfiable in a reflexive and transitive tree . therefore , every @xmath64-satisfiable formula is also @xmath84-satisfiable , and hence , every @xmath84-validity is also @xmath64-valid . therefore , @xmath64 is an extension of @xmath84 . since by proposition [ proposition : extensions of elementary logics and first - order implication ] , @xmath84 is an extension of @xmath1468 this implies that @xmath1469 therefore , we can assume that @xmath11 is not satisfied in every reflexive and transitive tree . now let @xmath763 be as determined by @xmath761 on input @xmath11 . since @xmath1466 and @xmath1467 are clauses in @xmath10 we know that @xmath1102 and @xmath1470 are elements of @xmath763 . if all elements in @xmath763 are of the form @xmath1102 or @xmath1471 then we know ( since @xmath67-transitivity is implied by @xmath69-transitivity ) , since @xmath11 is satisfied on every @xmath786-tree , that @xmath11 is satisfied in every reflexive and transitive tree , a contradiction . therefore , we now that @xmath1472 and hence by construction , @xmath761 reports @xmath0-membership . since by theorem [ theorem : horn conjunction classification ] the output of the algorithm is correct , we know that @xmath1473 and @xmath84 has the polynomial - size model property , as claimed . we further can show a @xmath1 upper bound for all universal horn logics which are extensions of the logic @xmath1474 , and hence , from theorem [ theorem : horn conjunction classification ] , conclude that these are all either solvable in @xmath0 ( and thus @xmath0-complete if they are consistent ) , or @xmath1-complete . [ corollary : above reflexive gives pspace ] let @xmath11 be a universal horn formula such that @xmath11 implies @xmath1466 . then @xmath1410 assume without loss of generality that @xmath761 determines the logic @xmath84 to have a @xmath1-hard satisfiability problem , otherwise the theorem holds trivially , since @xmath1411 . if @xmath11 implies @xmath1475 then the result follows from corollary [ corollary : ladner hardness result optimal ] . hence assume that this is not the case . note that in reflexive graphs , @xmath67-transitivity is equivalent to transitivity . also note that the conditions requiring a node to have a certain depth or height in a graph are always satisfied in a reflexive graph , because nodes here have infinite depth and height . therefore , if @xmath11 implies a formula of the form @xmath1476 for some @xmath1477 and some @xmath1478 then @xmath11 also implies @xmath1479 and due to the above , we can assume that this is not the case . thus , @xmath763 as determined by @xmath761 contains no condition of the form @xmath922 for any @xmath70 . the complexity result now follows from theorem [ theorem : pspace membership for non - trans logics ] . in a similar way , we can prove that all universal horn logics which imply a variant of symmetry give rise to a satisfiability problem in @xmath1 . a noteworthy difference in the prerequisites of corollary [ corollary : above reflexive gives pspace ] and corollary [ corollary : pspace for extensions of symm ] is that the former requires the reflexivity condition to be implied by the formula @xmath1244 while the latter only needs a `` near - symmetry''-condition as detected by @xmath761 . [ corollary : pspace for extensions of symm ] let @xmath11 be a universal horn formula such that @xmath1442 adds @xmath1480 to @xmath763 on input @xmath11 . then @xmath1481 . in particular , any universal horn logic which is an extension of @xmath1482 has a satisfiability problem solvable in @xmath1 . if @xmath1473 the claim trivially holds . hence , since @xmath1442 is correct due to theorem [ theorem : horn conjunction classification ] , we can assume that @xmath1442 returns @xmath1-hard , and @xmath1472 where @xmath763 is as determined by @xmath761 . since @xmath761 does not report @xmath0 , we know from its construction that @xmath1483 for all @xmath1484 . hence the complexity result follows from theorem [ theorem : pspace membership for non - trans logics ] . we analyzed the complexity of modal logics defined by universal horn formulas , covering many well - known logics . we showed that the non - trivial satisfiability problems for these logics are either @xmath0-complete or @xmath1-hard , and gave an easy criterion to recognize these cases . our results directly imply that ( unless @xmath1485 ) such a logic has a satisfiability problem in @xmath0 if and only if it has the polynomial - size model property . we also demonstrated that a wide class of the considered logics has a satisfiability problem solvable in @xmath1 . open questions include determining complexity upper bounds for the satisfiability problems for all modal logics defined by universal horn formulas . we strongly conjecture that all of these are decidable , and consider it possible that all of these problems are in @xmath1 . a successful way to establish upper complexity bounds is the guarded fragment @xcite . this does not seem to be applicable to our logics , since it can not be used for transitive logics , and we obtain @xmath1-upper bounds for all of our logics except those involving a variant of transitivity . the next major open challenges are generalizing our results to formulas not in the horn class , and allowing arbitrary quantification . initial results show that even when considering only universal formulas over the frame language , undecidable logics appear . an interesting enrichment of horn clauses is to allow the equality relation . preliminary results indicate that corollary [ corollary : horn cunjunction without algorithm ] holds for this more general case as well .

the period of enlightenment was a time when spanish physics , which had been lagging behind the level of the rest of europe , was able to recover . however , the political events of the end of the 18th - century , the war of independence , and then the reign of fernando vii ruined the panorama of spanish science . history was sadly to repeat itself with the spanish civil war of 1936 , which brought to nothing the efforts of physicists and mathematicians of the end of the 19th and the beginning of the 20th centuries to regain the time lost during the reign of fernando vii until the `` generation of 98 '' . part of the disastrous situation during the 19th - century can be traced back to the legislation and curricular plans of spain s universities . the faculty of exact , physical , and natural sciences was not created until 1857 , with the ley ( law ) moyano . physics had been relegated to a minor faculty , that of arts , as preparation for the major faculties , and formed part of the `` philosophical institutions '' which were studied in the following order : i. history of philosophy and elements of mathematics ; ii . logic and metaphysics ; iii . general physics ; iv . special physics . all was in a latin which had become progressively less suited to teaching , so that enlightened reformers always attempted to publish textbooks in spanish . physics texts in latin were still being imposed on students even up to the time of the absolutist period of fernando vii @xcite . after the death of fernando vii , an ordinance regulating printing was promulgated in 1834 which allowed a certain freedom in publication , in particular in scientific publishing . one of the consequences of the centralization of curricular plans , such as the plan pidal , was the impulse given to the production of textbooks associated with the new programs of study . they were original productions as well as translations of foreign authors , and formed the beginning of a national scientific output which was to be of greater or lesser quality according to each particular case . the study plans usually counselled the use of certain textbooks which most closely approached the spirit of the courses . the procedure consisted in appointing committees to decide on the most suitable texts which then put forward a number of them to be chosen from . in the meantime , europe was seeing a major change in vision towards atomism thanks to such personalities as herapath , waterston , clausius , van der waals , maxwell , and boltzmann @xcite . imponderable fluids had been abandoned as the explanation of physical phenomena , and mechanics , heat , electromagnetism , and optics had been unified by the concept of energy . teaching was based on a program of mechanical explanations @xcite . the goal of the present communication is to analyze the impact of modern ideas concerning energy and the constitution of matter on the textbooks of 19th - century spain . we have examined 45 works on general physics at secondary and university education levels , with publication dates from the mid 19th - century to the early 20th - century . one of the first things that struck us was the great similarity between the books of different authors . with centralism , teachers were encouraged to write their own textbooks and , indeed , were rewarded for doing so . at the same time , the administration s goal was for teaching to be uniform nationwide . the result was that textbooks were written essentially with the ministerio de fomento s ( ministry of development ) official program as the table of contents , and that the texts served solely to expound known science and in no way to serve as the basis for further research . there were those who , being interested in science teaching , protested about the policy concerning textbooks . representative of them was eduardo lozano y ponce de len . under the pseudonym l. opando y uceda , lozano published `` programas y libros de texto '' ( programs and textbooks ) in _ revista de la sociedad de profesores de ciencias _ @xcite , and under the pseudonym `` un extremeo '' the same article in the journal _ el magisterio extremeo _ ( _ the extremean schoolmaster _ ) @xcite . in this same journal and on the same theme , ildefonso fernndez snchez published another critical article @xcite . in his writings lozano asked for the programs to have reference strictly to the subject matter to be dealt with and not to the methods to be used , and that the programs should include suggestions of the books that were best suited to the subject , but without the obligation of following them . the immediate consequence of that educational policy was that the different editions of textbooks seemed rather to be reprintings : textbook `` immutability '' from one edition to another was astonishing . let us take two cases as example . the _ trait lmentaire de physique _ ( _ elementary treatise on physics _ ) by a. ganot was used in many spanish and european institutes ( secondary education ) and universities . the first spanish language edition dates from 1853 . the 18th spanish edition was printed in 1923 , and even as late as 1945 an edition was published in san sebastin @xcite . in all this time there were only small modifications made with respect to the first french edition . an even more extreme example , since it was a work with even more antiquated ways of putting the material than ganot , was the textbook of @xcite . the 2nd edition dates from 1851 @xcite and the 10th edition from 1870 @xcite . in these twenty years and eight new editions the work remained without a single significant change , notwithstanding the archaic ways of setting forth the subject matter of the first editions . a third example is less disappointing : _ elementos de fsica _ ( _ elements of physics _ ) by enrique iglesias ejarque . the first edition of 1897 @xcite has quite a modern manner of exposition . nevertheless , while the following editions introduce small amounts of additional material into the text , there are never any significant changes . the 8th edition dates from 1924 @xcite and the last that we can find a reference to is the 10th edition in 1933 . clearly one may conclude that physics textbooks had quite a long effective lifetime . while there was a certain degree of difficulty in introducing new material into the textbooks , the great problem was to eliminate content which was included by tradition even though it was antiquated . the case of eduardo snchez pardo , the translator of ganot s work , is significant . in the prologue to that popularly used book : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the editor d. crlos bailly - baillire being for his part desirous that our public in general , and the pupils of our institutes and faculties in particular , follow science in her latest advances , charged us with the translation of the latest edition of the cited treatise on physics . on taking on this commitment , with the object that the spanish edition be more complete , we judged it convenient , corresponding thereby also to the wishes of the editor , to conserve certain theories , the exposition of various experiments , and the description of some of the instruments or apparatus that had figured in earlier editions of the forementioned work , and that in the latest edition had been totally or partially suppressed . ( @xcite @xcite , p. v ) _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ another of the characteristics of spanish physics textbooks was their orientation to student success in tests and examinations . works that are halfway between textbook and simple program of the curriculum abound . secondary education teachers often encouraged the students to think little about the physical phenomena themselves but to learn definitions , laws , and descriptions of apparatus . as clear examples we could cite , amongst several works , the _ resumen de fsica y nociones de qumica _ ( _ a summary of physics and notions of chemistry _ ) @xcite , and _ definiciones , principios y leyes de la fsica _ ( _ definitions , principles , and laws of physics _ ) @xcite . hence , for most students learning was not meaningful but purely memorization . together with the lack of practical classes , this may have been one of the causes of spanish physics slow rate of development during the 19th - century . in order to analyze what we understand to have been the introduction of modern physics into spanish textbooks , we asked ourselves the following questions relative to the works that we consulted : 1 . [ q1]are imponderable fluids studied in the textbook ? 2 . [ q2 ] is the term caloric used to refer to heat ? 3 . [ q3]does the concept of energy appear in a general form ? [ q4]does the mechanical theory of heat appear ? [ q5]does the kinetic theory of gases appear ? affirmative responses to the first two questions would indicate traditionalism , and to the last three , modernity . we would classify a textbook as modern if questions [ q3 ] and [ q4 ] are answered affirmatively . if the kinetic theory of gases appears as well ( question 5 ) , then we would consider the textbook to be quite complete . table [ tablei ] lists the responses to these questions for the 19th - century textbooks to which we had access . the works are ordered by publication date . when the edition is not given , reference is to the first edition . most of the works consulted belong to the library of the real sociedad econmica de amigos del pas in badajoz ( spain ) . while not exhaustive , the sample is significant , since students of the city of badajoz in the 19th - century were prepared in this centre for the entrance examinations of the schools of engineering ( high - level technical degree courses , rather than purely engineering in the english language sense ) or military academies . . responses to the five questions posed in the text , according to the different works consulted . [ cols="<,^,^,^,^,^",options="header " , ] some comments are necessary concerning table [ tablei ] . the books that were published in spain in the 1840s studied imponderable fluids , and dealt neither with energy , nor with the mechanical theory of heat , nor with the kinetic theory of gases . this situation was natural since these concepts were only being developed in europe at this time . however some foreign authors already did not study imponderable fluids ( neither did they use the term caloric to refer to heat ) , and their spanish translations set an example for such fluids to be abandoned . this was the case with the textbook of deguin , translated by venancio gonzlez @xcite . the translator would write his own textbook later @xcite , and despite going through numerous editions , would never abandon imponderable fluids . another textbook of a french author is @xcite , translated by florencio martn castro , although here imponderable fluids were still being studied in the original . the handbook of @xcite presents physics as a science of nature in general , including subjects such as geology and geography . this was the concept of physics that existed in europe at the beginning of the century . on the positive side , the mechanical ideal which was at the base of the development of physics in the 19th - century did indeed have a reflection in spanish textbooks . one reads in the prologue of the work of @xcite : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ( ) effectively , all the phenomena attributed to caloric , to light , and to the electric fluid are mensurable and calculable effects ; all are attributed to forces , all consist of movements , and constitute , in sum , mechanics . ( @xcite @xcite , p. i ) _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the texts that we were able to consult from the 1850s are mainly editions of the work of @xcite ( @xcite @xcite , @xcite , @xcite ) . the principal characteristic of this text was its immutability , edition after edition . although it studied imponderable fluids , the term caloric was not used , and neither energy , nor the mechanical theory of heat , nor the kinetics of gases appear in the text . the other textbook that we consulted is that of @xcite . although imponderable fluids are studied and the term caloric is used , the author briefly explains the theory of ondulations ( as the mechanical theory of heat was first known ) , and says of that theory that `` it is the one which today seems more correct '' ( p. 173 ) . this work was awarded a prize in a public competition under the auspices of the real academia de ciencias ( gaceta de madrid , 9 september 1854 ) . neither the concept of energy in a general form nor the kinetic theory of gases appear in rodrguez s text . it is interesting too that in this decade there appeared a physics textbook written in latin which followed the scholastic tradition @xcite . the textbooks of spanish authors of the 1860s were still studying imponderable fluids . they were also still using the term caloric to refer to heat , except for the editions of @xcite ( @xcite , @xcite ) , as we noted before . one french - authored textbook , @xcite , did not study imponderable fluids or use the term caloric , but then neither did it use the energy concept or the theories we are looking for . the same was the case with the textbook of @xcite , commented on above . a possible reason for these to be missing is that the translator , ramn de la sagra , used the 7th french edition . the note of modernity is found in the textbook of @xcite , in which there appears an idea concerning the mechanical theory of heat , namely the hypothesis of ondulations , although for didactic purposes the theory of emissions ( caloric ) was preferred . thus , one reads : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the system of ondulations is the most scientific , and the most admitted in modern physics ; but that of emission lends itself more to demonstrations , for which reason it is generally preferred for the explanation of the phenomena of the caloric . ( @xcite @xcite , p. 204 ) _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the textbooks of the 1870s begin to abandon the traditional theses , and show signs of modernity . the most traditional are @xcite , an 8th edition , and @xcite . the latter already does not use the term caloric , and later editions were progressively modernized . the textbook of @xcite studies imponderable fluids , and uses the term caloric . in its treatment of radiant heat , however , the work seems very modern . as noted above for @xcite , the author prefers the caloric hypothesis for its simplicity in teaching : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the admissible hypothesis is at present that of ondulations , in the light of the advances in modern physics ; but as it simplifies the demonstrations , many physicists prefer the hypothesis of emission to explain the phenomena of heat . ( @xcite @xcite , p. 148 ) _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the textbooks with a more modern spirit are those of @xcite and @xcite . mximo fuertes acevedo s work neither studies imponderable fluids nor uses the term caloric . but energy does not appear as a general concept either , despite the mechanical theory of heat being explained . adolphe ganot s textbook was much used in europe . the first spanish edition is of the year 1853 . we consulted the 7th spanish edition @xcite . it contains a paragraph explaining the dynamic theory of gases : gases are described as formed by elastic molecules in motion , and the elasticity of a gas at a given volume is proportional to the _ vis viva _ ( total mass of the molecules multiplied by the square of their speed ) . while making use of such concepts as this ( today , of course , replaced by kinetic energy ) , the text does not deal with energy in a general form . as is to be expected with this perspective , imponderable fluids and the term caloric have been forgotten , and the mechanical theory of heat is studied . all the textbooks of the 1880s that we consulted have abandoned imponderable fluids , with the exception of the handbook of @xcite which is already in its 10th edition . of the other works , the most traditional is that of @xcite , since it neither deals with energy in a general way nor introduces the theories we are looking for . we then have the 2nd edition of the work of @xcite , with no substantial changes from the 1st edition . we also consulted a portuguese secondary education textbook , @xcite , which has a similar perspective to those of @xcite ( @xcite , @xcite ) . the textbook of @xcite is the first of the series of books that we consulted in which energy is dealt with in a general fashion . the mechanical theory of heat is studied , but the kinetic theory of gases has still not appeared . the 7th edition of the work of @xcite has almost nothing to do with the 2nd edition that we commented on above @xcite . now , energy and the mechanical theory of heat are presented . in the prologue to the 6th edition , also included in the 7th , one reads : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ with great insistence i have attempted in the treatment of heat to relate together all the phenomena of thermo - dynamic theory . ( @xcite @xcite , p. v ) _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the outstanding textbook of this decade is that written by @xcite , published in tarragona . the author held the chair of physics and chemistry in the instituto provincial of tarragona . this is a modern text which includes the kinetic theory of gases , and in general explains physical phenomena mechanistically . in the work s prologue , the author speaks about a book of his on mechanics published in 1885 , and gives great importance to this branch of physics : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the criterion that has inspired this treatise responds to the necessity already recognized by all to explain the subject of physics in a single course , always preceded by a short course of mechanics as foundation and basis of the former ( ) . ( @xcite @xcite , p. 3 ) _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ as a continuation of thermodynamics , @xcite describes the kinetic theory of gases by following the ideas of clausius and introducing the definition of free path of a molecule as the distance travelled between two consecutive collisions . he deduces mariotte s law , obtaining the formula @xmath0 , where @xmath1 is the pressure of the gas , @xmath2 its volume , @xmath3 the number of molecules , @xmath4 the mean velocity , and @xmath5 the mass of a molecule . he also deduces avogadro s hypothesis from the kinetic theory of gases @xcite . as an indication of the little that this personality and his work have been studied , there only appears one book of @xcite in the _ collected catalogue of spain s bibliographic heritage : 19th century _ this is a textbook on elementary chemistry @xcite . the main characteristic of the books of the 1890s is that they all now treat energy in a general form , as well as the mechanical theory of heat . the textbook of @xcite simultaneously uses imponderable fluids and the term caloric , but only as additional information . a more curious case is that of @xcite whose textbook uses the term caloric without studying imponderable fluids . the work of @xcite has the interest of studying heat and light side by side , to bring out their relationship as vibratory phenomena : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ( ) the molecular vibrations of the bodies , transmitted by the aether , produce the feelings of heat in the touch and of light in the sight . ( @xcite @xcite , p. 496 ) _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ other modern textbooks are those of @xcite , the pharmacist @xcite , @xcite , and @xcite . the small book of @xcite contains only definitions and principles . at no time is any idea developed , since the book s object is to serve as a collection of phrases for students to learn in preparation for their examinations . as part of these examination aids , the author prepared two plates which accompany the text , one on units and abbreviations of the decimal metric system , and another on physical units . the kinetic theory of gases does not appear explicitly , but some of its results do . for instance , one can read maxwell s law : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the viscosity of a gas measured by the coefficient of friction is independent of the density . ( @xcite @xcite , p. 60 ) _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ another modern work is that of @xcite . the author indicates on the first page in a footnote that the works that had been consulted were `` spanish : escriche , feli , muoz , rojas , and rodrguez largo . foreign : ganot , jamin , joubert , maxwell , and tyndall '' . the text contains a short paragraph on the theory of gases , in which bernouilli , clausius , and maxwell are cited . a kinetic interpretation of pressure is also given , but neither is the concept of mean free path introduced nor estimates of molecular speeds given . the textbook of @xcite , another modern work , has in the book dedicated to heat an article on the _ thermal constitution of gases _ , in which the kinetic theory is described . a kinetic interpretation of pressure is given , together with an explanation of high molecular velocities , including numerical estimates . a short section is dedicated to the _ height of the atmosphere _ , and another to the _ mean free path _ , in which one reads that `` crookes calls mean free path the space travelled by the molecule between two of those collisions '' . on the other hand , the kinetic theory of gases is not mentioned by @xcite . we also consulted some 20th - century books . they all had a modern perspective . the textbooks of @xcite and @xcite did not include the kinetic theory of gases . the 6th and 8th editions of @xcite ( @xcite , @xcite ) present improvements over the 1st edition . with respect to the kinetic theory of gases , a new paragraph is included to explain molecular velocities . the textbook of @xcite explains van der waals s equation of state . the last textbook that we wish to comment on is a translation of an italian book . it is the work of @xcite , a university textbook of modern physics where the kinetic theory of gases is developed completely . the topics in the book are surprising in their breadth and modernity . as a qualitative summary , figure [ fig1 ] shows the evolution of the responses to the questions that we formulated , according to the data listed in table [ tablei ] . spain s educational policies in the 19th - century encouraged a great similarity between the general physics textbooks of different authors , although there were voices such as that of eduardo lozano y ponce de len , in discord with this policy . this situation led to the differences between the successive editions of a textbook being minimal , and as a result , the effective lifetime of these textbooks was extremely long . a noteworthy example is the spanish translation of the work of a. ganot the 1st edition was published in 1853 and the last we know of was 1945 . likewise , there were difficulties in rooting out the antiquated content of physics textbooks . in many of them , modern and old theories shared the pages . a significant example was the case of eduardo snchez pardo , who amplified his translation @xcite with elements of old editions that had been discarded from consideration in new foreign editions . the result was to help make the recovery of spanish physics was a very slow process . with respect to the term caloric , it was used in the spain of this period with different acceptations : as an imponderable fluid responsible for thermal phenomena ( its original meaning ) , as a synonym for heat ( the latter being understood according to the mechanical theory of heat ) , and as the cause of the phenomena of heat , whatever their nature . also , during the analysis of the sources that were available , we found the orientation towards examination preparation to be excessive in the textbooks that we consulted . some are a simple cookbook of laws and physical phenomena that the students would have to learn if they were to pass their examination . hence , no interest was aroused in reflecting on content or in carrying out experiments in practical classes . certain textbooks shamelessly encouraged totally memoristic learning , so that content was not applied to new situations and was readily forgotten . another fact revealed by the analysis of the textbooks was the speed with which the mechanical theory of heat gained acceptance . this theory had become quite usual in many of the spanish physics textbooks well before the concept of energy had begun to be treated in a general form . while this fact seems disconcerting from today s viewpoint , it must be pointed out that both the mechanical theory of heat and the second law of thermodynamics had been proposed before the law of conservation of energy was generally accepted . the introduction of the kinetic theory into 19th - century spain occurred from approximately the 1870s onwards . in the 1880s , the theory already appears more developed in one of the textbooks that we consulted . this was the work of @xcite , which stands out for its mechanistic deductions of mariotte s law and avogadro s law . the topics dealt with concerning the kinetic theory in the textbooks that we consulted were : generalities of the theory , kinetic interpretation of pressure , molecular velocities , mean free path , height of the atmosphere , and the van der waals equation . lastly it should be remarked that we found no original contribution to kinetic theory in these spanish works of the end of the 19th - century . this work has been partially supported by the dges ( spain ) through grant no . pb97 - 1501 and by the junta de extremadura ( fondo social europeo ) through grant no . ipr98c019 . moreno gonzlez , a. : 1988 , _ una ciencia en cuarentena . evolucin de la fsica en la universidad y otras instituciones espaolas de mediados del siglo xviii a la crisis finisecular del xix _ , csic , madrid .

analyses of the data from reactor @xcite , accelerator @xcite , atmospheric @xcite and solar @xcite neutrino experiments conclusively demonstrate the oscillation of neutrino flavors . the results of these experiments can be interpreted by two independent mass squared differences ( between three active neutrinos ) . from the direct measurement of the invisible part of the decay width of @xmath5 boson ( _ i.e. _ , @xmath6 ) , the number of active neutrinos lighter than @xmath7 found to be @xmath8 @xcite and from the fit of the lep data to standard model prediction it found to be @xmath9 @xcite . thus , if an extra light neutrino exists , it should be a sterile neutrino ( singlet under the gauge symmetries of the standard model . ) historically the strongest hint for the existence of sterile neutrino came from the short baseline lsnd experiment @xcite . data of the lsnd experiment , with neutrino energy @xmath10 mev and baseline @xmath11 m , suggested a 3 + 1-scheme ( active+sterile ) with the new mass squared difference @xmath12 ev@xmath2 . the lsnd result has not been verified by the miniboone experiment @xcite and considerable efforts have gone into reconciling the null result of miniboone with the lsnd data @xcite . all the other data of the neutrino experiments can be interpreted by assuming only three active massive neutrinos without any need to introduce sterile neutrinos in the data analyses . however , sterile neutrinos can still be present in the yet not probed regions of the parameter space @xmath13 . these regions correspond to sterile neutrinos almost degenerate in mass with the active ones , with very tiny mass differences @xmath14 . the scenario of degenerate sterile neutrinos , the so - called `` pseudo - dirac '' ] scenario , has been proposed long time ago in @xcite and has been studied in the literature extensively @xcite . the prospect for the existence of light sterile neutrinos with masses nearly degenerate with the masses of active neutrinos is motivated in many theoretical extensions of the standard model @xcite . from the observational point of view , probing very small @xmath15 between sterile and active neutrinos needs very long baselines . neutrinos coming from the sun ( which is the farthest observed source of neutrinos with continuous emission ) set the bound @xmath16 ev@xmath2 on the active - sterile mass splitting . bounds from other performed or forthcoming experiments will be discussed in sect . [ bounds ] . in this paper we concentrate on the effects of almost degenerate sterile neutrinos on the expected flux of cosmic neutrinos coming from astrophysical sources . the new generation of km@xmath17 scale neutrino telescopes give a unique opportunity to probe the very tiny @xmath15 in pseudo - dirac scenario . the cosmic neutrinos from sources such as grbs @xcite , agn @xcite and type ib / c supernovae @xcite travel large distances over @xmath18 mpc before arriving at neutrino telescopes in earth . with such extremely long baseline , tiny mass squared differences as small as @xmath19 can be probed . the idea of using neutrino telescopes to discover the sterile neutrinos present in pseudo - dirac scenario was proposed in @xcite . in order to probe small values of @xmath15 , it has been suggested to look at distortions in the spectrum of @xmath20 from supernovae remnants in the average distance of @xmath21 kpc in @xcite and the spectrum of @xmath20 from galactic center in @xcite . the authors of @xcite evaluate the effect of the pseudo - dirac neutrinos on the flavor composition of the cosmic neutrinos ; _ i.e. _ , the deviation of the @xmath22 ( where @xmath23 is the flux of @xmath24 at earth ) from the expected value @xmath25 in the absence of sterile neutrinos . in the measurement of flavor composition of neutrinos in neutrino telescopes , the uncertainties in the relevant parameters and experimental limitations should be taken into account . for example , there are uncertainties in the mixing parameters of neutrinos and also in the spectrum of the arriving neutrinos . also , the current constructed or proposed neutrino telescopes , amanda@xmath26icecube @xcite , nemo @xcite , nestor @xcite , antares @xcite and km3net @xcite can not identify all three flavors of the active neutrinos . @xcite considers these uncertainties and experimental limitations in the analysis of the cosmic neutrinos in order to extract mixing parameters and flavor composition of neutrinos at the source . in this paper , by considering the aforementioned uncertainties and experimental limitations , we investigate the potential of neutrino telescopes in discovering the pseudo - dirac nature of neutrinos . the initial flavor ratio of the neutrinos at the source can also be a source of uncertainty in the calculation of event rates in neutrino telescopes . we discuss the robustness of our result to this kind of uncertainty . the paper is organized as follows . in sect . [ formalism ] , the pseudo - dirac scenario for massive neutrinos is reviewed . in sect . [ bounds ] , the current bounds on @xmath15 from various neutrino experiments are summarized ; and in sect . [ cosmic ] the effects of sterile neutrinos on the flavor composition of cosmic neutrinos are discussed . [ detection ] is devoted to the production mechanism of neutrinos at the source and their detection processes in the neutrino telescopes . various sources of uncertainties that enter the calculation of event rates in neutrino telescopes are enumerated . sect.[result ] , summarizes the results of the present analysis on the capability of neutrino telescopes to discriminate between pseudo - dirac and conventional scenarios . a summary of the results and the conclusions are given in sect . [ conclusion ] . a simple and economic way to generate mass for neutrinos in the sm is to add right - handed ( sterile ) neutrinos to the matter content of sm . in the presence of @xmath27 right - handed ( sterile ) fields @xmath28 ( @xmath29 ) , we define the following column matrix @xmath30 of @xmath31 left - handed fields @xmath32 where @xmath33 and @xmath34 is the charge conjugation operator . for majorana neutrinos which we consider here @xmath35 . here we consider models with at most three sterile neutrinos ( @xmath36 ) . in the basis @xmath30 , the generic mass term for neutrinos is @xmath37 the @xmath38 mass matrix @xmath39 is of the following form ( after electroweak symmetry breaking ) @xmath40where @xmath41 is the @xmath42 dirac mass matrix and @xmath43 and @xmath44 are the @xmath45 left - handed and @xmath46 right - handed majorana mass matrices , respectively . the non - vanishing elements of @xmath43 and @xmath44 violate lepton numbers while by assigning the lepton number @xmath47 to sterile neutrinos , @xmath41 conserves this symmetry . the left - handed mass matrix @xmath43 is not invariant under the sm gauge group @xmath48 and should be zero unless other new particles ( such as a new higgs triplet ) are present . the elements of @xmath44 can take a wide range of values , it can be as large as the gut scale @xmath49 gev which are preferred in _ see - saw _ mechanisms , or it can vanish like @xmath43 as a result of new gauge symmetries such as @xmath50 @xcite . the case @xmath51 and @xmath52 results in pure dirac neutrinos . in this case the six weyl neutrinos decompose into three pairs of neutrinos with degenerate masses . the active - sterile mixing angle in each pair is maximal @xmath53 , but the active neutrinos do not oscillate to their sterile partners because @xmath54 , where @xmath55 and @xmath56 are the masses of sterile and active neutrinos in the @xmath57-th pair , respectively . here we are interested in the case @xmath58 . the non - zero but very small values of the elements of @xmath43 and @xmath44 lift the degeneracy in mass at each pair . in this `` pseudo - dirac '' scenario , active - sterile mixing angle in each pair is @xmath59 and active - sterile oscillation can in principle occur due to very small but non - zero @xmath60 . to illustrate this point , let us consider the one generation example . in this case , the mass matrices @xmath43 , @xmath44 and @xmath41 in eq . ( [ massmatrix ] ) are numbers ( we assume that all the masses are real . ) in the pseudo - dirac limit , we obtain @xmath61 and @xmath62 . notice that in the pseudo - dirac scenario , neutrinos oscillate even in one generation , in contrast to pure dirac scenario where oscillation occurs only between generations . in general the @xmath63 symmetric mass matrix @xmath39 ( where @xmath31 ) can be diagonalized by @xmath64 , where @xmath65 is a @xmath63 unitary matrix . we choose the elements of @xmath65 such that @xmath66 . the mixing matrix @xmath67 appearing in the weak charged - current @xmath68 is a @xmath69 rectangular matrix with the elements @xmath70 , where @xmath71 is the @xmath72 diagonalizing unitary matrix of charged leptons mass matrix . in the case @xmath52 , the @xmath73 matrix @xmath67 can be parameterized by 12 mixing angles and 12 cp - violating phases ( 7 dirac phases+5 majorana phases . ) it has been shown in @xcite that in the pseudo - dirac limit @xmath58 and at first order of perturbation in the small parameters @xmath74 and @xmath75 , the mixing matrix @xmath67 has only three mixing angles ( responsible for oscillation between the pairs ) and three cp - violating phases ( 1 dirac phases+2 majorana phases ) . this fact can be seen from the explicit form of the matrix @xmath65 which diagonalizes the mass matrix @xmath39 ( in the pseudo - dirac limit ) @xcite : @xmath76 + where @xmath77 is the @xmath72 conventional neutrino mixing matrix of left - handed neutrinos , @xmath78 is the @xmath79 unitary matrix which diagonalizes the right - handed majorana mass matrix , @xmath80 is the @xmath81 identity matrix and the matrices @xmath82 ( @xmath36 ) are : @xmath83 + the flavor conversion probability between the active neutrinos @xmath84 is @xmath85 + using the explicit form of the matrix @xmath65 in eq . ( [ vnu ] ) , the probability @xmath86 becomes @xmath87 where @xmath88 and @xmath89 are the mass eigenvalues in the @xmath57-th pair of active and sterile neutrinos ; @xmath90 and @xmath91 are the elements of the @xmath72 mxixng matrix @xmath77 . notice that this relation reduces to the standard flavor conversion probability formula in the limit of pure dirac neutrinos @xmath92 ( @xmath93 ) . by setting @xmath92 for the active neutrino generations which do not have sterile partners , eq . ( [ probability ] ) also applies to cases with @xmath94 . using eq . ( [ probability ] ) in analyzing the data of oscillation experiments gives information on @xmath95 in each pair . in subsect . [ bounds ] we review the current bounds on active - sterile mass square differences and the prospect of future experiments to improve these bounds . in subsect . [ cosmic ] we discuss the implications of eq . ( [ probability ] ) on the flavor composition of cosmic neutrinos . an oscillation experiment with baseline @xmath96 and neutrino energy @xmath97 can probe mass square difference @xmath98 . if @xmath99 , the baseline is too short for flavor oscillation to take place ; on the other hand , if @xmath100 so many oscillations take place during the propagation and the oscillatory term should be averaged out . in both of these cases it is not possible to derive the value of @xmath15 in oscillation experiments . solar neutrino experiments with the baseline 1 au @xmath101 m and neutrino energy @xmath102 mev , can probe mass squared differences @xmath103 ev@xmath2 . these very small values of @xmath15 has been favored by the so - called `` vacuum oscillation solution '' of the solar neutrino problem , but as is well - known this solution has been ruled out by kamland @xcite . however , the sterile - active oscillation with mass square differences @xmath104 ev@xmath2 can still be present as a subdominant effect in solar data . the recent work @xcite updates the solar data and obtains @xmath105 ev@xmath2 ( at @xmath106 level ) for the sterile - active mass splitting . this bound is the most stringent bound on @xmath15 . the flavor composition of the neutrinos from core - collapse supernovae ( sne ) also can change due to an active - sterile oscillation from the sn to earth . the mean energy of the neutrinos from a sn explosion is @xmath107 mev . thus , a sn explosion at a distance of @xmath108 kpc can probe @xmath109 ev@xmath2 . the constraint from the data of the sn1987a data is not restrictive because of the low statistics and high uncertainties in the mechanism of sne explosion @xcite . construction of future mton water - erenkov detectors can dramatically improve the current bound or find a hint for sterile neutrinos hypothesizes in pseudo - dirac scenario @xcite . population of the sterile neutrinos in the early universe and their effects on the big bang nucleosynthesis ( bbn ) can change the abundance of light elements . the standard bbn , given the number of the relativistic particles @xmath110 and the baryon asymmetry @xmath111 , predicts the abundance of the light nuclei in the universe . assuming that the sterile neutrinos are produced only in the active / sterile oscillation and the initial abundance of sterile neutrinos at temperatures @xmath112 mev is zero , the tightest limit comes from the @xmath113he abundance : @xmath114 ev@xmath2 @xcite . two main non - oscillation neutrino experiments which probe neutrino masses kinematically are tritium beta decay and neutrinoless double beta decay ( @xmath115 ) experiments . among them , the @xmath115 decay is sensitive to the majorana or dirac nature of neutrinos . the rate of @xmath115 decay is proportional to the effective mass of the electron neutrino which is defined as @xmath116 + as mentioned after eq . ( [ massmatrix ] ) , in the limit @xmath51 ( pure dirac neutrino ) each dirac neutrino is the superposition of two majorana neutrinos with degenerate masses and opposite cp eigenvalues . it is easy to see that the majorana neutrinos in each pair interfere destructively in eq . ( [ mee ] ) ( @xmath92 ) which results in @xmath117 for pure dirac neutrinos . in the pseudo - dirac scenario with non - zero majorana masses and @xmath58 , the cancelation is not exact and @xmath118 but it is very small @xcite . thus , it seems that the observation of a positive signal in the next generation @xmath115 experiments , with sensitivities @xmath119 mev , will rule out the small values of @xmath43 and @xmath44 in the mass matrix of neutrinos and therefore pseudo - dirac scenario . however , two points should be considered . the first one is that the value of @xmath120 can still be significant if only one or two families of neutrinos have sterile partners . contribution of each family to the value of @xmath120 depends on the corresponding mixing matrix element @xmath90 . this means that , because of the small value of @xmath121 ( @xmath122 ) , presence or absence of a sterile neutrino with a mass almost degenerate with @xmath123 do not change the value of @xmath124 substantially ; but the case with two sterile neutrinos with a masses degenerate with @xmath125 and @xmath126 leads to an effective mass @xmath120 much smaller than its value in the absence of sterile neutrinos . the second point is that the dominant contribution to @xmath115 decay can come from new particles or physics beyond the sm , such as a @xmath127 interaction @xcite . it is shown in @xcite that a non - zero @xmath115 decay rate generates small @xmath43 through radiative corrections , which results in pseudo - dirac scenario for neutrino masses . considering these points , it is not easy to draw a conclusion on pseudo - dirac scenario from the results of the @xmath115 experiments . neutrinos arriving at neutrino telescopes from astrophysical sources travel distances of the order @xmath128 mpc . the flavor conversion probabilities over these large distances can be obtained by averaging out the oscillatory terms in eq . ( [ probability ] ) . two different scales of @xmath15 are involved in eq . ( [ probability ] ) , one is the atmospheric , @xmath129 ev@xmath2 , and solar , @xmath130 ev@xmath2 , and the other is the very small @xmath15 between the mass eigenstates in each pair of active and sterile neutrinos . as it is shown in @xcite the oscillatory terms depending on @xmath131 and @xmath132 should be completely averaged out over these large distances . the mass squared difference @xmath15 that can be probed by neutrinos with energy @xmath97 which propagate through distance @xmath96 is @xmath133 + thus , even for @xmath96 as large as 10 mpc oscillatory terms given by @xmath134 ev@xmath2 do not average out . after averaging out @xmath131 and @xmath135 , eq . ( [ probability ] ) becomes @xcite @xmath136 + where @xmath137 is the mass squared difference in the @xmath57-th pair . thus , if the initial flavor composition of neutrinos in the source is @xmath138 , the flavor composition of the neutrino beam arriving at earth will be @xmath22 , where @xmath139 + the average of the cosine factor for @xmath140 is @xmath141 . thus , if for all three pairs ( @xmath93 ) the condition @xmath140 is satisfied , all three @xmath23 in eq . ( [ flavor ] ) are multiplied by @xmath141 such that the flavor ratios @xmath22 do not change with respect to the flavor ratios in the pure dirac case @xmath142 . in this situation the only difference between the pure dirac and pseudo - dirac scenarios is that the number of neutrinos arriving at earth is reduced by half in the pseudo - dirac case . thus , it is very hard to verify pseudo - dirac scenario for distances or mass squared differences where @xmath140 for @xmath93 . that is because the estimation of the overall normalization of neutrino flux needs knowledge about the details of the neutrino production mechanism in the source which is not well understood . however , the pseudo - dirac scenario of neutrinos can be tested in neutrino telescopes for the cases that only one or two sterile neutrinos exist ( @xmath94 ) or the distance of the source or the mass square differences are such that all the three oscillatory terms given by @xmath143 do not average out . in these cases the average of one or two of the cosine factors in eq . ( [ flavor ] ) are @xmath141 and the other cosine factors can be replaced by @xmath144 ( we do not consider the special situations where @xmath145 ) . thus , measuring the deviations of the flavor ratios in eq . ( [ flavor ] ) from their standard values ( in the absence of nearly degenerate sterile neutrinos ) in neutrino telescopes can shed light on these cases . in the next section we discuss the details of the detection processes in the neutrino telescopes and the feasibility of identifying different neutrino flavors in these experiments . by taking into account the realistic measurable quantities in neutrino telescopes and the uncertainties in these measurements , we discuss to what extent it is possible to measure the flavor ratio of cosmic neutrinos and their deviations from the standard values . in this section we briefly discuss neutrino flavor identification in the km@xmath17 scale neutrino telescopes such as icecube or its counterparts in the mediterranean sea ; km3net , nemo , nestor and antares . a detailed description of the flavor tagging efficiencies in a typical neutrino telescope can be found in @xcite . here we summarize the main points relevant for the present analysis . particularly , for the first time , we take into account different sources of uncertainties in the calculation of event rates and also a more realistic analysis of the detectable events , such as the contribution of @xmath146 ( @xmath147 ) to the @xmath3-tracks . the flux of neutrinos arriving at earth can come from a single luminous point source or as a diffuse flux from sum over different sources at different distances . the advantage of the point sources is that the short period of the burst and the direction of incoming neutrinos can be used to reduce the background events ( especially when the source can be identified using a different method such as gamma photons for grbs ) . point sources with an intense neutrino flux detectable at km@xmath17-scale neutrino telescopes can take place in the close - by galaxies located at a distance of @xmath148 mpc and such a source of neutrinos yields about a few hundred neutrino events in icecube @xcite . discriminating between different flavors of neutrinos is a great challenge for neutrino telescopes . two types of events are completely distinguishable in the next generation of these experiments : @xmath3-track events and shower - like events . charged current ( cc ) and neutral current ( nc ) interactions of different flavors can contribute to each of these events . @xmath3-track events , which are the erenkov light radiated by muons propagating through the volume of the detector , get contributions from two sources : i ) @xmath3 ( @xmath149 ) produced in the cc interaction of @xmath20 ( @xmath150 ) ; ii ) cc interaction of @xmath146 ( @xmath147 ) which produce @xmath151 ( @xmath152 ) leptons and the subsequent _ leptonic _ decay of tau leptons @xmath153 ( @xmath154 ) produce @xmath3 ( @xmath149 ) . shower - like events have three sources : i ) nc interactions of all the three flavors of neutrinos ; ii ) cc interactions of @xmath155 and @xmath156 ; and iii ) cc interaction of @xmath146 ( @xmath147 ) which produce @xmath151 ( @xmath152 ) leptons and their subsequent _ hadronic _ decays . the exact formulae for calculating the rate for each of these events can be found in sect . 2 of @xcite . the threshold energy of the detection of @xmath3-tracks and showers in experiments such as icecube , respectively , is @xmath157 gev and @xmath158 tev @xcite . on the other hand , the mean free path of the neutrinos with energy @xmath159 tev becomes of the order of @xmath160 , the diameter of earth . the exact values of the @xmath161 , @xmath162 and @xmath163 depend on the details of the experiments , such as the geometry of the photomultipliers in the volume of the detector and the direction of the incoming neutrinos ; and a dedicated analysis can be done for each experiment . here , in order to avoid considering the absorption of neutrinos in the earth , we restrict the analysis to 100 gev @xmath164 tev . the realistic quantity that can be measured in neutrino telescopes is @xmath165 + the value of @xmath166 can be calculated if we know the initial flux and the flavor ratio of neutrinos at the source . the main mechanism of neutrino production at the astrophysical sources is the interaction of the accelerated proton by the ambient protons and photons . the decay of the secondary particles ( @xmath167 , @xmath168 , @xmath169 , ) produced in these @xmath170 and @xmath171 interactions generate neutrinos and muons . for example , the pion chain decays @xmath172 and @xmath173 generate neutrinos with the flavor ratio @xmath174 . the exact value of the flavor ratio @xmath138 depends on the spectrum of the parent particles and the properties of the production medium . a class of models based on the fermi acceleration mechanism for the particle in the source , predict a power - law spectrum for neutrinos @xmath175 + where @xmath176 is the spectral index and @xmath177 is a normalization factor . acceleration of particles through fermi acceleration mechanism @xcite results in @xmath178 for neutrino spectrum . however , non - linear effects change this value such that @xmath176 can take any value in the interval @xmath179 @xcite . it is shown in @xcite that for the case of pion decay chain and assuming @xmath178 , the initial flavor ratio is @xmath180 ( the difference with @xmath181 comes from the wrong polarization states of @xmath182 in the decay of @xmath167 ) . also the authors of @xcite show that inclusion of other secondary particles ( such as @xmath168 ) has a very little effect on this value . the @xmath182 generated in the decays of the secondary particles can substantially lose their energy before decay . in this case the neutrinos generated in the decay of muons do not contribute to the flux of neutrinos with 100 gev @xmath183 tev and the flavor ratio becomes @xmath184 . in the calculation of @xmath166 for different scenarios of neutrino production at source and propagation between source and earth , the uncertainties in the input parameters should be considered . uncertainties of the input parameters in the calculation of the @xmath3-track and shower - like event rates induce uncertainties in the value of @xmath166 . here we summarize these sources of uncertainties : mixing parameters : : flavor content of the neutrino beam arriving at earth depends on the mixing angles @xmath185 and dirac cp - violating phase @xmath186 through the @xmath187 factors in eq . ( [ flavor ] ) . an update on the values of these parameters and uncertainties in each of them can be found in @xcite , which are also listed in table [ para ] . spectral index : : as it is mentioned after eq . ( [ powerlaw ] ) , the spectral index @xmath176 can take any value in the interval @xmath179 . it is shown in @xcite that icecube can measure the spectral index @xmath176 with 10 % precision ( assuming @xmath188 gev @xmath189 sr@xmath190 yr@xmath190 and after one year of data - taking ) . @xmath191 ratio : : let us define @xmath192 for @xmath193 . it is obvious that @xmath194 in the pion decay chain , but the value of @xmath195 depends on the ratio of @xmath196 production and can take any value in the interval @xmath197 . to best of our knowledge , in the energy range we are interested here ( 100 gev @xmath164 tev ) , there is not any proposed or established method to measure @xmath195 . neutrino - nucleon cross section : : the rate of the @xmath3-track and shower - like events depends on the cc and nc neutrino - nucleon cross sections @xmath198 and @xmath199 . the current uncertainty in these cross sections is @xmath200 . however , the uncertainties in @xmath198 and @xmath199 have a very small effect on @xmath166 because of the cancelation between numerator and denominator of eq . ( [ ratior ] ) . .relevant parameters in the calculation of @xmath166 with their uncertainties . the current uncertainty column represents the @xmath106 uncertainty interval for mixing angles @xcite . the future uncertainty column shows the precisions that can be achieved in the forthcoming neutrino oscillation experiments , as described in the corresponding references . for @xmath201 it is assumed that its value is close to the current upper limit : @xmath202 . [ cols="^,^,^,^",options="header " , ] in table [ para ] we have listed the relevant parameters in the calculation of @xmath166 and their uncertainty intervals . the third and forth columns respectively correspond to the current and future uncertainties in these parameters . in addition to the uncertainties in the input parameters listed above , measurement of @xmath166 is also done with limited precision and has an uncertainty . it is shown in @xcite that by assuming that the flux of neutrinos is @xmath203 gev @xmath189 sr@xmath190 yr@xmath190 , the ratio @xmath166 can be measured with @xmath204 % precision after a couple of years of data - taking . now , by taking into account the uncertainties mentioned in this section , the question is to what extent it is possible to measure the deviations of the flavor composition @xmath22 from its value in the absence of pseudo - dirac sterile neutrinos . in the next section we will discuss the prospect of neutrino telescopes to measure these deviations . in this section we consider two sources with different initial flavor compositions : the pion and stopped - muon sources with flavor compositions @xmath205 and @xmath184 , respectively . the value of @xmath166 in eq . ( [ ratior ] ) in the absence of almost degenerate sterile neutrinos and assuming the best - fit values for the mixing angles , @xmath206 and @xmath207 , will be denoted by @xmath208 and @xmath209 , for pion and stopped - muon sources respectively . we assume the power - law spectrum eq . ( [ powerlaw ] ) for the neutrino production , with the spectral index @xmath178 . at the end , we investigate the robustness of the results with respect to the deviation of initial flavor ratios from the assumed values @xmath205 and @xmath184 . ( a)(b ) ( c)(d ) ( e)(f ) ( a)(b ) ( c)(d ) ( e)(f ) fig . ( [ 120 ] ) shows @xmath166 versus @xmath176 for pion source , where @xmath210 . the hatched area between the two red curves corresponds to the values that @xmath166 can take if the input parameters vary in the uncertainty intervals . the inputs have been varied in the current uncertainty intervals shown in the third column of table [ para ] . the two horizontal and vertical pairs of dashed - lines show the 7 % and 10 % precision intervals in the measurement of @xmath166 and @xmath176 , respectively . the rectangle created from the intersection of these dashed - lines corresponds to the region of parameter space @xmath211 where can be limited by the measurements in neutrino telescopes . thus , the points inside the rectangle represent the values of @xmath166 for the case of no sterile neutrino and consistent with 7 % ( 10 % ) precision in the measurement of @xmath166 ( @xmath176 ) . the green dashed - curves in fig . ( [ 120 ] ) show the values of @xmath166 in the presence of sterile neutrinos . for example , in fig . ( [ 120]-a ) the hatched region between the green dashed - curves corresponds to the case when the sterile neutrino mass is almost degenerate with the mass of the active neutrino @xmath125 such that the oscillatory term depending on @xmath212 in eq . ( [ probability ] ) can be averaged to @xmath141 and the flavor conversion probabilities become @xmath213 as can be seen , in all six parts of the fig . ( [ 120 ] ) the two hatched areas overlap . the overlapping of the hatched areas inside the dashed - line rectangles means that the presence of the sterile neutrinos can not be ruled out , even if the measurement of @xmath166 gives a value inside this rectangles . however , if the measurement of @xmath166 gives a value much different than @xmath214 , the existence of sterile neutrinos is favored . for example , the value @xmath215 is not possible in the no sterile case and this value is favored by the scenarios depicted in fig . ( [ 120]-b , c , f ) . fig . ( [ 120new ] ) is the same as fig . ( [ 120 ] ) with the exception that the input parameters have been varied in the future uncertainty intervals shown in the forth column of table [ para ] . we assumed ( 6% , 6% , 5% ) uncertainties for ( @xmath216 ) , which can be achieved in the forthcoming neutrino oscillation experiments @xcite . also , we assumed a large value for the 13-mixing angle @xmath202 which is near its present upper bound . the uncertainties of @xmath186 , @xmath195 , @xmath176 and @xmath166 are the same as in fig . ( [ 120 ] ) . as can be seen from fig . ( [ 120new ] ) , reducing the uncertainties of mixing angles results in a separation between the two hatched areas where there was an overlap in fig . ( [ 120 ] ) . for example , in fig . ( [ 120new]-f ) , the regions corresponding to no sterile neutrino case and the case with averaged second and third pairs are completely separated . this separation means that if the measurement of @xmath166 gives a value inside the rectangle in fig . ( [ 120new]-f ) , the existence of sterile neutrinos ( for the case of average on second and third pairs ) will be ruled out . on the other hand , a value of @xmath166 outside the rectangle can be interpreted as a signal for the existence of sterile neutrinos . however , recognizing which case is consistent with the measurement of @xmath166 should be done with care . to illuminate this point , let us assume that icecube has measured @xmath217 . according to the diagrams in fig . ( [ 120new ] ) this value of @xmath166 is consistent with three cases : i ) average on first pair ; ii ) average on first and second pairs ; and iii ) average on first and third pairs . notice that with the current uncertainties of the mixing angles ( depicted in fig . ( [ 120 ] ) ) it is not possible to conclude that @xmath217 is a signal of sterile neutrinos . figs . ( [ 010 ] ) and ( [ 010new ] ) show the dependence of @xmath166 on @xmath176 for the stopped - muon source ; where @xmath218 . the input parameters have been varied in the current uncertainty intervals in fig . ( [ 010 ] ) . comparing fig . ( [ 010 ] ) with fig . ( [ 120 ] ) , it is obvious that the hatched areas are wider for the stopped - muon source , which means that the recognition of sterile neutrinos is harder for this kind of sources . in contrast to the pion source , reducing the uncertainty of input parameters to the future uncertainty intervals do not lead to a separation between the hatched areas in fig . ( [ 010new ] ) . however , some cases can marginally be discriminated such as the case shown in fig . ( [ 010new]-d ) . ( a)(b ) ( c)(d ) ( e)(f ) ( a)(b ) ( c)(d ) ( e)(f ) ( a)(b ) in drawing figs . ( [ 120],[120new ] ) and figs . ( [ 010],[010new ] ) we have assumed initial flavor ratios @xmath205 and @xmath184 , respectively . however , it should be noticed that the initial flavor ratio can deviates from these values due to the interplay of different mechanisms in neutrino production or effects that have not been considered in the calculation of these values . for example , in the pion source , a _ part _ of the produced muons in the decays of @xmath167 ( not all of them ) can lose their energy before decay such that the initial flavor ratio of neutrinos takes a value between the two extreme cases of pion and stopped - muon sources . the question that arises here is that to what extent the results of this section ( for example the separation between hatched areas in fig . ( [ 120new]-f ) ) are robust against the deviations of the initial flavor ratio @xmath138 from the assumed values . to answer this question we consider particularly the case of fig . ( [ 120new]-f ) where the hatched areas are completely separated . also we use the following parametrization of the initial flavor ratio has been used in @xcite . ] @xmath219 the @xmath220 in the above parametrization comes from the fact that the number of prompt @xmath146 ( from the decays @xmath221 ) is very small and can be neglected . in fig . ( [ robustness]-a ) we have varied @xmath222 in eq . ( [ initial ] ) such that the separated regions in fig . ( [ 120new]-f ) begin to overlap . in drawing fig . ( [ robustness]-a ) we have varied @xmath223 . the initial flavor ratios @xmath224 corresponding to this interval are shown in fig . ( [ robustness]-b ) . as can be seen , for nearly large deviations of the initial flavor ratio from @xmath205 , the hatched areas remain separated , which means that the lack of knowledge about the exact value of the initial flavor ratio of neutrinos at the source do not affect substantially the capability of neutrino telescopes in probing pseudo - dirac neutrino scenario . the new generation of km@xmath17 neutrino telescopes opens a new window to study cosmos via the detection of high energy neutrinos predicted to be emitted from the astrophysical objects . because of the extremely large distance of the sources ( @xmath225 mpc ) the flavor oscillation is sensitive to the very tiny mass squared differences @xmath226 ev@xmath2 @xmath227 ev@xmath2 . the existence of sterile neutrinos with masses almost degenerate with the active ones such that the mass squared differences between the active and sterile neutrinos lie in the above region is hypothesized in many models ( the so - called pseudo - dirac scenario . ) we have studied the effect of these sterile neutrinos on the flux of cosmic neutrinos and discussed the capability of icecube to verify the existence of them . in the analysis we have considered different cases corresponding to the existence of sterile neutrinos degenerate in mass with different @xmath228 such that the oscillatory terms driven by these mass squared differences can be averaged out . the detection power of icecube ( and other proposed km@xmath17 neutrino telescopes ) , in the range of neutrino energies 100 gev @xmath229 tev , is limited to distinguishing two types of events : @xmath3-track and shower - like events . we have considered the ratio of these events , @xmath166 in eq . ( [ ratior ] ) , as the realistic quantity that can be measured in the icecube . we have studied the possibility of using the measured value of @xmath166 as a discriminator between the pseudo - dirac scenario and the scenario with no sterile neutrinos . we have considered various sources of uncertainties in our analysis . one part of these uncertainties comes from the imprecision of the neutrino oscillation experiments in the determination of mixing parameters . for this part we considered two sets of uncertainty intervals for the mixing parameters : i ) the current uncertainty intervals from the performed experiments ; and ii ) the future uncertainty intervals that will be achieved in the forthcoming experiments . both of these sets have been depicted in table [ para ] . among the mixing parameters , the measurable quantity @xmath166 is most sensitive to the exact value of the mixing angle @xmath230 . for the uncertainty of this parameter we have used the 6 % precision on @xmath231 @xcite depicted in table . the other part of uncertainties comes from the not completely known mechanism of neutrino production in the sources . many models of neutrino production predict a power - law spectrum for the cosmic neutrinos . however , the value of the spectral index in the power - law spectrum depends on the details of the neutrino production mechanism and can take values in the interval @xmath179 . also , the ratio of the number of electron anti - neutrinos to the number of electron neutrinos is not known . we have taken into account all these uncertainties in our analysis . the analysis has been done for two different initial flavor composition @xmath138 at the source : i ) pion source @xmath205 ; and ii ) stopped - muon source @xmath184 . it has been shown that with a higher precision of the mixing angles achievable in the forthcoming oscillation experiments , in certain cases it is possible to demonstrate or rule out the existence of sterile neutrinos hypothesized in the pseudo - dirac scenario . in these cases the regions corresponding to existence and absence of sterile neutrinos are well - separated in the parameter space for neutrinos coming from pion sources . for example , it is very promising to probe the case of two sterile neutrinos with masses almost degenerate with @xmath126 and @xmath123 ( see fig . ( [ 120new]-f ) ) . for neutrinos coming from stopped - muon sources , these regions mostly overlap such that their discrimination can not be done without ambiguity . also , the robustness of these results has been tested against the uncertainties in the initial flavor ratio of neutrinos at the source . it has been shown that for reasonably large variations of the initial flavor ratio around the expected value for pion source , @xmath205 ( see fig . ( [ robustness ] ) for clarification ) , the regions corresponding to the existence and the absence of sterile neutrinos remain separated . the author is grateful to y. farzan for useful discussions and for her valuable comments on the manuscript . also , i would like to thank h. firouzjahi for the careful reading of the manuscript and valuable comments . i would like to thank `` bonyad - e melli - e nokhbegan '' for partial financial support . k. eguchi _ et al . _ [ kamland collaboration ] , phys . * 90 * ( 2003 ) 021802 [ arxiv : hep - ex/0212021 ] ; m. apollonio _ et al . _ [ chooz collaboration ] , eur . j. c * 27 * ( 2003 ) 331 [ arxiv : hep - ex/0301017 ] . e. aliu _ et al . _ [ k2k collaboration ] , phys . rev . lett . * 94 * ( 2005 ) 081802 [ arxiv : hep - ex/0411038 ] ; d. g. michael _ et al . _ [ minos collaboration ] , phys . * 97 * ( 2006 ) 191801 [ arxiv : hep - ex/0607088 ] . s. n. ahmed _ et al . _ [ sno collaboration ] , phys . * 92 * ( 2004 ) 181301 [ arxiv : nucl - ex/0309004 ] ; g. alimonti _ et al . _ [ borexino collaboration ] , astropart . * 16 * ( 2002 ) 205 [ arxiv : hep - ex/0012030 ] . a. aguilar _ et al . _ [ lsnd collaboration ] , phys . d * 64 * ( 2001 ) 112007 [ arxiv : hep - ex/0104049 ] ; c. athanassopoulos _ et al . _ [ lsnd collaboration ] , phys . * 81 * ( 1998 ) 1774 [ arxiv : nucl - ex/9709006 ] . y. farzan , t. schwetz and a. y. smirnov , jhep * 0807 * ( 2008 ) 067 [ arxiv:0805.2098 [ hep - ph ] ] ; a. de gouvea and y. grossman , phys . d * 74 * ( 2006 ) 093008 [ arxiv : hep - ph/0602237 ] ; t. schwetz , jhep * 0802 * ( 2008 ) 011 [ arxiv:0710.2985 [ hep - ph ] ] ; v. barger , d. marfatia and k. whisnant , phys . d * 73 * ( 2006 ) 013005 [ arxiv : hep - ph/0509163 ] . an incomplete list is : e. j. chun , c. w. kim and u. w. lee , phys . d * 58 * ( 1998 ) 093003 [ arxiv : hep - ph/9802209 ] ; k. e. cahill , arxiv : hep - ph/9912416 ; k. r. s. balaji , a. kalliomaki and j. maalampi , phys . b * 524 * ( 2002 ) 153 [ arxiv : hep - ph/0110314 ] ; a. s. joshipura and s. d. rindani , phys . b * 494 * ( 2000 ) 114 [ arxiv : hep - ph/0007334 ] ; u. sarkar , phys . d * 59 * ( 1999 ) 037302 [ arxiv : hep - ph/9808277 ] ; a. geiser , phys . b * 444 * ( 1999 ) 358 [ arxiv : hep - ph/9901433 ] . j. k. becker , phys . * 458 * ( 2008 ) 173 [ arxiv:0710.1557 [ astro - ph ] ] ; k. mannheim , r. j. protheroe and j. p. rachen , phys . d * 63 * ( 2001 ) 023003 [ arxiv : astro - ph/9812398 ] ; k. mannheim , astropart . phys . * 3 * ( 1995 ) 295 . a. kappes , j. hinton , c. stegmann and f. a. aharonian , astrophys . j. * 656 * ( 2007 ) 870 [ erratum - ibid . * 661 * ( 2007 ) 1348 ] [ arxiv : astro - ph/0607286 ] ; s. gabici and f. a. aharonian , arxiv:0705.3011 [ astro - ph ] ; f. halzen , a. kappes and a. omurchadha , phys . d * 78 * ( 2008 ) 063004 [ arxiv:0803.0314 [ astro - ph ] ] ; f. vissani , astropart . * 26 * ( 2006 ) 310 [ arxiv : astro - ph/0607249 ] . k. enqvist , k. kainulainen and m. j. thomson , nucl . b * 373 * ( 1992 ) 498 ; x. shi , d. n. schramm and b. d. fields , phys . d * 48 * ( 1993 ) 2563 [ arxiv : astro - ph/9307027 ] ; d. p. kirilova and m. v. chizhov , nucl . b * 591 * ( 2000 ) 457 [ arxiv : hep - ph/9909408 ] .

suppose we are given a finite family of finite or countable sets @xmath1 with distinguished vertices @xmath2 for @xmath3 . the free product of the sets @xmath4 is given by @xmath5 , the set of all finite words of the form @xmath6 such that each letter is an element of @xmath7 and two consecutive letters arise not from the same @xmath4 . we consider a transient markov chain @xmath8 on @xmath9 starting at the empty word @xmath10 , which arises from a convex combination of transition probabilities on the sets @xmath4 . denote by @xmath11 the distribution of @xmath12 . we are interested in whether the sequence @xmath13/n$ ] converges , and if so , to compute this constant . if the limit exists , it is called the _ asymptotic entropy_. in this paper , we study this question for random walks on general free products . in particular , we will derive three different formulas for the entropy by using three different techniques . let us outline some results about random walks on free products : for free products of finite groups , mairesse and mathus @xcite computed an explicit formula for the rate of escape and asymptotic entropy by solving a finite system of polynomial equations . their result remains valid in the case of free products of infinite groups , but one needs then to solve an infinite system of polynomial equations . gilch @xcite computed two different formulas for the rate of escape with respect to the word length of random walks on free products of graphs by different techniques , and also a third formula for free products of ( not necessarily finite ) groups . the techniques of @xcite are adapted to the present setting . asymptotic behaviour of return probabilities of random walks on free products has also been studied in many ways ; e.g. gerl and woess @xcite , @xcite , sawyer @xcite , cartwright and soardi @xcite , and candellero and gilch @xcite . our proof of existence of the entropy envolves generating functions techniques . the techniques we use for rewriting probability generating functions in terms of functions on the factors of the free product were introduced independently and simultaneously by cartwright and soardi @xcite , woess @xcite , voiculescu @xcite and mclaughlin @xcite . in particular , we will see that asymptotic entropy is the rate of escape with respect to a distance function in terms of green functions . while it is well - known by kingman s subadditive ergodic theorem ( see kingman @xcite ) that entropy ( introduced by avez @xcite ) exists for random walks on groups whenever @xmath14<\infty$ ] , existence for random walks on other structures is not known a priori . we are not able to apply kingman s theorem in our present setting , since we have no ( general ) subadditivity and we have only a partial composition law for two elements of the free product . for more details about entropy of random walks on groups we refer to kaimanovich and vershik @xcite and derriennic @xcite . an important link between drifts and harmonic analysis was obtained by he proved that for symmetric finite range random walks on groups the existence of non - trivial bounded harmonic functions is equivalent to a non - zero rate of escape . karlsson and ledrappier @xcite generalized this result to symmetric random walks with finite first moment of the step lengths . this leads to a link between the rate of escape and the entropy of random walks , compare e.g. with kaimanovich and vershik @xcite and erschler and kaimanovich @xcite asked if drift and entropy of random walks on groups vary continuously on the probability measure , which governs the random walk . we prove real - analyticity of the entropy when varying the probabilty measure of constant support ; compare also with the recent work of ledrappier @xcite , who simultaneously proved this property for finite - range random walks on free groups . apart from the proof of existence of the asymptotic entropy @xmath15/n$ ] ( theorem [ thm : entropy ] ) , we will calculate explicit formulas for the entropy ( see theorems [ thm : entropy ] , [ th : entropy2 ] , [ thm : entropy - groups ] and corollary [ cor : entropy - dgf ] ) and we will show that the entropy is non - zero . the technique of our proof of existence of the entropy was motivated by benjamini and peres @xcite , where it is shown that for random walks on groups the entropy equals the rate of escape w.r.t . the greenian distance ; compare also with blachre , hassinsky and mathieu @xcite . we are also able to show that , for random walks on free products of graphs , the asymptotic entropy equals just the rate of escape w.r.t . the greenian distance ( see corollary [ cor : greenroe ] in view of theorem [ thm : entropy ] ) . moreover , we prove convergence in probability and convergence in @xmath16 ( if the non - zero single transition probabilities are bounded away from @xmath17 ) of the sequence @xmath18 to @xmath19 ( see corollary [ cor : entropy3 ] ) , and we show also that @xmath19 can be computed along almost every sample path as the limes inferior of the aforementioned sequence ( corollary [ cor : entropy2 ] ) . in the case of random walks on discrete groups , kingman s subadditive ergodic theorem provides both the almost sure convergence and the convergence in @xmath16 to the asymptotic entropy ; in the case of general free products there is neither a global composition law for elements of the free product nor subadditivity . thus , in the latter case we have to introduce and investigate new processes . the question of almost sure convergence of @xmath20 to some constant @xmath19 , however , remains open . similar results concerning existence and formulas for the entropy are proved in gilch and mller @xcite for random walks on directed covers of graphs . the reasoning of our proofs follows the argumentation in @xcite : we will show that the entropy equals the rate of escape w.r.t . some special length function , and we deduce the proposed properties analogously . in the present case of free products of graphs , the reasoning is getting more complicated due to the more complex structure of free products in contrast to directed covers , although the main results about existence and convergence types are very similar . we will point out these difficulties and main differences to @xcite at the end of section [ subsec : entropy ] . finally , we will link entropy with the rate of escape and the growth rate of the free product , resulting in two inequalities the plan of the paper is as follows : in section [ sec : preliminaries ] we define the random walk on the free product and the associated generating functions . in section [ sec : entropy ] we prove existence of the asymptotic entropy and give also an explicit formula for it . another formula is derived in section [ sec : dgf ] with the help of double generating functions and a theorem of sawyer and steger @xcite . in section [ sec : groups ] we use another technique to compute a third explicit formula for the entropy of random walks on free products of ( not necessarily finite ) groups . links entropy with the rate of escape and the growth rate of the free product . sample computations are presented in section [ sec : examples ] . let @xmath21 , where @xmath22 . for each @xmath23 , consider a random walk with transition matrix @xmath24 on a finite or countable state space @xmath4 . we assume that the sets @xmath4 are pairwise disjoint and we exclude the case @xmath25 ( see below for further explanation ) . the corresponding single and @xmath26-step transition probabilities are denoted by @xmath27 and @xmath28 , where @xmath29 . for every @xmath23 , we select an element @xmath30 of @xmath4 as the `` root '' . to help visualize this , we think of graphs @xmath31 with vertex sets @xmath4 and roots @xmath30 such that there is an oriented edge @xmath32 if and only if @xmath33 . thus , we have a natural graph metric on the set @xmath4 . furthermore , we shall assume that for every @xmath23 and every @xmath34 there is some @xmath35 such that @xmath36 . for sake of simplicity we assume @xmath37 for every @xmath23 and @xmath34 . moreover , we assume that the random walks on @xmath4 are _ uniformly irreducible _ , that is , there are @xmath38 and @xmath39 such that for all @xmath29 @xmath40 we set @xmath41 and @xmath42 . for instance , this property is satisfied for nearest neighbour random walks on cayley graphs of finitely generated groups , which are governed by probability measures on the groups . let @xmath43 for every @xmath23 and let @xmath44 . free product _ is given by @xmath45 the elements of @xmath9 are `` words '' with letters , also called _ blocks _ , from the sets @xmath46 such that no two consecutive letters come from the same @xmath4 . the empty word @xmath10 describes the root if @xmath47 and @xmath48 with @xmath49 and @xmath50 then @xmath51 stands for their concatenation as words . this is only a partial composition law , which makes defining the asymptotic entropy more complicated than in the case of free products of _ groups_. in particular , we set @xmath52 for all @xmath23 and @xmath53 . note that @xmath54 and @xmath30 as a word in @xmath9 is identified with @xmath10 . the _ block length _ of a word @xmath55 is given by @xmath56 . additionally , we set @xmath57 . the _ type _ @xmath58 of @xmath59 is defined to be @xmath60 if @xmath61 ; we set @xmath62 . finally , @xmath63 denotes the last letter @xmath64 of @xmath59 . the set @xmath9 can again be interpreted as the vertex set of a graph @xmath65 , which is constructed as follows : take copies of @xmath66 and glue them together at their roots to one single common root , which becomes @xmath10 ; inductively , at each vertex @xmath67 with @xmath68 attach a copy of every @xmath69 , @xmath70 , and so on . thus , we have also a natural graph metric associated to the elements the next step is the construction of a new markov chain on the _ free product_. for this purpose , we lift @xmath24 to a transition matrix @xmath71 on @xmath9 : if @xmath72 with @xmath73 and @xmath74 , then @xmath75 . otherwise we set @xmath76 . we choose @xmath77 with @xmath78 . then we obtain a new transition matrix on @xmath9 given by @xmath79 the random walk on @xmath9 starting at @xmath10 , which is governed by @xmath80 , is described by the sequence of random variables @xmath8 . for @xmath81 , the associated single and @xmath26-step transition probabilities are denoted by @xmath82 and @xmath83 . thus , @xmath80 governs a nearest neighbour random walk on the graph @xmath65 , where @xmath80 arises from a convex combination of the nearest neighbour random walks on the graphs @xmath31 . theorem 3.3 in @xcite shows existence ( including a formula ) of a positive number @xmath84 such that @xmath85 almost surely . the number @xmath84 is called the _ rate of escape w.r.t . the block length_. denote by @xmath11 the distribution of @xmath12 . if there is a real number @xmath19 such that @xmath86,\ ] ] then @xmath19 is called the _ asymptotic entropy _ of the process @xmath87 ; we write @xmath88 . if the sets @xmath4 are groups and the random walks @xmath24 are governed by probability measures @xmath89 , existence of the asymptotic entropy rate is well - known , and in this case we even have almost surely ; see derriennic @xcite and kaimanovich and we prove existence of @xmath19 in the case of general free products . our main tool will be the usage of generating functions , which we introduce now . the _ green functions _ related to @xmath24 and @xmath80 are given by @xmath90 where @xmath91 , @xmath92 and @xmath81 . at this point we make the _ basic assumption _ that the radius of convergence @xmath93 of @xmath94 is strictly bigger than @xmath95 . this implies _ transience _ of our random walk on @xmath9 . thus , we may exclude the case @xmath25 , because we get recurrence in this case . for instance , if all @xmath24 govern _ reversible _ markov chains , then @xmath96 ; see ( * ? ? ? * theorem 10.3 ) . furthermore , it is easy to see that @xmath96 holds also if there is some @xmath23 such that @xmath97 for all @xmath98 . the _ first visit generating functions _ related to @xmath24 and @xmath80 are given by @xmath99z^n\ \textrm { and}\\ f(x , y|z ) & : = & \sum_{n\geq 0 } { \mathbb{p}}\bigl[x_n = y,\forall m\leq n-1 : x_m\neq y \mid x_0=x\bigr]z^n,\end{aligned}\ ] ] where @xmath100 describes a random walk on @xmath4 governed by @xmath24 . the stopping time of the first return to @xmath10 is defined as for @xmath23 , define @xmath101\,z^n \ \textrm { and } \ \xi_i(z ) : = \frac{\alpha_i z}{1-\overline{h}_i(z)}.\ ] ] we write also @xmath102 , @xmath103 and @xmath104 . observe that @xmath105 ; see ( * ? ? ? * lemma 2.3 ) . we have @xmath106 for all @xmath92 ; see woess ( * ? ? ? * prop . thus , @xmath107 for @xmath108 and @xmath72 , define the stopping times @xmath109 and @xmath110 , which take both values in @xmath111 . then the _ last visit generating functions _ related to @xmath24 and @xmath80 are defined as @xmath112\,z^n,\\ l(x , y|z ) & : = & \sum_{n\geq 0 } { \mathbb{p}}\bigl[x_n = y , t_x > n \mid x_0=x\bigr]\,z^n.\end{aligned}\ ] ] if @xmath113 with @xmath114 then @xmath115 this equation is proved completely analogously to ( * ? ? ? if all paths from @xmath72 to @xmath116 have to pass through @xmath117 , then @xmath118 this can be easily checked by conditioning on the last visit of @xmath119 when walking from @xmath120 we have the following important equations , which follow by conditioning on the last visits of @xmath121 and @xmath120 , the first visits of @xmath122 and @xmath119 respectively : @xmath123 g(x , y|z ) & = & g(x , x|z)\cdot l(x , y|z ) = f(x , y|z)\cdot g(y , y|z ) . \end{array}\ ] ] observe that the generating functions @xmath124 and @xmath125 have also radii of convergence strictl bigger than @xmath95 . in this subsection we prove existence of the rate of escape with respect to a specific length function . from this we will deduce existence and a formula for the asymptotic entropy in the upcoming subsection . we assign to each element @xmath108 the `` length '' @xmath126 we extend it to a length function on @xmath9 by assigning to @xmath127 the length @xmath128 observe that the lengths can also be negative . e.g. , this can be interpreted as height differences . the aim of this subsection is to show existence of a number @xmath129 such that the quotient @xmath130 tends to @xmath131 almost surely as @xmath132 . we call @xmath131 the _ rate of escape w.r.t . the length function @xmath133_. we follow now the reasoning of ( * ? ? ? * section 3 ) . denote by @xmath134 the projection of @xmath12 to the first @xmath135 letters . we define the _ @xmath135-th exit time _ as @xmath136 moreover , we define @xmath137 , @xmath138 and @xmath139 . we remark that @xmath140 as @xmath132 , and consequently @xmath141 almost surely for every @xmath142 ; see ( * ? ? ? recall that @xmath143 is just the laster letter of the random word @xmath144 . the process @xmath145 is markovian and has transition probabilities @xmath146 for @xmath147 and @xmath148 ; see ( * ? ? ? * lemma 3.4 ) . this process is positive recurrent with invariant probability measure @xmath149 see ( * ? ? ? * section 3 ) . furthermore , the rate of escape w.r.t . the block length exists almost surely and is given by the almost sure constant limit @xmath150 ( see ( * ? ? ? * theorem 3.3 ) ) , where @xmath151 [ lemma : tildew - process ] the process @xmath152 is markovian and has transition probabilities @xmath153 furthermore , the process is positive recurrent with invariant probability measure @xmath154 _ remark : _ observe that the transition probabilities @xmath155 of @xmath152 do not depend on @xmath156 . therefore , we will write sometimes an asterisk instead of @xmath156 . by ( * ? ? ? * section 3 ) , the process @xmath157 is markovian and has transition probabilities @xmath158 where @xmath159 $ ] for @xmath160 . thus , @xmath161 is also markovian and has the following transition probabilities if @xmath147 : @xmath162 in the third equality we conditioned on the last visit of @xmath10 before finally walking from @xmath10 to @xmath163 and we remark that @xmath164 . a straight - forward computation shows that @xmath165 is the invariant probability measure of @xmath152 , where we write @xmath166 : @xmath167 now we are able to prove the following : [ prop : ell - existence ] there is a number @xmath168 such that @xmath169 define @xmath170 by @xmath171 . then @xmath172 . an application of the ergodic theorem for positive recurrent markov chains yields @xmath173 if the integral on the right hand side exists . we now show that this property holds . observe that the values @xmath174 are uniformly bounded from above for all @xmath175 : @xmath176 for @xmath177 , denote by @xmath178 the smallest @xmath98 such that @xmath179 . uniform irreducibility of the random walk @xmath24 on @xmath4 implies that there are some @xmath180 and @xmath181 such that for all @xmath182 , @xmath183 with @xmath184 we have @xmath185 for some @xmath186 . thus , for @xmath175 we have @xmath187 observe that the inequality @xmath188 holds if and only if @xmath189 . define the sets @xmath190 ) . we can now prove existence of @xmath191 : @xmath192 since @xmath193 ; see ( * ? ? ? * proof of prop . 3.2 ) . from this follows that @xmath194 tends to @xmath195 almost surely . the next step is to show that @xmath196 to prove this , assume now that we have the representations @xmath197 and @xmath198 . define @xmath199 & & \quad + \bigl(\|x_n\|-{\mathbf{k}}(n)\bigr ) \cdot \big| \log ( 1-\xi_{\max})\bigr|\end{aligned}\ ] ] @xmath200 & \leq & 3 \cdot m\cdot ( n-{\mathbf{e}_{{\mathbf{k}}(n)}}).\end{aligned}\ ] ] dividing the last inequality by @xmath26 and letting @xmath132 provides analogously to nagnibeda and woess ( * ? ? ? * section 5 ) that @xmath201 almost surely . recall also that @xmath202 and @xmath203 almost surely ; compare ( * ? ? ? * proof of theorem d ) and ( * ? ? ? * prop . 3.2 , thm . now we can conclude : @xmath204 we now compute the constant @xmath195 from the last proposition explicitly : @xmath205 we conclude this subsection with the following observation : [ cor : greenroe ] the rate of escape @xmath131 is non - negative and it is the rate of escape w.r.t . the greenian metric , which is given by @xmath206 . that is , @xmath207 by ( [ gl - equations ] ) , we get @xmath208 since @xmath209 it remains to show that @xmath210 is uniformly bounded in @xmath72 : for @xmath211 , the _ first visit generating function _ is defined as @xmath212\,z^n.\ ] ] therefore , @xmath213 since @xmath214 for all @xmath215 , @xmath216 and @xmath217 is continuous , stricly increasing and strictly convex , we must have @xmath218 , that is , @xmath219 . this finishes the proof . in this subsection we will prove that @xmath131 equals the asymptotic entropy , and we will give explicit formulas for it . the technique of the proof which we will give was motivated by benjamini and peres @xcite , where it is shown that the asymptotic entropy of random walks on discrete groups equals the rate of escape w.r.t . the greenian distance . the proof follows the same reasoning as in gilch and mller @xcite . recall that we made the assumption that the spectral radius of @xmath87 is strictly smaller than @xmath95 , that is , the green function @xmath220 has radius of convergence @xmath96 . moreover , the functions @xmath221 , @xmath222 , have radius of convergence bigger than @xmath95 . recall that @xmath223 for every @xmath222 . thus , we can choose @xmath224 such that @xmath225 for all @xmath222 . we now need the following three technical lemmas : [ lemma : prob - bound ] for all @xmath226 , @xmath227 denote by @xmath228 the circle with radius @xmath229 in the complex plane centered at @xmath17 . a straightforward computation shows for @xmath230 : @xmath231 let be @xmath232 . an application of fubini s theorem yields @xmath233 since @xmath234 is analytic on @xmath228 , we have @xmath235 for all @xmath236 . thus , @xmath237 iterated applications of equations ( [ l - equation ] ) and ( [ gl - equations ] ) provide @xmath238 since @xmath239 , we obtain @xmath240 [ cut - lemma ] let @xmath241 , @xmath242 , @xmath243 be sequences of strictly positive numbers with @xmath244 . assume that @xmath245 and that @xmath246 for all @xmath247 . then @xmath248 . under the made assumptions it can not be that @xmath249 for every indeed , assume that this would hold . choose any @xmath250 . then there is a subseqence @xmath251 with @xmath252 . moreover , there is @xmath253 such that @xmath254 for all @xmath255 . but this implies @xmath256 the last inequality holds for every @xmath250 , yielding that @xmath257 , a contradiction . thus , there is some @xmath258 such that @xmath259 for all @xmath260 . we get for all @xmath260 : @xmath261 taking limits yields that @xmath262 tends to @xmath263 , since the leftmost and rightmost side of this inequality chain tend to @xmath263 . for the next lemma recall the definition of @xmath264 from @xmath265 . [ lemma : sum - bounds ] for @xmath98 , consider the function @xmath266 defined by @xmath267 then there are constants @xmath268 and @xmath269 such that @xmath270 for all @xmath98 and @xmath72 . assume that @xmath271 . recall from the proof of corollary [ cor : greenroe ] that we have @xmath272 . therefore , @xmath273 that is @xmath274 for the upper bound , observe that , by uniform irreducibility , @xmath72 with @xmath271 can be reached from @xmath10 in @xmath275 steps with a probability of at least @xmath276 , where @xmath277 from ( [ def : uniform - irred ] ) is independent from @xmath120 . thus , at least one of the summands in @xmath278 has a value greater or equal to @xmath279 . thus , @xmath280 . now we can state and prove our first main result : [ thm : entropy ] assume @xmath96 . then the asymptotic entropy exists and is given by @xmath281 by ( [ gl - equations ] ) we can rewrite @xmath131 as @xmath282 since @xmath283 we have @xmath284 the next aim is to prove @xmath285 \leq \ell$ ] . we now apply lemma [ cut - lemma ] by setting @xmath286 by lemma [ lemma : prob - bound ] , @xmath287 therefore , @xmath288 decays faster than any geometric sequence . applying lemma [ cut - lemma ] yields @xmath289 by lemma [ lemma : sum - bounds ] , we may apply the dominated convergence theorem and get : @xmath290 recall that _ shannon s inequality _ gives @xmath291 for every finitely supported probability measure @xmath292 on @xmath9 . we apply now this inequality by setting @xmath293 : @xmath294 now we can conclude with fatou s lemma : @xmath295 thus , @xmath296 $ ] exists and the limit equals @xmath131 . the rest follows from ( [ ell0-limit ] ) and ( [ ch - formel ] ) . we now give another formula for the asymptotic entropy which shows that it is strictly positive . [ th : entropy2 ] assume @xmath96 . then the asymptotic entropy is given by @xmath297 _ remarks : _ observe that the sum on the right hand side of theorem [ th : entropy2 ] equals the entropy rate ( for positive recurrent markov chains ) of @xmath152 , which is defined by the almost sure constant limit @xmath298 where @xmath299 is the joint distribution of @xmath300 . that is , @xmath301 . for more details , we refer e.g. to cover and thomas ( * ? ? ? * chapter 4 ) . at this point it is essential that we have defined the length function @xmath133 with the help of the functions @xmath302 and not by the greenian metric . for a moment let be @xmath303 . then : @xmath304 we now replace @xmath120 by @xmath305 in the last equation : since @xmath130 tends to @xmath19 almost surely , the subsequence @xmath306 converges also to @xmath19 . since @xmath307 and @xmath308 , we get @xmath309 where @xmath310 and @xmath311 . by positive recurrence of @xmath152 , an application of the ergodic theorem yields @xmath312 whenever @xmath313 . obviously , for every @xmath314 @xmath315 since @xmath316 we get @xmath317 whenever @xmath313 . in particular , @xmath318 since @xmath319 by ( * ? ? ? * section 4 ) . it remains to show that it can not be that @xmath320 . for this purpose , assume now @xmath320 . define for @xmath258 the function @xmath321 by @xmath322 then @xmath323 observe that @xmath324 as @xmath325 . since @xmath326 and @xmath320 by assumption there is for every @xmath327 and almost every trajectory of @xmath328 an almost surely finite random time @xmath329 such that for all @xmath330 @xmath331 on the other hand side there is for every @xmath332 , every small @xmath333 and almost every trajectory an almost surely finite random time @xmath334 such that for all @xmath335 @xmath336 furthermore , since @xmath337 and @xmath338 there is an almost surely finite random time @xmath339 such that for all @xmath340 and all @xmath341 and @xmath342 @xmath343 choose now @xmath344 . then we get the desired contradiction , when we substitute in equality ( [ q - entropy ] ) the vertex @xmath120 by @xmath305 with @xmath340 , divide by @xmath345 on both sides and see that the left side is in @xmath346 and the rightmost side is bigger than @xmath347 . this finishes the proof of theorem [ th : entropy2 ] . [ cor : entropy2 ] assume @xmath96 . then we have for almost every path of the random walk @xmath87 @xmath348 recall inequality ( [ equ : liminf - h ] ) . integrating both sides of this inequality yields together with the inequality chain ( [ eq : entropy - final ] ) that @xmath349 providing that @xmath350 for almost every realisation of the random walk . the following lemma gives some properties concerning general measure theory : [ lemma : measuretheory ] let @xmath351 be a sequence of non - negative random variables and suppose that @xmath352 almost surely and @xmath353 = c$ ] . then the following holds : 1 . @xmath354 , that is , @xmath355 converges in probability to @xmath263 . 2 . if @xmath355 is uniformly bounded then @xmath356 , that is , @xmath357 as @xmath132 . first , we prove convergence in probability of @xmath351 . for every @xmath358 , there is some index @xmath359 such that for all @xmath360 @xmath361 furthermore , due to the above made assumptions on @xmath362 there is for every @xmath363 some index @xmath364 such that for all @xmath365 @xmath366 > 1-\delta_2.\ ] ] since @xmath367 it must be that for every arbitrary but fixed @xmath333 , every @xmath368 and for all @xmath26 big enough @xmath369\cdot ( c-\delta_1 ) + { \mathbb{p}}\bigl[z_n > c + { \varepsilon}\bigr]\cdot ( { \varepsilon}+\delta_1 ) \leq \int z_n\,d{\mathbb{p}}\leq c+\delta_1,\ ] ] or equivalently , @xmath370 \leq \frac{c+\delta_1 -{\mathbb{p}}\bigl[z_n > c-\delta_1\bigr]\cdot ( c-\delta_1)}{{\varepsilon}+\delta_1}.\ ] ] letting @xmath371 we get @xmath372 \leq \frac{2\delta_1}{{\varepsilon}+\delta_1}.\ ] ] since we can choose @xmath373 arbitrarily small we get @xmath370 \xrightarrow{n\to\infty } 0 \quad \textrm { for all } { \varepsilon}>0.\ ] ] this yields convergence in probability of @xmath355 to @xmath263 . in order to prove the second part of the lemma we define for any small @xmath333 and @xmath98 the events @xmath374 \ \textrm { and } b_{n,{\varepsilon}}:=\bigl[|z_n - c| > { \varepsilon}\bigr].\ ] ] for arbitrary but fixed @xmath333 , convergence in probability of @xmath355 to @xmath263 gives an integer @xmath375 such that @xmath376<{\varepsilon}$ ] for all @xmath377 . since @xmath378 is assumed to be uniformly bounded , we get for @xmath379 : @xmath380 thus , we have proved the second part of the lemma . we can apply the last lemma immediately to our setting : [ cor : entropy3 ] assume @xmath96 . then we have the following types of convergence : 1 . convergence in probability : @xmath381 2 . assume that there is @xmath382 such that @xmath383 whenever @xmath384 . then : @xmath385 setting @xmath386 and applying lemma [ lemma : measuretheory ] yields the claim . note that the assumption @xmath383 yields @xmath387 . the assumption of the second part of the last corollary is obviously satisfied if we consider free products of _ finite _ graphs . the reasoning in our proofs for existence of the entropy and its different properties ( in particular , the reasoning in section [ subsec : entropy ] ) is very similar to the argumentation in @xcite . however , the structure of free products of graphs is more complicated than in the case of directed covers as considered in @xcite . we outline the main differences to the reasoning in the aforementionend article . first , in @xcite a very similar rate of escape ( compare with proposition [ prop : ell - existence ] ) is considered , which arises from a length function induced by last visit generating functions . while the proof of existence of the rate of escape in @xcite is easy to check , we have to make more effort in the case of free products , since @xmath388 is not necessarily bounded for @xmath34 . additionally , one has to study the various ingridients of the proof more carefully , since non - trivial loops are possible in our setting in contrast to random walks on trees . secondly , in @xcite the invariant measure @xmath389 of our proof collapses to @xmath390 , that is , in @xcite one has to study the sequence @xmath391 , while in our setting we have to study the more complex sequence @xmath392 ; compare with lemma [ lemma : tildew - process ] and proposition [ prop : ell - existence ] . in this section we derive another formula for the asymptotic entropy . the main tool is the following theorem of sawyer and steger ( * ? ? ? * theorem 2.2 ) : [ sawyer ] let @xmath393 be a sequence of real - valued random variables such that , for some @xmath394 , @xmath395 where @xmath396 and @xmath397 are analytic for @xmath398 and . denote by @xmath399 and @xmath400 the partial derivatives of @xmath156 with respect to @xmath401 and @xmath163 . then @xmath402 setting @xmath403 and @xmath404 the expectation in theorem [ sawyer ] becomes @xmath405 we define for @xmath222 , @xmath406 : @xmath407 finally , @xmath408 is defined by @xmath409 with these definitions we have @xmath410 and @xmath411 . simple computations analogously to ( * ? ? ? * lemma 4.2 , corollary 4.3 ) yield @xmath412 we now define @xmath413 and @xmath414 and apply theorem [ sawyer ] by differentiating @xmath415 and evaluating the derivatives at @xmath416 : @xmath417 [ cor : entropy - dgf ] assume @xmath96 . then the entropy can be rewritten as @xmath418 @xmath419 in this section let each @xmath4 be a finitely generated group @xmath420 with identity @xmath421 . we assume that the @xmath4 s are pairwise disjoint . the free product is again a group with concatenation ( followed by iterated cancellations and contractions ) as group operation . we write @xmath422 . suppose we are given a probability measure @xmath89 on @xmath423 for every @xmath222 governing a random walk on @xmath420 , that is , @xmath424 for all @xmath425 . let @xmath426 be a family of strictly positive real numbers with @xmath427 . then the random walk on the free product @xmath428 is defined by the transition probabilities @xmath429 , where @xmath430 analogously , @xmath431 denotes the @xmath26-th convolution power of @xmath292 . the random walk on @xmath432 starting at the identity @xmath433 of @xmath432 is again denoted by the sequence @xmath87 . in particular , the radius of convergence of the associated green function is strictly bigger than @xmath95 ; see ( * ? ? ? * theorem 10.10 , corollary 12.5 ) . in the case of free products of groups it is well - known that the entropy exists and can be written as @xmath434 see derriennic @xcite , kaimanovich and vershik @xcite and blachre , hassinsky and mathieu @xcite . for free products of _ finite _ groups , mairesse and mathus @xcite give an explicit formula for @xmath19 , which remains also valid for free products of countable groups , but in the latter case one needs the solution of an infinite system of polynomial equations . in the following we will derive another formula for the entropy , which holds also for free products of _ infinite _ groups . we set @xmath435 . observe that transitivity yields @xmath436 . thus , @xmath437 first , we rewrite the following expectations as @xmath438 thus , @xmath439 recall that @xmath440 almost surely . that is , @xmath12 converges almost surely to a random infinite word @xmath441 of the form @xmath442 , where two consecutive letters are not from the same @xmath443 . denote by @xmath444 the first letter of @xmath441 . let be @xmath445 . for @xmath446 , the integrand in @xmath447 is constant : if @xmath448 then @xmath449 and if @xmath450 then @xmath451 by ( * ? ? ? * section 5 ) , for @xmath23 and @xmath452 , @xmath453 = 1-(1-\xi_i)\,g_i(o_i , o_i|\xi_i ) \quad \textrm { and}\\ { \mathbb{p}}[x_\infty^{(1)}=g ] & = & f(o_i , g|\xi_i ) \ , ( 1-\xi_i)\,g_i(o_i , o_i|\xi_i ) = ( 1-\xi_i)\,g_i(o_i , g|\xi_i).\end{aligned}\ ] ] recall that @xmath454 for each @xmath455 . we get : [ thm : entropy - groups ] whenever @xmath456 for all @xmath222 , that is , when all random walks on the factors @xmath420 have finite single - step entropy , then the asymptotic entropy @xmath19 of the random walk on @xmath432 is given by @xmath457,\ ] ] where @xmath458 consider the sequence @xmath459 . if this sequence converges , its limit must equal @xmath19 . by the above made considerations we get @xmath460 \log \frac{f_i(o_i , gg'|\xi_i)}{f_i(o_i , g'|\xi_i)}\biggr],\end{aligned}\ ] ] if the sum on the right hand side is finite . we have now established the proposed formula , but it remains to verify finiteness of the sum above . this follows from the following observations : @xmath461 is finite . observe that @xmath462 for @xmath463 . thus , @xmath464 this proves claim a. @xmath465 is finite . observe that @xmath466 . obviously , @xmath467 for @xmath468 set @xmath469 . then : @xmath470\cdot \big| \log g_i(o_i , g'|\xi_i)\bigr| \\ & = & \sum_{g'\in\gamma_i^\times } ( 1-\xi_i)\cdot g_i(o_i , g'|\xi_i)\cdot \big| \log g_i(o_i , g'|\xi_i)\bigr|\\ & = & \sum_{g'\in\gamma_i^\times } ( 1-\xi_i)\cdot \sum_{n\in n(g ' ) } \mu_i^{(n)}(g')\cdot \xi_i^n\cdot \big| \log g_i(o_i , g'|\xi_i)\bigr|\\ & \leq & \sum_{g'\in\gamma_i^\times } ( 1-\xi_i)\cdot \sum_{n\in n(g ' ) } \mu_i^{(n)}(g')\cdot \xi_i^n\cdot \max\bigl\lbrace -\log \bigl(\mu_i^{(n)}(g')\,\xi_i^n\bigr ) , -\log(1-\xi_i)\bigr\rbrace\\ & \leq & ( 1-\xi_i ) \cdot \sum_{n\in n(g ' ) } n\,\xi_i^n \cdot\underbrace{\frac{-1}{n } \sum_{g'\in\gamma_i } \mu_i^{(n)}(g ' ) \log \mu_i^{(n)}(g')}_{(\ast ) } - ( 1-\xi_i)\,\log \xi_i \sum_{n\geq 1 } n\,\xi_i^n\\ & & \quad - ( 1-\xi_i)\,\log ( 1-\xi_i ) \sum_{n\geq 1 } \xi_i^n.\end{aligned}\ ] ] recall that @xmath471 together with kingman s subadditive ergodic theorem implies existence of a constant @xmath472 with @xmath473 thus , if @xmath98 is large enough , the sum @xmath474 is in the interval @xmath475 for any arbitrarily small @xmath333 . that is , the sum @xmath474 is uniformly bounded for all @xmath98 . from this follows that the rightmost side of the last inequality chain is finite . furthermore , we have for each @xmath455 with @xmath476 : @xmath477\cdot \big| \log g_i(o_i , gg'|\xi_i)\bigr|\\ & = & \sum_{g'\in\gamma_i^\times } ( 1-\xi_i)\cdot g_i(o_i , g'|\xi_i)\cdot \bigl| \log g_i(o_i , gg'|\xi_i)\bigr|\\ & = & \sum_{g'\in\gamma_i^\times } ( 1-\xi_i ) \cdot\sum_{n\in n(g ' ) } \mu_i^{(n)}(g')\cdot \xi_i^n\cdot \bigl| \log g_i(o_i , gg'|\xi_i)\bigr|\\ & \leq & \sum_{g'\in\gamma_i^\times } ( 1-\xi_i)\cdot \sum_{n\in n(g ' ) } \mu_i^{(n)}(g')\cdot \xi_i^n \cdot \max\bigl\lbrace -\log \bigl(\mu_i(g)\,\mu_i^{(n)}(g')\,\xi_i^{n+1}\bigr),-\log(1-\xi_i)\bigr\rbrace\\ & \leq & -(1-\xi_i ) \cdot \sum_{n\in n(g ' ) } \xi_i^n \cdot \sum_{g'\in\gamma_i } \mu_i^{(n)}(g ' ) \log \mu_i^{(n)}(g ' ) - ( 1-\xi_i)\cdot \log \xi_i \cdot \sum_{n\geq 1 } ( n+1)\,\xi_i^n\\ & & \quad - \log \mu_i(g)- \log ( 1-\xi_i).\end{aligned}\ ] ] if we sum up over all @xmath156 with @xmath478 , we get : @xmath479 convergence of @xmath480 follows from ( [ convergence - hi ] ) , @xmath481 converges since @xmath105 and @xmath482 is convergent by assumption @xmath471 . this finishes the proof of claim b , and thus the proof of the theorem . erschler and kaimanovich @xcite asked if drift and entropy of random walks on groups depend continuously on the probability measure , which governs the random walk . ledrappier @xcite proves in his recent , simultaneous paper that drift and entropy of finite - range random walks on free groups vary analytically with the probability measure of constant support . by theorem [ thm : entropy - groups ] , we are even able to show continuity for free products of finitely generated groups , but restricted to nearest neighbour random walks with fixed set of generators . [ cor : analyticity ] let @xmath420 be generated as a semigroup by @xmath483 . denote by @xmath484 the set of probability measures @xmath89 on @xmath483 with @xmath485 for all @xmath486 . furthermore , we write @xmath487 . then the entropy function @xmath488 is real - analytic . the claim follows directly with the formula given in theorem [ thm : entropy - groups ] : the involved generating functions @xmath489 and @xmath490 are analytic when varying the probability measure of constant support , and the values @xmath491 can also be rewritten as @xmath492 where @xmath493 . this yields the claim . _ remarks : _ 1 . corollary [ cor : analyticity ] holds also for the case of free products of _ finite _ graphs , if one varies the transition probabilities continously under the assumption that the sets @xmath494 remain constant : one has to rewrite @xmath491 as power series in terms of ( finitely many ) @xmath495 and gets analyticity with the formula given in theorem [ thm : entropy ] . 2 . analyticity holds also for the drift ( w.r.t . the block length and w.r.t . the natural graph metric ) of nearest neighbour random walks due to the formulas given 3 . the formula for entropy and drift given in mairesse and mathus @xcite for random walks on free products of _ finite _ groups depends also analytically on the transition probabilities . in this section we consider the case of free products of _ finite _ sets @xmath1 , where @xmath4 has @xmath496 vertices . we want to establish a connection between asymptotic entropy , rate of escape and the volume growth rate of the free product @xmath9 . for @xmath497 , let @xmath498 be the set of all words of @xmath9 of block length @xmath26 . the following lemmas give an answer how fast the free product grows . the sphere growth rate w.r.t . the block length is given by @xmath499 where @xmath500 is the perron - frobenius eigenvalue of the @xmath501-matrix @xmath502 with @xmath503 for @xmath504 and @xmath505 otherwise . denote by @xmath506 the @xmath507-diagonal matrix , which has entries @xmath508-vector with all entries equal to @xmath95 . thus , we can write @xmath509 let @xmath510 and @xmath511 be eigenvectors of @xmath269 w.r.t . the perron - frobenius then @xmath512 are some constants independent from @xmath26 . thus , @xmath513 recall from the perron - frobenius theorem that @xmath514 for each @xmath182 ; in particular , @xmath515 . we also take a look on the natural graph metric and its growth rate . for this purpose , we define @xmath516 that is , the set of all vertices in @xmath9 which are at distance @xmath26 to the root @xmath10 w.r.t . the natural graph metric . we now construct a new graph , whose adjacency matrix allows us to describe the exponential growth of @xmath517 as @xmath132 . for this purpose , we visualize the sets @xmath1 as graphs @xmath518 with vertex sets @xmath1 equipped with the following edges : for @xmath29 , there is a directed edge from @xmath120 to @xmath119 if and only if @xmath33 . consider now directed spanning trees @xmath519 of the graphs @xmath518 such that the graph distances of vertices in @xmath520 to the root @xmath30 remain the same as in @xmath31 . we now investigate the free product @xmath521 , which is again a tree . we make the crucial observation that @xmath522 can be seen as the directed cover of a finite directed graph @xmath523 , where @xmath523 is defined in the following way : 1 . the vertex set of @xmath523 is given by @xmath524 with root @xmath10 . the edges of @xmath523 are given as follows : first , we add all edges inherited from one of the trees @xmath519 , where @xmath10 plays the role of @xmath30 for each @xmath23 . secondly , we add for all @xmath23 and every @xmath525 an edge from @xmath120 to each @xmath526 , @xmath70 , whenever there is an edge from @xmath30 to @xmath119 in @xmath527 . the tree @xmath522 can be seen as a _ periodic tree _ , which is also called a _ tree with finitely many cone types _ ; for more details we refer to lyons @xcite and nagnibeda and woess @xcite . now we are able to state the following lemma : the sphere growth rate w.r.t . the natural graph metric defined by @xmath528 exists . moreover , we have the equation @xmath529 , where @xmath530 is the perron - frobenius eigenvalue of the adjacency matrix of the graph @xmath523 . since the graph metric remains invariant under the restriction of @xmath9 to @xmath522 and since it is well - known that the growth rate exists for periodic trees ( see we have existence of the limit @xmath531 . more precisely , @xmath532 tends to the perron - frobenius eigenvalue of the adjacency matrix of @xmath523 as @xmath132 . for sake of completeness , we remark that the root of @xmath522 plays a special role ( as a cone type ) but this does not affect the application of the results about directed covers to our case . for @xmath533 , we write @xmath534 . now we can prove : the volume growth w.r.t . the block length , w.r.t . the natural graph metric respectively , is given by @xmath535 for ease of better readability , we omit the subindex @xmath533 in the following , since the proofs for @xmath536 and @xmath537 are completely analogous . choose any small @xmath538 . then there is some @xmath539 such that for all @xmath540 @xmath541 write @xmath542 . then for @xmath543 : @xmath544{c_\varepsilon + \sum_{k = k_\varepsilon}^n \lambda^k e^{k\varepsilon } } = \lambda e^{\varepsilon } \sqrt[n]{\frac{c_\varepsilon}{\lambda^n e^{n\varepsilon } } + \sum_{k = k_\varepsilon}^n \frac{1}{\lambda^{n - k } e^{(n - k)\varepsilon}}}\\ & \leq & \lambda e^{\varepsilon } \sqrt[n]{\frac{c_\varepsilon}{\lambda^n e^{n\varepsilon } } + ( n - k_\varepsilon+1 ) } \xrightarrow{n\to\infty } \lambda e^{\varepsilon}.\end{aligned}\ ] ] in the last inequality we used the fact @xmath545 . since we can choose @xmath538 arbitrarily small , we get @xmath546 . analogously : @xmath547{c_\varepsilon + \sum_{k = k_\varepsilon}^n \lambda^k e^{-k\varepsilon } } = \lambda \sqrt[n]{\frac{c_\varepsilon}{\lambda^n } + \sum_{k = k_\varepsilon}^n \frac{e^{-k\varepsilon}}{\lambda^{n - k } } } \xrightarrow{n\to\infty } \lambda e^{-\varepsilon}.\ ] ] that is , @xmath548 . for @xmath533 , define @xmath549 by @xmath550 and @xmath551 . then the limits @xmath552 exist ; see ( * ? ? ? * theorem 3.3 , section 7.ii ) . now we can establish a connection between entropy , rate of escape and volume growth : [ cor : inequalities ] @xmath553 and @xmath554 . let be @xmath533 and @xmath555 . then there is some @xmath556 such that for all @xmath557 @xmath558 that is , @xmath559 if we let @xmath26 tend to infinity and make @xmath560 arbitrarily small , we get the claim . finally , we remark that an analogous inequality for random walks on groups was given by and more generally for space- and time - homogeneous markov chains by kaimanovich and woess ( * ? ? ? * theorem 5.3 ) . consider the graphs @xmath561 and @xmath562 with the transition probabilities sketched in figure [ entropy - bsp2 ] . we set @xmath563 . for the computation of @xmath84 we need the following functions : @xmath564 \xi_1(z ) & = & \frac{z/2}{1-\frac{z}{2}\frac{\xi_2(z)^2}{2}\frac{1}{1-\xi_2(z)^3/2 } } , & \quad & \xi_2(z ) & = & \frac{z/2}{1-\frac{z}{2}\frac{\xi_1(z)^2}{2}\frac{1}{1-\xi_1(z)^2/2}}. \end{array}\ ] ] simple computations with the help of ( * ? ? ? * section 3 ) and mathematica allow us to determine the rate of escape of the random walk on @xmath565 as @xmath566 . for the computation of the entropy , we need also the following generating functions : @xmath567 thus , we get the asymptotic entropy as @xmath568 . [ entropy - bsp2 ] and @xmath562,title="fig:",width=226 ] consider the free product @xmath570 of the _ infinite _ groups @xmath571 with @xmath572 and @xmath573 for each @xmath574 . we set @xmath575 , @xmath576 , @xmath577 and @xmath578 for @xmath579 . define @xmath580\,z^n,\\ \hat f(b|z ) & : = & \sum_{n\geq 1 } { \mathbb{p}}\bigl[y_n = b , \forall m < n : \lambda(y_m)<1\,\bigl|\ , y_0=(0,0)\bigr]\,z^n,\end{aligned}\ ] ] where @xmath581 is a random walk on @xmath582 governed by @xmath583 . the above functions satisfy the following system of equations : @xmath584 from this system we obtain explicit formulas for @xmath585 and @xmath586 . we write @xmath587 for @xmath588 . to compute the entropy rate we have to solve the following system of equations : @xmath589 moreover , we need the value @xmath590 . this value can be computed analogously to ( * ? ? ? * section 6.2 ) , that is , @xmath591 has to be computed numerically from the equation @xmath592 solving this equation with mathematica yields @xmath593 . to compute the entropy we have to evaluate the functions @xmath594 at @xmath595 for each @xmath596 . for even @xmath98 , we have the following formulas : @xmath597 for odd @xmath98 , @xmath598 moreover , we define @xmath599 $ ] . this probability can be computed by conditioning on the first step and solving @xmath600 that is , @xmath601 . observe that we get the following estimations : @xmath602 these bounds allow us to cap the sum over all @xmath603 in ( [ equ - mathcalf ] ) and to estimate the tails of these sums . thus , we can compute the entropy rate numerically as @xmath604 . the author is grateful to frdric mathus for discussion on the problems and several hints regarding content and exposition . 99 a. avez . entropie des groupes de type fini . _ c.r.acad.sci.paris sr . , 275:a1363a1366 , 1972 . i. benjamini and y. peres . tree - indexed random walks on groups and first passage percolation . _ probab . theory related fields _ , 98(1 ) : 91112 , 1994 . s. blachre and p. hassinsky and p. mathieu . asymptotic entropy and green speed for random walks on countable groups . _ ann . of probab . _ , 36(3):11341152 , 2008 . e. candellero and l. gilch . phase transitions for random walk asymptotics on free products of groups . submitted , current version available at arxiv:0909.1893v3 , 2009 . d.i . cartwright and p.m. soardi . random walks on free products , quotients and amalgams . _ nagoya math . j. _ , 102:163180 , 1986 . t. cover and j. thomas . _ elements of information theory_. wiley & sons , 2nd edition , 2006 . y. derriennic . quelques applications du thorme ergodique sous - additif . _ astrisque _ , 74:183201 , 1980 . a. erschler . on the asymptotics of drift . _ j. of math . sciences _ , 121(3):24372440 , 2004 . a. erschler and v. kaimanovich . continuity of entropy for random walks on hyperbolic groups . _ in preparation _ , 2010 . p. gerl and w. woess . local limits and harmonic functions for nonisotropic random walks on free groups . theory rel . fields _ , 71:341355 , 1986 . l. a. gilch . rate of escape of random walks on free products . _ j. aust . _ , 83(i):3154 , 2007 . l. a. gilch and s. mller . random walks on directed covers of graphs . _ j. of theoret . _ , doi 10.1007/s10959 - 009 - 0256 - 0 , 2009 . y. guivarch . sur la loi des grands nombres et le rayon spectral dune marche alatoire . _ astrisque _ , 74:4798 , 1980 . kaimanovich and a.m. vershik . random walks on discrete groups : boundary and entropy . _ ann . of prob . _ , 11:457490 , 1983 . v. kaimanovich and w. woess . boundary and entropy of space homogeneous markov chains . _ ann . of prob . _ , 30:323363 , 2002 . a. karlsson and f. ledrappier . linear drift and poisson boundary for random walks . _ pure appl . , 3:10271036 , 2007 . the ergodic theory of subadditive processes . _ j. royal stat . b _ , 30:499510 , 1968 . s. lalley . finite range random walk on free groups and homogeneous trees . _ ann . of prob . _ , 21(4):20872130 , 1993 . f. ledrappier . analyticity of the entropy for some random walks . _ preprint , available at arxiv:1009.5354v1 _ , 2010 . r. lyons , with y. peres . _ probability on trees and networks_. in preparation . current version available at http://mypage.iu.edu/ rdlyons/ , 2010 . j. mairesse and f. mathus . random walks on free products of cyclic groups . _ j. london math . _ , 75(1):4766 , 2007 . _ random walks and convolution operators on free products . _ phd thesis , new york univ . , t. nagnibeda and w. woess . random walks on trees with finitely many cone types . _ j. theoret . _ , 15:399438 , 2002 . s. sawyer . isotropic random walks in a tree . _ zeitschrift f. wahrscheinlichkeitstheorie _ , 42:279292 , 1978 . s. sawyer and t. steger . the rate of escape for anisotropic random walks in a tree . theory rel . fields _ , 76:207230 , 1987 . long range estimates for markov chains . _ , 109:225252 , 1985 . d. voiculescu . addition of certain non - commuting random variables . _ j. funct . _ , 66:323346 , 1986 . w. woess . nearest neighbour random walks on free products of discrete groups . _ _ , 5-b:961982 , 1986 . w. woess . _ random walks on infinite graphs and groups_. cambridge university press , 2000 .

in 1917 , einstein replaced the euclidean three - space by the three - sphere and so introduced the first spherical manifold to describe the spatial part of the universe @xcite . more generally , three - manifolds with non - euclidean topology have in recent years found applications in cosmology , in particular in relation to the multipole analysis of the cosmic microwave background ( cmb ) radiation . cosmological models with positive curvature are related to the three - sphere . the three - sphere covers all spherical three - manifolds , including the platonic manifolds considered here . for the poincars dodecahedral three - manifold , the first orthonormal basis required for the multipole analysis was constructed in @xcite , using lie algebraic methods and drawing on work by klein 1884 @xcite . compared to other manifolds , the platonic three - manifolds have the advantage that their homotopies have been derived and classified in 2003 by everitt @xcite . the homotopies are fundamental for the mathematical analysis of the topology of manifolds @xcite . the connection of multipole - resolved cmb measurements to cosmic topology is provided by selection rules for the harmonic analysis . surprisingly , there is no systematic account that links the homotopy of platonic manifolds to deck groups and to their harmonic analysis and selection rules . our aim is to close this gap . we start by relating the three - sphere to the group @xmath0 in section [ sec:3sphere ] . the representations of the group @xmath0 yield the wigner polynomials introduced in section [ sec : wigner ] , with unitary actions resulting in a multipole decomposition discussed in section [ sec : unitary ] . in section [ sec : grids ] , we index the wigner basis polynomials by points on an @xmath1-grid in two dimensions in preparation for deriving topological selection rules . the platonic spherical polyhedra are constructed in sections [ sec : coxeter ] and [ sec : unimodular ] by use of four spherical coxeter groups for the tetrahedron , cube , octahedron and dodecahedron . the platonic homotopy groups from @xcite are expressed in sections [ sec : homotopy ] and [ sec : synopsis ] by the gluings of faces and directed edges of the platonic polyhedra . in the central section [ sec : deck ] , we use the isomorphism of homotopy and deck groups . any topological spherical three - manifold appears on its universal cover @xmath2 in the form of a tiling . any tile is an image of the spherical manifold . the tiles are related to one another by deck actions from the deck group . we construct from the homotopies of @xcite the isomorphic deck actions on the three - sphere @xmath2 and , from them determine , the deck groups for the platonic manifolds . the deck groups are factorized in section [ sec : point ] by the point symmetry groups of the platonic polyhedra . we handle fixpoints under rotations in point groups by the new construction of multiply - connected topological orbifolds . whereas a topological manifold strictly excludes fixpoints under deck actions , orbifolds allow for such fixpoints @xcite , @xcite pp . 652714 . for the harmonic analysis on orbifolds , we employ in section [ sec : orbi ] wigner polynomials with restrictions to the subgrid representation of topological selection rules . the resulting subbases appear as denumerable sets of wigner polynomials on subgrids , as shown in figure [ fig : subgrids ] and extend earlier work @xcite . the @xmath1-grid representation allows in section [ sec : harmonic ] for a very transparent graphical display of the restriction from general bases to subbases of a given topology . finally in section [ sec : cmb ] , we present the application of topology to cosmology . in contrast to einstein s closure of the spatial part of space - time on the simply - connected three - sphere , we discuss the closure on spherical orbifolds and incorporate a wider variety of multiply - connected cosmologies . our harmonic analysis on orbifolds yields for each orbifold specific multipole selection rules and , moreover , predicts topological correlations between different multipole orders @xmath3 of the cmb . in three - dimensional topology , the three - sphere @xmath2 is the simply - connected universal cover @xcite ( p. 290 ) of spherical topologies . the points of the three - sphere , seen in euclidean space @xmath4 with covariant coordinates , are : @xmath5 they are in one - to - one correspondence to the elements of the unitary unimodular group @xmath0 . from real coordinates of @xmath4 : @xmath6 , we pass to a @xmath7 unimodular unitary matrix @xmath8 in the form : @xmath9 = \left [ \begin{array}{ll } x_0-ix_3&x_2-ix_1\\ -x_2-ix_1&x_0+ix_3\\ \end{array } \right ] , \ : { \rm det } ( u)=1\ ] ] as angular coordinates on @xmath0 , we can use the euler angles @xmath10 @xcite ( pp . 68 ) . in these , the matrix @xmath11 takes the form : @xmath12\end{aligned}\ ] ] the measure of integration in the euler angles becomes @xcite ( p. 62 ) , @xmath13 summary : the three - sphere @xmath2 corresponds one - to - one to the group @xmath0 . from this correspondence , the isometries of @xmath2 inherit a unitary structure . this provides the full representation theory @xcite of the group @xmath0 as a tool for spherical topology . the finite irreducible unitary representations of @xmath0 were studied by wigner @xcite . for unitarity and the irreducibility of representations , we refer to his monograph . we now introduce the wigner polynomials . we shall show that they span the harmonic analysis on the three - sphere @xcite and play an important part in topological analysis . let the matrix @xmath14 equation ( [ u1 ] ) act from the left on the complex column vector : @xmath15\\ \rightarrow u(z_1,z_2)\left[\begin{array}{l } a\\ b \end{array}\right]\ ] ] in the bargmann hilbert space of analytic functions @xcite in two complex variables @xmath16 , the monomials : @xmath17 are normalized with the measure : @xmath18 with @xmath1 either an integer or half - integer , under the action equation ( [ u5 ] ) , the monomials in equation ( [ u6 ] ) carry irreducible representations of @xmath0 and transform as : @xmath19 the @xmath20 coefficients in this equation are the wigner @xmath21-functions @xcite ( equation ( 4.1.10 ) ) . wigner @xcite has shown that the @xmath21-functions are the unitary irreducible representations of @xmath0 . we replace euler angles by four complex variables from equation ( [ u1 ] ) in the arguments of the @xmath21 . following @xcite , we term the resulting complex valued polynomials wigner polynomials . from equation ( [ u8 ] ) , they are given by : @xmath22^{1/2}\nonumber\\\nonumber & & \sum_{\upsigma } \frac{(j+m_2)!(j - m_2)!}{(j+m_1-\upsigma)!(m_2-m_1+\upsigma)!\upsigma!(j - m_2-\upsigma ) ! } \\ & & \times ( -1)^{m_2-m_1+\upsigma } z_1^{j+m_1-\upsigma}\overline{z}_2^{m_2-m_1+\upsigma}z_2^{\upsigma } \overline{z}_1^{j - m_2-\upsigma } \vspace{-10pt } \label{u9}\end{aligned}\ ] ] all exponents of the four complex variables in equation ( [ u8a ] ) are integers and must be non - negative . this restricts the summation over the integer @xmath23 . two important properties of these wigner polynomials arise under inversion and under complex conjugation of the matrix @xmath11 , @xmath24 we prove that the wigner polynomials vanish under the laplacian on @xmath4 and , so , are harmonic . consider first the case @xmath25 . this implies @xmath26 , @xmath27^{1/2 } ( z_1)^{j+m_2}(z_2)^{j - m_2}\ ] ] the polynomials equation ( [ u11 ] ) are analytic in @xmath28 , and so , by the cauchy riemann differential equations , vanish under the laplacian @xmath29 on @xmath4 . from the particular polynomials equation ( [ u11 ] ) , we can lower the eigenvalue @xmath30 to any value @xmath31 for fixed degree @xmath32 by the application of the left lowering operator @xmath33 of @xmath34 , given from @xcite ( equation ( 35 ) ) by : @xmath35 \vspace{-6pt}\ ] ] by repeated application of this lowering operator , we can reach all of the @xmath20 wigner polynomials for fixed degree @xmath32 . the lowering operator commutes with the laplacian @xmath29 on @xmath4 and , so , can not change its eigenvalue zero . it follows that all of the wigner polynomials equation ( [ u8a ] ) are harmonic . the action of the rotation group @xmath36 is expressed by left and right actions on @xmath0 . when restricted to conjugation , its action reduces to rotations only of the three coordinates @xmath37 ; see equation ( [ u14db ] ) . the orthogonality of the wigner polynomials , expressed in the euler angles , is @xcite ( p. 62 ) @xmath38 summary : the wigner polynomials , equation ( [ u6 ] ) , are harmonic , orthonormal and homogeneous of degree @xmath39 . their transformation properties under @xmath0 are well known from the theory of angular momentum in quantum mechanics @xcite . the group of isometries of @xmath2 is @xmath36 . the action of this group is isomorphic to the direct product of a left and a right group @xmath40 acting on @xmath41 in the form : @xmath42 the left and right action groups in equation ( [ u2 ] ) commute and act as : @xmath43 the subgroup @xmath44 in equation ( [ u2 ] ) is generated by @xmath45 . the direct product form equation ( [ u2 ] ) , contrary to what is asserted in @xcite ( p. 277 ) , in general does not extend to subgroups of @xmath36 and , so , can not yield their classification . a counter - example is provided by the deck group @xmath46 of the tetrahedral manifold ( equation ( [ f12 ] ) ) , which entangles left and right actions . among the actions described by equation ( [ u3 ] ) are the conjugation actions : @xmath47 which form a subgroup @xmath48 . the action equation ( [ u3b ] ) preserves the trace , @xmath49 and therefore operates only on the subspace with coordinates @xmath50 . the unitary structure of the three - sphere governs its isometries and introduces the representation theory of @xmath0 . in particular , we obtain for the action of @xmath36 on a wigner polynomial equation ( [ u9 ] ) : @xmath51 the expression in the last line is the matrix element of the irreducible representation @xmath52 of @xmath36 for the pair @xmath53 . the degree @xmath32 stays fixed under the action of @xmath36 , while the pairs @xmath54 take @xmath20 values . the actions by conjugation equation ( [ u3b ] ) of the subgroup @xmath34 reduce into irreducible form upon transforming the wigner polynomial basis ; compare also @xcite , with wigner coefficients of @xmath0 @xcite ( pp . 3152 ) , into a spherical basis by : @xmath55 equation ( [ u14b ] ) implies that the irreducible representation @xmath56 of @xmath57 when reduced to the subgroup @xmath48 contains any irreducible representation @xmath58 of @xmath48 once and only once for @xmath59 . the transformation inverse to equation ( [ u14b ] ) is : @xmath60 the two transformations equations ( [ u14b ] ) and ( [ u14c ] ) will play a crucial role in the recursive construction of the harmonic analysis for orbifolds in section [ sec : projharmonic ] . in cosmic topology , section [ sec : cmb ] , the spherical basis equation ( [ u14b ] ) in terms of spherical harmonics @xmath61 is required for the multipole analysis of physical observables , like the cmb radiation . it is given by : @xmath62 where @xmath63 denotes the gegenbauer polynomial . as a result of equation ( [ u14b ] ) , the conjugation action of @xmath48 on the spherical basis becomes : @xmath64 identical to the irreducible action of the rotation operator @xmath65 on the usual spherical harmonics @xmath66 in @xmath67 @xcite ( pp . the coordinate @xmath68 in @xmath4 is unchanged under this action , and so , we find : summary : the action of the rotation @xmath69 on wigner polynomials is given by the representation @xmath52 , equation ( [ u14 ] ) . the rotations @xmath70 act only on the first three coordinates , and the spherical basis equation ( [ u14b ] ) yields the decomposition of wigner polynomials into irreducible subbases characterized by the multipole order @xmath71 . to display the topological selection rules for wigner polynomials , we concentrate on their discrete labels @xmath73 . both of them are an integer or half - integer . these labels form an integer plus a half - integer @xmath72 @xmath1-grid , both of spacing @xmath74 , on a @xmath72 plane . from the point of view of representations of @xmath0 , the grid points fix subrepresentations of left and right subgroups @xmath75 . consider the set of wigner polynomials as being attached to the grid points @xmath54 . at a given grid point @xmath54 , the labels @xmath76 of all wigner polynomials attached to it have the denumerable range @xmath77 . conversely , the wigner polynomials for fixed degree @xmath32 occur with values @xmath73 on a centered square @xmath78 . in section [ sec : subgrids ] , we shall see that harmonic analysis on topological orbifolds selects wigner bases on subgrids . summary : a denumerable set of wigner polynomials @xmath79 with the index pair @xmath54 and degree @xmath80 is associated with any single point @xmath54 of a @xmath72 @xmath1-grid . geometric orbifolds @xmath81 , @xmath82 are shown in the top panels and @xmath83 , @xmath84 in the bottom panels . the yellow color marks the duplex form of the orbifolds as pairs of coxeter simplices , glued with triangular faces in blue color . the axes of the covering rotations @xmath85 for each orbifold are marked by thin lines . the parts of these axes that bound the glue triangles carry fixpoints of integer order @xmath86 . , scaledwidth=80.0% ] in this section , we turn from the three - sphere to the spherical manifolds of platonic polyhedra shown in figure [ fig : duplices ] . to describe their geometrical transformation properties , we use spherical coxeter groups . coxeter groups @xmath87 are generated by weyl reflections in ( hyper-)planes of euclidean space @xmath88 . any weyl reflection @xmath89 is characterized by a unit vector @xmath90 , normal to the reflection ( hyper-)plane . the action of the weyl reflection @xmath89 on @xmath91 is given by : @xmath92 a weyl reflection is a euclidean isometry of determinant @xmath93 . a coxeter group @xmath87 is generated by weyl reflections @xmath94 . for a fixed coxeter group @xmath87 , we use the short - hand notation @xmath95 . @xmath87 is graphically denoted by a coxeter diagram , built from circles for each generator and connected by a sequence of lines . the diagram allows one to construct the group @xmath87 from its weyl reflection generators and their relations @xcite ( pp . 1223 ) , @xcite . the lines between circles carry integer numbers @xmath96 . lines without numbers are short - hand for @xmath97 . the number @xmath98 between two successive reflections @xmath99 implies the relation : @xmath100 the number @xmath98 is the order of the rotation generated by @xmath101 . in tables [ tab : coxa ] and [ tab : coxb ] , we list the four spherical coxeter groups of the tetrahedron , cube , octahedron and dodecahedron , each with four generators , their diagrams and quadruples of unit vectors @xmath102 . these coxeter groups produce tilings @xcite of the three - sphere @xmath2 by platonic polyhedra . .[tab : coxa ] diagrams of four spherical coxeter groups @xmath87 of order @xmath103 , four platonic polyhedra @xmath104 and seven deck groups @xmath105 of order @xmath106 according to @xcite ( p. 134 ) . in the table , @xmath107 denotes a cyclic , @xmath108 the quaternion , @xmath109 the binary tetrahedral and @xmath110 the binary icosahedral group . the symbols @xmath111 are adapted from @xcite . [ cols="<,<,<,<,<,<",options="header " , ] the enumeration of faces and directed edges of the platonic polyhedra is shown in figure [ fig : platonicpoly ] . in table [ tab : homo ] , we list the gluing of faces and directed edges as given in @xcite . homotopic gluings relate faces and edges of a single polyhedron . a new geometric view of topologies emerges on the universal covering manifold @xmath2 . a general theorem , given by seifert and threlfall @xcite ( pp . 181203 ) , proves for topological manifolds the isomorphism between the first homotopy group and the group of deck or covering actions on the universal cover . we implement this theorem for the cubic spherical manifold @xmath82 and convert its homotopic gluings from table [ tab : homo ] into deck actions between neighboring copies of a single proto - cube with fixed face and edge enumeration . the deck rotations , which generate the tiling of @xmath2 , form the topological deck group . the eight cubic copies tile the three - sphere in the form of the eight - cell shown in figure [ fig : platonicpoly ] in a projection from @xcite ( p. 170 ) . this means that the topological deck group must be of order eight . for the homotopy of @xmath82 , we make full use of the prescriptions of table [ tab : homo ] . any quadratic face of the proto - cube is surrounded by four directed edges given in figure [ fig : platonicpoly ] . these four directed edges in turn determine the face and , moreover , give its orientation with respect to a four - fold axis perpendicular to the face . we use square diagrams as given in equation ( [ u54 ] ) , formed from four directed edges , to denote edges around a face plus their orientation . this orientation must be respected in the homotopic gluing of faces and their edges given in table [ tab : homo ] . it suffices to convert only the homotopic face gluings of a proto - cube , the glue generators , into deck actions . then , these deck actions by multiplication generate the full deck group . in @xmath67 , we choose the coordinates @xmath112 normal to the three faces labeled @xmath113 in figure [ fig : platonicpoly ] . positive rotations by an angle @xmath114 around these normals we denote as @xmath115 . before turning to the manifold @xmath82 in equation ( [ u55 ] ) , we first treat in equation ( [ u54 ] ) a simple reference glue @xmath116 of two opposite faces of the proto - cube . we convert this glue into a deck action of two copies of the proto - cube and write it algebraically in equation ( [ u54 ] ) as a product of weyl reflections and inversions . by @xmath117 , we denote the inversions in @xmath118 . we find this action in the form : @xmath119 \leftarrow \left[\begin{array}{lll } & \overline{10}&\\ \overline{6}&&12 \\ & \overline{11 } & \end{array}\right ] \\ \nonumber & & = w_4 { \cal j}_3= w_4 w_0 { \cal j}_4\end{aligned}\ ] ] the product of weyl reflections and inversions in equation ( [ u54 ] ) combines into an overall unimodular rotation matrix . for convenience , we introduced the additional weyl reflection operator @xmath120 , @xmath121 . now , we return to the cubic manifold @xmath82 . by use of two rotations , we convert its first face gluing , @xmath122 , into the reference deck action equation ( [ u54 ] ) , @xmath123 by similar conjugations as in equation ( [ u54 ] ) , we convert the two other face gluings of the manifold @xmath82 in equation ( [ u52 ] ) into deck actions , @xmath124 from products of two weyl reflection operators in equation ( [ u56 ] ) , we pass to elements @xmath125 given in table [ tab : coxb ] by equation ( [ b3 ] ) . as shown in @xcite , the deck generator in equation ( [ u55 ] ) generates a cyclic group @xmath126 of order eight , which paves the eight - cell tiling of @xmath2 by copies of the prototile . the eight - cell admits a second inequivalent cubic homotopy @xmath83 @xcite . its deck group derived in @xcite is the quaternion group @xmath108 . we give the generating elements in equation ( [ qg ] ) . for the other platonic homotopies studied by everitt @xcite , the conversions to deck actions are carried out in @xcite . summary : the homotopic gluings for platonic polyhedra from @xcite are listed in table [ tab : homo ] . we convert them into actions of deck groups @xmath127 on the covering three - sphere . they generate tilings by @xmath106 tiles of the covering three - sphere . the spherical platonic polyhedra cover the three - sphere by deck transformations . by use of its point symmetry group @xmath128 , we can decompose a single platonic polyhedron under @xmath128 into a fundamental domain and its orbit . we must remove two obstacles towards a topological interpretation : \(i ) to keep the orientation under deck actions , we restrict the point group @xmath128 to proper rotations . for the coxeter group action , its fundamental domain is a simplex bounded by four weyl reflection planes . when we restrict to rotations to preserve orientation , the actions belong to the unimodular subgroup of @xmath87 of determinant one , which we denote as @xmath129 . when passing to @xmath129 , we must extend the fundamental simplex domain of @xmath87 for @xmath129 to a duplex , consisting of a simplex and its image under a reflection , see figure [ fig : duplices ] . \(ii ) a topological manifold strictly excludes fixpoints under deck actions . here , they appear on the rotation axes of the point groups . we must extend our topological notion from a manifold to an orbifold @xcite , which allows for fixpoints of finite order . the notion of orbifolds is explained in more detail in @xcite ( pp . 652714 ) . in figure [ fig : duplices ] , we give the construction of orbifolds by gluing pairs of coxeter simplices into duplices for the four platonic polyhedra . the fixpoints appear on the edges of the blue glue triangles for the two coxeter simplices . to cover the full three - sphere by orbifolds , we proceed in two steps . first of all , any platonic polyhedron under its rotational point symmetry group @xmath128 is paved by @xmath130 copies of spherical orbifolds . these spherical orbifolds we take as duplices glued from a coxeter simplex and its mirror image . copies of the duplices shown in figure [ fig : duplices ] under the point group @xmath128 tile the platonic polyhedra , but have fixpoints on their edges . to cover the full three - sphere , we augment the rotational point symmetry group @xmath128 by the operations of the full deck group @xmath127 of the platonic manifold . these two groups share only the identity element , @xmath131 . the product @xmath132 of point and deck operations @xmath133 , acting on a proto - orbifold , generates as the deck group the unimodular subgroup @xmath134 of the coxeter group . the images of the proto - duplex under all of these products cover the three - sphere . summary : the topological orbifolds for platonic spherical polyhedra are duplices ( figure [ fig : duplices ] ) , glued from two coxeter simplices . their deck groups are the unimodular subgroups @xmath135 . each orbifold deck group is factorized as @xmath136 , into the point symmetry group @xmath128 and the deck group @xmath127 for the platonic polyhedron . as the general basis for all of these topologies , we take the harmonic wigner polynomials . to adapt them to specific orbifolds , we shall present in the following subsections different routes . by graphical means , we shall illuminate on @xmath72 @xmath1-subgrids ; see figure [ fig : subgrids ] , denumerable sets of wigner polynomials for orbifolds , selected from the full harmonic set on the three - sphere . more details of the harmonic analysis on the platonic orbifolds are given in section [ sec : harmonic ] . subgrids of integer / half - integer black / red points @xmath137 are spanned by heavy black / red vectors . the marked grid points select the bases @xmath138 for a spherical orbifold @xmath111 from the square grid formed by thin lines . coordinate transformations are given in equation ( [ u14e ] ) . tetrahedral case @xmath81 , coordinates @xmath139 , integer black and half - integer red grid points . cubic case @xmath82 , coordinates @xmath139 , with integer black grid points . cubic case @xmath83 , coordinates @xmath140 , vertical broken mirror line , integer black grid points . octahedral case @xmath141 , coordinates @xmath140 , horizontal broken mirror line , integer black grid points.,scaledwidth=51.0% ] the unitary structure of the three - sphere allows one to reduce any given deck action to a pure phase transformation . a general rotation operator @xmath142 can be brought to a simple form by unitary diagonalizations of the matrices @xmath143 , @xmath144 , \updelta_r=\left [ \begin{array}{ll } \exp(i\upgamma ) & 0\\ 0 & \exp(-i\upgamma ) \end{array}\right ] \vspace{4pt}\end{aligned}\ ] ] to find the angular parameters @xmath145 of the diagonal form , one can avoid the explicit diagonalization equation ( [ u14da ] ) . for any @xmath41 , use the trace relation : @xmath146 to determine @xmath147 . we now pass from @xmath140 to a new matrix @xmath139 by the @xmath36 rotation : @xmath148 the matrix @xmath139 represents a new system of coordinates on @xmath2 . it follows from equation ( [ u14da ] ) that to the action of @xmath149 on the coordinates @xmath140 , there corresponds the action @xmath150 on the coordinates @xmath139 . if we pass in a wigner polynomial from the coordinates @xmath151 to euler angles , we find in this parametrization : @xmath152 that is , a pure phase transformation . the condition of invariance of a wigner polynomial @xmath153 under the diagonal action of the operator @xmath150 now takes the form of a phase condition : @xmath154 summary : any deck action on wigner polynomials by a rotation can be converted into a pure phase transformation equation ( [ u14q ] ) . any deck group with a single - generator can be treated in this way . the harmonic analysis must be restricted to the modes admissible for a specific spherical topology . a denumerable set of wigner polynomials @xmath79 with the index pair @xmath54 and degree @xmath80 is achieved by choosing from the harmonic wigner basis a subbasis invariant under the relevant point and deck group . since all deck actions are of finite order , the angles @xmath155 in equation ( [ u14q ] ) must be integral fractions of @xmath156 of the forms @xmath157 . for groups with a single generator , rewrite equation ( [ xx1 ] ) in the form : @xmath158 the solutions @xmath159 of equation ( [ xx2 ] ) for @xmath160 form integer multiples of a grid vector @xmath161 . solutions of equation ( [ xx2 ] ) for @xmath162 are on lines parallel to @xmath163 . we choose a second vector @xmath164 to a grid point outside , but next to the line spanned by @xmath163 . then , the vectors @xmath165 form the basis of an @xmath1-subgrid , whose points yield all solutions of equations ( [ xx1 ] ) and ( [ xx2 ] ) . this case applies to the orbifolds @xmath81 and @xmath82 with cyclic deck groups @xmath46 , @xmath126 . in figure [ fig : subgrids ] , we mark the subgrids for the orbifolds @xmath81 , @xmath82 , @xmath83 , @xmath141 . note that the topological rules equation ( [ xx2 ] ) , given in terms of subgrids , select enumerable sets of harmonic widner basis polynomials of unrestricted degree @xmath32 . the subgrid method can be extended to orbifolds when at most two wigner polynomials with index pairs @xmath166 $ ] or @xmath167 $ ] linearly combine into an invariant basis polynomial . we call these two cases vertical and horizontal mirror pairs . it then suffices to mark one partner of this pair and to give the algebraic phase for its partner . these cases apply for the orbifolds @xmath82 and @xmath83 , illustrated in figure [ fig : subgrids ] . for more general deck groups , we establish the identity representation by application to a wigner polynomial @xmath21 of degree @xmath32 the projection operator @xmath168 equation ( [ u57 ] ) to the identity representation of the deck group @xmath127 , @xmath169 by equation ( [ u14 ] ) , this application produces for any degree @xmath32 a linear combination of wigner polynomials . if this can be normalized , it contributes to the harmonic analysis . for platonic spherical manifolds , these projections were carried out in @xcite . in many cases , it is possible to display the results of the projection equation ( [ u57 ] ) on an @xmath1-subgrid : the non - vanishing wigner polynomials selected by projection are located on a subgrid , dependent on the manifold , and spanned by two specific grid vectors @xmath165 . we display subgrids in figure [ fig : subgrids ] . the advantage of the subgrid method is that different topologies can be directly compared in terms of their subgrids . summary : the wigner polynomials invariant under a single - generator deck group are located on a subgrid spanned by two grid vectors @xmath165 . any subgrid point @xmath54 stands for an enumerable set of invariant wigner polynomials of degree @xmath170 . the spherical manifolds @xmath171 with cyclic deck groups @xmath172 belong to this set . the basis of the harmonic analysis for a specific topology can be recursively computed , as we shall show . we have seen that the deck group of the orbifold factorizes into the symmetry group @xmath128 of the platonic manifold and its deck group @xmath127 . the basis functions for the harmonic analysis must be invariant under the deck group @xmath132 , which is possible only if they are invariant under both @xmath128 and @xmath127 . the two projectors for the groups @xmath128 and @xmath127 commute @xcite . we start for the cubic orbifold @xmath83 the first loop of a recursive construction of basis functions of increasing degree @xmath32 : \(i ) the action of symmetry groups @xmath173 of the platonic manifolds is known from the euclidean setting in @xmath67 . the point group @xmath174 is a subgroup of the rotation group @xmath175 ; therefore , any @xmath128-invariants must appear for fixed @xmath71 as linear combinations of spherical harmonics @xmath176 . standard spherical harmonics @xmath177 with lowest label @xmath178 and invariant under @xmath174 can be found in @xcite ( pp . 436438 ) for the cubic group and in @xcite ( pp . 105114 ) . for several point groups @xmath128 , the onset @xmath179 of spherical harmonics invariant under @xmath128 is listed in @xcite , table 3 . the linear combinations of lowest @xmath178 can be expressed in the spherical basis equation ( [ u14b ] ) of @xmath36 . for the cubic point group @xmath174 , the lowest linear combination of spherical harmonics given in table [ tab : cubicl ] has @xmath180 . \(ii ) to also project on an invariant under @xmath127 , which is @xmath181 for the orbifold @xmath83 , we transform with equation ( [ u14c ] ) from the spherical basis back to the wigner basis and then apply the projector equation ( [ u57 ] ) with the matrix element : @xmath182 once the result of ( ii ) is transformed back into the spherical basis with equation ( [ u14b ] ) , we recover linear combinations with new values of multipole order @xmath183 , which , if non - vanishing , can be normalized . these new states must still be invariant under the point group @xmath128 . it follows that they must contain @xmath128-invariant linear combination of spherical harmonics with @xmath184 . for the cubic point group @xmath174 , we demonstrate this result in table [ tab : cubicl ] with new cubic invariants for @xmath185 and @xmath186 . with them , we enter the next loop with steps ( i ) and ( ii ) in search of new invariant basis functions for the deck group @xmath187 . this recursive method uses only well - known wigner coefficients of @xmath0 , but avoids any new projection to invariants under @xmath128 . the new basis functions in table [ tab : cubicinv ] contain linear combinations of different multipole orders @xmath188 . this implies that topology can enforce linear correlations between amplitudes of different multipole order , which has important implications for the analysis of the cosmic microwave background . @xmath189 @xmath190 in table [ tab : cubicinv ] , we exemplify the first basis functions for the cubic orbifold @xmath83 . summary : the recursive method , by use of loops with steps ( i ) and ( ii ) , yields a basis of the harmonic analysis of any platonic orbifold . invariance under a topological deck group will imply in general the correlation between multipoles of different orders @xmath3 . depending on the orbifold , we apply one of the methods given in the previous section to derive the basis of each platonic orbifold . the deck group of @xmath81 from @xcite is the cyclic group @xmath191 of order five . its single generator is the rotation : @xmath192 where the vector @xmath193 of the weyl reflection @xmath194 of the coxeter group from table [ tab : coxb ] by equation ( [ b3 ] ) determines the unitary matrix @xmath195 ( equation ( [ b2 ] ) ) , and equation ( [ b3 ] ) is used to convert products of two weyl reflection operators into rotations . equation ( [ f12 ] ) demonstrates that the entanglement of left and right actions in the group @xmath46 is enforced by the geometry of the tetrahedron . we follow section [ sec : wigner ] and replace the old coordinates @xmath140 by the new ones @xmath196 . by the trace relation equation ( [ u14f ] ) for @xmath143 , we find the diagonal forms : @xmath197 , \:\ : \updelta_r= \left [ \begin{array}{ll } \exp(i\:6\pi/5 ) & 0\\ 0 & \exp(-i\:6\pi/5)\\ \end{array}\right ] \vspace{3pt}\ ] ] to project a basis state invariant under @xmath46 , it suffices to make it invariant under the generator equation ( [ f12 ] ) of @xmath46 , since then , it will be invariant under any power of this generator . we choose a fixed wigner polynomial and find in the coordinates @xmath139 as action the phase transformation : @xmath198 to have invariance under the generator of @xmath46 , we require : @xmath199 this condition is fulfilled on the subgrid points marked in figure [ fig : subgrids ] , @xmath81 by black and by red circles . the subgrids are spanned by the black vectors @xmath200 from an integer or by the same red vectors from a half - integer grid point . the subgrid points form a selection from all grid points . the subbasis consists of all wigner polynomials associated with the subgrid points . the deck group is @xmath126 generated by : @xmath201,\:\ : g_r= \left[\begin{array}{ll } 0 & \overline{a}\\ -a & 0\\ \end{array } \right],\ : a=\exp(i\ : 2\pi/8)\ ] ] again , the left and right actions are entangled . the subbasis of the harmonic analysis is of the horizontal mirror type ( see section [ sec : mirror ] ) given by the linear combination of two wigner polynomials : @xmath202 , \\ \nonumber & & j \in z,\ : \text{for } \ : m_1 \ : \text{even},\ : 0 < m_2\leq j\end{aligned}\ ] ] the deck group is the quaternionic group @xmath108 , acting from the left , with elements : @xmath203 , \ : g_l=-{\bf j } = \left [ \begin{array}{ll } 0&-1\\ 1 & 0\\ \end{array}\right],\ : g_l=-{\bf i}= \left [ \begin{array}{ll } -i&0\\ 0 & i\\ \end{array}\right]\ ] ] the subbasis is of vertical mirror type , given by a linear combination of two wigner polynomials : @xmath204\\ & & \text{for}\:j\:\text{odd } , \geq 3 , \text{for}\:m_1\:\text{even } , 0<m_1\leq j , -j\leq m_2\leq j\end{aligned}\ ] ] the deck groups here are of the order @xmath207 . we give the basis construction to the orbifold @xmath141 with deck group @xmath208 acting from the left and from the right , respectively . for the subbasis in new coordinates @xmath139 , we find a horizontal mirror symmetry and , from @xcite ( p. 26 ) , the following polynomials and mirror phases : @xmath209 , \nonumber \\ \text{for}\:j \:\text{even } , m_2=0 : & \upphi^j_{\uprho,0}&= d^j_{\uprho,0}(u(\tilde{x } ) ) , \nonumber \\ \text{for}\:j \:\text{even } , \geq 2,\ : 0 < m_2 \leq j : & \upphi_{\uprho , m_2}&= [ d^j_{\uprho , m_2}(u(\tilde{x}))+ d^j_{\uprho ,- m_2}(u(\tilde{x}))]\nonumber \\ \end{aligned}\ ] ] the octahedral manifolds @xmath205 , @xmath206 are analyzed in @xcite . an analysis of invariant polynomials on poincar s dodecahedral manifold is given in @xcite , based on the work of klein @xcite . as an application of the derived subbases , we consider the observable effects of multiply - connected universes . the topology of the cosmos has its roots in albert einstein s work @xcite . in 1917 , einstein presented his pioneering paper on general relativity , cosmology and gravitation @xcite ( pp . 160164 ) . he communicates three fundamental ideas : \(i ) space - time manifold and metric : space and time are unified into a single 4d space - time manifold , with three space coordinates @xmath112 and one time coordinate @xmath210 . following riemann , this four - manifold carries a pseudo - euclidean metric with the space - time squared distance : @xmath211 in the second line , we adopt the usual splitting into time and space . space - time by equation ( [ e1 ] ) is distinct from @xmath4 used in previous sections . \(ii ) gravity and geodesics : einstein gives field equations for newton s gravity . they link the second derivatives of the metric tensor @xmath212 with respect to the coordinates @xmath213 linearly to the energy - momentum tensor @xmath214 . geodesics , the shortest lines allowed by the metric equation ( [ e1 ] ) , are the lines followed by massive test particles . for velocities small compared with the velocity @xmath215 of light , einstein s field equations reduce to newton s laws for the gravitational potential in the presence of masses and the differential equations for geodesics to newton s equations of motion . in general , the metric and the energy - momentum tensor with einstein become observables of astrophysics . \(iii ) topology : the connectivity of the space - time manifold is the subject of cosmic topology . the classes of possible closed paths are the basic concept of homotopy . the view of space - time as a manifold , with gravitation obtained from local differential equations , presents topology as an observable . when time is split off as in equation ( [ e1 ] ) , homotopy refers to the @xmath216 space part of space - time . in cosmology , three possible curvatures of the universe are distinguished : hyperbolic space with negative , euclidean space with zero and spherical space with positive average curvature . the present astrophysical data favor positive curvature and , hence , a spherical spatial topology , as discussed here . one important observable is the spatial fluctuations of the cosmic microwave background . the cosmic microwave background ( cmb ) is an observed uniform thermal black - body radiation of a present temperature of 2.725 kelvin with a peak frequency of 160.2 ghz , discovered in 1964 by a. penzias and r. wilson @xcite . according to big bang cosmology , its origin is dated back to the early epoch of photon decoupling , when neutral atoms were formed at very high density and temperature . its amplitude today is observed by the wilkinson probe and up to 2013 by the planck satellite @xcite . after a non - trivial cleaning of the measured temperature for global astrophysical influences , such as the wolfe sachs effect and the local foreground of the solar system and our galaxy , the cmb spatial temperature fluctuations are expanded in multipoles , shown for the lowest values of @xmath71 in figure [ fig : cmb ] . cmb temperature power - spectrum as a function of multipole order @xmath71 ; data from @xcite ( figure 35 ) . note the weak amplitudes of some of the lowest multipoles.,scaledwidth=50.0% ] the connection of multipole - resolved cmb measurements to cosmic topology is provided by selection rules for the harmonic analysis . the different topologies discussed here should imprint their selection rules on the multipole amplitudes of the cmb . the cubic spherical manifold @xmath82 discussed here has been used for detailed predictions of the cmb temperature fluctuation map @xcite . in addition , the ratio of different multipole - moments encodes the specific topology ; see table [ tab : cubicinv ] for the cubic case . both topological signatures can be revealed with the methods given here , which yields regular tilings of the three - sphere . according to @xcite , section 9.3 on p. 36 , the observed low amplitudes and correlations of the first cmb multipoles ( figure [ fig : cmb ] ) might indicate a violation of statistical isotropy or might indicate the presence of topological selection rules . besides regular tilings , also random tilings of spherical three - manifolds have been analyzed in the literature @xcite . we have constructed the orthonormal basis sets for four platonic spherical topological orbifolds , required for performing a harmonic analysis on specific topologies . the mathematical framework used starts from the homotopy or fundamental groups and yields isomorphic deck actions and deck groups on the three - sphere . the point symmetry groups @xmath128 of the platonic manifolds lead from manifolds to orbifolds . the orthonormal wigner polynomials provide a basis for the harmonic analysis on @xmath0 . by deriving the subgrid of each platonic manifold on the space spanned by the indices of the wigner polynomials , we obtain selection rules for the harmonic analysis . moreover , the coefficients of the multipole expansion encode the underlying topology . in contrast to other approaches , no numerical orthogonalization of basis sets is required , and the connection between topology and geometry of space is transparent from the introduction of the homotopies . finally , we discussed the harmonic analysis of the cmb and present signatures of the underlying topology of multiply - connected universes encoded in the cmb fluctuations . einstein , a. osmologische betrachtungen zur allgemeinen relativittstheorie . in _ sitzungsberichte der kniglich preuischen akademie der wissenschaften ( berlin ) _ ; 1917 ; pp . kramer , p. an invariant operator due to f klein quantizes h poincar s dodecahedral 3-manifold . , _ 38 _ , doi:10.1088/0305 - 4470/38/16/004 . kramer , p. multipole analysis in cosmic topology . in _ symmetries in nature : symposium in memoriam marcos moshinsky _ ; american institute of physics ( aip ) publishing : college park , md , usa , 2010 ; volume 1323 , pp .

chemical reaction is one of the most remarkable quantum mechanical phenomena that made biological life possible on earth , which can also be manipulated to discover novel `` green '' chemistry , and to avoid `` toxic '' technologies , provided we can understand the reaction microscopically to a certain extent . however , within the current formalism of chemical kinetics and the laws of thermodynamics , a chemical reaction between two reactant atoms ( for a given chemical potential , temperature , pressure and concentration ) occurs when they collide with each other with enough energy to overcome an arrhenius - type activation energy , to form a(several ) new bond(s ) @xcite . here , we attempt to go beyond this macroscopic collision - thermodynamic based description by incorporating some microscopic physics ( with respect to atomic energy level spacings ) to understand why and how a chemical reaction proceeds for a given set of conditions . the chemical reaction considered here may need the right reaction coordinates to form the transition states . therefore , for two atoms ( either isolated or from two different molecules ) to chemically react , they need to have the correct reaction coordinates , and even after fulfilling this condition , these atoms still need to have an overall strong attractive interaction ( either before or after the collision ) to form the required transition states . for example , the transition state is by definition gives equal probability for the reaction to go either way ( to form reactant atoms , or the reaction product ) . therefore , there should be a microscopic mechanism that is responsible for an effective attraction between reactant atoms such that the reaction product is more likely than the formation of reactant atoms , even after forming the transition state . the standard attractive vdw interaction is a long - range type , and it is too weak ( because it is proportional to 1/@xmath0 ) to initiate any chemical reaction . on the other end , we have the electron(from atom 1)-ion(from atom 2 ) coulomb attraction , which is the strongest , but can not lead to an attraction due to electron(from atom 1)-electron(from atom 2 ) coulomb repulsion . apparently , collisions among the reactant atoms can overcome this electron - electron coulomb repulsion if the atoms are weakly interacting . collisions can induce electronic excitation and/or polarization , which then can activate ( i ) the electron - electron coulomb repulsion or ( ii ) the attraction due to ion - ion vdw and electron - ion coulomb interactions . as a consequence , we need to find the microscopic physical origin of a chemical reaction with respect to the above strong attractions , which goes beyond the present - day thermodynamics ( chemical potential ) and energetic - collision knowledge . here , we prove the existence of a stronger attraction ( stronger than the usual vdw type ) that can be used to understand why and how a chemical reaction can proceed ( with further assistance from collisions ) by overcoming the electron - electron coulomb repulsion . in other words , we set out to prove the existence of a stronger vdw interaction , which is also shown to be further enhanced by the electron - ion coulomb attraction to activate the chemical reaction . in particular , we expose that these stronger attractions are polarization - induced effect , which can be precisely associated to atomic energy - level spacings , and these stronger attractions are also responsible to overcoming the said repulsion ( or the activation energy ) . once the repulsion is overcome , either after , or before the collision , then it is straightforward to understand why and how a chemical reaction between two reactant atoms can proceed . we stress here that we do need atom - atom collisions for strongly interacting atoms ( with isotropic polarization ) , and also for weakly polarized atoms . however , after the collision , the above attractions still need to be activated for a chemical reaction to proceed . it is to be noted here that this stronger vdw attraction has also been partly predicted computationally and indirectly @xcite by means of the hermansson blue - shifting hydrogen bond @xcite , the arunan composite hydrogen - vdw bond @xcite and the non - covalent carbon bond @xcite . in the section , additional notes prior to conclusions , we will show that ramachandran , sasisekharan and ramakrishnan @xcite have already discovered the notion of stronger non - covalent bonds ( stronger than the vdw bond ) between atoms . this means that the hermansson blue - shifting hydrogen bond , the hydrogen - vdw composite bond and the non - covalent carbon bond are special cases of the ramachandran bond , in the absence of wavefunction overlapping or chemical reaction . hence , the ramachandran bond is a non - chemical bond that does not require chemical reaction . however , to pin down the origin of a stronger vdw and an attractive electron - ion coulomb interactions that overcome the electron - electron repulsion ( due to polarization ) , we definitely need analytic solutions and some abstract analyses . therefore , we exploit the crude quasi one - dimensional ( 1d ) drude model within the ionization energy theory ( iet ) @xcite using the energy - level spacing renormalization group method @xcite to arrive at accurate and exact results . the `` quasi '' here implies that one can realign the vectors in a one - dimensional system to account for the two and three dimensional cases in which the atoms are treated as real atoms . in our further analyses ( i and ii ) , we unequivocally show why and how the stronger vdw interaction is also consistent with respect to the second - order perturbation theory and the @xmath1 multipole expansion . therefore , its existence is not due to some interpretational issues arising exclusively from the crude drude model hamiltonian . the relevant hamiltonian for the one - dimensional two - atom system ( strictly neutral atoms , but with polarizable valence electrons ) plotted schematically in fig . [ fig:1 ] is given by @xcite @xmath2 , \label{eq:2}\end{aligned}\ ] ] where @xmath3 are the electrons momenta , and the ground state energy for the above hamiltonian is ( taking @xmath4 and after some nontrivial substitutions ) @xcite @xmath5 , \label{eq:3}\end{aligned}\ ] ] where @xmath6 is the electron mass and @xmath7 is the interaction constant ( that changes for different atoms ) . note here that @xmath8 and @xmath9 are indeed vectors , but we have aligned the atoms such that their electronic polarization are maximum along a straight line where @xmath10 and @xmath11 . we now prove why this alignment is valid for three - dimensional cases ( imagine two polarizable spheres)suppose @xmath8 ( from sphere 1 ) is aligned along the @xmath12 axis , so that @xmath13 . the induced polarization due to @xmath8 causes @xmath14 where @xmath15 , @xmath16 , @xmath17 is the angle in the @xmath18-plane and @xmath19 denotes the angle in the @xmath20-plane . it is easy to verify that if @xmath21 and @xmath22 , then there is an additional induced polarization affecting @xmath14 . if such a case exists , then we just need to take this additional polarization into account ( that affects @xmath9 ) , and realign @xmath9 ( from sphere 2 ) with respect to this new @xmath23 . in this second case , @xmath24 causes @xmath25 where @xmath15 and @xmath16 . this means that , @xmath17 and @xmath19 remains the same such that @xmath26 $ ] and @xmath27 $ ] . for one - dimensional cases , @xmath28 , while for a two - dimensional case , one has @xmath26 $ ] and @xmath29 . @xmath30 in eq . ( [ eq:3 ] ) , @xmath31 is the planck constant divided by 2@xmath32 , @xmath33 is the distance between two neutral and identical atoms . the subscripts 1 and 2 refer to electron 1 and 2 bounded to nuclei 1 and 2 , respectively . the ordinary hamiltonian denoted by @xmath34 consists of two kinetic energy terms , and two semiclassical harmonic - oscillator type electron - ion potential terms , and they are associated to two electrons bounded to their respective nucleus ( see fig . [ fig:1 ] ) . the interaction hamiltonian , @xmath35 captures all the coulomb interaction between the two atoms depicted in fig . [ fig:1 ] . note that @xmath35 does not consider the coulomb forces within the atom ( between the valence electrons and their respective nucleus ) , which have been taken into account by @xmath34 . now , one can renormalize the interaction potential constant , @xmath7 such that @xcite @xmath36 } , \label{eq:4}\end{aligned}\ ] ] where @xmath37 , in which , @xmath38 is the bohr radius of atomic hydrogen , @xmath39 and @xmath40 are the electron charge and the permittivity of free space , respectively . the above renormalization procedure is also related ( exactly ) to the shankar renormalization technique @xcite . the renormalized interaction potential constant , @xmath41 ( from eq . ( [ eq:4 ] ) ) replaces the standard @xmath7 in @xmath34 to capture the changes to @xmath7 whenever one changes the type of atom . whereas , the variable , @xmath42 denotes the atomic energy level spacing when @xmath43 , which is proportional to @xmath42 when @xmath44 . this proportionality ( see eq . ( [ eq : in14b ] ) ) has been proven in ref . @xcite , and we do not allow any electronic wavefunction overlapping . if the drude model hamiltonian @xcite is independent of @xmath7 , then one should renormalize the frequency , such that @xcite , @xmath45}$ ] . since the factor @xmath46}$ ] is a constant for a given system and for a given set of conditions , one will obtain the identical vdw interaction energy ( see eq . ( [ eq:6 ] ) ) such that this energy is proportional to @xmath47}$ ] , @xmath48 regardless whether one uses the @xmath7-dependent hamiltonian given by eq . ( [ eq:2 ] ) or the @xmath49-dependent drude model hamiltonian , provided that the condition , @xmath50 } \gg e^2/2\pi\epsilon_0r^3 $ ] is satisfied . if @xmath50 } \gg e^2/2\pi\epsilon_0r^3 $ ] is not satisfied , then one may obtain @xmath51 where @xmath52 denotes a positive integer ( @xmath53 ) . the proper renormalization procedure ( with explanation ) are given in ref . @xcite . we now expose some details on the meaning of @xmath42 and the ionization energy approximation , which can be understood from the iet - schr@xmath54dinger equation @xcite , @xmath55\psi(\textbf{r},t ) = h_{\rm iet}\psi(\textbf{r},t ) = ( e_0 \pm \xi)\psi(\textbf{r},t ) , \label{eq : in14}\end{aligned}\ ] ] where @xmath56 , @xmath57 denotes the planck constant and @xmath6 is the mass of electron , and @xmath58 is the time - dependent many - body wavefunction . the eigenvalue , @xmath59 is the real ( true and unique ) energy levels for a given quantum system ( for a given molecule ) . here , @xmath60 is a constant because it represents the energy level spacings at zero temperature , and in the absence of any external disturbances . on the other hand , @xmath61 is the so called ionization energy . therefore , all we need to do now is to find a way to obtain @xmath62 . we make use of the ionization energy approximation to approximate @xmath62 . this approximation hinges on this proportionality , @xmath63 as stated earlier , @xmath62 represents the real ( unique and true ) energy level spacings of a particular quantum system , and it can be regarded as a generalized electronic energy gap . in other words , @xmath62 can only be obtained if we know the real many body wavefunction ( @xmath58 ) , not some guessed wavefunctions , because only the true wavefunction can represent the electrons properly , and consequently , can capture the real electronic properties of a quantum matter . in molecules , @xmath64 is the energy level spacing between an occupied level ( in the highest occupied molecular orbital ( homo ) ) and an empty level ( in the lowest unoccupied molecular orbital ( lumo ) ) . using this approximation , we can now predict the changes to @xmath62 if we know @xmath65 . the values for @xmath65 can be readily obtained from the experimental atomic spectra . the ionization energy approximation becomes exact if the quantum matter are atoms or ions ( @xmath66 from eq . ( [ eq : in14b ] ) ) . the approximation given in eq . ( [ eq : in14b ] ) has been proven the indirect proof reads , @xmath67 , while the second formal proof is direct based on logic and the excitation probability of electrons and holes within the ionization energy based fermi - dirac statistics @xcite . from here onwards , any variable found to wear a tilde implies a renormalized parameter . subsequently , @xmath68 ( from eq . ( [ eq:3 ] ) ) can be written in the form ( after making use of the series , @xmath69 ) @xmath70\sqrt{\frac{\tilde{k}}{m } } , \label{eq:5}\end{aligned}\ ] ] in which , we consider only the first three terms by imposing the condition , @xmath71 \gg e^2/2\pi\epsilon_0r^3 $ ] , in other words @xmath72 for @xmath73 and @xmath74 where @xmath75 is the set of even natural numbers , and finally , eq . ( [ eq:5 ] ) also satisfies @xmath76 . using @xmath77 we can carry out the subsequent algebraic rearrangements of eq . ( [ eq:5 ] ) . after subtracting @xmath78 , eq . ( [ eq:5 ] ) will lead us to @xmath79 . \label{eq:6}\end{aligned}\ ] ] this is the standard vdw interaction energy in its renormalized form ( due to the renormalizing factor ( renormalizer for short ) , @xmath80 $ ] ) , and note the popular vdw factor , @xmath81 . the original ( unrenormalized ) vdw formula is given in the curly bracket . however , even in the presence of this renormalizer , eq . ( [ eq:6 ] ) is only valid for weakly interacting systems because of this condition @xmath71 \gg e^2/2\pi\epsilon_0r^3 $ ] . we will explain why this is so in the following section . the term , @xmath78 is the renormalized ground state energy ( @xmath82 ) when @xmath44 ( in the absence of coulomb interactions between charges from different atoms or molecules ) . one can observe the correct trend from eq . ( [ eq:6 ] ) , in which , for a given @xmath33 , when @xmath42 shoots to infinity ( valence electrons with zero polarization ) , @xmath83 , as it should be . in contrast , @xmath84 maximum , if @xmath42 obtains an allowable minimum value . note here that taking @xmath85 is physically not acceptable as this means that the valence electrons are not bound to their respective nucleus , and they are free . we now prove the existence of a stronger attraction between two interacting atoms . the above standard renormalized vdw attraction ( see eq . ( [ eq:6 ] ) ) is weak because @xmath86 such that the polarization is either small or , if the polarization is large , then it is approximately isotropic ( see fig . [ fig:2](a ) ) because the atoms are weakly interacting , made sure by the condition , @xmath50 } \gg e^2/2\pi\epsilon_0r^3 $ ] or @xmath87 . in the case of strongly interacting two - atom system , a larger induced anisotropic polarization is obtainable ( see fig . [ fig:2](b ) ) . here , [ fig:2](a ) depicts a weakly interacting system with an isotropically polarized atom , while fig . [ fig:2](b ) and ( c ) depict two types of strongly interacting systems with anisotropic polarization . in particular , ( a ) depicts isotropic polarization that is valid for weakly interacting atoms , ( b ) is for strongly interacting atoms , which induce an anisotropic polarization and also an electron - electron coulomb repulsion between atoms , while ( c ) contains a mixture of easily - polarizable and least - polarizable atoms , giving rise to an asymmetric polarization . here , ( a ) and ( b ) need additional assistance to initiate the strong attractions , namely , collisions between reactant atoms . in contrast , ( c ) may not need any external `` help '' , hence could be spontaneous . to induce such an anisotropic polarization , we need a stronger interaction between these atoms such that the condition , @xmath88 is not violated ( to make sure eq . ( [ eq:3 ] ) is still valid ) . in other words , we need a new condition , @xmath50 } = e^2/2\pi\epsilon_0r^3 $ ] , not @xmath50 } \gg e^2/2\pi\epsilon_0r^3 $ ] to impose a large anisotropic polarization . the new condition allows a large intermolecular interaction energy if @xmath42 is small ( see eq . ( [ eq:6 ] ) ) . the only analytic relation with the smallest @xmath42 is coincidently @xmath50 } = e^2/2\pi\epsilon_0r^3 $ ] where @xmath50 } < e^2/2\pi\epsilon_0r^3 $ ] is physically not allowed . after making use of this new condition , @xmath50 } = e^2/2\pi\epsilon_0r^3 $ ] , we can rewrite eq . ( [ eq:3 ] ) to obtain ( subject to the renormalization procedure developed in ref . @xcite ) @xmath89 - \hbar\tilde{\omega}_0 . \label{eq:8}\end{aligned}\ ] ] unlike eq . ( [ eq:6 ] ) , which was approximated using eq . ( [ eq:5 ] ) , eq . ( [ eq:8 ] ) is obtained exactly from eq . ( [ eq:3 ] ) without any approximation and therefore , eq . ( [ eq:5 ] ) is irrelevant when @xmath50 } = e^2/2\pi\epsilon_0r^3 $ ] . however , the proper question here is that whether eq . ( [ eq:5 ] ) when summed after taking @xmath50 } = e^2/2\pi\epsilon_0r^3 $ ] ( not @xmath72 for @xmath73 ) gives eq . ( [ eq:8 ] ) . to see that it does , we rewrite eq . ( [ eq:5 ] ) accordingly @xmath90 - \hbar\tilde{\omega}_0 \nonumber \\&= & \bigg[\frac{\hbar\tilde{\omega}_0}{2}\sum_{\texttt{p } = 0}^{\infty}\frac{2\cdot(2\texttt{p})!}{(1 - 2\texttt{p})(\texttt{p}!)^2(4^{\texttt{p } } ) } - \hbar\tilde{\omega}_0\bigg ] = \hbar\tilde{\omega}_0\bigg[\frac{\sqrt{2}}{2 } - 1\bigg ] , \label{eq:6a}\end{aligned}\ ] ] as it should be . here @xmath91 , @xmath92 and we have used @xmath93 obtained from ref . @xcite . equation ( [ eq:6a ] ) unequivocally proves that the original vdw formula is for atoms with small polarization such that these atoms are weakly interacting . equations ( [ eq:6 ] ) and ( [ eq:8 ] ) are true for any distance between two nuclei ( @xmath33 ) , for as long as there is no wavefunction overlapping . the difference here is that eq . ( [ eq:6 ] ) is valid only for systems with small polarization ( or large isotropic polarization ) and/or for weakly interacting atoms . whereas , eq . ( [ eq:8 ] ) is for systems with large anisotropic polarization and/or for strongly interacting atoms . this means that , by definition , eq . ( [ eq:6 ] ) is not applicable for strongly interacting systems , while eq . ( [ eq:8 ] ) is inapplicable for weakly interacting systems . for example , for weakly interacting atoms ( with small polarization or large isotropic polarization ) , we need to invoke the condition , @xmath50 } \gg e^2/2\pi\epsilon_0r^3 $ ] into eq . ( [ eq:3 ] ) to obtain eq . ( [ eq:6 ] ) . on the other hand , we should impose @xmath50 } = e^2/2\pi\epsilon_0r^3 $ ] into eq . ( [ eq:3 ] ) to obtain eq . ( [ eq:8 ] ) for strongly interacting atoms ( giving rise to large anisotropic polarization ) . another point to note here is that eq . ( [ eq:6 ] ) can be regarded as valid for large @xmath33 , while eq . ( [ eq:8 ] ) becomes valid for small @xmath33 , which will be explained in the section further analyses ii . we now move on and rearrange eq . ( [ eq:8 ] ) to arrive at @xmath94^{\frac{1}{2}}\bigg(\frac{1}{\sqrt{2 } } - 1\bigg)\frac{1}{r^{3/2 } } , \label{eq:1}\end{aligned}\ ] ] @xmath95 } , \label{eq:1.2}\end{aligned}\ ] ] by using @xmath96 interchangeably . here , we should not assume that @xmath42 or @xmath41 has become @xmath33-dependent from the condition , @xmath96 . in fact , this condition simply means that the magnitude of @xmath41 is equal to the magnitude of @xmath97 . both @xmath41 and @xmath97 vary independently . obviously , @xmath98 and @xmath99 are negative guaranteed by the inequality @xmath100 , hence these interactions are always attractive , while the proportionality , @xmath101 implies it is of a stronger - type compared to the renormalized standard vdw interaction , @xmath102 ( see fig . [ fig:1 ] ) . it is worth noting that the proportionality @xmath103 is identical to the vdw interaction energies derived for two different geometric conditions @xcite . in particular , the relevant geometric conditions are ( a ) two infinite - length cylinders with radii @xmath104 and @xmath105 , separated by @xmath33 , and ( b ) a cylinder ( radius @xmath104 and with infinite length ) separated from a two - dimensional sheet ( negligible thickness but with infinite length and width ) at a distance @xmath33 . contrary to eq . ( [ eq:1 ] ) , both ( a ) and ( b ) must satisfy the condition , @xmath106 , in which , these classical objects are very close to each other @xcite . in contrast , eqs . ( [ eq:1 ] ) and ( [ eq:1.2 ] ) are for two spherical quantum objects ( see fig . [ fig:1 ] ) that satisfy conditions , @xmath4 and @xmath50 } = e^2/2\pi\epsilon_0r^3 $ ] ( large polarization or weakly bound valence electrons ) . this large polarization will give rise to an anisotropic polarization in the presence of strong interactions between atoms . however , the imposed condition @xmath4 does not in any way , implies eq . ( [ eq:1.2 ] ) is inapplicable for any @xmath33 other than @xmath4 . for example , what eq . ( [ eq:1.2 ] ) tells us is that for a given @xmath33 such that @xmath4 , if @xmath42 is large , then one obtains a large attractive interaction . now , the only way to change the value of @xmath42 is to change the type of atoms because each atom has a unique set of ionization energies or energy level spacings . therefore , when we extrapolate this scenario for any @xmath33 such that @xmath107 , then such an extrapolation does not technically and physically change the effect of the renormalizer or @xmath42 . in other words , large @xmath42 still gives rise to a large attractive vdw interaction , regardless whether the atoms are close to each other or far apart . unfortunately , eq . ( [ eq:1 ] ) does not provide any new physical insight other than , that it is stronger than the original vdw interaction . whereas , eq . ( [ eq:1.2 ] ) is indeed a function of the microscopic variable , @xmath42 , but physically not yet appropriate because eq . ( [ eq:1.2 ] ) does not allow large anisotropic polarization effect the repulsive interaction between the anisotropically polarized electrons from two identical atoms separated by @xmath33 , will give rise to an effective repulsion , not an attractive interaction ( see fig . [ fig:2](b ) ) . as a consequence , eq . ( [ eq:1.2 ] ) certainly can not lead to a stronger vdw attraction for large anisotropic polarization , due to the @xmath39-@xmath39 repulsion between anisotropically polarized atoms . warning : eq . ( [ eq:6 ] ) is valid when @xmath50 } \gg e^2/2\pi\epsilon_0r^3 $ ] ( valid for the least polarizable or weakly interacting atoms ) , whereas eq . ( [ eq:1.2 ] ) is valid for @xmath50 } = e^2/2\pi\epsilon_0r^3 $ ] ( valid for easily polarizable or strongly interacting atoms ) . for example , the condition , @xmath50 } \gg e^2/2\pi\epsilon_0r^3 $ ] , which allows us to invoke the solution in series ( see eq . ( [ eq:5 ] ) ) imposes the least polarizability condition into the atomic system . in contrast , @xmath50 } = e^2/2\pi\epsilon_0r^3 $ ] by definition allows large polarization . we stress here that both eqs . ( [ eq:6 ] ) and ( [ eq:1.2 ] ) ignore the existence of @xmath39-@xmath39 coulomb repulsion , @xmath108 between polarized electrons . now , the easily polarizable atoms ( that obey eq . ( [ eq:1.2 ] ) ) satisfies the anisotropic polarization ( see fig . [ fig:2](b ) ) . here , the polarized electrons will also activate the repulsive @xmath39-@xmath39 coulomb interaction ( @xmath108 ) between the polarized atoms . therefore , we require @xmath109 for an overall strong attraction . in fact , the existence and importance of @xmath108 between any two highly anisotropically polarizable entities has been analytically proven and discussed in refs . however , the effect of anisotropic polarization , induced by strongly interacting atoms , which gives rise to the @xmath39-@xmath39 repulsive interaction ( @xmath108 ) has been neglected in the standard ( for both renormalized and original ) vdw formula , given in eq . ( [ eq:6 ] ) . consequently , eq . ( [ eq:1 ] ) , which allows stronger vdw attractive interaction needs to be made compatible with eq . ( [ eq:1.2 ] ) in the presence of large anisotropic polarization . in order to do so , we need to do something to counter the repulsive effect coming from @xmath108 . we know ( [ eq:1.2 ] ) allows large polarization but requires both atoms to be least polarizable ( to avoid @xmath108 ) such that the nucleus of atom 1 ( left - hand side ( l.h.s ) of fig . [ fig:1 ] ) attracts the valence electron of atom 2 ( right - hand side ( r.h.s ) of fig . [ fig:1 ] ) , and vice versa . thus , eq . ( [ eq:1.2 ] ) , which ignores @xmath108 , can not lead to a stronger vdw attraction . however , the strength of this attraction can be greatly enhanced if the polarizabilities of these two atoms are made to be strongly asymmetric ( see fig . [ fig:2](c ) ) , _ i.e. _ , these atoms are strongly interacting . meaning , one atom is anisotropically polarized in the presence of another . in particular , if one of the atoms has large @xmath42 , while the other has small @xmath42 ( @xmath110 ) then this will guarantee the existence of @xmath111 because @xmath112 is also negative by definition . in this case , the electron - ion ( e - ion ) coulomb attraction ( @xmath112 ) is between the polarized valence electron of atom 1 ( easily polarizable ) and the nucleus of atom 2 ( difficult to be polarized ) . to justify the stronger vdw attraction given in eq . ( [ eq:1 ] ) , we require asymmetric polarizabilities between two atoms with strong interactions ( see fig . [ fig:2](c ) ) . in this case , an asymmetric - polarization can be properly introduced into eq . ( [ eq:1.2 ] ) to obtain @xmath113 equation ( [ eq:1.3 ] ) is the one that correctly corresponds to eq . ( [ eq:1 ] ) in the presence of large asymmetric polarization ( see fig . [ fig:2](c ) ) . our notation , @xmath114 is justified in the section , additional notes ( see the last paragraph ) . the second term on the r.h.s of eq . ( [ eq:1.3 ] ) represents the effect of two identical and weakly interacting atoms coming face - to - face ( see fig . [ fig:1 ] ) , hence the factor @xmath115 . whereas , @xmath112 captures the coulomb attraction between a polarized electron from a polarizable atom ( induced by the non - polarizable atom ) and the nucleus of a non - polarizable atom . the screening effect is included by default whenever we deal with @xmath42 ( due to eq . ( [ eq : in14b ] ) ) to predict the polarizability @xcite ( or any physical properties ) from the ionization energy theory . in any case , the atoms depicted in fig . [ fig:1 ] are least polarizable ( weakly interacting ) , and they affect each other equally with approximately isotropic polarization . while , @xmath112 is defined as the interaction between the polarized electron of an easily - polarizable atom and the nucleus of the least polarizable nearest - neighbor atom . therefore , we do not need to include the factor @xmath115 in @xmath112 due to its definition given above , in which the induced anisotropic polarization is unidirectional , from the polarized valence electron of a strongly polarizable atom , to the nucleus of the least polarizable atom . in the absence of renormalization , the new condition reads @xmath116 , and one obtains an exact copy of eq . ( [ eq:1 ] ) for @xmath98 and @xmath117 here , eq . ( [ eq:10 ] ) physically correctly captures the same effect as eq . ( [ eq:1.2 ] ) . in particular , for a given separation @xmath33 , large @xmath7 implies small polarization , which is exactly identical to large @xmath42 that causes small polarization . hence , two least polarizable atoms ( weakly interacting ) can give rise to the standard vdw attractive interaction , for instance , when @xmath118 maximum ( from eq . ( [ eq:10 ] ) ) , or @xmath119 maximum ( from eq . ( [ eq:1.2 ] ) ) then both @xmath120 and @xmath99 increase to a maximum value due to zilch coulomb repulsion between the polarized electrons ( isotropically polarized atoms ) . as stated earlier , eq . ( [ eq:1 ] ) is independent of any microscopic variable ( that can be related to the atomic structure ) , and as such it is not useful , except it correctly predicts the inverse proportionality between @xmath121 and @xmath122 . on the other hand , the changes in @xmath7 ( from eq . ( [ eq:10 ] ) ) is arbitrary , for example , we will never know how to implement the changes in @xmath7 accordingly for different atoms in a given system . in contrast , and as proven earlier , only eqs . ( [ eq:1.2 ] ) and ( [ eq:1.3 ] ) give us the option to accurately extract the microscopic information on two competing interactions , between the vdw attraction and the @xmath39-@xmath39 coulomb repulsion ( due to anisotropically polarized atoms ) simultaneously , for strongly interacting atoms ( _ via _ @xmath42 ) . equation ( [ eq:1 ] ) is shown to vary as @xmath123 , and this stronger vdw attraction can only be understood by adding an attractive coulomb interaction ( due to asymmetric polarization between two strongly interacting atoms ) to a standard vdw attraction ( eq . ( [ eq:6 ] ) ) , or to a stronger vdw attraction ( eq . ( [ eq:1.2 ] ) ) . therefore , due to this added coulomb attraction , one can not obtain eq . ( [ eq:1 ] ) using the 1/@xmath33 expansion since each individual energy term in this expansion represents a specific interaction ( coulomb , dipole , quadrupole , and so on ) without any additional interaction ( that gives rise to anisotropic polarization ) added to 1/@xmath33 expansion @xcite . meaning , each specific interaction is determined by a specific @xmath124 and varies as 1/@xmath125 where @xmath126 , which is applicable for weakly interacting systems . recall here that a weakly interacting system means there will be no induced anisotropic polarization , even though these atoms can have large isotropic polarization due to intense laser . to understand this point , we derive eq . ( [ eq:6 ] ) using the second - order perturbation theory ( the first - order energy is zero ) @xcite . the renormalized second - order energy @xmath127 + \big[\tilde{e}^{(0)}_{s}(2 ) - \tilde{e}^{(0)}_{t}(2)\big ] } , \label{eq:11}\end{aligned}\ ] ] where ( 1 ) and ( 2 ) refer to atom 1 and 2 ( see fig . [ fig:1 ] ) , while the subscripts , @xmath128 and @xmath129 denote the ground and excited eigenvalues ( @xmath130 ) and eigenstates ( @xmath131 ) , respectively . moreover , @xmath132 after taking @xmath4 . we first note @xmath133 , \label{eq:12}\end{aligned}\ ] ] where @xmath134 is the ground - state eigenvalue , while @xmath135 is the first excited state eigenvalue for the harmonic oscillator - like two atomic system sketched in fig . [ fig:1 ] . subsequently , using eq . ( [ eq:12 ] ) and @xmath136 ^ 2 , \label{eq:13}\end{aligned}\ ] ] we can derive eq . ( [ eq:6 ] ) from eq . ( [ eq:11 ] ) following griffiths @xcite the renormalization procedure introduced here does not disturb the derivation in any way . as predicted earlier , @xmath137 , and is always negative determined entirely by the eigenvalue - denominator ( ground state eigenvalue @xmath138 excited state eigenvalue = @xmath139 ) in eq . ( [ eq:12 ] ) . the numerator in eq . ( [ eq:11 ] ) is positive - definite . interestingly , the same conclusions ( without renormalization ) can be obtained from ref . for example , the coefficient @xmath140 [ see eq . ( 4.3 ) in ref . @xcite ] represents the vdw attraction , and is always negative due to @xmath141 because @xmath142 , @xmath143 and @xmath144 . here , @xmath145 and @xmath146 denote atom 1 and 2 , respectively , @xmath147 and @xmath148 represent both the principal and the angular momentum quantum numbers such that @xmath149 , while @xmath150 are the respective eigenvalues . to understand the above inequalities , assume @xmath145 and @xmath146 are two hydrogen atoms , and therefore , @xmath151 and @xmath152 are the ground - state eigenvalues for these atomic hydrogen , respectively , while @xmath153 and @xmath154 are the respective excited - state eigenvalues for hydrogen atoms . since , @xmath15513.6 ev @xmath156 , @xmath1573.4 ev @xmath158 , then we can immediately see why the denominator [ in eq . ( 4.6 ) of ref . @xcite ] is always negative ( as predicted earlier ) for as long as the polarization is from a lower energy level ( strongly bounded electron ) to a higher energy level ( weakly bounded electron ) . add to that , @xmath159 originates from the interaction potential operator , @xmath160 [ see eqs . ( 2.13 ) and ( 4.3 ) in ref . @xcite ] . in our atomic system , we had @xmath161 [ see eq . ( [ eq:11x ] ) ] . importantly , regardless whether @xmath162 is negative or positive , @xmath140 is guaranteed to be negative due to the eigenvalue - inequalities given above , and note that @xmath163 , which is positive - definite . as anticipated , @xmath162 is a specific interaction potential ( @xmath164 ) , which was identified to give the vdw attraction without any additional induced anisotropic polarization added to @xmath162 . therefore , similar to eq . ( [ eq:6 ] ) , @xmath162 excludes the @xmath39-@xmath39 coulomb repulsion in the presence of large anisotropic polarization , both in the ground and excited states or in the mixed states . the large @xmath39-@xmath39 repulsion due to large anisotropic polarization ( see fig . [ fig:2](b ) ) has been excluded by invoking certain initial conditions used in ref . @xcite to write the potential , @xmath165 the initial conditions were @xmath166 such that most of the charges are distributed within an atomic or a molecular radius . this means that the allowable polarization is either small or increase isotropically ( see fig . [ fig:2](a ) ) . isotropically increasing polarization implies that the amount of increased polarization is essentially identical in all physical directions , namely , in @xmath12 , @xmath167 and @xmath168 axes . now , these conditions need to be imposed so that eq . ( 2.14 ) in ref . @xcite is valid . rightly so , chang has correctly invoked the above conditions even for large atomic or molecular polarization such that the amount of increased polarization is three - dimensionally isotropic . this in turn means that the systems studied there were for weakly interacting atoms or molecules . if the atoms or molecules are strongly interacting , then any polarization that increased isotropically has got to be invalid . in addition , the condition that requires the polarization to increase isotropically can not hold for @xmath169 . hence , what we have proven earlier is that isotropic polarization is inapplicable for @xmath169 ( for small internuclear distances ) because we need to take the anisotropic polarization effect into account , which has been included in our formalism [ see the first term in eq . ( [ eq:1.3 ] ) ] . moreover , one should also note that the isotropic polarization is not applicable for strongly interacting systems , even if @xmath166 . thus , eq . ( [ eq:1.3 ] ) is derived for strongly interacting systems that allow asymmetric polarization for @xmath166 ( see fig . [ fig:2](c ) ) . in contrast , isotropic polarization can only occur for systems that are weakly interacting and @xmath166 . it is physically incorrect to enforce isotropic polarization for @xmath169 or for strongly interacting systems . in particular , for strongly interacting systems with @xmath170 , there will definitely be a significant amount of induced anisotropic polarization on one atom or molecule that is caused by its neighboring atoms or molecules , which then gives rise to this strong asymmetric polarization . this is what we have addressed in this paper when we derived eq . ( [ eq:1.3 ] ) . one may correctly wonder if there is an effective @xmath39-@xmath39 coulomb repulsion between two isotropically polarized atoms or molecules . the effective repulsion will only be activated when the electrons in both atoms or molecules are polarized anisotropically in the presence of strong interaction between these atoms or molecules , and/or when @xmath169 . for example , atoms or molecules will be isotropically polarized in the gas - phase , if these atoms or molecules are weakly interacting in the presence of high temperature or intense laser . this isotropic polarization can be large . in this case , eqs . ( [ eq:2 ] ) and ( [ eq:3 ] ) are applicable where one just starts with a new set of radii for atom 1 and 2 , and again obtains eq . ( [ eq:6 ] ) . the new set of radii are larger than the original radii because the electrons are now polarized isotropically ( see fig . [ fig:2](a ) ) . in fact , one can go on and improve eq . ( [ eq:6 ] ) by incorporating other leading terms due to spin - orbit coupling and quadrupole - quadrupole interaction as carried out by chang @xcite . but this additional effects can not induce the @xmath39-@xmath39 coulomb repulsion , which is due to anisotropic polarization , which occurs exclusively when there is a strong interaction between atoms or molecules or when the average internuclear distance , @xmath169 ( that eventually will induce the strong interaction ) . now , we should know that the standard vdw theory can not handle anisotropically polarized atoms or molecules even though it has incorporated other leading terms , namely , @xmath171 ( quadrupole - quadrupole ) and @xmath172 ( magnetic interaction or due to spin - orbit coupling ) . these terms do improve the potential given in eq . ( [ eq:14 ] ) by incorporating the above stated physical mechanisms , and can be used to improve eq . ( [ eq:11x ] ) , but they ( the other leading terms ) do not cause the @xmath39-@xmath39 coulomb repulsion to exist . for example , these other leading terms are there to improve the equation for @xmath140 or eq . ( [ eq:6 ] ) by incorporating the appropriate correction terms such that each term represents a specific physical effect , namely , @xmath172 represents magnetic interaction , whereas @xmath171 is due to quadrupole - quadrupole interaction . in this case , the coulomb repulsion is not a correction term , but originates from the @xmath39-@xmath39 repulsion due to polarized electrons . this means that , the first term in eq . ( [ eq:1.3 ] ) is the leading term that represents the asymmetric polarization , and is responsible for the stronger vdw attraction in strongly interacting systems . in particular , the above @xmath39-@xmath39 coulomb repulsion , which is due to the large anisotropic polarization , caused by the strongly interacting atoms or molecules , can only be overcome by the first term in eq . ( [ eq:1.3 ] ) . as a matter of fact , ( 2.14 ) in ref . @xcite has been enforced to make sure that the standard vdw theory presented in ref . @xcite is for isotropic polarizations , which does not take large anisotropic polarization effect into account . for example , see the spherical harmonics given on page 139 in ref . @xcite and eq . ( 2.15 ) in ref . @xcite to observe the expansion for @xmath173 that is isotropic by definition , which has been enforced in ref . @xcite . for example , the standard vdw theory given in ref . @xcite is valid for weakly interacting system and for @xmath166 such that the polarizations are ( assumed to be ) approximately isotropic . on the other hand , the vdw theory with asymmetric polarization developed here is valid for strongly interacting systems , and for @xmath170 . in summary , a new van der waals ( vdw ) interaction of a stronger type is proven to exist analytically by generalizing the original vdw interaction . this generalization includes three types of polarizations between two atoms , namely , isotropic , anisotropic and asymmetric polarizations . we show that an isotropic polarization is only valid for @xmath174 ( @xmath175 ) , which is a well known result within the original vdw theory . but the other two polarizations ( anisotropic and asymmetric ) are shown to be valid for all @xmath33 . recall here that @xmath33 is the internuclear distance between two atoms , while @xmath176 is the atomic radius for two identical atoms . using the above generalization , we went on to explain why and how the original vdw attraction does not lead to an attractive interaction between two highly polarizable atoms due to large coulomb repulsion between polarized electrons . this explains why @xmath88 is a necessary condition for systems with isotropic polarizations , in accordance with the original vdw formalism . however , for @xmath177 , the two atomic system should be considered as strongly interacting , giving rise to an anisotropic or an asymmetric polarization . in these cases , there is no requirement on @xmath33 such that these polarizations are valid for any @xmath33 . here , we briefly reiterate the consequences of our main result , which are important for an overall understanding and also to avoid confusion within quantum chemistry . if one employs a full electronic hamiltonian ( within the quantum chemical methods ) for different interaction strengths , then it is indeed possible to calculate the reaction barriers , and the reaction rates accurately by means of the variational principle and approximate wavefunctions @xcite . on the other hand , the theory presented here has been used to predict ( exactly ) which atoms ( from two molecules ) can react to form a peptide bond , in accordance with the experimental observations using the principle of maximum interaction @xcite . here , we did not attempt to calculate the activation energies because our methodology , by definition , does not allow such calculations to be carried out because our theory does not have variationally adjustable parameters and does not rely on wavefunctions directly . therefore , our motivation here is to discover the precise microscopic physical origin responsible for a chemical reaction between two atoms , which is still unclear within the quantum chemical calculations due to the introduction of variationally adjustable interaction terms and wavefunctions . here , we have derived the relevant interactions ( coulomb and van der waals types ) that control a particular chemical reaction where such interactions can not be obtained by renormalizing or adjusting the coefficient , @xmath140 ( discussed earlier ) , which is also in agreement with refs . @xcite . for example , we obtained the stronger vdw interaction by summing all the terms in eq . ( [ eq:6a ] ) ( with or without renormalization ) , and by adding the electron(from atom 1)-ion(from atom 2 ) coulomb attraction . in fact , @xmath178 is a natural consequence of eq . ( [ eq:2 ] ) for @xmath4 ( see eq . ( [ eq:11x ] ) ) , which then leads to the standard vdw term ( if we only take the first three terms in eq . ( [ eq:6a ] ) ) . in this work , the stronger interaction ( see eq . ( [ eq:1.3 ] ) ) is shown to go beyond adjusting or renormalizing eq . ( [ eq:6a ] ) , which has been used to understand precisely how two reactant atoms can overcome the activation energy via collisions ( for weakly interacting atoms ) , or without it ( for strongly interacting atoms with induced anisotropic polarization ) . in fact , we also have justified that the electron(from atom 1)-electron(from atom 2 ) coulomb repulsion is at least , one of the main sources that defines the origin of activation energy . we now realized that the discovery of stronger attraction between non - chemically bonded ( or simply nonbonded ) atoms has been reported much earlier by ramachandran , sasisekharan and ramakrishnan @xcite where nonbonded atoms in amino acids ( c , n , o and h ) can approach much closer than what is allowed by the standard vdw bond . ramachandran correctly and properly reasoned that @xcite the above nonbonded atoms in amino acids and peptides can approach closer ( with shorter distance between nonbonded atoms ) leading to the formation of triplehelical structure in collagen . he further proved that such a stronger attraction due to allowed bond rotation , did play an important role in polypeptide chain s conformations , limited only by the steric - hindrance , which led him , sasisekharan and ramakrishnan to invent the ramachandran plot . in view of this fact , we now set the record straight by stating that eq . ( [ eq:1.3 ] ) also rediscovers the ramachandran bond ( in the absence of chemical reaction ) such that the standard vdw , hydrogen , non - covalent carbon and other similar bonds are special cases . using the ionization energy theory and the energy - level spacing renormalization group method , we obtained the relevant analytic functions to formally expose the limitation in the original vdw formalism . in particular , the coulomb repulsion between polarized electrons has been neglected by requiring small polarization , which has been implicitly embedded into the original formalism by requiring the interaction potential constant ( @xmath7 ) to be much greater than the coulomb interaction between atoms ( molecules ) . or , by imposing a large isotropic polarization by invoking spherical harmonics . after taking the large anisotropic polarization effect into account , we managed to obtain a new vdw formula of a stronger type , which is proportional to @xmath123 . however , the new formalism also leads to a formula that is proportional to @xmath179 $ ] where @xmath42 is the energy level spacing of a given molecule or system . here , large @xmath42 means large @xmath7 , in accordance with the original vdw theory . therefore , large anisotropic polarization does not favor large attractive interaction at all , because of the @xmath39-@xmath39 coulomb repulsion effect ( due to polarized electrons ) . to overcome this intrinsic problem , large asymmetric polarization between two interacting molecules was invoked to expose the existence of a stronger vdw interaction , in which , the above repulsion has been physically converted into a coulomb attraction _ via _ an asymmetric polarization . for example , the attraction between the polarized electrons ( of an easily anisotropically polarizable atom ) and a nearest neighbor nucleus ( of the least polarizable atom ) gives rise to this asymmetric polarization . this means that , the large vdw interaction ( due to @xmath123 ) can be justified with the large asymmetric polarization effect . therefore , there exists a stronger attractive vdw interaction between two strongly interacting atoms if they are polarized asymmetrically such that one of the atom or molecule is anisotropically polarized due to interaction with its nearest least - polarizable neighbor . the overall strong attractions that come from the stronger ion - ion vdw and electron - ion coulomb interactions determine the possibility of a chemical reaction between two or more neutral atoms , for a given set of conditions ( chemical potential , temperature , pressure and concentration ) . in particular , we have established that ( i ) a chemical reaction proceeds as a result of some stronger attractions between reactant atoms , ( ii ) the attractions are due to stronger vdw and electron - ion coulomb interactions , ( iii ) systems that contains a mixture of least and easily polarizable reactant atoms induce asymmetric polarization , which then leads to the stronger attractions , and ( iv ) these stronger attractions are the ones that eventually overcome the so - called activation energy . however , it turns out to be that this stronger attraction has been discovered and proven to exist much earlier by ramachandran , sasisekharan and ramakrishnan . rightly so , the bond caused by this generalized stronger attraction ( see eq . ( [ eq:1.3 ] ) ) is ( and should be ) called the ramachandran bond . we have shown here that this stronger attraction can then lead to chemical reaction , if the conditions stated above are met @xcite . i am grateful to madam sebastiammal innasimuthu , mr . arulsamy innasimuthu , madam amelia das anthony , mr . malcolm anandraj and mr . kingston kisshenraj for their financial support and kind hospitality between august 2011 and august 2013 . i also would like to thank the referees for directing me to ref . @xcite , and for pointing out all the missing explanation . special thanks to mr . mir massoud aghili yajadda ( csiro , lindfield ) and mr . alexander jeffrey hinde ( the university of sydney ) for providing some of the listed references . compendium of chemical terminology , 2nd ed . ( the `` gold book '' ) . compiled by a. d. mcnaught and a. wilkinson . blackwell scientific publications , oxford ( 1997 ) . xml on - line corrected version : http://goldbook.iupac.org ( 2006- ) created by m. nic , j. jirat , b. kosata ; updates compiled by a. jenkins . isbn 0 - 9678550 - 9 - 8 . doi:10.1351/goldbook . e. arunan , g. r. desiraju , r. a. klein , j. sadlej , s. scheiner , i. alkorta , d. c. clary , r. h. crabtree , j. j. dannenberg , p. hobza , h. g. kjaergaard , a. c. legon , b. mennucci , d. j. nesbitt , _ pure appl . chem . _ , 2011 , * 83 * , 1619 . e. arunan , g. r. desiraju , r. a. klein , j. sadlej , s. scheiner , i. alkorta , d. c. clary , r. h. crabtree , j. j. dannenberg , p. hobza , h. g. kjaergaard , a. c. legon , b. mennucci , d. j. nesbitt , _ pure appl . _ , 2011 , * 83 * , 1637 . a. d. arulsamy , _ pramana j. phys . _ , 2010 , * 74 * , 615 ; a. d. arulsamy , phd thesis , _ many - body hamiltonian based on ionization energy concept : a renormalized theory to study strongly correlated matter and nanostructures _ , the university of sydney , australia ( 2009 ) .

graphene appears to be the only known mono - atomic two - dimensional ( 2d ) crystal and apart from the intrinsic interest it engenders , it is becoming more and more also a focus of possible and desired advanced technological applications @xcite . it is for these reasons that in the past several years we have witnessed a veritable explosion of theoretical and experimental interest in graphene @xcite . the nobel prize for physics in 2010 only consolidated this trend . graphene differs fundamentally from other known 2d semiconductors because of its unique electronic band structure , viz . the monoatomic sheet of carbon atoms arranged in a honeycomb lattice leads to an electron band structure that displays quite unusual properties @xcite . the fermi surface is reduced to just two points in the brillouin zone and the value of the band gap is reduced to zero . the energy dispersion relation for both the conduction and the valence bands are linear at low energy , namely less than 1 ev , meaning that the charge carriers behave as relativistic particles with zero rest mass . the agent responsible for many of the interesting electronic properties of graphene sheets is the non - bravais honeycomb - lattice arrangement of carbon atoms , which leads to a gapless semiconductor with valence and conduction @xmath0-bands . states near the fermi energy of a graphene sheet are described by a massless dirac equation which has chiral band states in which the honeycomb - sublattice pseudospin is aligned either parallel or opposite to the envelope function momentum . the dirac - like wave equation leads to both unusual electron - electron interaction effects and to unusual response to external potentials . when the graphene sheet is chemically doped with either acceptor or donor impurities its carrier mobility can be drastically decreased @xcite . because of its 2d periodic structure graphene is closely related to single wall carbon nanotubes , being in fact a carbon nanotube rolled out into a single 2d sheet @xcite . the main difference between the electronic properties of single wall carbon nanotubes and graphene is that the former show circumferential periodicity and curvature that leave their imprint also in the electronic spectrum and consequently also van der waals ( vdw ) interactions @xcite . on the other hand , graphite appears to be the poor cousin of graphene though it is the stable form of carbon at ordinary temperatures and pressures . many efforts have been invested into understanding its structural and electronic details ( for an account see ref . various known modifications of graphite differ primarily in the way the mono - atomic two - dimensional graphene layers stack . their stacking sequence in terms of commonality is aba for the bernal structure , aaa for simple hexagonal graphite or abc for the rhombohedral graphite @xcite . graphene layers in graphitic systems are basically closed shell systems and thus have no covalent bonding between layers which makes them almost a perfect candidate to study long(er ) ranged non - bonding interactions . indeed , they are stacked at an equilibrium interlayer spacing of about 0.335 nm and are held together primarily by the non - bonding long range vdw interactions @xcite . therefore the interaction between graphene layers can be described as a balance between attractive vdw dispersion forces and corrugated repulsive ( pauli ) overlap forces @xcite , following in this respect closely the paradigm of nano - scale interactions @xcite . besides a few notable exceptions @xcite , until 2009 many electronic and optical properties of graphene could be explained within a single - particle picture in which electron - electron interactions are completely neglected . the discovery of the fractional quantum hall effect in graphene @xcite represents an important hallmark in this context . by now there is a large body of experimental work @xcite showing the relevance of electron - electron interactions in a number of key properties of graphene samples of sufficiently high quality . because of band chirality , the role of electron - electron interactions in graphene sheets differs in some essential ways @xcite from the role which it plays in an ordinary 2d electron gas . one important difference is that the contribution of exchange and correlation to the chemical potential is an increasing rather than a decreasing function of carrier density . this property implies that exchange and correlation increases the effectiveness of screening , in contrast to the usual case in which exchange and correlation weakens screening @xcite . this unusual property follows from the difference in sublattice pseudospin chirality between the dirac model s negative energy valence band states and its conduction band states @xcite , and in a uniform graphene system is readily accounted for by many - body perturbation theory . . the thickness of both ( left and right ) layers are equal to @xmath1 . in view of later generalizations , we have labeled the left semi - infinite vacuum layer with @xmath2 , the graphene layers with @xmath3 , the intervening vacuum layer with @xmath4 and the right vacuum layer with @xmath5 . the form of the dielectric functions of the graphene layers is given in eq . ( [ eq : dielectric - function - graphene - sheet]).,width=188 ] in this work we focus our efforts on the vdw dispersion component of the graphene stacking interaction . dispersion forces can be formulated on various levels @xcite giving mostly consistent results for their strength and separation dependence . in the context of graphene stacking interactions , the problem can be decomposed into the calculation of the dielectric response of the carbon sheets and the subsequent calculation of the vdw interactions either via the quantum - field - theory - based lifshitz approach , as advocated in this paper , by means of the electron correlation energy @xcite or the non - local vdw density functional theory @xcite . one can show straightforwardly that in fact the non - local van der waals functional approach of the density functional theory and the lifshitz formalism are in general equivalent @xcite . specifically we will calculate the vdw - casimir interaction free energy , per unit area between two graphene sheets as a function of the seperation between them , in a system composed of * two apposed undoped or doped graphene sheets , * an undoped or a doped graphene layer over a semi - infinite substrate , and * a multilayer ( infinite ) array of graphene sheets . in the latter case we will investigate the many - body non - pairwise additive effects in the effective interaction between two sheets within a multilayer array . we should note that non - pairwise additive effects are ubiquitous in the context of vdw interactions @xcite often leading to non - trivial properties of macromolecular interactions . in this case they will lead to variations in the equilibrium stacking separation as a function of the number of layers in a graphitic configuration . in the calculation of the vdw - casimir free energy we will employ the dielectric response function of a single graphene layer calculated previously @xcite . the geometry of the system composed of two parallel graphene layers with thicknesses @xmath1 , facing each other in a bilayer arrangement at a separation @xmath6 , is shown schematically in fig . [ fig : schematic - two - graphene - sheet ] . in view of later generalizations we label the left semi - infinite vacuum space as @xmath2 , graphene sheets as @xmath3 , the intervening layer as @xmath4 and the right semi - infinite vacuum space by @xmath5 . in order to calculate the vdw - casimir dispersion interaction free energy in the planar geometry we use the approach of ref . @xcite where it has been calculated _ exactly _ for a multilayer planar geometry . the thus derived general form of the interaction free energy per unit area includes retardation effects and is therefore valid for any spacing between the layers . for the system which is shown schematically in fig . [ fig : schematic - two - graphene - sheet ] , the vdw - casimir interaction free energy is obtained in the lifshitz form as @xmath7 , \label{eq : free - energy - two - sheets}\ ] ] where @xmath8 stands for the graphene - graphene interaction free energy as a function of the layer spacing @xmath6 ( normalized in such a way that it tends to zero at infinite interlayer separation ) . in the above lifshitz formula the @xmath9 summation is over the transverse wave vector and the @xmath10 summation ( where the prime indicates that the @xmath11 term has a weight of @xmath12 ) is over the imaginary matsubara frequencies @xmath13 where @xmath14 is the boltzman constant , @xmath15 is the absolute temperature , and @xmath16 is the planck constant devided by @xmath17 . all the quantities in the bracket depend on @xmath9 as well as @xmath18 . the other quantities entering the lifshitz formula are defined as @xmath19 with @xmath20 where @xmath21 quantifies the dielectric discontinuity between homogeneous dielectric layers in the system , where a layer labeled by @xmath22 is located to the left hand side of the layer labeled by @xmath23 ( for details see ref . @xcite ) . also @xmath24 for each electromagnetic field mode within the material @xmath23 is given by @xmath25 where @xmath26 is the speed of light in vacuo , @xmath27 is the magnitude of the transverse wave vector , and @xmath28 and @xmath29 are the dielectric function and the magnetic permeability of the @xmath23-th layer at imaginary frequencies , respectively . for the sake of simplicity we assume that for all layers @xmath30 and that the dielectric function for vacuum layers equals to @xmath31 for all frequencies . note that @xmath32 is standardly referred to as the vdw - london transform of the dielectric function and is defined via the kramers - kronig relations as @xcite @xmath33 it characterizes the magnitude of spontaneous electromagnetic fluctuations at frequency @xmath34 . in general @xmath35 is a real , monotonically decaying function of the imaginary argument @xmath34 ( for details see parsegian s book in ref . @xcite ) . in order to proceed one needs the vdw - london transform of the dielectric function of all layers in the system . the detailed @xmath27- and @xmath36-dependent form of the dielectric function for undoped and/or doped graphene layers are introduced in secs . [ two - undoped - graphene ] and [ two - doped - graphene ] . we employ the response function of a graphene layer from refs . @xcite which for doped graphene assumes the form @xmath37 , \label{eq : response - function - graphene - sheet}\end{aligned}\ ] ] where @xmath38 , @xmath39 is the fermi velocity in graphene layer and @xmath40 is the chemical potential , @xmath41 the fermi energy , and @xmath42 is the fermi momentum , where @xmath43 and @xmath44 is the the average electron density . to begin with , we assume that two layers are decoupled and ignore the interlayer coulomb interaction . the vdw - london dispersion transform of the dielectric function on the level of the _ random phase approximation _ ( rpa ) is then given by @xmath45 where @xmath46 is the ( transverse ) 2d fourier - bessel transform of the coulomb potential , @xmath47 , @xmath48 is the electric charge of electron , @xmath49 is the permittivity of the vacuum and @xmath50 is the average of the dielectric constant for the surrounding media which is equal to 1 for vacuum . in what follows we furthermore assume that to the lowest order the dielectric properties of the graphene layers are not affected by the variation of the separation between them . this assumption is also consistent with the lifshitz theory that presumes complete independence of the dielectric response functions of the interacting layers . for undoped graphene layer , i.e. @xmath51 , the expression for the vdw - london dispersion transform of the dielectric function simplifies substantially and assumes the form @xmath52 where @xmath53 is the electromagnetic fine - structure constant @xmath54 . it should be noted that , in general , a model going beyond the rpa is necessary in order to account for enhanced correlation effects that would be present in an undoped system @xcite . in this paper , however , we restrict ourselves to the rpa approximation and analyze its predictions in detail . , and at temperature @xmath55 . the functional dependence of the interaction free energy on @xmath6 is compared with the following scaling forms : @xmath56 and @xmath57 as the separation increases.,width=321 ] the functional dependence of the interaction free energy of the system per unit area , eq . ( [ eq : free - energy - two - sheets ] ) , is presented in fig . [ fig : energy - two - undoped - graphene - sheet ] as a function of the separation between two graphene layers . we assume that the graphene layers are immersed in vacuo and both of them have the same thickness @xmath31 as well as equal susceptibilities . note that in all cases considered in this paper the interaction free energies as defined in eq . ( [ eq : free - energy - two - sheets ] ) are negative reflecting attractive vdw - casimir force between graphene layers in vacuum . for the sake presentation , we shall plot the absolute value ( magnitude ) of the free energy in all cases . as one can discern from fig . [ fig : energy - two - undoped - graphene - sheet ] the general dependence of the vdw - casimir interaction free energy on the separation between the graphene layers has the scaling form of a power law , @xmath58 , with a weakly varying separation - dependent scaling exponent , @xmath59 . this scaling exponent can be defined standardly as @xcite @xmath60 for two undoped graphene sheets we observe that at small separations the functional dependence of the free energy on interlayer spacing yields the scaling exponent @xmath61 for smallest values of the separation . the scaling exponent then steadily increases to @xmath62 , then @xmath63 , and finally at asymptotically large separations it reverts back to @xmath62 . this variation in the scaling exponent for the separation dependence of the interaction free energy can be rationalized by invoking some well known results on the vdw interaction in multilayer geometries ( see e.g. the relevant discussions in ref . @xcite ) . for example , for two semi - infinite layers the interaction free energy should go from the non - retarded form characterized by @xmath64 for small spacings , through retarded @xmath61 form for larger spacings and then back to zero - frequency - only form that also scales with @xmath64 but with a different prefactor than the non - retarded form . for two infinitely thin sheets , on the other hand , we have the non - retarded @xmath62 form for small separation , followed by the retarded @xmath63 form for larger spacings and then reverting back to zero - frequency - only term with @xmath62 scaling , but again with a different prefactor than the non - retarded limit . furthermore , the transitions between various scaling forms and the locations of the transition regions are not universal but depend crucially on the characteristics of the dielectric spectra and can thus be quite complicated , sometimes not yielding any easily discernible regimes with a quasi - constant scaling exponent @xmath10 . reading fig . [ fig : energy - two - undoped - graphene - sheet ] with this in mind we can come up with the following interpretation of the calculated separation dependence : for small separation the interaction free energy is dominated by the @xmath62 dependence except in the narrow interval very close to vanishing separation where the dependence asymptotically levels off at @xmath65 form . this form is consistent with the non - retarded interaction between two very thin layers , see above , for small but not vanishing separations . for vanishing interlayer separations the final leveling - off of the scaling exponent is due to the fact that the system is approaching the limit of two semi - infinite layers where in principle @xmath66 , but in reality the finite thickness of the graphene sheets is way too small to observe this scaling in its pure form . all we can claim is that for vanishing interlayer spacings the scaling exponent drops below the @xmath62 value , valid for two infinitely thin layers . for larger values of the interlayer separations we then enter the retarded regime with @xmath63 scaling exponent , again valid strictly for two infinitely thin layers . the retarded regime finally gives way to the regime of asymptotically large spacings where the interaction free energy limits towards its form given by the zero frequency term in the matsubara summation and characterized by @xmath62 scaling dependence . obviously the numerical coefficient in the small separation non - retarded and asymptotically large separation regimes ( both with @xmath62 ) are necessarily different . the interaction free energy scaling with the interlayer separation is thus completely consistent with the vdw - casimir interactions between two thin dielectric layers for all , except for vanishingly small , separations where finite thickness effects of the graphene sheets leave their mark in a smaller value of the scaling exponent that should ideally approach the value valid for a regime of interaction between two semi - infinite layers . the numerical value of the interaction free energy per unit area at @xmath67 is about @xmath68 . that means that for two graphene layers with surface area of @xmath69 the magnitude of the free energy is about @xmath70 at 1 nm separation ; at @xmath71 it is about @xmath72 and at @xmath73 about @xmath74 for the same surface area . for a doped graphene layer the vdw - london dispersion transform of the dielectric function can be read off from eqs . ( [ eq : dielectric - function - graphene - sheet ] ) and ( [ eq : response - function - graphene - sheet ] ) as @xmath75 \nonumber\\ & & \hspace{-2.9 cm } + 2 \sqrt{\frac{4 \pi \rho / g}{q } } \big ( a_2 ^ 2+b_2 ^ 2 \big)^{1/4 } \cos \big [ \frac{1}{2 } \arg ( a_2 + \imath b_2 ) \big ] \nonumber\\ & & \hspace{-2.9 cm } - \frac{\xi_n } { v q } \big ( a_2 ^ 2+b_2 ^ 2 \big)^{1/4 } \sin \big [ \frac{1}{2 } \arg ( a_2 + \imath b_2 ) \big ] \bigg\ } , \label{eq : dielectric - function - doped - graphene - sheet}\end{aligned}\ ] ] with the following coefficients @xmath76 , at temperature @xmath55 . the functional dependence of the free energy on the interlayer spacing is compared with scaling forms @xmath77 and @xmath78 in various regimes of separation.,width=321 ] with this dielectric response function we again evaluate the vdw - casimir interaction free energy of the system per unit area , eq . ( [ eq : free - energy - two - sheets ] ) , as a function of the separation between two graphene layers @xmath6 as shown in fig . [ fig : energy - two - doped - graphene - sheet ] . the interaction free energy has again a scaling form with a scaling exponent varying with the separation between the layers . it is clear that in this case it is much more difficult to partition the variation of the scaling exponent into clear - cut piecewise constant regions . as it can be seen from fig . [ fig : energy - two - doped - graphene - sheet ] at small separations the form of the functional dependence of the interaction free energy has @xmath79 . then for increasing spacings there follows a relatively broad regime with @xmath80 , followed eventually by the scaling form with @xmath64 for @xmath81 . one needs to add here that only the scaling regime of @xmath64 for asymptotically large separations and an intermediate regime with @xmath80 are clearly discernible . we can gain some understanding of these regimes by comparing with the various exact limits in the layer geometry as before . such comparison is however not as straightforward as before . the asymptotic @xmath64 regime is easiest to rationalize : it has the same scaling form as the _ finite temperature _ vdw - casimir interaction between two metallic sheets at asymptotically large separations . the presence of free charges would in fact be a reasonable characterization of doped graphene layers . for smaller separations we then enter the regime dominated by the retardation effects with @xmath82 form , and finally for vanishing separations we approach the regime of @xmath83 . recent calculations of vdw interactions between thin metallic layers indeed lead to exactly this exponent for small layer separations @xcite . the doped graphene sheet results would thus indicate that the dependence of the vdw - casimir interactions free energy on the separation could be rationalized in terms of interactions between two thin metallic sheets . for comparison we have also plotted the interaction free energy between two _ ideal _ metallic sheets which exhibits a much stronger attractive interaction free energy , i.e. @xcite @xmath84 \nonumber \\ & = & - \frac{k_b t \zeta ( 3)}{8 \pi b^2 } + 2 k_b t \times \\ & & \hspace{1cm}\times \sum_{n=1}^{\infty } \int \frac{d^2 p}{(2 \pi)^2 } \ln \big [ 1- e^{-2 b \sqrt{p^2 + ( \xi_n / c)^2}}\big ] . \nonumber\end{aligned}\ ] ] the numerical value of the energy per unit area at @xmath67 is about @xmath85 which is equal to @xmath86 for surface area of @xmath69 ; at @xmath71 it is about @xmath87 and at @xmath73 it is about @xmath72 for the same surface area . ) immersed in vacuo as a function of the interlayer spacing @xmath6 , at temperature @xmath55 . as seen for all separations the magnitude of the interaction free energy for doped graphene layers is greater than that of the undoped one . , width=321 ] , for the system composed of two doped graphene layers with different electron densities immersed in vacuum as a function of @xmath88 for different interlayer separations @xmath89 and @xmath90 ( from top to bottom ) at temperature @xmath55 . note that @xmath91 has been calculated for @xmath92 . , width=321 ] it is instructive to compare the interaction between two graphene layers in the undoped and doped cases . for this purpose we have plotted the interaction free energy of the system for both cases in fig . [ fig : energy - two - undoped - doped - graphene - comparison ] . the electron density in doped graphene is assumed to be @xmath93 ( solid curve ) . the dashed curve is the interaction free energy of the system composed of two undoped graphene layers . as it can be seen , for all separations the magnitude of the interaction free energy for doped graphene layers is more than that of the undoped one ( note again that the interaction free energies as defined in eq . ( [ eq : free - energy - two - sheets ] ) are negative in the present case due to attractive vdw - casimir force between graphene layers in vacuum ) . at separation @xmath67 the vdw - casimir interactions for doped graphene is about 30 times the magnitude of the interaction for undoped graphene , while at the separation of @xmath94 this ratio is about 1600 . this means that the attractive interaction between graphene layers is enhanced when the contribution of the electron density in the dielectric function of the graphene layers is taken into account . this same trend was observed also in the work of sernelius @xcite and is clearly a consequence of the fact that the largest value of vdw - casimir interactions is obtained for ideally polarizable , i.e. metallic layers . the closer the system is to this idealized case , the larger the corresponding vdw - casimir interaction will be . let us investigate also the effect of asymmetry of doped graphene sheets on vdw - casimir interactions between them . introducing the dimensionless parameter @xmath95 , where @xmath96 is the electron density of the @xmath23-th graphene layer ( @xmath97 ) , we find that the interaction free energy depends on the asymmetry in the system . when @xmath98 , the electron densities are the same for both graphene layers and we have a symmetric case , whereas @xmath99 means that one of the graphene layers is undoped while the other one is doped , leading to an asymmetric case . in fig . [ fig : energy - two - different - doped - graphene ] we have plotted the rescaled free energy @xmath100 of the system composed of two graphene layers as a function of @xmath101 for different values of the interlayer separation . the magnitude of the electron density for one of the layers has been fixed at @xmath102 while that of the other layer , @xmath103 , varies . as seen from this figure , the curves show a monotonic dependence on @xmath101 with a stronger interaction at smaller values of @xmath101 . note that at large separations the curves tend to coincide and will become indistinguishable . the asymmetry effects are therefore largest at small separations between the interacting graphene layers . . the top solid line shows the general form of the hamaker coefficient as defined via eq . ( [ eq : hamaker - coefficient - general - form - two - graphene - layers ] ) , the dashed line shows the limiting large - distance form , eq . ( [ eq : hamaker - coefficient - large - distances - two - graphene - layers ] ) , and the dotted line shows the limiting small - distance form , eq . ( [ eq : hamaker - coefficient - small - distances - two - graphene - layers ] ) . , width=321 ] the general form of the hamaker coefficient , @xmath104 , for a system composed of two dielectric layers of finite thickness ( fig . [ fig : schematic - two - graphene - sheet ] ) when retardation effects are neglected is defined via @xcite @xmath105 . \label{eq : hamaker - coefficient - general - form - two - graphene - layers}\ ] ] at large separations @xmath106 , the hamaker coefficient , @xmath107 , can be obtained from @xmath108 while at small separations @xmath109 , it can be read off from @xmath110 in fig . [ hamaker - two - graphene - layers ] , we show the hamaker coefficient as a function of the layer separation for a system of two undoped graphene layers . we show the general form of the hamaker coefficient from eq . ( [ eq : hamaker - coefficient - general - form - two - graphene - layers ] ) ( solid line ) as well as the limiting forms at small ( dotted line , eq . ( [ eq : hamaker - coefficient - small - distances - two - graphene - layers ] ) ) and large ( dashed line , eq . ( [ eq : hamaker - coefficient - large - distances - two - graphene - layers ] ) ) separations . the large - distance limiting form obviously coincides with the general form at separations beyond @xmath111 . the small - distance limiting form tends to the general form at small separations but given that the thickness of the layers is only about @xmath31 , it is expected to merge with the general form in sub - ngstrm separations . for doped graphene , a similar analysis to eq . ( [ eq : hamaker - coefficient - general - form - two - graphene - layers ] ) is not possible because the corresponding general expression for the interaction free energy is missing . in this section we study the interaction between a graphene layer and a semi - infinite dielectric substrate as depicted schematically in fig . [ fig : schematic - substrate - graphene - sheet ] . for this system the free energy per unit area is @xmath112,\nonumber\\ \label{eq : free - energy - substrate - and - graphen - sheet}\end{aligned}\ ] ] where @xmath113 now stands for the interaction free energy between the substrate and the graphene layer . we have excluded the explicit dependence of the quantities in the bracket on the imaginary matsubara frequencies , but they are the same as in eq . ( [ eq : delta - two - sheets ] ) . ( labeled by @xmath3 ) apposed to a substrate ( labeled by @xmath2 ) at a separation @xmath6 . we have labeled the intervening layer ( assumed to be vacuum ) with @xmath4 , and the right one with @xmath5 . the dielectric function of the graphene layer is defined via eq . ( [ eq : dielectric - function - graphene - sheet ] ) and for substrate via eq . ( [ eq : epsilon - substrate]).,width=188 ] in order to gain insight into the magnitude of the vdw - casimir interaction free energy and for the sake of simplicity , we assume that the semi - infinite substrate is made of sio@xmath114 which has the vdw - london dispersion transform of the dielectric function of the form @xcite @xmath115 where the values of the parameters , c@xmath116 = 1.098 , c@xmath117 = 1.703 , @xmath118 rad / s , and @xmath119 rad / s have been determined from a fit to optical data @xcite . the static dielectric permittivity of sio@xmath114 is then obtained as @xmath120 . a characteristic feature of the vdw - london transform of sio@xmath114 is thus that it contains two relaxation mechanisms . the first one is due to electronic polarization and the second one is due to ionic polarization . all calculations of the vdw - casimir interaction free energy are done at room temperature ( @xmath55 ) . substrate and an undoped graphene layer , eq . ( [ eq : free - energy - substrate - and - graphen - sheet ] ) , plotted at temperature @xmath55 as a function of the separation @xmath6 . the functional dependence of the free energy on @xmath6 is compared with scaling forms @xmath121 and @xmath122 in various regimes of separation.,width=321 ] substrate and a doped graphene layer , eq . ( [ eq : free - energy - substrate - and - graphen - sheet ] ) is plotted at temperature @xmath55 as a function of the separation @xmath6 . here the functional dependence of the interaction free energy for the graphene doping electron density @xmath123 ( solid line ) is compared with scaling forms @xmath124 and @xmath78 in various regimes of separation . we also include the same plot for the doping electron density @xmath125 ( dashed line ) . , width=321 ] using the vdw - london transforms of the dielectric functions given in the preceding sections , one can now calculate the free energy , eq . ( [ eq : free - energy - substrate - and - graphen - sheet ] ) , for an undoped graphene layer next to a semi - infinite sio@xmath114 substrate . the results are shown in fig . [ fig : energy - substrate - undoped - graphene - slope ] . at small separations , the free energy varies with a scaling exponent @xmath64 , while at larger separations one can distinguish the scaling regimes @xmath61 , and @xmath62 , finally approaching the asymptotic limit at large separations with @xmath61 . this sequence of interaction free energy scalings can be rationalized as follows : at asymptotically large separations we are at the zero - frequency matsubara term for a semi - infinite layer and a thin sheet . this case is right in between the asymptotically large separation limit for two semi - infinite layers ( @xmath64 ) and two infinitely thin layers ( @xmath62 ) . for smaller spacings we then progressively detect contributions from higher matsubara terms which lead to scaling exponent @xmath62 that corresponds to a retarded form of the interaction free energy and then for yet smaller spacings the scaling exponent reverts to @xmath61 non - retarded form of the vdw interaction between a semi - infinite substrate and an infinitely thin sheet . for vanishing spacings the finite thickness of the sheet starts playing a role and eventually , we approach the @xmath64 scaling for two semi - infinite layers . the magnitude of the interaction free energy per unit area at @xmath67 is about @xmath126 . it means that for a system with surface area of @xmath69 this value is about @xmath127 . at @xmath71 it is about @xmath128 and at @xmath73 it is about @xmath129 for the same surface area . substrate as a function of the distance between them , @xmath6 . the top solid line shows the general form of the hamaker coefficient as defined via eq . ( [ eq : hamaker - coefficient - general - form - substrate - graphene ] ) , the dashed line show the limiting large - distance form , eq . ( [ eq : hamaker - coefficient - large - distances - substrate - graphene ] ) , and the dotted line shows the limiting small - distance form , eq . ( [ eq : hamaker - coefficient - small - distances - substrate - graphene]).,width=321 ] the vdw - casimir interaction free energy eq . ( [ eq : free - energy - substrate - and - graphen - sheet ] ) for a system composed of a doped graphene layer next to a semi - infinite sio@xmath114 substrate is shown fig . [ fig : energy - substrate - doped - graphene - slope ] for two values of the electron density @xmath130 ( dashed line ) and @xmath131 ( solid line ) . at small separations the free energy shows the @xmath83 scaling , while for larger spacings it shows a scaling exponent @xmath61 , approaching the @xmath66 limit for asymptotically large separations . in this respect the case of a thin doped graphene sheet apposed to a semi - infinite substrate is very similar to the case of two thin doped layers , except that the retarded regime covers a smaller interval of spacings . this means that the metallic nature of one of the interacting surfaces is enough to switch the behavior of the interaction free energy completely towards the case of two metallic interacting surfaces . this case has in fact not yet been thoroughly discussed in the literature . the changes in the slope appear to occur at the same values of the interlayer spacing when the electron density decreases ( compare dashed and solid curves ) . the magnitude of the free energy for the electron density @xmath132 ( dashed line ) and surface area of @xmath69 at @xmath67 is about @xmath133 , while at @xmath71 it is about @xmath134 , and at @xmath73 is about @xmath135 . these values increase for larger electron densities , e.g. , for @xmath131 ( solid line ) and the same surface area , the free energy magnitude is @xmath136 at @xmath67 , while it is about @xmath137 at @xmath71 and about @xmath138 at @xmath73 . the general form of the hamaker coefficient , @xmath139 , for a system composed of a dielectric layer of finite thickness apposed to a semi - infinite dielectric substrate ( fig . [ fig : schematic - substrate - graphene - sheet ] ) when retardation effects are neglected is defined via @xcite @xmath140 . \label{eq : hamaker - coefficient - general - form - substrate - graphene}\ ] ] at large separations @xmath106 , the hamaker coefficient , @xmath141 , can be obtained from @xmath142 while at small separations @xmath109 , it can be read off from @xmath143 in fig . [ hamaker - substrate - and - graphene ] , we again show the hamaker coefficient as a function of the layer separation for a system comprising a sio@xmath114 substrate and an undoped graphene layer . we show the general form of the hamaker coefficient from eq . ( [ eq : hamaker - coefficient - general - form - substrate - graphene ] ) ( solid line ) as well as the limiting forms at large ( dashed line , eq . ( [ eq : hamaker - coefficient - large - distances - substrate - graphene ] ) ) and small ( dotted line , eq . ( [ eq : hamaker - coefficient - small - distances - substrate - graphene ] ) ) separations . the large - distance limiting form obviously coincides with the general form at separations beyond @xmath71 but again since the thickness of the graphene layer is only about @xmath31 , the small - distance limiting form is expected to merge with the general form in sub - ngstrm separations . in this section we shall use the lifshitz formalism in order to study the many - body vdw interactions in a system composed of @xmath144 layers of graphene . the layers are separated from each other by @xmath145 layers of vacuum and are bounded at the two ends by two semi - infinite dielectric slabs as depicted in fig . [ fig : n - layer - new ] . the thickness of each graphene layer is @xmath1 while the separation between two successive layers is @xmath6 . we have labeled the left semi - infinite dielectric medium with @xmath2 , the right one with @xmath5 ( which both will be assumed to be vacuum ) , the graphene layers with @xmath3 and the vacuum layers with @xmath4 . following ref . @xcite , one can calculate the vdw - casimir part of the interaction free energy , @xmath146 , in an explicit form for any finite @xmath145 . interestingly , it turns out that for very large values of @xmath145 the vdw - casimir free energy can be written as a linear function of @xmath145 so that the interaction free energy @xmath146 becomes @xcite layers of graphene ( @xmath3 ) with equal thicknesses @xmath1 separated from each other by @xmath145 layers of vacuum ( @xmath4 ) with equal thicknesses of @xmath6 . the two semi - infinite substrates in the left and right end of the system are labeled by @xmath2 and @xmath5 and will be assumed to be vacuum . , width=321 ] @xmath147 where @xmath148 can be interpreted as an _ effective _ pair interaction between two neighboring layers in the stack and is given by @xmath149 , \label{eq : att - vdw - sheets - per - number - of - layers}\end{aligned}\ ] ] with @xmath150 defined as @xmath151 + \delta^4 \big ( e^{-2 \kappa_a a } - e^{-2 \kappa_b b } \big)^2 . \nonumber\\ \label{eq : cal - g}\end{aligned}\ ] ] here @xmath152 is @xmath153 where @xmath154 and @xmath155 are @xmath156 for simplicity we have dropped the explicit dependence on the matsubara frequency in all the above expressions . we stress that @xmath148 is an effective pair interaction between two neighboring graphene layers in a multilayer geometry and is in general _ not _ equal to @xmath157 in eq . ( [ eq : free - energy - two - sheets ] ) , which is valid for two interacting layers in the absence of any other neighboring layers . the difference between these two interaction free energies thus encodes the non - pairwise additive effects in the interaction between two layers due to the presence of other vicinal layers . , plotted as a function of the separation , @xmath6 , between two successive undoped ( bottom black dot - dashed line ) and doped ( top black solid line ) graphene layers for a system composed of infinitely many layers of undoped / doped layers as schematically depicted in fig . [ fig : n - layer - new ] . the temperature of the system is @xmath55 , the thickness of the graphene layers is fixed at @xmath31 and we have chosen the dielectric function for the undoped case from eq . ( [ eq : epsilon - undoped - graphene - sheet ] ) and for doped case ( with the doping electron density chosen as @xmath158 ) from eq . ( [ eq : dielectric - function - doped - graphene - sheet ] ) . the functional form of the free energy for the undoped case is compared with scaling forms @xmath56 and @xmath57 in various regimes of separation . for the doped case , the free energy is compared with the scaling forms @xmath159 and @xmath122.,width=321 ] let us first consider the case of @xmath144 undoped graphene layers . in this case the vdw - casimir interaction free energy per unit area and per number of layers , @xmath160 ( black dot - dashed line ) , has been plotted in fig . [ fig : energy - doped - and - undoped - n_layers - per - area - number - of - layer ] as a function of the separation between the layers , @xmath6 . the temperature of the system is chosen as @xmath55 , the thickness of the graphene layers is @xmath31 and we have used the dielectric function given by eq . ( [ eq : epsilon - undoped - graphene - sheet ] ) for each undoped graphene sheet . the value of @xmath161 at @xmath67 is about @xmath162 which is @xmath163 when the surface area is equal to @xmath69 . at @xmath71 the value of @xmath164 for the same surface area is about @xmath72 and at @xmath73 is about @xmath165 . the scaling of @xmath160 for different values of the interlayer spacing @xmath6 is shown in fig . [ fig : energy - doped - and - undoped - n_layers - per - area - number - of - layer ] . it shows the scaling exponent @xmath61 at vanishing separations while at finite yet small separations it is characterised by @xmath62 , continuously merging into a @xmath63 form and finally attaining the @xmath62 form . the rationalization of this sequence of scaling exponents is exactly the same as in the case of two isolated layers and will thus not be repeated here . one can directly compare the reduced free energy , @xmath166 , with that of a system composed of only two undoped graphene layers of the same thickness @xmath1 , @xmath167 ( i.e. , comparing the results in fig . [ fig : energy - two - undoped - graphene - sheet ] with the corresponding results in fig . [ fig : energy - doped - and - undoped - n_layers - per - area - number - of - layer ] ) . this is shown in fig . [ fig : energy - nlayer - over - twolayer - doped - and - undoped ] ( black dot - dashed line and green dots ) , where apparently the results nearly coincide . however , by inspecting the ratio between these two interaction free energies it turns out that in the multilayer system the interaction free energy per layer is slightly more attractive than in the case of a two - layer system . this difference thus stems directly from the many - body effects which in this case augment the binding interaction in a graphitic stack . , for the system composed of @xmath144 layers of undoped ( bottom black dot - dashed line ) and doped ( top black solid line ) graphene is compared with the magnitude of the interaction free energy per unit area , @xmath168 , for the system composed of only two undoped ( green dots ) and doped ( red dots ) graphene layers as a function of the separation between the layers , @xmath6 . the inset shows the ratio of these two quantities for both undoped ( black dot - dashed line ) and doped ( black solid line ) cases . , width=321 ] in fig . [ fig : energy - doped - and - undoped - n_layers - per - area - number - of - layer ] the magnitude of the vdw - casimir interaction free energy for doped graphene layers per unit area and per number of layers , @xmath160 , is plotted ( black solid line ) as a function of the separation between layers , @xmath6 . the vdw - london dispersion transform of the dielectric function for each graphene sheet is chosen as in eq . ( [ eq : dielectric - function - doped - graphene - sheet ] ) . we have fixed the density of electrons for all the graphene layers as @xmath123 . the value of @xmath161 at @xmath67 is about @xmath169 , which is about @xmath170 when the surface area is equal to @xmath69 . at @xmath71 the value of @xmath164 is about @xmath171 and at @xmath73 it is about @xmath172 . the scaling exponents of the interaction free energy dependence for different regions of interlayer spacings are illustrated in fig . [ fig : energy - doped - and - undoped - n_layers - per - area - number - of - layer ] . the scaling exponent is @xmath173 for small separations while at larger separations it tends towards the value @xmath61 . it is thus exactly the same as in the case of two isolated doped graphene sheets , see fig . [ fig : energy - two - doped - graphene - sheet ] , except that in the multilayer geometry we have not shown the same range of separations as for two isolated layers . the comparison of vdw - casimir interaction free energies in the case of two isolated doped graphene sheets with the effective interaction between two graphene sheets in a multilayer system ( i.e. , comparing @xmath174 in fig . [ fig : energy - two - doped - graphene - sheet ] with @xmath164 in fig . [ fig : energy - doped - and - undoped - n_layers - per - area - number - of - layer ] ) is made in fig . [ fig : energy - nlayer - over - twolayer - doped - and - undoped ] . as shown by the inset the interaction free energy is again slightly more attractive within a multilayer . comparing the results in the inset of fig . [ fig : energy - nlayer - over - twolayer - doped - and - undoped ] shows that in average the many - body effects are stronger in the doped multilayer case than in the undoped case . note also that a direct comparison between the undoped and doped systems ( fig . [ fig : energy - doped - and - undoped - n_layers - per - area - number - of - layer ] ) shows that for all separations the free energy magnitude for a doped multilayer ( solid line ) is more than that of the undoped one ( dot - dashed line ) and thus the interaction is more attractive in the former case . at separation @xmath67 the magnitude of the doped free energy is about 34 times larger , while at the separation @xmath71 it is about 1730 larger than that of the undoped one . in this work we have studied the vdw - casimir interaction between graphene sheets and between a graphene sheet and a substrate . we calculated the interaction free energy via the lifshitz theory of vdw interactions that takes as an input the dielectric functions , or better their vdw - london transform , of isolated layers . within this approach it would be inconsistent to take into account any separation dependent coupling between the dielectric response of the layers . this need possibly not be the case for some other approximate approaches to vdw interactions as in , e.g. , the vdw augmented density functional theory ( see the paper by langreth et al . in ref . @xcite ) . by inserting the _ random phase approximation _ dielectric function of a graphene layer into the lifshitz theory we are thus in a position to evaluate not only the pair interaction between two isolated graphene sheets , but also between a graphene sheet and a semi - infinite substrate of a different dielectric nature ( sio@xmath114 in our case ) as well as the effective interactions between two graphene sheets in an infinite stack of graphene layers . all these cases that have been analyzed and discussed above are relevant for many realistic geometries in nano - scale systems @xcite and thus deserve to be studied in detail . in the three cases studied we found the following salient features of the vdw - casimir interaction dependence on the separation between the interacting bodies : * in a system composed of two graphene layers we demonstrated that the vdw - casimir interactions in the case of undoped graphene show scaling exponents identical to those displayed in the case of interacting thin dielectric layers . in the doped case the scaling exponents are consistent with vdw - casimir interactions between two thin metallic layers . * in a system composed of a semi - infinite dielectric substrate and an undoped graphene layer the vdw - casimir interactions display scaling exponents expected for this asymmetric geometry . for a doped graphene layer the exponents revert to the previous case of two doped graphene layers . * in a multilayer system composed of many graphene sheets the vdw - casimir interaction scaling exponents are the same as in the case of two isolated layers but the interactions are stronger due to many body effects as a consequence of the presence of other layers in a stack . in order to describe the correlation effects especially at low doping or the interlayer coupling on a more systematic level , one needs to go beyond the standard random phase approximation by incorporating more sophisticated theoretical models for the dielectric response function which would be worth exploring further in the future . the main motivation for a detailed study of vdw - casimir interaction between graphene sheets in graphite - like geometries is the fact that graphitic systems belong to closed shell systems and thus display no covalent bonding , so that any bonding interaction is by necessity of a vdw - casimir type . its detailed characterization is thus particularly relevant for this quintessential nano - scale system @xcite . we would like to thank b. sernelius for providing us with a preprint of his work . r.p . acknowledges support from arrs through the program p1 - 0055 and the research project j1 - 0908 . a.n . acknowledges support from the royal society , the royal academy of engineering , and the british academy . j.s . acknowledges generous support by j. stefan institute ( ljubljana ) provided for a visit to the institute and the department of physics of iasbs ( zanjan ) for their hospitality . a. bostwick , t. ohta , t. seyller , k. horn and e. rotenberg , nature phys . * 3 * , 36 ( 2007 ) ; z. q. li , e. a. henriksen , z. jiang , z. hao , m. c. martin , p. kim , h. l. stormer and d. n. basov , nature physics * 4 * , 532 ( 2008 ) . brar et al . , , 036805 ( 2010 ) ; e.a . henriksen et al . , _ ibid _ * 104 * , 067404 ( 2010 ) ; a. luican , g. li , and e.y . andrei , , 041405(r ) ( 2011 ) ; k.f . mak , j. shan , and t.f . heinz , , 046401 ( 2011 ) ; f. ghahari et al . , _ ibid _ * 106 * , 046801 ( 2011 ) . m. polini , r. asgari , y. barlas , t. pereg - barnea , and a. h. macdonald , solid state commun . * 143 * , 58 ( 2007 ) ; m. polini , r. asgari , g. borghi , y. barlas , t. pereg - barnea , and a. h. macdonald , , 081411(r ) ( 2008 ) ; a. qauimzadeh , r. asgari , phys . rev . b * 79 * , 075414 ( 2009 ) ; new j. phys . * 11 * 095023 ( 2009 ) ; a. qauimzadeh , n. arabchi and r. asgari , solid state commun . * 147 * , 172 ( 2008 ) . v. a. parsegian , _ van der waals forces _ ( cambridge university press , cambridge , 2005 ) ; m. bordag , g. l. klimchitskaya , u. mohideen , and v. m. mostepanenko , _ advances in the casimir effect _ ( oxford university press , new york , 2009 ) ; d. c. langreth et al . , j. phys . : condens . matter * 21 * , 084203 ( 2009 ) ; t. emig , int . j. mod a * 25 * , 2177 ( 2010 ) ; j. schwinger , l. l. deraad , jr . , and k. a. milton , ann . phys . ( n.y . ) , * 115 * , 1 ( 1978 ) . b. wunsch , t. stauber , f. sols , and f. guinea , new j. phys . * 8 * , 318 ( 2006 ) ; x .- f . wang and t. chakraborty , phys . b * 75 * , 041404 ( 2007 ) ; e.h . hwang and s. das sarma , phys . b * 75 * , 205418 ( 2007 ) .

in general relativity , a black hole may be formed during the evolution from asymptotically flat initial data where none is present . consider a one - parameter family of regular asymptotically flat initial data . it is not difficult to find such families which form a black hole for some range of the parameter ( strong data ) but disperse for another range ( weak data ) . the boundary between the two regimes is the black hole threshold . in what is called type ii critical collapse , the black hole mass can be made arbitrarily small by adjusting the parameter @xmath0 of the initial data to its critical value @xmath1 . near the threshold , the final black hole mass @xmath2 then scales as @xmath3 where @xmath4 is a constant . @xmath4 depends on the family , but the transcendental number @xmath5 is universal it depends on the type of matter but not on the family of initial data . type ii critical collapse was originally discovered by choptuik in the spherically symmetric massless scalar field @xcite , but has since been found in many simple matter systems in spherical symmetry , and also in axisymmetric gravitational waves @xcite . a review of the field is @xcite . type ii critical phenomena can be described in dynamical systems terms : the phase space of the system has ( at least ) two attracting fixed points , namely black holes and dispersion . the boundary between the two basins of attraction , the critical surface , contains a critical point : it is an attractor within the boundary surface , and a repeller only in the one remaining direction . this means that it must have precisely one unstable linear perturbation , with the property that adding a bit of that perturbation with one sign leads to collapse , while adding it with the opposite sign leads to dispersion . in type ii critical collapse the critical point is either a discretely self - similar ( dss ) or a continuously self - similar ( css ) spacetime . type ii critical collapse is interesting , among other things , because the maximum value of the curvature in a subcritical evolution , and the maximal value of the curvature outside the black hole in a supercritical evolution , both diverge as @xmath6 as the fine - tuning is improved . ( this is basically the same result as the black hole mass scaling , and similar results hold for any curvature invariant . ) from the dynamical systems picture it is clear that the end point of type ii critical collapse in the limit of perfect tuning of @xmath0 to its critical value @xmath1 is not a `` zero mass black hole '' but the critical solution itself . this solution has a naked singularity . it is therefore interesting to examine the global spacetime structure of the critical solution , and in particular its cauchy horizon . here we do this for the spherically symmetric massless scalar field , where the critical solution is dss . we focus on this system because css can be viewed as a limiting case of dss , and because the critical solution in the most interesting system in which type ii critical phenomena have been found , axisymmetric pure gravity , is also dss @xcite . in section [ section : kinematics ] we discuss the global structure of choptuik s critical solution kinematically . section [ section : scalarfield ] sets out the field equations for the real massless scalar field in spherical symmetry , in coordinates adapted to our problem , and describes the mathematical structure of the solution at the cauchy horizon of the singularity . sections [ section : numerics ] , [ section : continuation ] and [ section : globalimages ] show the results of our numerical integration of the critical solution , and section [ section : conclusions ] contains our conclusions . some technical details have been removed to appendices . a spacetime is discretely self - similar ( dss ) if there is a conformal isometry @xmath7 of the spacetime such that @xmath8 the value of the dimensionless `` logarithmic scale period '' @xmath9 is a geometric property of the spacetime , independent of coordinates . it is often useful to work in coordinates adapted to the symmetry at hand . a generic self - similar and spherically symmetric metric can be written as @xmath10 where @xmath11 is the metric on the unit 2-sphere , and where @xmath12 , @xmath13 , @xmath4 and @xmath14 are functions of @xmath15 and @xmath16 . this metric is dss if and only if they are periodic in @xmath15 with period @xmath9 . it is continuously self - similar ( css ) if these functions are completely independent of @xmath15 . we assume that the signature is @xmath17 , and that the metric is non - degenerate . this leads to the inequality @xmath18 . we also assume @xmath19 for definiteness . any four - dimensional spacetime splits into a product of a two - dimensional spacetime ( the reduced manifold ) and a round two - sphere of area @xmath20 . the area radius @xmath21 is a scalar in the reduced manifold . here the coordinates on the reduced manifold are @xmath15 and @xmath16 , and the area radius is given by @xmath22 . geodesics in the reduced spacetime are radial ( constant @xmath23 and @xmath24 geodesics in the full spacetime ) . the hawking mass @xmath25 is defined by @xmath26 . it is a scalar on both the full and the reduced spacetime . from @xmath25 we define the two dimensionless scalars @xmath27 and @xmath28 . it is easy to show that a spherical surface where @xmath29 is a closed trapped surface , and one where @xmath30 is an apparent horizon . in a dss spherical spacetime , @xmath31 and @xmath32 are periodic in @xmath15 . radial null geodesics which are invariant under the symmetry ( [ dssgeometric ] ) are called self - similarity horizons ( ssh ) . they are the key to determining the causal structure . all coordinate systems of the form ( [ spherical_metric ] ) are related by coordinate transformations of the form @xmath33 where @xmath24 and @xmath34 are periodic in @xmath15 with period @xmath9 . we use this coordinate freedom to make all lines in the reduced manifold where @xmath35 into lines of constant @xmath16 . ( these can be either regular centers or central singularities . ) we also make all sshs into lines of constant @xmath16 where @xmath36 . in order to discuss the global structure of the choptuik spacetime and its possible continuations we briefly review the kinematical results of @xcite . in a spherically symmetric dss spacetime , two kinds of singularities can be distinguished . from dimensional analysis it can be seen that the kretschmann scalar scales as @xmath37 for constant @xmath16 , and therefore the set @xmath38 is a central ( because @xmath21 scales as @xmath39 ) curvature singularity . we call this the kinematical singularity . geometrically , this singularity is either a point or a null line in the reduced spacetime . there are two types of self - similarity horizons that in @xcite we have called fans and splashes . the kinematical singularity is null if there is at least one splash . additional central singularities can arise where @xmath35 for all @xmath15 . ( we call these dynamical . ) because @xmath15 takes all values up to @xmath40 they are connected to the kinematical singularity . there are at most two of them , connected to the kinematical singularity at its ends . topologically , they are lines in the reduced manifold . all known type ii critical solutions in spherical gravitational collapse can be defined by the properties of self - similarity ( css or dss ) , analyticity at the past lightcone , and the requirement that they have a single unstable perturbation mode . a generic spherically symmetric dss scalar field solution is singular either at the center or the past lightcone . imposing analyticity at both the center and the past lightcone defines a nonlinear pde boundary value - problem which admits at most discrete solutions @xcite . only one such solution has been found , and it empirically turns out to have only one unstable mode , and to agree perfectly with the critical solution found previously in collapse simulations by choptuik @xcite . the global structure of the choptuik solution up to the future light cone of the kinematical singularity ( which is a cauchy horizon ) is sketched in fig . [ figure : prech ] , together with the @xmath16 and @xmath15-lines of the three coordinate patches that we shall use in the numerical calculations . this structure is the same as for all other known type ii critical solutions in spherical symmetry . these solutions have a regular center @xmath41 in the past . as @xmath16 increases , the @xmath16 lines are at first timelike , so that @xmath42 . they become null at the fan @xmath43 where @xmath36 , @xmath44 and @xmath45 . as @xmath16 increases further , they are spacelike , so that @xmath46 spacelike . somewhere in the spacelike region @xmath13 changes sign . the @xmath16-lines become null again at the second fan @xmath47 , where @xmath36 , @xmath48 and @xmath49 . approaching the kinematical singularity @xmath38 from the range of `` angles '' @xmath50 it is a single point . @xmath51 and @xmath52 are its past and future lightcones . are shown dashed ( we have assumed @xmath53 while in reality @xmath54 ) . note that the numerical domain is @xmath55 ( shaded ) , with periodic boundary conditions . we have illustrated the three coordinate patches we use for numerical work : in the past patch between @xmath56 and @xmath51 @xmath15-lines are spacelike . in the outer patch between @xmath51 and @xmath52 @xmath15-lines are timelike . in the future patch beyond @xmath52 they are null . the three patches together cover the entire spacetime without overlapping , and the coordinates @xmath16 and @xmath15 are continuous at the interfaces.,width=264 ] we demonstrate below _ numerically _ that the curvature at the cauchy horizon @xmath47 is finite in choptuik s scalar field critical solution . furthermore all geodesics cross it in finite affine parameter . the spacetime can therefore be extended beyond , but this continuation is not unique . mathematically speaking we shall see that the solution is not analytic at the cauchy horizon in the limit coming from the past , and so there is no preferred analytic continuation to the future . ( the curvature is only @xmath57 from the past . ) the family of continuations in which the curvature is @xmath57 across the cauchy horizon and which are dss is parameterized by one free periodic function @xmath58 . physically speaking this function can be interpreted as data on the naked singularity which determine the continuation , in addition to the null data on the cauchy horizon . we now discuss the possible continuations that are allowed kinematically . and @xmath52.,height=302 ] in the simplest case , the @xmath16-lines to the future of the cauchy horizon are all timelike ( @xmath42 ) until a timelike singularity @xmath35 is reached . in our classification this is a dynamical singularity , while the kinematical singularity is a single point . this conformal structure is shown in fig . [ figure : timelikesingular ] . if @xmath59 , or equivalently @xmath60 as @xmath61 , the conformal structure is the same , but with the singularity replaced by a regular center . nolan @xcite has drawn spacetime diagrams for spherically symmetric css spacetimes in which the point singularity at the origin of the cauchy horizon is only the starting point of an ingoing central null singularity . in our classification this is an extended kinematical singularity , which requires the existence of at least one splash , that is a line where @xmath36 again . [ figure : spacelikesingular ] shows the simplest generic possibility , in which the splash is followed by a spacelike dynamical singularity , which covers part of the naked null singularity . this spacetime structure can actually be realized in spherical css scalar field solutions @xcite . the same , and more exotic structures , can also be realized in spherical css perfect fluid solutions @xcite . contrary to our expectations , we have found that none of these exotic possibilities are realized as spherical dss continuations of the choptuik solution . there is a unique dss continuation with a regular center , and all other dss continuations have a timelike singularity . in hindsight , the reason appears to be that the null data on the cauchy horizon are extremely weak . with stronger data ( not associated with the choptuik spacetime ) we find different kinds of continuations . and @xmath52 and a splash at @xmath62 . note that this extension is _ not _ realized as a dss continuation of the choptuik solution . , height=302 ] in spherical symmetry , there are four algebraically independent einstein equations , which can be taken to be @xmath63 , @xmath64 , @xmath65 and @xmath66 . the fourth of these is a combination of derivatives of the first three and can therefore be disregarded . the first three equations contain first and second derivatives of @xmath14 , but only first derivatives of @xmath12 , @xmath13 , and @xmath4 . in the following we assume that @xmath14 is a given function of @xmath16 and @xmath15 . the einstein equations then determine three independent linear combinations of the six first derivatives of @xmath12 , @xmath13 and @xmath4 . a non - degenerate basis of this three - dimensional space is @xmath67 , @xmath68 and @xmath69 . ( with @xmath70 given , this last term contains only first derivatives of @xmath12 , @xmath13 and @xmath4 . ) here we investigate massless real scalar field matter . the scalar field @xmath71 obeys the wave equation @xmath72 and the einstein equations can be written as @xmath73 the wave equation , when written as a second order pde , will in general contain both first derivatives of all four metric coefficients . however , by writing the wave equation in a geometrically defined first - order form , all metric derivatives except one can be eliminated . define two null derivative operators @xmath74 and @xmath75 with the usual convention that both point towards the future and that @xmath74 points inwards while @xmath75 points outwards . normalize them by imposing that @xmath76 and @xmath77 . we define the null derivatives of the scalar field as @xmath78 and @xmath79 . the massless wave equation in spherical symmetry can then be written as @xmath80 with the scalar @xmath81 . using the einstein equations , the curvature can be given in terms of @xmath82 and @xmath83 : @xmath84 ( other components of the ricci tensor vanish ) and @xmath85 the most general form of the scalar field compatible with dss is @xmath86 where @xmath87 and @xmath88 is a global constant . in the choptuik solution , @xmath89 empirically , so that @xmath71 is periodic in @xmath15 . this means that @xmath82 and @xmath83 have zero average in @xmath15 . moreover , @xmath82 and @xmath83 in the choptuik solution obey @xmath90 and so for @xmath83 . ( this of course implies zero average . ) as a consequence @xmath91 and so for other suitable metric fields . we shall assume these extra symmetries in our numerical work , but all our analytical expressions remain valid if these symmetries are dropped . we now describe three coordinate patches that cover the critical solution . we demand that both the past and the future lightcones of the singularity occur at lines of constant @xmath16 . this makes it easier to impose regularity at the center and past lightcone , and to investigate the behavior at the future lightcone ( cauchy horizon ) . it also allows us to match the coordinate patches without overlap . subject to these requirements , we have tried to make the field equations in each patch as simple as possible . based on the three patches , it is straightforward to construct a single smooth coordinate system covering the whole spacetime ( see section [ section : globalimages ] ) , but using it from the beginning would unnecessarily complicate our numerical work . we summarize a number of coordinate systems for spherically symmetric css and dss spacetimes , and their advantages and disadvantages , in the appendix . on the past patch , which extends from the regular center to the past lightcone , we write the metric in terms of two free functions @xmath92 and @xmath93 as @xmath94 for @xmath95 . here @xmath96 is the scalar we defined above . these coordinates can be derived from the schwarzschild - type metric @xmath97 through the coordinate change @xmath98 and @xmath99 , and defining @xmath100 . the remaining gauge freedom of time relabelling is fixed by imposing that the past lightcone is on a constant @xmath43 line . note that we do not follow the convention of @xcite that @xmath101 at the center , which in @xcite forced us to introduce an additional free function of time in the definition of @xmath16 . the einstein equations are @xmath102a^{-2}-1 , \end{aligned}\ ] ] and the matter equations are @xmath103-{xu_{,\tau}}\over x(f+x ) } , \\ v_{,x}&=&{f[(1-a^2)v+u]+{xv_{,\tau}}\over x(f - x)},\end{aligned}\ ] ] at the regular center @xmath104 we impose elementary flatness , that is the absence of a conical singularity . in order to do this , we define @xmath105 and @xmath106 and impose that both are regular even functions of @xmath16 at @xmath104 . at the past lightcone we have @xmath107 , which by our gauge choice means that @xmath108 there . we also impose the physical regularity condition @xmath109 on the past lightcone . the conditions of dss , regularity at the center and regularity at the past lightcone select a solution . the equations on the past patch are form - invariant under the linear coordinate transformation @xmath110 , @xmath111 , and @xmath32 , @xmath82 and @xmath83 unchanged . in the numerical results presented here we have set @xmath112 on the past patch . note that the regularity condition ( [ past_patch_regularity ] ) is coordinate - independent , as @xmath82 , @xmath83 and @xmath32 and @xmath21 are all scalars and @xmath113 on the lightcone . on the outer patch , which extends from the past to the future lightcone , we write the metric in terms of @xmath93 , @xmath114 , and the auxiliary function @xmath115 as @xmath116 where @xmath117 is a function of @xmath15 only . @xmath118 is the scalar defined above . we fix the remaining gauge freedom by imposing that the past lightcone @xmath119 occurs at @xmath43 , and the future lightcone @xmath120 at @xmath47 . putting both lightcones on a line of constant @xmath16 requires @xmath115 to be non - constant . there is no outer coordinate patch that does not have at least one function like @xmath115 . as @xmath113 everywhere on the outer patch , @xmath15 is continuous between the past and outer patches . the metric equations are @xmath121\ , a^{-2}-1,\end{aligned}\ ] ] and the matter equations are @xmath122 note that the @xmath123 constraint equation is again linear in @xmath124 . the equations on the outer patch are form - invariant under the linear coordinate transformation @xmath125 , @xmath126 , and @xmath32 , @xmath127 , @xmath82 and @xmath83 unchanged . in the numerical results presented here we have set @xmath128 and @xmath129 on the outer patch . we shall see that at the cauchy horizon the solution is mildly singular . naive finite differencing breaks down there . instead we expand the generic solution around the cauchy horizon in terms of two free periodic functions @xmath130 and @xmath131 , and match this expansion to the numerical evolution at a small finite distance to the past of the cauchy horizon . before describing the full expansion , we focus on the origin of the singular behavior . equation ( [ outer_u ] ) becomes singular at the future lightcone because the denominator of the right hand side vanishes there . in contrast to the past lightcone , we do not have any freedom left to enforce the vanishing of the numerator as well . therefore we expect the solution to have some kind of singularity at @xmath52 . we rewrite ( [ outer_u ] ) @xmath132 where we have defined the metric function @xmath133 @xmath134 is positive on the outer patch and vanishes on the future light cone . the characteristics @xmath135 of eq . ( [ u_singular_eq ] ) are given by @xmath136 recall that we impose the gauge condition that the ch is at @xmath47 , or @xmath137 . we define the shorthand @xmath138 . we now make one fundamental assumption , namely that the spacetime admits regular null data on the cauchy horizon @xmath139 . this assumption is clearly necessary if we want to continue the spacetime through the ch , but here we make it simply because we have not been able to find a more general ansatz . we shall see later that it is sufficiently general to be matched to the critical solution that we have obtained numerically on the past patch . we therefore assume that @xmath83 is continuous , or @xmath140 by substituting this into eq . ( [ outer_a ] ) in the limit @xmath120 we find that @xmath32 is also continuous and @xmath141 obeys @xmath142 because we impose periodic boundary conditions in @xmath15 this ode has a unique solution . the physical significance of this is that the null data @xmath130 determine the geometry of the ch . similarly , from ( [ u_singular_eq ] ) in the limit @xmath143 we find that @xmath82 must be continuous , and @xmath144 is the unique periodic solution of @xmath145 this condition follows from the assumption of dss . finally , from ( [ ddef ] ) , ( [ outer_first ] ) and ( [ outer_second ] ) we find that @xmath134 is once differentiable , so that @xmath146 , and @xmath147 is given by @xmath148 at this point we introduce more shorthand notation . if @xmath149 is any periodic function ( with period @xmath9 ) , let @xmath150 be its average value , and let @xmath151 be its oscillatory part . let @xmath152 be the definite integral @xmath153 where @xmath154 is chosen so that @xmath152 has vanishing average . we can now integrate ( [ ucharacteristics ] ) for the @xmath82-characteristics to leading order and obtain @xmath155 we see that on a characteristic @xmath156 as @xmath157 . because @xmath158 is periodic in @xmath15 with period @xmath9 , an infinite number of oscillations in @xmath159 at constant @xmath15 pile up at the cauchy horizon @xmath139 . we can solve ( [ u_singular_eq ] ) to leading order by the method of characteristics . the general solution is @xmath160 where @xmath161 is an arbitrary periodic function with period @xmath9 and @xmath162 rewriting ( [ a0eqn ] ) as @xmath163 we see that @xmath164 , and so we can express @xmath165 and @xmath166 as @xmath167 our initial assumption that @xmath82 and @xmath83 are continuous at the light cone is therefore justified either if @xmath168 , so that @xmath169 , or if @xmath170 . we shall show numerically that @xmath165 is small but positive on the ch . in this case @xmath82 and @xmath83 are just @xmath57 , and the scalar field is therefore @xmath171 . in spherical symmetry the riemann tensor is determined completely by the ricci tensor , which in turn is quadratic in the partial derivatives of @xmath71 , see ( [ ricci ] ) . the curvature is therefore quadratic in @xmath82 and @xmath83 and so is @xmath57 . a similar analysis , with @xmath82 and @xmath83 interchanged , applies to the past lightcone . in the notation we have introduced here we can describe the past lightcone by saying that @xmath172 there ( because the null data @xmath173 on the past lightcone are large , @xmath174 ) but that the free coefficient @xmath175 vanishes identically ( because we have imposed analyticity as a boundary condition ) . we can not apply fuchsian techniques to our system of equations because they require the _ simultaneous _ vanishing on @xmath139 of the coefficients of @xmath176 and @xmath177 in equation ( [ u_singular_eq ] ) . however , the form ( [ uepsilon ] ) of the leading terms in @xmath82 suggests that the full non - linear solution can be written as a regular part , containing only integer powers of @xmath159 , plus a singular part which contains powers of @xmath178 . we can in fact construct a formal solution near the cauchy horizon in the form of an asymptotic double series : @xmath179 where @xmath180 stands for @xmath82 , @xmath83 , @xmath32 and @xmath127 , and @xmath181 here @xmath165 and @xmath182 are the quantities defined above in terms of @xmath130 . the expansion depends on the two free periodic functions @xmath183 and @xmath161 . by function counting we can therefore match any dss solution to this expansion . the first non - integer term appears in each variable at the following orders : @xmath184 in the previous section we obtained the coefficients of expansions ( [ expansion0][expansion3 ] ) up to @xmath185 . stopping there , the first order we neglect is @xmath186 . this truncation already depends on both free functions @xmath130 and @xmath131 and shows the singular behavior . it is also a sensible truncation numerically because @xmath165 turns out to be very small in the choptuik solution . going further , for the same reason there would be no point in including @xmath186 terms without also including all @xmath187 terms . it turns out that we need to go to @xmath188 . we have used the expansion to that order to check convergence . the expressions are given in appendix [ singular_expansion ] . our analysis of the possible continuations of the critical solution in section [ section : kinematics ] has shown that we can cover the entire future of the cauchy horizon in a single patch if we make @xmath15 an ingoing null coordinate . this means setting @xmath189 and @xmath49 . in order to put the center @xmath190 at a known coordinate location , we also set @xmath191 . we choose @xmath192 here so that @xmath16 increases as we extend the spacetime away from the cauchy horizon . we parameterize this metric in terms of the scalar @xmath32 and a coefficient @xmath180 ( not the same as @xmath180 in the past patch ) : @xmath193 regularity of the metric requires @xmath118 and @xmath194 . the field equations are @xmath195a^{-2}-1 , \\ u_{,x}&=&{f[(1-a^2)u+v]-xu_{,\tau}\over x(f+x ) } , \\ v_{,x}&=&{u+(1-a^2)v\over x}. \label{futureeq5}\end{aligned}\ ] ] at the cauchy horizon @xmath196 we impose the coordinate condition @xmath197 . the equations are invariant under @xmath110 , @xmath111 . we use this to set @xmath198 on this patch . as a consequence @xmath113 on the lightcone , and so @xmath15 is continuous between the outer and future patches . we find an asymptotic expansion around the cauchy horizon in terms of two free functions @xmath130 and @xmath131 . @xmath130 is given by the null data on the cauchy horizon , and so is the same as on the outer patch . @xmath165 is therefore the same on both sides . @xmath173 obeys the same ode , ( [ u0eqn ] ) , as on the outer patch , and so is the same function . there is no need , however , to make @xmath131 the same on both sides , as we would not gain any differentiability by doing so . instead we consider @xmath131 as free `` data on the naked singularity '' , and we shall find experimentally how the global structure of the spacetime is influenced by this choice . we have carried out a brand new numerical computation of the choptuik spacetime , improving the precision of our previous calculations @xcite by roughly four orders of magnitude . this was mainly needed to assert without doubt that @xmath165 is different from zero , even though it is extremely small . essentially , our new scheme uses shooting methods on the @xmath16 axis , instead of relaxation methods . this allows us to improve the treatment of the regular singular points of the equations ( the center and the lightcones ) by using taylor expansions . we still work with pseudo - spectral fourier techniques in @xmath15 because the solution is periodic . we shall see however that the particular structure of the fourier transform of the choptuik spacetime poses an unexpected problem when combining taylor and fourier expansions . in this section we explain the numerical scheme and present the results for the first two patches , with special emphasis on convergence properties and error analysis . for definiteness , let us suppose we work on the past patch . following @xcite we discretize our @xmath9-periodic fields @xmath199 , where @xmath200 stands for any of the set @xmath201 , using @xmath202 equidistant points in one period : @xmath203 for @xmath204 . in this way we transform our 1 + 1 pde problem for @xmath199 into an ode problem for the modes @xmath205 . the essential idea of pseudo - spectral methods is to carry out algebraic operations pointwise on the @xmath206 and @xmath15-differentiation / integration on the @xmath207 , switching from one to the other with a fast fourier transform algorithm . the main drawback of the method is the aliasing problem : pointwise products of fields ( nonlinearities ) generate high frequency modes which can not be sampled with only @xmath202 points . we partially solve that problem by doubling the @xmath207 ( padding with zeros ) before going to the @xmath206 , carrying out the necessary algebraic operations on the doubled @xmath206 and going back to the @xmath207 , then halving the @xmath207 and thus throwing away high frequency noise . we have tried other possibilities , such as padding with @xmath208 or @xmath209 instead of @xmath202 zero fourier components , or extrapolating the fourier coefficients using the observed fact that high frequency modes have a simple exponential dependence on frequency ( see below ) , but the results are not improved . aliasing can only be reduced by going to higher @xmath202 . from a numerical point of view , we are only safe from aliasing when the amplitude of the modes we are cutting off is below machine precision . because all our fields are real @xmath210 . furthermore , the metric fields @xmath32 and @xmath180 are even [ in the sense @xmath211 and therefore their fourier transform only contains even @xmath212 modes , while the matter fields @xmath82 and @xmath83 are odd [ in the sense @xmath90 ] and therefore their fourier transform only contains odd @xmath212 modes . taking this symmetry into account an even or odd field @xmath200 sampled with @xmath202 points per period is encoded by @xmath214 independent non - zero complex modes . as we have 4 independent variables @xmath215 , the ode system we solve comprises @xmath202 complex or @xmath216 real variables . in our calculations we have used @xmath217 and @xmath218 . previous investigations used @xmath219 . see appendix c of ref . @xcite for a complete discussion of our fourier pseudo - spectral method . we can not cross the lightcones during the integration of the ode in the @xmath16-axis because they are regular singular points of the equations . therefore we perform consecutive shooting calculations on the past , outer and future patches , in this order because we need information from the first to shoot the second and from the second to shoot the third . the issue of error propagation becomes very important . again , we describe the past patch for definiteness . given the equations ( [ pastmetriceqs][pastmattereqs ] ) and free data @xmath220 , @xmath221 at the center , we calculate the solution at @xmath222 slightly larger than @xmath56 using a second - order power expansion [ leaving errors of order @xmath223 . from these data we integrate the ode system forward in @xmath16 using finite differencing , up to @xmath224 . in the same way , given free data @xmath225 and the gauge condition @xmath226 , we calculate the solution at @xmath227 slightly smaller than @xmath51 [ this time with errors of order @xmath228 and integrate backward in @xmath16 up to the same @xmath229 . finally we use newton s method to look for the free data which brings the mismatch between both integrated solutions at @xmath229 down to a minimum , typically of order @xmath230 . ( this is the machine precision of @xmath231 reduced by a factor @xmath232 due to the calculations at each step and a factor @xmath233 from the ode - integration along @xmath234 points . ) we use a grid that becomes logarithmic near the regular singular points , with maximum stepsize @xmath235 at some intermediate position . that grid is constructed using the transformation @xmath236 from a grid of equidistant points in @xmath237 between the values @xmath238 and @xmath239 corresponding to @xmath222 and @xmath240 respectively . near the two endpoints we have @xmath241 and @xmath242 . we integrate on a fixed grid in @xmath16 rather than using a variable stepsize method in order to check for convergence with @xmath235 . this gives us a good estimate of the underlying discretization error . concerning the ode - integrator , we have tried several runge - kutta methods , both explicit and implicit ( gauss - legendre ) , with different convergence orders @xcite . in general , implicit methods are better suited to our problem than explicit methods , particularly for high @xmath202 , because high frequencies make the problem stiffer . we choose an irk2 method ( implicit , 2 stages , order 4 ) , which is a compromise between the accuracy of a high order method and the clarity of convergence of a low order method . our implicit schemes are implemented by iteration until @xmath243-differences between successive iterations converge below @xmath244 . we can not get closer to actual machine precision ( @xmath231 ) due to the accumulation of roundoff error . an implicit step typically takes 5 to 15 iterations to converge . the natural choice for the coordinate of the center is @xmath245 , but there is no preferred value for the past lightcone coordinate and we choose @xmath112 . with the set of parameters @xmath246 our newton s method converges to the free data plotted in fig . [ freedata ] ( see also table [ fouriertable ] ) , with a value @xmath247 which improves the precision of our previous result @xmath248 more than four orders of magnitude . the metric and matter fields integrated from those free data are given in fig . [ 3dinnerplots ] . best free data on the singular points : @xmath249 ( continuous line ) , @xmath250 ( dotted line ) and @xmath251 ( dash - dot line ) . ] [ cols="^,<,<,<,<,<,<",options="header " , ] finally , it is interesting to see that going to @xmath188 is not essential for obtaining an accurate result . table [ eps_conv ] compares the results for @xmath165 using a zeroth - order expansion ( that is , we only include the terms of orders @xmath252 and @xmath178 ) with those from the first order expansion ( which also includes orders @xmath159 and @xmath253 ) . it is clear that they both converge to the same number , even though with very different rates of convergence . we can not continue the solution as flat empty minkowski spacetime after the cauchy horizon because we have a small amount of ingoing radiation there , given by @xmath254 . the simplest continuation to look for is one with a regular timelike center , so that the conformal diagram is the same as for minkowski spacetime . in this case both @xmath82 and @xmath83 are small on the entire future patch , and we can obtain an approximate solution in perturbation theory around minkowski space , using the magnitude @xmath255 of @xmath256 as the small parameter . to leading order in @xmath165 we obtain the dalembert solution on flat spacetime : @xmath257 + o(\epsilon^{3/2 } ) , \label{upert } \\ v & = & \epsilon^{1/2 } \left[-g'(\tau ) + ( 1+x)\frac{f(\hat\tau)+g(\tau)}{x } \right ] + o(\epsilon^{3/2 } ) , \label{vpert}\end{aligned}\ ] ] with @xmath258 . in order to match the null data on the cauchy horizon @xmath259 , we need @xmath260 . recall that @xmath254 is given numerically by ( [ v_f_numerical ] ) . if we want to have a regular solution at the center we need @xmath261 . the solution is then completely determined , and it is clear that a nonlinear solution exists of which this is the leading order , and which can be found numerically . because the null data @xmath256 on the cauchy horizon are dss , it appears highly unlikely that there is another continuation with a regular timelike center that is not dss . we have obtained the ( probably ) unique regular continuation by shooting from expansions around the cauchy horizon and a regular center . the free data for the shooting algorithm ( given by @xmath262 at the ch and @xmath180 , @xmath176 at the center ) were obtained from the flat spacetime approximation . in this case we use an irk1 integrator and @xmath263 with very good convergence , that we do not show again . the fields @xmath32 , @xmath180 , @xmath82 and @xmath83 are shown in fig . [ choptuon_future ] . we now consider the other possible continuations , in particular those of figs . [ figure : timelikesingular ] and [ figure : spacelikesingular ] . before performing a numerical search of the possibilities in the @xmath131 space , we study the equations of motion in the future patch . we proceed in four steps : [ [ a - necessary - condition - for - another - ssh - is - a22 . ] ] a necessary condition for another ssh is @xmath264 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in order to obtain fig . [ figure : spacelikesingular ] , or any even more exotic continuation , we must have a self - similarity horizon before the central singularity occurs at @xmath265 . a self - similarity horizon is a dss line @xmath266 ( periodic ) where @xmath36 . the only factor in @xmath267 that can vanish ( while the metric is regular ) is @xmath268 . we have @xmath269 and so @xmath270 at the cauchy horizon @xmath197 and @xmath271 because @xmath272 is small , and so @xmath273 and @xmath42 at least to the immediate future of the cauchy horizon . if there are more self - similarity horizons to the future of the cauchy horizon , we must have @xmath197 again there , and therefore @xmath274 in some intermediate region . @xmath275 must increase from @xmath276 to @xmath264 in order to achieve this . [ [ once - a21-for - any - tau - a2to-0-for - that - tau . ] ] once @xmath277 for any @xmath15 , @xmath278 for that @xmath15 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + @xmath32 is given by the constraint ( [ a_constraint ] ) which reduces at the cauchy horizon to @xmath279 and this means that @xmath280 there . now @xmath32 obeys the evolution equation @xmath281 recall that in the future patch @xmath192 and increases towards the future . therefore @xmath282 is a repeller in the absence of matter , but any outgoing radiation @xmath82 drives @xmath32 to smaller values . hence , once @xmath32 becomes smaller than 1 for some value of @xmath15 , it will afterwards decrease to 0 for that @xmath15 whatever happens . because @xmath283 , this means that we are in a negative mass regime and remain there until we reach the central singularity @xmath104 , which is therefore timelike and has negative infinite mass at least for this @xmath15 . the singularity can not be reached with a finite value of @xmath284 . [ [ is - a21-at - the - singularity - possible - at - least - for - some - values - of - tau ] ] is @xmath285 at the singularity possible at least for some values of @xmath15 ? + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + if we choose @xmath131 much larger than @xmath286 , the @xmath287 term will drive @xmath32 to @xmath288 almost immediately . however , we know that the regular solution does not have negative mass and therefore it seems plausible that functions @xmath289 close to the regular case could generate a singularity with positive mass . we shall now argue that this is not true . @xmath32 can only increase from its ch value if @xmath290 is very small . in order to explore this regime we return to the approximation of perturbing around minkowski spacetime , but now dropping the assumption that @xmath261 . the result can be summarized as : @xmath291 ^ 2 \\ & + & \epsilon \ ; o(x^{-1 } ) + o(\epsilon^{3/2 } ) . \nonumber\end{aligned}\ ] ] the function @xmath292 is positive and always smaller than 2 , and is independent of @xmath14 . all singular terms vanish for @xmath293 . for every other @xmath14 , the function @xmath32 becomes smaller than 1 , which corresponds to negative mass , as the center @xmath104 is approached . this is due to the divergent terms @xmath294 in ( [ upert ] ) and ( [ vpert ] ) . the crucial point is that @xmath275 is integrated from @xmath295 and therefore the term in the perturbation expansion that makes the mass nonzero has coefficient @xmath296 , and so the mass can not be positive . the negative mass regime is reached while perturbation theory still applies , and we have seen that afterwards the mass must decrease indefinitely . with these arguments we have ruled out the possibility of having positive mass at the singularity for very large @xmath131 and for very small ( in particular , close to the regular case ) @xmath131 . however we could be missing some intermediate regime where perturbation theory does not apply . for these intermediate cases we observe that the minimum of @xmath32 is far from 1 ( and therefore we can not apply perturbation theory ) , but its maximum is positive and close to 1 , and only becomes negative in the vicinity of the singularity . there could be solutions where @xmath32 is positive at the singularity for some values of @xmath15 . numerically , however , we do not find that . we have performed a large number of numerical evolutions in this intermediate regime starting from @xmath297 , for @xmath298 and @xmath299 . even though it is difficult to evolve the system near the singularity , we always observe a final decay of @xmath32 to 0 . see fig . [ negativemass ] . therefore we believe that , starting from small null data @xmath256 at the ch it is not possible to form another ssh . therefore , either we have a regular center or a timelike singularity , and this singularity always has negative mass . in the approach to the center with @xmath300 we always have @xmath301 with both @xmath302 finite . it should be stressed that this scenario is a consequence of the small amplitude of the null data @xmath256 on the cauchy horizon in the choptuik solution . we have performed other evolutions starting from large null data on the cauchy horizon ( not @xmath256 of the choptuik solution ) where @xmath275 increases beyond 2 . maximum and minimum of @xmath32 for different initial data functions @xmath131 from numerical evolutions of the nonlinear equations . note that the @xmath16 axis becomes logarithmic when approaching the boundaries , while the @xmath32 axis is linear between the dotted lines at @xmath303 , @xmath304 , and logarithmic elsewhere . the initial @xmath131 have been obtained multiplying the regular data ( black ) by different constants : 0 ( red ) , -1 ( blue ) and 5 ( green ) . clearly , the larger @xmath131 is , the sooner @xmath32 decays . ] [ [ how - is - the - singularity - approached ] ] how is the singularity approached ? + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we find that in all numerical continuations @xmath305 , @xmath306 , @xmath82 and @xmath83 tend to finite values and @xmath307 as the singularity @xmath104 is approached . if we assume that @xmath305 and @xmath308 as @xmath309 then the @xmath310 and @xmath311 equations to leading order give @xmath312 making the ansatz that @xmath313 we find from the @xmath176 and @xmath314 equations to the leading order , @xmath315 that @xmath316 ( the expansion also holds in the special case @xmath101 . ) the next order , @xmath317 , gives @xmath318 the metric in the future patch contains a residual gauge freedom worth one periodic function of @xmath15 . near the singularity @xmath104 , we can fix this gauge freedom by setting @xmath319 to an arbitrary value . ( in our numerical evolutions starting at the cauchy horizon , the gauge has of course been fixed already by setting @xmath320 at the cauchy horizon . ) @xmath144 and @xmath321 are then physical free data which determine @xmath141 and all other coefficients of the expansion . ( [ constraint ] ) can be solved for @xmath322 only if the right - hand side has vanishing average , which means that @xmath323 can not be set completely freely . ) this expansion is therefore generic in the sense of depending on two free functions after the gauge has been fixed . the behavior just described is what we observe numerically for large values of the free data @xmath324 at the cauchy horizon . for small values of @xmath324 we find @xmath325 , that is @xmath326 . note that by our ansatz of exact dss with @xmath89 , @xmath82 must be an odd function of @xmath15 with zero average . we observe in fact that @xmath158 goes to a fundamental frequency square wave of amplitude @xmath327 , that is , in half of each @xmath15-period @xmath328 , and @xmath329 in the other half . this is shown in fig . [ choptuon_singular ] . as the center @xmath104 is approached with @xmath330 , we observe empirically that the @xmath15-derivatives become dynamically negligible . this means that different values of @xmath15 effectively decouple , that at each point @xmath15 the evolution equations become an ode system in @xmath16 , while the constraint becomes algebraic , and that the spacetime becomes locally css . it also means that the evolution equations become `` velocity - dominated '' in the sense that all derivatives transversal to the singularity ( here in spherical symmetry , this is only the @xmath15-derivative ) become dynamically irrelevant compared to the one derivative running into the singularity ( here , the @xmath16-derivative ) . it is known that generic spacelike singularities in general relativity with massless scalar field matter are velocity - dominated @xcite . here we find this to be the case only in the limit of small data @xmath324 . as this class of continuations seems to be locally css near the singularity , it is interesting to study the exactly css solutions from the point of view of a dss ansatz . starting from a generic dss scalar field ( [ generalphi ] ) we introduce a new ( gauge - dependent ) variable @xmath331 which coincides with @xmath332 at the future lightcone @xmath333 , and obeys the equation @xmath334 in exactly css solutions of the system the metric functions and @xmath302 are independent of @xmath15 , and hence @xmath335 is constant with value @xmath336 . the css solution with @xmath89 was found in closed form by roberts @xcite and is described in appendix [ appendix : roberts ] . the css solutions with @xmath337 were studied numerically by brady @xcite . we find that the small @xmath324 continuations locally approach a roberts solution ( with @xmath89 ) with a value of the parameter @xmath0 of the roberts solution that depends on @xmath15 , and not , as one might have expected , one of brady s solutions . this is discussed in appendix [ appendix : roberts ] . figures [ 3dinnerplots ] , [ choptuon_outer ] , [ choptuon_future ] and [ choptuon_singular ] show in a very detailed way the structure of the choptuik spacetime . however it is difficult to get from them an idea of what it looks like globally . in this final section before the conclusions we present a number of additional figures that will fill this gap . we do it in two very different ways . as shown in fig . [ figure : prech ] , our three @xmath338 patches match continuously , but we do not expect the resulting global coordinate system to be differentiable at the interfaces between the patches . the critical spacetime itself , however , is differentiable ( analytic at the past lightcone and @xmath171 at the ch ) , and it must be possible to construct global coordinate systems which are at least @xmath339 . one simple possibility is synchronous slicing plus area gauge : @xmath340 we add the gauge condition @xmath341 on the past lightcone of the singularity . the coordinate transformation @xmath342 can be easily integrated and it is shown on the left panel of fig . [ coordinates ] for a single @xmath9-period in @xmath15 . another simple possibility is double null coordinates : @xmath343 with gauge condition @xmath344 at the center , where @xmath345 is the time coordinate constructed in the previous paragraph . the coordinate change @xmath346 is shown on the right panel of fig . [ coordinates ] for a single @xmath9-period in @xmath15 . the fields @xmath32 , @xmath82 and @xmath83 in double null coordinates are shown in fig . [ fieldsuv ] . as these fields are spacetime scalars , they should be analytic in these coordinates at the past lightcone , and @xmath171 at the cauchy horizon . in the plot it looks as if they are only @xmath57 at the cauchy horizon , but that is due to a lack of resolution in the plot : the slopes on the two sides of the cauchy horizon are dominated by @xmath347 and @xmath348 , which are discontinuous at the resolution of the plot , but close enough to the cauchy horizon , the slope becomes @xmath349 , which is continuous . we shall now consider the spherical dss scalar field as an infinite - dimensional dynamical system where @xmath16 is the `` time '' coordinate . the dynamical variables in this system are @xmath350 , @xmath351 and @xmath149 ( or @xmath352 in the outer patch ) . the variable @xmath353 is not independent , but given by a constraint . however , many solutions of the dynamical system correspond to the same spacetime , namely all that are related by the coordinate transformations ( [ dssgaugefreedom ] ) . the pair of periodic functions @xmath350 and @xmath351 describes a closed , possibly self - intersecting curve in the @xmath354 plane . an entire evolution in @xmath16 gives a surface that is topologically a cylinder , which we may call a phase portrait of the solution . clearly the surface itself is invariant under the coordinate change @xmath355 . if we consider @xmath15 as our `` space '' coordinate in the usual `` 3 + 1 '' split , then by looking at the phase portrait we have eliminated the `` spatial '' gauge freedom . the `` slicing '' freedom @xmath356 however does change the shape of the phase portrait , so it is not completely gauge - invariant . the choptuik spacetime up to the ch is given in fig . [ dynbefore ] and its regular continuation is shown in fig . [ dynafter ] . phase - portrait of the regular continuation of the choptuik spacetime from the ch at @xmath198 to the regular center at @xmath357 . again , from @xmath259 to @xmath358 we have distorted the axis to show the infinite number of oscillations that pile up there . from @xmath358 to @xmath359 we use a logarithmic axis and from this middle point to @xmath360 we use a different logarithmic axis . ] imposing css means that the system is independent of @xmath15 and hence the phase - portrait in @xmath361 reduces to a line that can be easily projected on the @xmath354 plane . then the whole evolution of the system can be described by this curve in the @xmath354 plane , which is now completely gauge - invariant . this is essentially what brady has done @xcite . in order to understand the singular continuations of the critical spacetime , we first look at phase portraits of the roberts spacetime . fig . [ roberts ] shows the phase - lines of the roberts solution for several values of @xmath0 in both branches ( see also appendix [ appendix : roberts ] ) . we see that the shape of the curves become constant for very small values of @xmath0 , but just translated along the @xmath362-axis . on the other hand fig . [ dynsingular ] gives the phase - portrait two of our singular evolutions , for small and large initial @xmath131 , respectively . in both cases as @xmath301 squeezes the phase cylinder into the diagonal , following a roberts solution in the former case , and not doing so in the latter . this figure demonstrates why it is not possible to obtain @xmath363 in the continuations of the choptuik solution . the roberts spacetime with @xmath364 does have @xmath363 on the singularity , @xmath365 ( minkowski ) has @xmath366 , and all other @xmath0 have @xmath278 . the @xmath364 roberts solution has the phase line @xmath367 line . to reach @xmath363 approach a solution must approach this line from the very beginning , because it is unstable . this requires @xmath83 of order @xmath327 at least and on the cauchy horizon we only have @xmath83 of order @xmath368 . phase - lines of the roberts spacetime . the colors encode different values of @xmath0 , for both branches : 1 ( red ) , 0.99 ( orange ) , 0.9 ( yellow ) , 0.5 ( light green ) , 0.1 ( dark green ) , 0.01 ( light blue ) , 0.001 ( dark blue ) and 0.0001 ( purple ) . the future ( past ) branch of the singularity is denoted with thick ( thin ) lines starting with @xmath369 ( @xmath370 ) . note that the vertical lines @xmath367 corresponding to @xmath364 are unstable . note also that the smaller @xmath0 is , the longer the curves stay close to the unstable flat line @xmath371 . ] phase - portrait of two singular continuation of the choptuik spacetime ( upper : initial @xmath289 of order @xmath368 ; lower : initial @xmath131 of order @xmath232 ) , together with a few roberts solutions for @xmath372 . the @xmath373 solution almost coincide with the external envelope of the upper phase - portrait and sets a maximum for the values that are realized in this solution . the lower phase - portrait is shown only for @xmath374 . for @xmath375 it is essentially on the diagonal in the @xmath354 plane although the amplitude of the oscillations is not the same as in the roberts solution . note that for small @xmath16 this phase portrait has been truncated also in @xmath82 , because @xmath82 becomes large . for @xmath376 the phase portrait is essentially on the @xmath82-axis . ] in a companion paper @xcite we investigate the constraints that the kinematic assumptions of spherical symmetry and discrete self - similarity impose on the causal structure . the key elements of our analysis were the self - similarity horizons radial null geodesics that are mapped onto themselves by the self - similarity . these come in two types : a `` fan '' connects a point singularity to a piece of null infinity , while a `` splash '' connects a piece of null singularity to a point at null infinity . all spherically symmetric and self - similar spacetimes can be enumerated in terms of a sequence of fans and splashes . we found that the singularity is central and consists of a middle segment which is either a point or null line , flanked by two segments which can be timelike , spacelike , or null lines , or which can be absent . there are never two or more disconnected singularities . in this paper , we have focussed on scalar field matter and in particular choptuik s critical solution in gravitational collapse . this solution is found as an intermediate attractor in the evolution of regular asymptotically flat scalar field initial data near the black hole threshold @xcite . independently , it can be constructed as the solution of a nonlinear pde boundary value problem from the assumptions of spherical symmetry , discrete self - similarity , and analyticity at the past center and past lightcone of the singularity @xcite . it can then be uniquely continued up to the future lightcone ( cauchy horizon ) of the singularity . @xcite found that the scalar field oscillates roughly as @xmath377 as the cauchy horizon @xmath139 is approached . it was suggested on theoretical grounds that these oscillations are damped by a factor @xmath178 with @xmath169 , but the numerical value of @xmath165 was too small to distinguish it from zero numerically . we have repeated the numerical calculations of ref . @xcite from scratch and have increased the accuracy by more than four orders of magnitude . we have found the positive value for @xmath165 given in eq . ( [ epsilon_value ] ) . therefore the scalar field is continuous on the cauchy horizon , with null data given in eq . ( [ v_f_numerical ] ) . the structure of the fields near the cauchy horizon is quite complicated , and has been discussed in subsections [ section : singularch ] and [ section : expansion ] . all possible dss continuations beyond the cauchy horizon are determined by the ( fixed ) null data on the cauchy horizon , and one ( arbitrary ) periodic function @xmath161 that can be thought of as data emerging from the naked singularity . ( in the absence of the naked singularity , null data on the cauchy horizon would alone determine the continuation . ) there is a unique dss continuation that has a regular timelike center except for a naked point singularity . for all other values of the arbitrary function @xmath161 , the continuation has a timelike central singularity with infinite negative mass . in particular , neither fig . [ figure : spacelikesingular ] , nor the even more exotic spacetime diagrams found in self - similar perfect fluid solutions @xcite can arise as continuations of the choptuik solution ( although they can probably arise in other dss scalar field solutions ) . the global structure of the choptuik solution is of interest partly because of the connection between critical collapse and cosmic censorship . choptuik s work has established that ( assuming the validity of general relativity at arbitrarily high curvatures ) a naked singularity can be formed in the spherical einstein - scalar field system from generic regular and asymptotically flat initial data by fine - tuning any one parameter to the black hole threshold . solutions with a naked singularity therefore form a subset of codimension one ( the black hole threshold in the space of initial data ) in all solutions arising from regular data . the choptuik spacetime is an attractor on the critical surface and therefore , assuming that it is actually a global attractor on the surface , we can conclude that every naked singularity will have , at least in a neighborhood of the singularity , the structure of the choptuik spacetime . we can now summarize this structure as follows : the curvature at the cauchy horizon of the singularity is finite but not differentiable , with an infinite number of damped oscillations piling up as the cauchy horizon is approached . the continuation of the spacetime beyond the cauchy horizon is not unique , but we have shown that if it is dss then the naked singularity is either a single point or timelike with infinite negative mass . the spacetime near the timelike singularity is locally css and velocity - dominated . the question remains whether this structure is stable against perturbations which break self - similarity and/or spherical symmetry . because the background is spherically symmetric and periodic in @xmath15 , all such perturbations are of the form @xmath378 , where @xmath379 is periodic , times a suitable scalar , vector or tensor spherical harmonic . we have shown in the past @xcite that all but one of these modes decay , with @xmath380 . ( the one growing mode determines the critical exponent for the black hole mass . ) the functions @xmath379 were calculated explicitly only between the regular center and past lightcone , but as they are by construction analytic at the past lightcone , they can be analytically continued up to the cauchy horizon , with the same @xmath381 . what needs to be done is to check that the functions @xmath379 remain bounded as @xmath16 approaches the cauchy horizon . we expect that this is true . nolan and waters @xcite have investigated a massless minimally coupled , non - spherically symmetric scalar field propagating on a class of spherically symmetric css spacetimes , and find that its gradient remains finite at the cauchy horizon . the structure of the full perturbation equations is similar . as the functions @xmath379 can not be analytic at the cauchy horizon ( because the background is not ) , the spectrum of @xmath381 need not be the same in the continuation beyond the cauchy horizon , and is likely to depend on the choice of continuation . we have finally given , for the first time , a number of global images of the choptuik spacetime , trying to minimize the gauge content of the pictures . we would like to thank james vickers for helpful discussions . this research was supported by epsrc grant gr / n10172/01 . in this appendix we review a variety of coordinate systems defined by giving @xmath14 as a function of @xmath15 and @xmath16 . this class covers the coordinate systems used in refs . @xcite , but does not include the lagrangian fluid coordinates used in @xcite . we classify the gauges within this class in three stages . 1 ) we fix @xmath70 . 2 ) we impose an algebraic relation between @xmath12 , @xmath13 and @xmath4 . 3 ) we parameterize @xmath12 , @xmath13 and @xmath4 in terms of two functions , say @xmath32 and @xmath180 . ( it will turn out to be useful to always use the scalar @xmath32 defined above as one of the two parameters . ) in a dss spacetime , the unknowns are periodic in @xmath15 . we therefore solve the einstein equations by `` evolving '' in @xmath16 , with periodic boundary conditions in @xmath15 . when we use the einstein equations to eliminate the metric derivative @xmath382 , @xmath82 and @xmath83 obey a pair of transport equation of the form @xmath383 where the dots stand for a known function of @xmath82 , @xmath83 , @xmath32 and @xmath180 but not their derivatives . by a suitable parameterization of @xmath12 , @xmath13 and @xmath4 in terms of @xmath32 and @xmath180 , the einstein equations can always be brought into the form of two odes in @xmath16 , which for our purposes are evolution equations , and one ode in @xmath15 , which for our purposes is a constraint : @xmath384 this last equation can be made linear in @xmath124 ( that is , in @xmath31 ) by a suitable choice of @xmath180 , at least in all three coordinate systems we use . this inhomogeneous linear equation can then be solved uniquely for @xmath32 in terms of @xmath82 , @xmath83 and @xmath180 , using the fact that @xmath82 , @xmath83 and @xmath180 are periodic in @xmath15 and that we require @xmath32 to be periodic , too . in a css solution , where nothing depends on @xmath15 , the @xmath123 constraint becomes an algebraic equation linking @xmath82 , @xmath83 , @xmath32 and @xmath180 . the regular center is both timelike and an @xmath16 line , but @xmath15 is finite there . this requires @xmath35 at the center , and @xmath386 elsewhere . the simplest choice is @xmath385 . on a past patch , it is possible to choose the @xmath15 lines to be timelike , null or spacelike , or to change signature . we consider only the first two possibilities [ [ bondi - past - patch ] ] bondi past patch + + + + + + + + + + + + + + + + if we choose the @xmath15-lines to be null everywhere , this means @xmath189 . we then have @xmath49 between the past center and the past light cone , and @xmath45 beyond the past lightcone , so that @xmath12 and @xmath13 both change sign at the past lightcone . @xmath385 and @xmath189 has been used by brady @xcite to investigate css solutions with scalar field matter . he used the bondi metric coefficients that are traditionally called @xmath387 and @xmath388 as parameters . in a dss solution this gives a constraint for a combination of @xmath387 and @xmath388 , as well as evolution equations for @xmath387 and @xmath388 . the best parameterization we have found uses the mass function @xmath32 and the bondi metric coefficient @xmath389 : @xmath390 where the upper sign applies when @xmath15 is an outgoing null coordinate , and the lower sign applies when @xmath15 is an ingoing null coordinate , assuming that @xmath194 . this gives an ode evolution equation and a linear ode constraint for @xmath124 , and an ode evolution equation for @xmath180 . [ [ schwarzschild - past - patch ] ] schwarzschild past patch + + + + + + + + + + + + + + + + + + + + + + + + if the @xmath15-lines are spacelike everywhere , this implies @xmath391 . a natural algebraic condition to impose is to make the @xmath15 lines orthogonal to the @xmath21 lines . this means @xmath392 . @xmath15 is then @xmath393 where @xmath394 is the schwarzschild time coordinate . a possible parameterization is in terms of the metric coefficients often called @xmath32 and @xmath395 in schwarzschild coordinates . ( @xmath32 is the scalar we have defined above ) . the best parameterization we have found is by @xmath32 and @xmath100 , and is given in ( [ pastpatch ] ) above . it gives ode evolution equations for @xmath32 and @xmath180 , and a linear ode constraint equation for @xmath124 . the transport equations for @xmath82 and @xmath83 are linear in @xmath82 and @xmath83 because @xmath382 reduces to a function of @xmath32 , @xmath15 and @xmath16 . the lightcones are at @xmath396 ( although of course only one lightcone can be covered at one time by a past patch ) . on an outer patch that stretches from the past to the future lightcone the @xmath15 lines must be timelike at least somewhere , and it is possible to make them timelike everywhere . the simplest choice for @xmath14 compatible with all these possibilities is @xmath14=1 . [ [ schwarzschild - outer - patch ] ] schwarzschild outer patch + + + + + + + + + + + + + + + + + + + + + + + + + assuming that the @xmath15 lines are timelike everywhere means @xmath398 . we have @xmath49 on the past lightcone and @xmath45 on the future lightcone , so that @xmath13 has to change sign somewhere between . this `` @xmath13-surface '' is where the @xmath15 and @xmath16 lines are orthogonal . in @xcite one of us used a coordinate system in the outer patch that was based on schwarzschild coordinates . as in the past patch based on schwarzschild coordinates , this meant @xmath399 . however , when we impose @xmath399 together with @xmath397 , we face an unexpected coordinate singularity at @xmath104 . in particular , if we again use @xmath32 and @xmath100 to parameterize the metric , we obtain an ode constraint and an ode evolution equation for @xmath32 as before on the past patch , but for @xmath180 we obtain an equation of the form @xmath400 in order to make the coordinates regular at @xmath104 , we need to impose the vanishing of the numerator of this equation there , and that is effectively what was done in @xcite . but this introduces an additional boundary condition just to keep the coordinate system regular on a surface where the solution itself is perfectly regular , and that is why we do not use these coordinates here . [ [ best - buy - outer - patch ] ] best buy outer patch + + + + + + + + + + + + + + + + + + + + a better algebraic condition to combine with @xmath397 is @xmath401 . a workable parameterization is in terms of the mass function @xmath32 and @xmath13 itself . this gives ode evolution equations for @xmath32 and @xmath13 , and a ( nonlinear ) ode constraint for @xmath32 . the lightcones are at @xmath402 . if we replace @xmath13 by @xmath403 , we have the parameterization given above in ( [ outerpatch ] ) , and the ode constraint for @xmath32 becomes linear in @xmath124 , while the wave equation becomes linear in @xmath82 and @xmath83 . [ [ a - c - unworkable ] ] @xmath404 : unworkable + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + it is compatible with @xmath397 to assume that the @xmath15 lines `` bend round '' to become spacelike at large and small @xmath16 . means this that @xmath4 changes sign . a natural choice is that @xmath12 and @xmath4 change sign together , and we can impose this by setting @xmath404 . we have not found a way of parameterizing this choice in a way that brings the einstein equations into the standard form ( [ standardform ] ) . parameterizing with @xmath13 and @xmath31 gives two roots for @xmath12 . using @xmath12 and @xmath31 gives two roots for @xmath13 . [ [ x - and - tau - orthogonal - unworkable ] ] @xmath16 and @xmath15 orthogonal : unworkable + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we can also force the `` bending round '' by making @xmath15 and @xmath16 lines orthogonal everywhere , or @xmath405 . this implies that , while they are independent , @xmath12 and @xmath4 change sign together . parameterizing by @xmath12 and @xmath4 we obtain an evolution equation and a constraint for @xmath12 and a constraint for @xmath4 . but the constraint for @xmath4 is homogeneous , @xmath406 , so that @xmath4 can not be periodic . therefore this coordinate system is not compatible with css or dss . [ [ forcing - the - lightcones - unworkable ] ] forcing the lightcones : unworkable + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + instead of finding a working outer patch and then forcing the two lightcones to fall on the lines @xmath43 and @xmath47 by imposing @xmath407 , we can make @xmath12 a given function of @xmath16 . the simplest such choice is @xmath408 , which vanishes at @xmath409 . ( the two lightcones will be distinguished by the sign of @xmath13 ) . a parameterization that brings the einstein equations into standard form is @xmath410 this gives an evolution equation and a constraint for @xmath127 and a linear constraint for @xmath411 . however , the right - hand sides of all three equations are of the form @xmath412 . furthermore , at @xmath413 the numerator of @xmath414 depends on the matter through @xmath415 , while the numerator of @xmath416 depends on the matter through @xmath417 . ( the numerator of @xmath418 is proportional to that of @xmath414 . ) that means that we would have to impose two independent regularity conditions at @xmath413 . together these would fix @xmath82 and @xmath83 completely . therefore this coordinate system is not sufficiently generic near the line @xmath405 . from ( [ expansion1],[expansion2 ] ) we see that @xmath419 the regular coefficients to @xmath186 are as follows : @xmath420 is the unique periodic solution of @xmath421 and @xmath422,\\ a_1&=&-{\xi\over 2}(u_0 ^ 2-v_0 ^ 2 ) , \\ b_1&=&-{\xi'\over a_0}-{\xi\over 2a_0}\left(-3+a_0 ^ 2 + 2v_0 ^ 2\right),\\ b_2&=&-{a_1\over 2 a_0}b_1 \nonumber \\ & & -{\xi\over 4a_0 } \left(2a_0a_1-b_1u_0 ^ 2+b_1v_0 ^ 2 + 4v_0v_1\right ) , \\ d_1&=&-{a_0 b_1\over \xi } .\end{aligned}\ ] ] the regular coefficients to higher orders continue in this style , with @xmath423 the solution of a linear inhomogeneous ode , while @xmath424 , @xmath425 and @xmath426 are given algebraically . the nonvanishing singular coefficients up to @xmath427 are @xmath428 we only give two examples of how these coefficients are derived . substituting the ansatz into the @xmath310 equation and isolating the terms of @xmath429 in the result , we obtain @xmath430 we solve this for all @xmath15 and @xmath182 by setting @xmath431 and by making @xmath432 the unique solution of the ode @xmath433 if we substitute the ansatz into the @xmath176 equation ( [ u_singular_eq ] ) and isolate the terms of @xmath434 , we find @xmath435 because @xmath436 , the derivatives @xmath437 cancel out . taking also into account that @xmath438 ( this is an accident at this particular order ) , the equation can be rewritten as @xmath439\hat u_{1+\epsilon}^{(i ) } \nonumber \\ + \left[-\epsilon d_2 - 2a_0a_2)\check u_\epsilon\right ] \hat u_\epsilon + \left[-kd_2\check u_\epsilon\right ] \hat u_\epsilon ' + \left[\check v_{1+\epsilon}-2a_0\check a_{1+\epsilon}u_0\right ] \hat a_{1+\epsilon}=0.\end{aligned}\ ] ] the terms in square brackets all depend on @xmath15 and each multiplies a different known function of @xmath182 . to solve this for all @xmath15 and @xmath182 , we assign one term @xmath440 to each function of @xmath182 : @xmath441 the corresponding coefficients @xmath442 are the unique solutions of the odes @xmath443 where the source terms @xmath444 can be read off directly from ( [ u1pe ] ) . the calculation of the other terms proceeds in a similar manner at all orders . note that the functions of type @xmath445 , where @xmath180 stands for @xmath83 , @xmath32 and @xmath127 , are given algebraically and @xmath446 obeys an ode . for @xmath82 it is the other way around . the limit @xmath447 of our expansion exists . the regular @xmath448 terms in the series vanish identically , and so do some of the @xmath449 terms but not all of them . @xmath349 and @xmath347 for example vanish because they are proportional to @xmath173 , but @xmath348 is proportional to @xmath450 , and so encodes the curvature to @xmath186 . in the limit @xmath451 the curvature components proportional to @xmath287 are no longer continuous at the ch because they are now periodic in @xmath452 , but for the same reason they remain bounded . the components proportional to @xmath453 and @xmath454 are still continuous even then because @xmath83 is @xmath186 . @xmath32 is also still continuous , with @xmath282 on the cauchy horizon . in the notation of @xcite , the roberts solution @xcite is given by @xmath455 , \\ \label{roberts3 } \phi(u , v ) & = & \frac{1}{\sqrt{16\pi g } } \log \frac{(1-p)v - u}{(1+p)v - u},\end{aligned}\ ] ] with @xmath0 a constant parameter . @xmath365 is minkowski spacetime with zero scalar field , and without loss of generality @xmath456 . only the regions @xmath457 are physical and without loss of generality , we consider the right side ( @xmath458 ) of the spacetime . for @xmath459 , the lines @xmath460 are central curvature singularities : the mass function @xmath461 diverges on them . the line @xmath462 , @xmath463 is timelike and has ( infinite ) negative mass , and for @xmath463 forms the past branch of the roberts singularity . the line @xmath464 is timelike with negative mass for @xmath465 , null with zero mass for @xmath364 , and spacelike with positive mass @xmath466 . for @xmath467 it forms the future segment of the singularity . some of our solutions become asymptotically css and have @xmath468 , therefore approaching the roberts spacetime . we know that the solutions develop a timelike , negative mass singularity , which must correspond to one of the two branches of the roberts singularity . in this appendix we study the issue of which of the two branches is actually approached in our numerical evolutions and for what values of @xmath0 . our scalars @xmath82 and @xmath83 are @xmath469 where both denominators are positive in the region of interest . it is clear that @xmath0 is covariantly defined by @xmath82 and @xmath83 on the light cones @xmath470 or @xmath471 , respectively . however , at the singularities @xmath460 the scalars take values @xmath472 for any @xmath473 in the roberts solution . if we use an expansion using a generic self - similar coordinate @xmath159 such that @xmath474 with the singularity at @xmath139 [ that is , @xmath475 : @xmath476 therefore @xmath0 is covariantly given by the ratio of the rates at which @xmath82 and @xmath83 approach their values at the singularity : @xmath477 this determines both @xmath0 and which branch of the singularity we approach locally : the upper sign applies for @xmath478 ( past branch , turned upside down ) and the lower sign for @xmath479 ( future branch of the singularity ) . we can relate the coordinates @xmath480 of ( [ roberts1]-[roberts3 ] ) to our coordinate @xmath16 through @xmath481 this expression applies only for negative @xmath16 and so only covers the future branch of the singularity . we obtain the expressions for the past branch exchanging the role of the functions @xmath82 and @xmath83 . that explains why exchanging @xmath82 and @xmath83 in eq . ( [ uvratio ] ) amounts to a change of branch . it also shows that we have to change branch four times per @xmath9-period as we move through the four quadrants determined by the lines @xmath482 and @xmath483 . c. gundlach , living reviews in relativity * 2 * , 4 ( 1999 ) , published electronically at http://www.livingreviews.org . for an updated version see c. gundlach , critical phenomena in gravitational collapse , to be published in phys . , preprint gr - qc/0210101 .

the current understanding of topological matter @xcite is mostly based on a single - particle description . a paradigmatic example is the integer quantum hall effect : in the absence of interactions , the hall conductivity has a deep topological interpretation@xcite , which explains its quantization and stability . a more recent example is provided by the classification of time - reversal invariant insulators @xcite , which , again , relies on the properties of the noninteracting bloch functions . understanding the effect of interactions on topological matter has become a very active area of research @xcite . a natural model in which to explore such issues is the _ haldane - hubbard model_. the haldane - hubbard model describes spin-1/2 electrons on the honeycomb lattice , interacting via a local hubbard interaction of strength @xmath0 . the electrons hop between nearest neighbor sites with hopping strength @xmath1 , and between next - to - nearest neighbor sites with alternating hopping parameters @xmath2 : the phases @xmath3 describe a transverse magnetic field , with zero net flux through the honeycomb plaquette . finally , the system is also exposed to a staggered chemical potential , with strength @xmath4 on the two triangular sublattices . in the absence of interactions @xcite this model shows , depending on the value of its parameters , a trivial insulating phase with vanishing transverse conductivity @xmath5 , or a quantum hall phase with @xmath6 . these topological phases are separated by two critical curves in the @xmath7 plane , intersecting at the crossing points @xmath8 and @xmath9 . along the critical curves , the energy bands touch at a conical intersection ; at the crossing points , there are two such conical intersections , as in standard graphene . indeed if @xmath10 the system describes graphene with short range interaction . from a theoretical viewpoint , the haldane topological phases have been argued to emerge in pure graphene sheets by spontaneous mass generation , due to the strong , unscreened coulomb repulsion @xcite . from an experimental viewpoint , the haldane model has been realized in ref.@xcite , and the topological phase transition has been observed . the inclusion of a tunable hubbard interaction seems to be accessible by the present technology . therefore , studying its effects on the transport coefficients is of fundamental importance for the next generation of cold atom experiments . so far , the properties of the haldane - hubbard model have been investigated mostly via mean - field , variational and numerical analyses@xcite . concerning the _ transverse conductivity _ , topological arguments for interacting systems@xcite ensure that , away from the critical curves , @xmath11 can only take integer values , in units of @xmath12 ( here @xmath13 are the elements of the conductivity matrix , in the limit of zero frequency and zero temperature ) . however , its specific value at a given point in the phase diagram can be different from the corresponding non - interacting value , in particular in the vicinity of the critical curves ( at weak coupling , far from the critical lines , the conductivity is known to be independent of the interaction @xcite ) . the relevant question here is to distinguish between two scenarios : the first , in which small interactions are not able to generate new phases and their main effect is a shift of the critical curves , as found in certain 3d topological insulators @xcite ; and a second one , characterized by the emergence of a novel , interaction - induced , topological phase , like the one corresponding to @xmath14 , predicted for the haldane - hubbard model in ref.@xcite . regarding the _ longitudinal conductivity _ , in the absence of interactions it is equal to @xmath15 , for all the values of @xmath16 _ on _ the critical lines , with the exception of the crossing points @xmath17 , where it is equal to @xmath18 ( of course , away from the critical lines @xmath19 ) . there are no topological arguments ensuring that the _ critical _ longitudinal conductivity should remain quantized when the interaction is switched on : therefore , the relevant question here is whether the interaction introduces corrections breaking this exact quantization or not . this question is related to a similar one discussed in the context of graphene , in which recent experiments @xcite showed that the optical longitudinal conductivity is essentially universal , and in excellent agreement with the value computed for the non - interacting model @xcite ; on the contrary , the interaction produces dramatic effects on other physical quantities , such as the fermi velocity @xcite . on the theoretical side , the universal behavior in graphene is still not completely understood @xcite , see , in particular , ref.@xcite for a recent review . a rigorous result for short range interaction @xcite showed that , in order to get exact universality of graphene s longitudinal conductivity , one needs to fully take into account the non linear correction to the bands , even if such terms are irrelevant in the renormalization group ( rg ) sense . in this paper we compute the conductivity matrix of the haldane - hubbard model via exact rg methods , close to and at the critical lines , for weak interactions . we take lattice effects into account , and we exploit _ lattice _ symmetries in order to reduce the number of independent running couplings , in a way similar to ref.@xcite for graphene . the use of exact rg methods is motivated by the fact that the computation of conductivity is extremely sensitive to the choice of the regularization scheme@xcite . even though they are irrelevant in the rg sense , lattice and interaction effects produce , in general , _ finite corrections _ to the physical observables , and they must be taken into account in order to prove or disprove the emergence of new interaction - induced topological phases , as well as to address the issue of universality of the critical longitudinal conductivity . by choosing the chemical potential @xmath20 so to fix the fermi energy half - way between the valence and conduction bands , the band gap can only close at the two fermi points @xmath21 , where @xmath22 is the _ valley _ index . we prove that , close to criticality , the interacting euclidean two - point function is : & & s_2(k_0,p_f^+k ) = + & & = - ik_0 z_1,r - m_r & v_r(-ik_1+k_2 ) + v_r(ik_1+k_2 ) & ik_0 z_2,r+m_r ^-1-.1truecm(1+r(k_0,k))where the error term @xmath23 is subleading in the effective mass @xmath24 , in the matsubara frequency @xmath25 and in the quasi - momentum @xmath26 . the parameters @xmath27 , depend non - trivially on the valley index @xmath28 and on the interaction . in particular , the renormalized mass @xmath29 reads : [ fpm ] m_r,=w 3t_2-f_(u , w , ) where @xmath30 is the valley index , and @xmath31 is expressed in the form of a _ convergent _ renormalized series , whose first non - trivial order is given by eq . the dressed critical lines , defined by the condition that the renormalized mass vanishes , are also modified by the interaction , see fig.[figaltezza ] . similarly , the fermi velocity @xmath32 and the wave function renormalizations @xmath33 have non - trivial interaction corrections and , remarkably , @xmath34 and @xmath35 are _ different _ , as shown in fig.[fig1 - 2 ] . all these non - universal renormalizations are _ absent _ in effective relativistic descriptions : by neglecting the ( irrelevant ) non - linear corrections to the energy bands , one would obtain a nambu - jona lasinio model , in which lorentz and chiral symmetry would imply the invariance of @xmath32 , the invariance of the critical lines and @xmath36 . however , these extra symmetries are broken by the lattice , and the renormalization of the effective parameters is a physical signature of many body interaction that should be visible in real systems , e.g. , in cold atom experiments . these non - universal parameters also enter the computation of the conductivity : remarkably , they are related by exact _ lattice ward identities _ , which induce non - trivial cancellations and imply subtle universality properties , as stated in the following theorem . we recall that @xmath13 are the elements of the kubo conductivity matrix , in the limit of zero frequency and zero temperature . we also denote by @xmath37 their values on the renormalized critical curves . , @xmath38 , and different values of @xmath0 . @xmath39 , where the values of @xmath40 are reported in the figure . for @xmath41 the system is a topological insulator ( ti ) , while for @xmath42 the system is a trivial , normal , insulator ( ni ) . ] * theorem . * _ there exists @xmath43 such that for @xmath44 , the system is massless if and only if the right side of eq . this condition defines two renormalized critical curves intersecting at @xmath17 , separating two non - trivial topological phases , characterized by transverse conductivity @xmath45 , from two standard insulating phases , see fig.[figaltezza ] . on the renormalized critical curves , the critical longitudinal conductivity @xmath46 , @xmath47 , is quantized : if @xmath48 , _ ii^ = , while @xmath49 at @xmath17 . _ thus , the critical lines acquire non - universal , interaction - dependent corrections , but they still separate topological regions labelled by @xmath50 from the trivial ones , labelled by @xmath51 , see fig.[figaltezza ] . new intermediate phases characterized by the quantum number @xmath52 are _ rigorously excluded _ at weak coupling : the universality class of the topological transition remains unchanged . the effect of the repulsive interaction is to _ enlarge _ the topologically non - trivial region , see fig.[figaltezza ] . this enhancement agrees with the numerical findings of ref.@xcite and is presumably a sign that repulsive interactions in graphene - like systems can favor the spontaneous generation of the topological insulating phase @xcite . even if not protected by any topological argument , the critical longitudinal conductivity @xmath53 is _ exactly universal _ and equal to half the one of graphene , on the whole critical line , with the exception of the special crossing points @xmath8 and @xmath9 , at which the value of @xmath53 is the same obtained for interacting graphene@xcite , namely @xmath54 : each dirac cone contributes with a universal quantity @xmath55 to the critical longitudinal conductivity . of course , away from the critical curves , the longitudinal conductivity is exactly zero . our results are in agreement with a low energy description in terms of an effective action that includes a non - trivial chern - simons term , whose coefficient ( the hall conductivity ) is proportional to the difference of the signs of the renormalized masses , @xmath56 , rather than the bare ones , as one would get in the relativisitic approximation @xcite . the theory that we develop is _ non perturbative _ , in the sense that it allows us to express all the correlations and transport coefficients in terms of _ convergent _ series . as it will appear from the analysis , our non - perturbative bounds on the correlation functions , once combined with ward identities , allow us to conclude the universality of the conductivity , without exploiting explicit cancellations at all orders . we have not tried to optimize the estimate for the radius of the convergence domain , which , therefore , is expected to be far from the values of @xmath0 where interaction - induced phase transitions might take place . however , we believe that the range of validity of our convergent expansions could be improved by combining our analysis with numerical techniques , as it is done , for instance , for the stability of kam tori in classical mechanics . finally , we stress that our analysis only requires the interaction to be short - ranged , we considered the hubbard interaction just for the sake of definiteness . the paper is organized as follows . in section [ sec2 ] we define the haldane - hubbard model , and derive the exact lattice ward identities for its correlation functions . in section [ sec3 ] we perform an exact renormalization group analysis for the correlations , we classify the allowed running coupling constants by the exact lattice symmetries of the system , and we compute the decay of the correlations at large distances , as well as the renormalized critical line . in section [ sec4 ] we prove the quantization of the hall conductivity across the critical line , and the universality of the critical longitudinal conductivity . the haldane - hubbard model describes interacting fermions on the honeycomb lattice @xmath57 , which can be understood as the superposition of two triangular sublattices @xmath58 and @xmath59 ; see fig . the triangular sublattice @xmath58 is generated by the basis vectors _ 1 = ( 3 , -),_2 = ( 3 , ) . with each sublattice , we introduce fermionic creation and annihilation operators @xmath60 , @xmath61 , where @xmath62 is the spin degree of freedom , @xmath63 . the hamiltonian is : [ eq : h ] h = h_0 + uv - n , where : @xmath64 is the noninteracting hamiltonian , @xmath65 is the hubbard interaction and @xmath66 fixes the chemical potential . the noninteracting hamiltonian is@xcite : [ eq : h0 ] h_0 & = & -_=_x , y t_1 + & & - _ = _ x , y + & & + w_= ; the first sum is over nearest - neighbours on @xmath57 , while the second is over next - to - nearest neighbours . each site on @xmath58 is connected to its three nearest - neighbours on @xmath59 by the vectors : _ 1 = ( 1 , 0),_2 = ( -1 , ) , _ 3 = ( -1 , - ) . the next - to - nearest neighbour hopping parameter @xmath67 is defined as : t_2(x , x + _ i ) = e^it_2,t_2(x , x - _ i ) = e^-it_2 , for @xmath68 . explicitely ( see fig . [ fig3 ] ) : _ 1 = _ 1 - _ 2,_2 = _ 2,_3 = -_1 . the hubbard interaction term is , as usual : [ eq : v ] v = _ x , where the sum ranges over the full honeycomb lattice ; the density operator @xmath69 is : n_x , = \ { cc a^+_x , a^-_x , & + b^+_x , b^-_x , & . , in terms of which we also have @xmath70 . the factors @xmath71 in eq . ( [ eq : v ] ) amount to a redefinition of @xmath72 , and simplify the functional integral representation of the model ( see section [ sec : funcint ] ) . we denote the finite volume version of @xmath73 by @xmath74 , with periodic boundary conditions . the finite volume and finite temperature gibbs state is : _ , l=\{e^-h_l}/\{e^-h_l } , and we let _ = _ l_,l , = _ _ , l . _ correlations , current and conductivity . _ it is convenient to define ^+_x,= ( a^+_x , , b^+_x+_1,),^-_x , = ( ^+_x,)^ ; also , for any inverse temperature @xmath75 , we let @xmath76 be their evolution at ` imaginary time ' @xmath77 . for general @xmath78 , we extend @xmath79 anti - periodically ( of anti - period @xmath75 ) beyond the basic interval @xmath80 . the fourier transform of the fields is defined as @xmath81 , where @xmath82 is the brillouin zone @xcite . the 2-point function is s_2(,)&= & + & = & _ re^-i(-)s_2(),where @xmath83 , @xmath84 , @xmath85 is the fermionic time - ordering operator ( which orders imaginary times in decreasing order @xcite ) , and @xmath86 , where @xmath25 is the matsubara frequency . note that @xmath87 is a @xmath88 matrix ( with indices in the ` sublattice ' space ) and its definition is independent of the choice of @xmath89 . the current is defined via the peierls substitution ( see app.[app.ver ] ) , and is equal to j_p(x_0)=_= _ ^+_k+p,(x_0)m(k , p)^-_k,(x_0).[eq : curr]the two components @xmath90 , @xmath47 , of @xmath91 are the bare vertex functions , which are @xmath88 matrices , with elements labelled by the spinor indices . for the explicit expression of the bare vertex functions , see app.[app.ver ] . the current - current and the vertex correlations are defined , respectively , as & & k_()=_rdx_0 e^-i p_0x_0 , + & & g_(,)=_rdx_0_rdy_0 e^-i p_0x_0+i(k_0+p_0)y_0 + & & 1.2truecm , [ eq : kmn ] where @xmath92 , @xmath93 with m_0(p ) = 1 & 0 + 0 & e^-ip_1 ; the labels @xmath94 refer to the components of the current defined in eq . ( [ eq : curr ] ) . moreover , @xmath95 is the trace per unit volume , and the semi - colon indicates that the expectation is truncated . for later reference , we also introduce the vertex function : _ ( , ) = s_2 ^ -1(+)g_(,)s_2 ^ -1(),[eq : vf]where @xmath96 is the inverse of the 2-point function , thought of as a @xmath88 matrix . finally , the d.c . kubo conductivity is defined in terms of the current - current correlation , in units such that @xmath97 , as : _ ij = - _ p_00^+ , [ eq : sijdef]where @xmath98 and @xmath99 is the area of the fundamental cell . _ ward identities . _ the continuity equation for the lattice current eq . , when averaged against an arbitrary number of field operators , implies exact identities among correlation functions ( ward identities ) , valid for any value of the interaction @xmath0 . in particular , the one relating the 2-point and the vertex functions , which will play an important role in the following , reads as follows : _ = 0 ^ 2 ( i)^_,0p_g_(,)=s_2(+)m_0(p)- m_0(p ) s_2().[1.14 ] if we derive this equation with respect to @xmath100 , compute the result at @xmath101 and recall the definition of the vertex function , we find : _ ( , * 0*)=(-i)^_,0_s_2 ^ -1()+.[eq : wi-2]in the following , @xmath102 will be denoted simply by @xmath103 . _ the non - interacting case . _ if @xmath104 , the band structure and the phase diagram can be computed explicitly : the bloch hamiltonian is @xcite & & h_0(k ) = [ hbloch ] + & & = -2t_2_1(k ) + m(k ) & -t_1^*(k ) + -t_1(k ) & -.3truecm -2t_2_1(k ) - m(k ) where @xmath105 and & & _ i=1 ^ 3(k_i),_2(k ) = _ i=1 ^ 3 ( k_i ) + & & m(k ) = w - 2t_2 _ 2(k ) . the corresponding energy bands are @xmath106 to make sure that the energy bands do not overlap , we assume that @xmath107 . the two bands can only touch at the _ fermi points _ @xmath108 , which are the two zeros of @xmath109 , around which @xmath110 . the condition that the two bands touch at @xmath111 , with @xmath112 , is that @xmath113 , with m_ m(p_f^ ) = w + 3t_2.therefore , the unperturbed critical curves are given by the values of @xmath7 such that : w=33 t_2 , which correspond to the dotted curves in fig.[figaltezza ] . fixing the chemical potential in such a way that the fermi energy lies in between the two bands , = -2t_2_1(p_f^)=-3 t_2 , the system passes from a semi - metallic behavior , when @xmath16 is on the critical line , to an insulating behavior , characterized by the exponential decay of correlations , when @xmath114 . the insulating phase consists of four disconnected regions in the @xmath16 plane , two of which are ` topologically trivial ' , while the other two have non - zero hall conductivity , see fig.[figaltezza ] : more precisely , if @xmath114 , _ 12 = , = ( m_- ) - ( m_+ ) . we now construct the interacting correlations and phase diagram , by using a convergent renormalized expansion , in the spirit of ref.@xcite . in this section , we introduce the functional integral formulation of the model , discuss the exact lattice symmetries of the fermionic action , and describe the infrared integration , including the study of the flow of the running coupling constants . one of the main results of this section is the equation for the interacting critical line . we are interested in the semi - metallic and insulating regimes of the interacting system . we , therefore , set the chemical potential accordingly ( its value will be different , in general , from the unperturbed one ) : = -2 t_2_1(p_f^)- , where @xmath115 ( the shift of the chemical potential ) must be chosen as a function of @xmath116 so that the renormalized propagator either has a linear , ` conical ' , infrared singularity ( along the interacting critical line ) , or is gapped ( in the insulating phase ) . the generating function @xmath117 for correlations , in which @xmath118 is the external field conjugated to the fermionic fields , and @xmath119 is the external field conjugated to the current , can be written as the following grassmann integral : e^(f , a)=,[eq.11 ] where : @xmath120 , with @xmath121 and @xmath122 , is a two - component grassmann spinor ( it is the grassmann counterpart of @xmath79 ) , whose components will be denoted by @xmath123 , with @xmath124 ; @xmath125 is the fermionic gaussian integration with propagator g(,)=_r e^-i(- ) g(),[prop]where , letting @xmath126 , g()= & -ik_0+r(k)+ m(k ) & -t_1 ^*(k ) + & -t_1 (k ) & -i k_0+r(k)- m(k ) ^-1and , at contact , @xmath127 should be interpreted as @xmath128 $ ] ; v()=_r dx_0_x_a _ = 1,2(un^_,n^_,+_= , n^_,),where @xmath129 ; and , finally , & & ( , f)=_rdx_0_x_a_=(^+_,f^-_,+f^+_,^- _ , ) , + & & ( j , a)=_r^3 a_,_,,where @xmath130 , in which @xmath131 are the _ bare _ vertex functions , namely : @xmath132 , and , if @xmath47 , @xmath133 are the two components of the ( matrix - valued ) vector @xmath91 defined in and following lines . in terms of these definitions , the correlations can be re - expressed as s_2 ( , ) & = & ( 0,0 ) , + k _ ( , ) & = & ( 0,0 ) , and of suitable linear combinations of g_2,1;(,,)=(0,0 ) . we now compute the generating function eq . via a renormalized expansion , which is convergent uniformly close to ( and even on ) the critical line . note that , on this line , the grassmann integral has an infrared problem . in order to resolve and re - sum the corresponding singularities , we proceed in a multi - scale fashion . first of all , we distinguish the ultraviolet modes , corresponding to large values of the matsubara frequency , from the infrared ones , by introducing two compactly supported cut - off functions , @xmath134 , supported in the vicinity of the fermi points ( more precisely , we let @xmath135 , where @xmath136 is a smoothed out characteristic function of the ball of radius @xmath137 , with @xmath137 equal to , say , @xmath138 , and @xmath139 ) and by letting @xmath140 . we correspondingly split the propagator in its ultraviolet and infrared components : g(,)=g^(1)(,)+_= e^-ip_f^(x - y)g_^(0)(,)[eq:4 ] where @xmath141 and @xmath142 are defined in a similar way as eq . , with @xmath143 replaced by @xmath144 and by @xmath145 , respectively . we then split the grassmann field as a sum of two independent fields , with propagators @xmath146 and @xmath147 : @xmath148 and we rewrite the grassmann gaussian integration as the product of two independent gaussians : @xmath149 . by construction , the integration of the ` ultraviolet ' field @xmath150 does not have any infrared singularity and , therefore , can be performed in a straightforward manner , thus allowing us to rewrite the generating function @xmath151 as the logarithm of p(d^(0))e^-v^(0)(^(0))+b^(0)(^(0 ) , f , a),[eq : v0 ] where @xmath152 and @xmath153 are , respectively , the effective potential and the effective source ( which depend explicitly on , respectively , @xmath154 and @xmath155 ) , @xmath156 is independent of @xmath154 ( and depends explicitly on @xmath157 ) , and @xmath158 . both @xmath152 and @xmath153 are expressed as series of monomials in the @xmath159 fields , whose kernels ( given by the sum of all possible feynman diagrams with fixed number and fixed location of the external legs ) are _ analytic functions _ of the interaction strength , for @xmath0 sufficiently small . the proof of their analyticity is based on a determinant expansion and on a systematic use of the gram - hadamard bounds , see ref.@xcite . before tackling the multi - scale integration of the infrared modes , we make a digression about the symmetry structure of the effective potential , and in particular of its local parts : the purpose is to classify the possible relevant and marginal coupling constants . in the case @xmath160 ( standard graphene model ) the lattice symmetries severely constrain the form of the quadratic terms in the effective potential : in particular , the interaction does not shift the chemical potential , nor does it generate a mass@xcite . in the general case ( @xmath161 ) the model is invariant under the following symmetry transformations ( since they do not mix the spin indices , for notational convenience we temporarily drop the spin labels from the formulas ) . we discuss the symmetries in the _ absence _ of external fields , since we will use them only to infer the structure of the relevant and marginal contributions to the effective potential @xmath152 . once the structure of these terms is known , the structure of the marginal contributions to the effective source @xmath153 can be computed by using the ward identity . \(1 ) _ discrete rotations : _ _ ^-e^ik(_3-_1)n_-_t^- , _ ^+_t^+e^-ik(_3-_1)n_- where , denoting the pauli matrices by @xmath162 , we defined n_-=(1-_3)/2,t=(k_0,e^-i_2k ) ; that is , @xmath163 is the spatial rotation by @xmath164 in the counter - clockwise direction . \(2 ) _ complex conjugation _ : ^_^_- , combined with cc^*,- , where @xmath165 is a generic constant appearing in @xmath125 or in @xmath166 . \(3 ) _ horizontal reflections _ : ^-__1 ^ -_r_h,^+_^+_r_h_1 , with r_h=(k_0,-k_1,k_2),(w,)(-w,- ) \(4 ) _ vertical reflections _ : ^_^_r_v , with r_v=(k_0,k_1,-k_2),- . \(5 ) _ particle - hole _ : ^-_i^+,t_p,^+_i^-,t_p , with p=(k_0,-k_1,-k_2),- . note that , at fixed @xmath167 , the theory is invariant under the transformations ( 1 ) , ( 2)+(4 ) , and ( 2)+(5 ) . in particular , these transformations leave the quadratic part @xmath168 of the effective potential @xmath169 invariant . this means that : w_2()&=&e^-ik(_1-_2)n_- w_2(t^-1)e^ik(_1-_2)n_- + & = & w_2^*(-k_0,-k_1,k_2)[symm ] + & = & w_2^(-k_0,k_1,k_2).as we will see in the next section , the values of @xmath170 and of its derivatives at the fermi points define the _ effective coupling constants_. by , we find , for @xmath171 , -.2truecm w_2(_f^)&=&e^-i3n_-w_2(_f^)e^i3n_- + & = & w_2^*(_f^)= w_2^(_f^),which implies that w_2(_f^)=_+__3,[eq : w2_loc0]for two _ real _ constants @xmath172 and @xmath173 . if we derive with respect to @xmath174 and compute the result at @xmath175 , we find : _ w_2(_f^)&=&e^-i3n_-t_w_2(_f^)e^i3n_- + & = & ( -r_v)_w_2^*(_f^)[symm_der ] + & = & ( -p)_w_2^(_f^),where @xmath176 ( resp . @xmath177 ) is the diagonal matrix with diagonal elements @xmath178 ( resp . @xmath179 ) . by using , it is straightforward to check that _w(_f^ ) = -i z_1, k_0 & -u_(-i k_1 + k_2 ) + -u_(i k_1 + k_2 ) & -i z_2, k_0 , [ eq:6]where @xmath180 , and @xmath181 are _ real _ constants . in conclusion , for general values of @xmath167 , the linearization of @xmath170 at @xmath175 is parametrized by 5 real constants , namely @xmath182 and @xmath183 , the first two are relevant coupling constants , and the other three are marginal . note that , in general , the values of these constants depend on @xmath184 ( therefore , there are 5 of them at @xmath185 and 5 more at @xmath186 ) . note also that , in general , @xmath187 , i.e. , the wave function renormalization depends explicitly on the spinor index , an effect that can be checked explicitly at second order in perturbation theory ( see below ) , and can not be explained purely in terms of the relativistic approximation of the model around the fermi points . note that there are special points in the @xmath188 plane , for which the model has more symmetries , and where the number of independent couplings is smaller than in the general case . for instance , if @xmath189 , the model is invariant under all the 5 symmetry transformations listed above , in which case it is straightforward to see that & & _ =_-,_=0,u_ = u_- , + & & z_1, = z_2, = z_1,- = z_2,- . a similar discussion applies to the case @xmath190 , @xmath191 . finally , if @xmath192 , the model is invariant under the following additional symmetry transformation ( see also ref.@xcite ) : ^-_,-i _ 1_3 ^ -_-r_v,,^+_,-i^+_-r_v,_3_1 , which implies that @xmath193 so that , in particular , _ =0,z_1, = z_2, . a similar discussion applies to @xmath194 . let us now describe the integration of the infrared fields . we shall focus on the semi - metallic behavior of the system at , or very close to , a generic point of the critical line . moreover , since we are interested in the behavior of the current - current correlations around @xmath101 , we shall assume that the external field @xmath195 is supported in the vicinity of the origin ( in particular , we assume that it vanishes in the vicinity of @xmath196 , @xmath171 ) . by dimensional considerations , the quadratic terms in the effective action are _ relevant _ , and , the ones corresponding to the renormalization of the mass are of particular importance . the flow of the effective mass tends to diverge linearly under the rg iterations , which signals that , in general , the location of the critical lines is changed by the interaction . in order to construct a convergent expansion , we need to dress the mass , after which we determine the location of the renormalized critical lines , which is given by the condition that the dressed mass vanishes . more in detail , we proceed as follows . we perform the integration of the infrared modes in iteratively , by decomposing the fermionic fields as @xmath197 as @xmath198 , where @xmath199 is a grassmann field whose propagator is supported on the momenta @xmath174 such that @xmath200 , and by integrating the fields @xmath201 step by step . after the integration of the modes on scales @xmath202 , we rewrite the generating function @xmath117 as the logarithm of p(d^(h))e^-v^(h)(^(h))+b^(h)(^(h ) , f , a),[eq : vh ] where @xmath203 and @xmath204 are , respectively , the effective potential and source terms , to be defined inductively in the following . moreover , @xmath205 is the grassmann gaussian integration with propagator ( diagonal with respect to the @xmath62 and @xmath184 indices ) g^(h)_(,)&=&p(d^(h))^-(h)_,,^+(h)_,, + & = & e^-i(- ) g_^(h)( ) , where @xmath206 and , letting @xmath207 , @xmath208 $ ] and @xmath209 ( here @xmath136 is the cutoff function defined a few lines before ) , g_^(h)()=_h( ) a_1, , h( ) & b^*_ , h( ) + b_ , h( ) & a_2, , h( ) ^-1,[eq : prop.2]with & & a_, , h()=-ik_0z_, , h+r_(k)+(-1)^-1 m_ , h(k ) , + & & b_ , h()=-v_ , h ( ik_1+k_2)+s_(k).[eq : prop.3 ] in which @xmath210 , @xmath211 and @xmath212 are , respectively , the wave function renormalizations , the effective mass and effective velocities , to be defined inductively in the following . their initial values are : z_j , ,0=1,m_,0(k)=m(k+p_f^),v_,0=32t_1 . in order to clarify the inductive definition of the effective potential , source , etc , we now describe the integration step at scale @xmath213 . we start from , where @xmath214 is a sum of even monomials in the @xmath215 fields , whose kernels of order @xmath216 are denoted by @xmath217 ( for notational simplicity , we temporarily drop the space - time , spin , spinor and valley indices of the fermionic fields ) . similarly , we denote the kernels of @xmath204 of order @xmath216 in @xmath215 , @xmath218 in @xmath118 and @xmath219 in @xmath119 , by @xmath220 . the scaling dimension of the kernels @xmath221 and @xmath220 is ( see ref.@xcite ) d=3-n - m - q,[eq : scdim]with the convention that @xmath222 corresponds to relevant , @xmath223 to marginal , and @xmath224 to irrelevant operators . note that the only relevant terms are those with @xmath225 , and the only marginal terms are those with @xmath225 and @xmath226 ( note that , by construction , @xmath227 is positive and even ) . in particular , the effective electron - electron interaction , corresponding to the case @xmath228 and @xmath229 , is _ irrelevant_. in order to define a convergent renormalized expansion , we need to re - sum the relevant and marginal terms . for this purpose , we split @xmath203 and @xmath204 into their _ local _ and _ irrelevant _ parts ( here , for simplicity , we spell out the definitions only in the @xmath230 case , the general case is treatable analogously , along the lines of , e.g. , sect . 12 of ref.@xcite , or ref.@xcite ) : @xmath231 and @xmath232 , where , denoting the quadratic part of @xmath203 by @xmath233 and the part of @xmath204 of order @xmath234 in @xmath235 by @xmath236 we let : & & v^(h)()=_=_= + & & ^+_,, [ w^(h)_2;(*0*)+_w^(h)_2;(*0*)]^-_,, , and & & b^(h)(,0,a)=_=_=_=0 ^ 2 + & & a_,^+_+,, w^(h)_2,1;,(*0*,*0*)^-_,, . by the symmetries discussed in the previous section ( see , in particular , and ) & & v^(h)()=_=,where @xmath237 are real constants . moreover , by using the ward identity , we find that & & b^(h)(,0,a)=_=_=_=0 ^ 2 + & & a_,^+_+,, _ , , h^-_,,,[eq : lbh]where _ 0, , h&=&-_=1 ^ 2(z_, , h+z_, , h)n _ + _ 1, , h&=&-(v_ , h+u_ , h)_2 + _ 2, , h&=&-(v_ , h+u_ , h)_1 in which @xmath238 and @xmath239 are the standard pauli matrices . once the effective potential and source have been split into local and irrelevant parts , we combine the part of @xmath240 in the second line of with the gaussian integration @xmath205 , thus defining a dressed measure @xmath241 whose propagator @xmath242 is analogous to @xmath243 , with the only difference that the functions @xmath244 , @xmath245 in - are replaced by a_, , h-1()&=&-ik_0z_, , h-1()+r_(k ) + & & + ( -1)^-1 m_ , h-1( ) , + b_ , h-1()&=&-v_ , h-1( ) ( ik_1+k_2)+s_(k ) , with & & z_, , h-1()=z_, , h+z_, , h_h( ) , + & & m_ , h-1()=m_ , h(k)+_ , h_h( ) , + & & v_ , h-1()=v_ , h+u_ , h_h().now , by rewriting the support function @xmath246 in the definition of @xmath242 as @xmath247 , we correspondingly rewrite : @xmath248 , where @xmath249 is defined exactly as in - , with @xmath213 replaced by @xmath250 , and @xmath251 defined by the flow equations : & & z_, , h-1=z_, , h+z_, , h , + & & m_ , h-1(k)=m_ , h(k)+_ , h,[eq : flow.1 ] + & & v_ , h-1=v_ , h+u_ , h.at this point , we integrate the fields on scale @xmath213 , and define : & & e^-v^(h-1)()+b^(h-1)(,f , a)+w^(h)(f , a)=c_hp(d^(h ) ) + & & e^-f_^(h)(^(h)+)+v^(h)(^(h)+)+b^(h)(^(h)+ , f , a),where @xmath252 is the gaussian integration with propagator @xmath253 , @xmath254 , and @xmath255 . finally , letting @xmath256 , we obtain the same expression as , with @xmath213 replaced by @xmath250 . this concludes the proof of the inductive step , corresponding to the integration of the fields on scale @xmath213 . the integration procedure goes on like this , as long as the two effective masses @xmath257 are small , as compared to @xmath258 . if we are not exactly at the ` graphene point ' @xmath189 , i.e. , if we are close to , or at , any other point on the critical line but the origin , then after a while we reach a scale @xmath259 at which @xmath260 ( possibly , @xmath261 , in the case that @xmath262 is of order 1 , i.e. , if @xmath167 are far enough from the graphene point ) . note that , once we reach scale @xmath259 , the field @xmath263 is massive ` on the right scale ' @xmath264 . at that point , we integrate out the field @xmath263 in a single step , and we are left with a ( chiral ) theory , whose only dynamical degree of freedom is @xmath265 , with @xmath266 . from that scale on , we integrate @xmath267 in a multi - scale fashion , analogous to the one discussed above , with the important difference that only the running coupling constants corresponding to the valley index @xmath268 continue to flow . the multi - scale integration goes on until we reach a scale @xmath269 such that @xmath270 , at which point we can integrate out the remaining degrees of freedom in a single step . the criticality condition , i.e. , the condition that the system is on the ( renormalized ) critical line , corresponds to the condition that @xmath271 . the multi - scale integration described in the previous section defines a flow for the effective chemical potential @xmath272 , the effective mass @xmath273 , the effective wave function renormalization @xmath274 , and the effective fermi velocity @xmath212 . the flow of @xmath275 and @xmath212 is driven by eqs.([eq : flow.1 ] ) , while @xmath276 where @xmath277 is the ( @xmath115-component of the ) beta function , which is defined in terms of the sum of all the local quadratic contributions in renormalized perturbation theory , and should be thought of as a function of @xmath0 and of the sequence of the effective coupling constants . remember that the flow drives the effective couplings with both @xmath278 and @xmath279 , up to the scale @xmath259 ; then the flow of the couplings with @xmath280 is stopped , and only the couplings with @xmath268 continue to flow until scale @xmath269 ( possibly @xmath271 ) . the multi - scale procedure is well defined , and the effective potentials are , step by step , given by _ convergent _ expansions , provided : ( i ) @xmath0 is small enough , ( ii ) @xmath281 remain small for all scales , and ( iii ) @xmath282 remain close to their initial ( bare ) values , for all scales . note that , in order for condition ( ii ) to be valid , we need to properly fix the initial condition on the chemical potential , as discussed in the following . in addition , note that , once that the flows of @xmath274 and @xmath212 are controlled , then the marginal contributions to the effective source term @xmath283 are automatically under control , thanks to and following lines . the key fact , which allows us to control the flow of the effective couplings , is that , since the electron - electron interaction is _ irrelevant _ , with scaling dimension @xmath284 ( cf . with ) , then the scaling dimensions of all diagrams with at least one interaction vertex can be effectively improved by one , see ref.@xcite . in particular , @xmath285 , for any @xmath286 and a suitable constant @xmath287 , and similarly for the beta functions of @xmath274 and @xmath212 . [ the reason why we lose , in general , an @xmath288 in the decay exponent as @xmath289 , is that we need to use a little bit of decay @xmath290 in order to sum over all diagrams and scales , see ref.@xcite for details . ] in order to guarantee that the flow of the chemical potential remains bounded , we fix the initial data ( via a fixed point theorem , such as the contraction mapping theorem ) so that @xmath291 , in the limit as @xmath292 . thanks to the dimensional gain of @xmath293 , due to the irrelevance of the interaction , we actually find that @xmath294 tends to zero , as @xmath289 , exponentially fast : @xmath295(const.)@xmath296 . once we imposed that @xmath294 remains bounded for all scales @xmath297 , we can a posteriori check that @xmath298 is also bounded for all scales @xmath299 : in fact , the beta function @xmath300 , for @xmath301 , can be rewritten as @xmath302 $ ] , where the difference in square brackets can be straightforwardly shown to be proportional to @xmath303 [ if all the masses @xmath304 were zero , then the model would be symmetric under the exchange of @xmath184 in @xmath305 , as in ref.@xcite , see also section [ sec.symm ] above ; therefore , the difference @xmath306 between the contributions with different valley indices must be proportional to a mass term @xmath307 , which is smaller than ( const.)@xmath308 . therefore , the flow of @xmath298 , for @xmath301 , remains close to the one of @xmath294 ( which is uniformly bounded for all scales ) , up to terms that are proportional to @xmath303 and , therefore , are bounded by ( const.)@xmath309 ( here @xmath310 is the dimensional amplification factor arising from the scaling dimension @xmath311 of the chemical potential terms , while @xmath293 is the dimensional gain coming from the irrelevance of the interaction ) . recalling that @xmath312 , we find that @xmath295(const.)@xmath296 , for all scales @xmath301 . finally , once the chemical potential is fixed so that @xmath313(const.)@xmath296 , we immediately infer that the beta functions of @xmath274 and @xmath212 are bounded by ( const.)@xmath296 , as well : therefore , their flows converge exponentially fast , and the dressed values of @xmath274 and @xmath212 are analytic functions of @xmath0 , analytically close to their bare values . the discussion in the previous section guarantees that , once the chemical potential is properly fixed , then the flows of the chemical potential , wave function renormalizations , and fermi velocity converge exponentially fast . the values of the chemical potential , as well as of the dressed wave functional renormalizations , dressed fermi velocity , and dressed critical lines are expressed in terms of convergent expansions ( they are _ analytic _ functions of @xmath0 ) , which are dominated by the first non trivial order in perturbation theory , provided @xmath0 is not too large ( note that the condition of convergence of the renormalized expansion is _ uniform _ in the gap , and is valid , in particular , _ on the critical line _ ) . the explicit lowest order contributions to the chemical potential @xmath115 , to the renormalized fermi velocity @xmath314 and the wave function renormalizations @xmath315 on the renormalized critical line @xmath271 are the following : 1 . chemical potential : = -_=1 ^ 2g_(+_f^_2)g_()g_(+ ) ; 2 . fermi velocity : v_r=32 t_1-iu^2_k_1g_12(+_f^_2)g_12()g_21(+);[eq : vr ] 3 . wave function renormalizations : z_,r=1+iu^2_k_0g_(+_f^_2)g_()g_(+).[eq : zrr ] moreover , the equation for the critical line @xmath271 reads : m__2=2,where @xmath316 , @xmath317 and @xmath318 where defined after . this is a fixed point equation for @xmath319 , whose solution leads to the plot in fig.[figaltezza ] . note that , as discussed in sect.[sec.symm ] , there is no symmetry reason why @xmath34 should be equal to @xmath35 . actually , an explicit computation shows that @xmath320 is different from zero along the critical line , unless we are at one of the highly symmetric points @xmath321 or @xmath192 , see fig.[fig1 - 2 ] , where we plot the value of @xmath320 on the critical line at second order in @xmath0 , for two different values of @xmath0 . in this section we compute the jump discontinuity of the hall conductivity across the critical line , as well as the value of the longitudinal conductivity on the same line , and prove a universality result for both of them , i.e. , we prove that their values are quantized and exactly independent of the interaction strength @xmath0 . note that this fact is highly non - trivial , due to the unusual renormalization of the fermi velocity and of the wave function renormalizations , which depends explicitly on the spinor index and break the asymptotic relativistic invariance of the propagator : the cancellations behind universality need to take lattice ( and , therefore , rg - irrelevant ) effects into account , and do not follow from asymptotic relativistic computations . we stress that our result is exact at all orders of the ( convergent , renormalized ) expansion for the conductivity . one key ingredient used in the proof is the lattice ward identity , which is rigorously valid ( without any sub - leading correction ) , thanks to the exact lattice symmetries and the fact that the correlations appearing at both sides can be computed in terms of convergent expansions , following from the multi - scale construction described above . here we compute the universal jump discontinuity of the hall conductivity across the renormalized critical line . for the moment , we assume not to be at the graphene points @xmath322 and @xmath190 , @xmath323 ; we shall discuss later the ( straightforward ) adaptation to these special cases . therefore , the goal is to compute : = _ m_r0^+ _ 12 - _ m_r0 ^ -_12,where @xmath324 is the mass gap of the dressed propagator . the condition that we are not at a graphene point means that @xmath325 should be kept finite as @xmath326 . using the definition , as well as the fact that @xmath327 is differentiable in @xmath100 outside the critical line , we can rewrite @xmath328.\ ] ] the interacting current - current correlation can be computed via the multiscale renormalized expansion discussed in sect . [ sec.rg ] : in particular , proceeding as in ref.@xcite , among the contributions to @xmath329 we can distinguish the dominant contribution , coming from the ` dressed bubble ' , from the sub - dominant one , which is the sum over all the renormalized diagrams with at least one interaction term . thanks to the irrelevance of the interaction , these sub - dominant diagrams have a dimensional gain ( of order @xmath258 on scale @xmath213 ) , which makes the corresponding contribution to @xmath327 _ differentiable _ at @xmath101 , in the limit @xmath326 . in particular , they give zero contribution to @xmath330 . the dominant contribution to @xmath327 ( i.e. , the ` dressed bubble ' ) is k_ij^()&=&-2\{s_2()_i ( , ) + & & s_2(+)_j(+,-)},where @xmath331 is the vertex function defined in , and the factor 2 in front of the integral takes into account the spin degrees of freedom . both @xmath332 and @xmath333 are given by convergent renormalized series , which depend on the details of the microscopic model . the finite contribution to the jump - discontinuity of @xmath334 across @xmath335 comes from the integration over @xmath174 in the vicinity of @xmath336 , since the rest is continuous as @xmath326 . for the same reason , for the purpose of computing @xmath330 , we can replace @xmath337 by @xmath338 , and @xmath332 by its linearization @xmath339 at @xmath336 , |s( ) = -ik_0 z_1,r+m_r & -v_r(-ik_1+_2 k_2 ) + -v_r(ik_1+_2 k_2 ) & -ik_0 z_2,r - m_r ^-1,[eq : linprop ] where @xmath340 and @xmath32 are _ analytic _ functions of @xmath0 , for @xmath0 small , whose expansions at second order in @xmath0 are given explicitly by - . recall that , a priori , @xmath341 are complicated infinite series in @xmath0 . thus , a direct computation of the jump - discontinuity , starting from the expression of the dressed bubble and from the feynman rules for the generic term in the renormalized expansions for @xmath340 , @xmath32 and @xmath342 , would be hopeless . the key fact is that , thanks to the ward identity , [ eq : wivertex ] _ i(_f^_2)=_k_i|s^-1(*0 * ) that is , [ eq : wivertex2 ] _ 1(_f^_2)=-v_r_2,_2(_f^_2)=-_2v_r_1 . therefore , & & = ( _ m_r0^+-_m_r0 ^ -)_|k| _ r + & & \ { |s( ) _ 1|s^-1(*0 * ) _ 0|s()_2|s^-1(*0*)},[eq : delta]where we used that @xmath343 , and we denoted by @xmath288 a small , arbitrary , positive constant . using the identity _ 0 |s( ) @xmath344 by @xmath345 ( which is allowed , for the purpose of computing @xmath330 , simply because the difference is continuous at @xmath335 ) , we can further rewrite @xmath346 as & & = -(_m_r0^+-_m_r0 ^ -)_|k| _ r + & & \ { |s( ) _ 1|s^-1(*0 * ) |s()_0|s^-1(*0*)|s()_2|s^-1(*0*)}.the integral over @xmath25 can be evaluated explicitly and , after a straightforward computation , we get & & = _ m_r0^+m_r + & & _ |k| dk^-3/2.thus , introducing [ eq : vmtilde ] v_r = , m_r = m_r , we see that @xmath330 can be rewritten as , performing the change of variables @xmath347 : & = & _ m_r0^+ m_r _ |k|v_r dk^-3/2 + & = & _ m_r0^+ _ |k|v_r / m_r dk^-3/2 + & = & , where we recall that the result is expressed in units such that @xmath97 . therefore , the cancellation between the parameters @xmath32 , @xmath34 , @xmath35 gives a universal result . finally , at the graphene points , the analogous computation gives twice the same value , because of an extra factor 2 coming from the valley degeneracy . a similar discussion as the one in the previous subsection can be repeated for the _ longitudinal _ conductivity on the renormalized critical line . the point here , as compared to the computation of @xmath330 in the previous subsection , is to take first the limit @xmath326 , and then @xmath348 ( recall the definition of conductivity , eq . ) . once again , we assume for definiteness not to be exactly at the graphene point ( a similar discussion applies there , too ) . note that , by the very definition of current - current correlations , @xmath349 is _ even _ in @xmath350 . therefore , all the contributions to @xmath349 that are differentiable in @xmath350 give zero contribution to the longitudinal conductivity on the critical line . by repeating a strategy analogous to the one that led us to , for the purpose of computing the longitudinal conductivity on the critical line , we can : ( i ) replace the full current - current correlation by its dominant contribution ( from the ` dressed bubble ' ) ; ( ii ) restrict the integration over the loop momenta in the vicinity of @xmath336 ; ( iii ) linearize the propagators and vertex functions around @xmath336 ; ( iv ) use the ward identity eq . ( [ eq : wivertex ] ) to replace the vertex functions by the derivatives of the inverse two - point function . after these replacements , we get ( denoting the value of the longitudinal conductivity on the critical line by @xmath351 ) : @xmath352,\ ] ] with f(,p_0)&=&\ { |s_0( ) _ i |s_0 ^ -1(*0 * ) & = & v_r^2\ { |s_0( ) _ i |s_0(k_0+p_0,k ) _ i } where @xmath353 is the linearized propagator , computed at @xmath335 , and the last step follows from ( [ eq : wivertex ] ) , ( [ eq : wivertex2 ] ) . by evaluating the integral over @xmath25 explicitly , and setting @xmath354 as in eq . ( [ eq : vmtilde ] ) , the computation of @xmath351 reduces to the contribution of just one dirac cone to the longitudinal conductivity of noninteracting graphene @xcite , with fermi velocity @xmath355 . thus , proceeding as in ref.@xcite , we get , in units such that @xmath97 : _ ii^ = 12_p_00^+_0^v_r dx=18 . notice that , as for graphene , the fermi velocity ( in general a nontrivial function of the hubbard interaction strength @xmath0 ) disappears , thus yielding a universal result . the analogous computation performed at the graphene points gives twice the same value , in agreement with the result of ref.@xcite . we studied the haldane - hubbard model by rigorous renormalization group techniques . our analysis predicts that the critical lines separating the distinct topological phases are modified non - trivially by the hubbard interaction , in particular that the non - trivial topological phase , characterized by the topological quantum number @xmath41 , is enlarged by weak repulsive interactions . moreover , our results rule out the presence of new interaction - induced topological phases in the vicinity of the phase boundaries . such predictions may be verified experimentally in optical lattice realizations of the system@xcite , where the on - site interaction can be produced and tuned by means of feshbach resonances . concerning numerical simulations , our results agree with those of ref.@xcite . the interaction affects the relativistic structure of the two - point function by non - universal renormalizatized coefficients , which differ from those obtained by approximate treatments of the system based on the effective dirac theory . in particular , we find that there are two different wave function renormalizations , one for each pseudo - spin index . despite the non - universal renormalization of the two - point function and of the vertex functions , lattice ward identities guarantee the quantization and the universality of the conductivity matrix at the critical line . concerning the transverse conductivity @xmath11 , its quantization follows from topological arguments ; however , these arguments do not provide any information regarding which values @xmath11 might take . for instance , numerical and mean - field analyses predict that , at intermediate coupling strengths , new topological phases might appear , corresponding to the values @xmath356 , which are not present in the noninteracting theory . our exact analysis rules out such new phases at small coupling . the second part of our result focuses on the critical longitudinal conductivity @xmath357 ( away from criticality @xmath358 is trivially zero ) . in constrast to @xmath11 , this quantity is not protected by any topological argument . nevertheless , we show that it is universal : all interaction and lattice corrections disappear . each dirac cone contributes with a universal quantum of conductivity @xmath55 ; in particular , at the doubly critical points where the two critical curves cross ( see fig . [ figaltezza ] ) , the critical longitudinal conductivity is @xmath54 , which is the same value measured in graphene @xcite . our results require the interaction to be weak and short - range ; instead , different features are expected in the presence of long - range interactions . for instance , it is known that , at the graphene point , long - range interactions have dramatic effects on several physical properties @xcite , and their role on the renormalization of the optical conductivity is still actively debated @xcite . we expect such effects to have profound implications for the haldane - hubbard model , especially in the proximity of the critical lines separating the different topological phases . we plan to investigate this issue in future work . the work of a.g . has been carried out thanks to the support of the a*midex project hypathie ( no anr-11-idex-0001 - 02 ) funded by the `` investissements davenir '' 25 french government program , managed by the french national research agency ( anr ) , and by a c.n.r.s . visiting professorship spent at the university of lyon-1 . the work of m.p . has been carried out thanks to the support of the nccr swissmap . in order to define the current , we couple the electron gas to an external vector potential @xmath359 , by multiplying the hopping strength from @xmath360 to @xmath361 by an extra phase factor @xmath362 ( peierls substitution ) . we denote by @xmath363 the modified hamiltonian , and let the ( paramagnetic ) current be @xmath364 , where @xmath47 label the two ( orthogonal ) coordinate directions @xmath365 and @xmath366 . an explicit computation leads to , with @xmath367 and , defining @xmath368 , & & m_11(k , p)=-it_2_j=1 ^ 3_=_j_p_je^i(-k_j ) , + & & m_12(k , p)=-it_1_j=1 ^ 3_j_p_je^-ik(_j-_1 ) , + @xmath369 and @xmath370 . in this appendix , we discuss some of the details of the numerical computations from which figs.[figaltezza]-[fig1 - 2 ] were produced . the program used to carry them out is available online @xcite , has been named hhtop , and is released under an apache license . the source code includes a documentation file , in which the computations are described in greater detail . the numerical computations carried out in this work involve numerical evaluations of integrals . the algorithm that was used to carry these out is based on _ gauss - legendre _ quadratures , by which , given an integer @xmath371 , an integral is approximated by a discrete sum with @xmath372 terms : @xmath373 where @xmath374 are the roots of the @xmath372-th legendre polynomial @xmath375 , and @xmath376 if @xmath118 is an analytic function , then one can show that the remainder @xmath377 decays exponentially in @xmath372 . however , in order to compute the difference of the wave - function renormalizations , we need to compute the integral of an integrand that , instead of being analytic , is a class-2 gevrey function ( a class-@xmath378 gevrey function is a @xmath379 function whose @xmath216-th derivative is bounded by @xmath380 , so that analytic functions are class-1 gevrey functions ) . the remainder @xmath377 can be shown to be bounded , if @xmath118 is a class-@xmath378 gevrey function with @xmath381 and @xmath372 is large enough ( independently of @xmath118 and @xmath378 ) , by @xmath382 for some @xmath383 , that only depend on @xmath118 . for a proof of this statement , see lemma a3.1 in the documentation of hhtop@xcite . in short , this estimate is obtained by expanding @xmath118 in chebyshev polynomials , and using a theorem of a.c . curtis and p. rabinowitz@xcite that shows that , if @xmath118 is the @xmath384-th chebyshev polynomial , then @xmath377 is bounded uniformly in @xmath384 . the decay of the coefficients of the chebyshev expansion of class-@xmath378 gevrey polynomials allows us to conclude . there is a single , minor , pitfall in the numerical evaluation of @xmath387 : we wish to use gauss - legendre quadratures ( see app.[app.gauss_legendre ] ) to carry out the computation , but the integrand in ( [ eq.fnumeric ] ) is not smooth : indeed , if @xmath388 , then its second derivative diverges at @xmath389 due to the divergence of the derivative of @xmath390 . however , by switching to polar coordinates @xmath391 , this singularities is regularized , that is , the integrand becomes a smooth function of @xmath392 and @xmath393 . at this point , there is yet another danger to avoid : while the integrand is smooth , the upper bound of the integral over @xmath392 is a function of @xmath393 , which is , due to the rhombic shape of @xmath394 , only smooth by parts . the integral over @xmath393 must , therefore , be split into parts in which the bounds of the integral over @xmath392 are smooth . this can be done very easily using the @xmath395 rotation symmetry . once both of these traps have been thwarted , gauss - legendre quadratures yield very accurate results . in order to compute the correction to the critical line , we solve @xmath396 for @xmath397 and @xmath398 . for the sake of clarity , we have made the @xmath16 dependence of @xmath387 explicit . to solve ( [ eq.mass1 ] ) , we fix @xmath398 , and use a newton algorithm to compute the critical value of @xmath397 : we set @xmath399 , and compute @xmath400 provided @xmath401 is not too far from the solution of ( [ eq.mass1 ] ) , @xmath402 converges _ quadratically _ ( i.e. @xmath403 , in which the constant depends on the supremum of @xmath404 , which is bounded ) to the solution of ( [ eq.mass1 ] ) . at second order in @xmath0 , @xmath320 is @xmath405 where @xmath406 the computation is carried out on the critical line , that is , when @xmath407 . the integrals over @xmath350 and @xmath408 can be carried out explicitly : @xmath409\hskip15pt\cdot \left(\frac{(\xi_p+\xi_q+\xi_f)(\frac{m_p}{\xi_p}+\frac{m_q}{\xi_q}-\frac{m_f}{\xi_f}-\frac{m_pm_qm_f}{\xi_p\xi_q\xi_f})z}{(z^2-(\xi_p+\xi_q+\xi_f)^2)^2}\right ) \end{array } \label{eq.z1-z2}\ ] ] where , using the definitions of @xmath317 , @xmath410 and @xmath109 after and after , @xmath411 , @xmath412 , @xmath413 , @xmath414 , @xmath415 and @xmath416 , @xmath417 , @xmath418 . the numerical evaluation of the integral in ( [ eq.z1-z2 ] ) involves a similar difficulty to that in ( [ eq.fnumeric ] ) : the integrand has divergent derivatives if any of the following conditions hold : @xmath419 , @xmath420 or @xmath421 . these singularities can not be regularized by changing @xmath422 and @xmath423 to polar coordinates , since @xmath424 is a singular function of the polar coordinates of @xmath422 and @xmath423 ( due to the fact that it behaves , asymptotically , as @xmath425 approaches @xmath426 , as @xmath427 , which has divergent second derivatives ) . however , there are coordinates , which we call _ sunrise coordinates _ ( since @xmath428 is the value of the so - called _ sunrise _ feynman diagram ) , which regularize these singularities . their expression is rather long , and will not be expounded here ; the interested reader is invited to consult the documentation file bundled with the source code of hhtop@xcite . once written in terms of the sunrise coordinates , the integral in ( [ eq.z1-z2 ] ) can be computed using gauss - legendre quadratures very accurately . c. l. kane and e. j. mele , phys . 95 , 146802 ( 2005 ) . j. e. moore and l. balents , phys . b 75 , 121306(r ) ( 2007 ) . r. roy , phys . b 79 , 195321 ( 2009 ) . l. fu and c. l. kane , phys . rev . b 74 , 195312 ( 2006 ) . d. carpentier , p. delplace , m. fruchart and k. gawedzki , phys . 114 , 106806 ( 2015 ) . d. carpentier et al . , nucl . b 896 , 779 - 834 ( 2015 ) . a. kitaev , in advances in theoretical physics : landau memorial conference , edited by v. lebedev and m. feigelaman , aip conf . proc.,vol . 1134 ( aip , melville , ny , 2009 ) , pp . a. p. schnyder , s. ryu , a. furusaki , and a. w. w. ludwig , phys . rev . b 78 , 195125 ( 2008 ) . s. raghu , x. qi , c. honerkamp , s. zhang , phys . 100 , 156401 ( 2008 ) . c. weeks , m. franz , phys b 81 , 085105 ( 2010 ) . p. ghaemi , j. cayssol , d. n. sheng , a. vishwanath , phys . rev . 108 , 266801 ( 2012 ) . d. a. abanin and d. a. pesin , phys . 109 , 066802 ( 2012 ) . j. gonzlez , j. high energ . 07 , 175 ( 2013 ) . b. roy and i. herbut , phys . b 88 , 045425 ( 2013 ) . m. daghofer and m. hohenadler , phys . b 89 , 035103 ( 2014 ) . j. he et al . , phys . rev . b 84 , 035127 ( 2011 ) . j. he , y. liang , s .- kou , phys . b 85 , 205107 ( 2012 ) . zhu et al . , j. phys . : cond . mat . 26 , 175601 ( 2014 ) . wu , n. li , and s .- p . kou , eur . j. b 88 , 255 ( 2015 ) . w. zheng , h. shen , z. wang and h. zhai , phys . b 91 , 161107(r ) . gu , k. li and j .- x . li , arxiv:1512.05118 . j. wu , j. p. l. faye , d. senechal and j. maciejko , phys . b 93 , 075131 ( 2016 ) . j. he , s .- kou , y. liang and s. feng , phys . rev . b 83 , 205116 ( 2011 ) . j. maciejko and a. ruegg , phys . rev . b 88 , 241101(r ) ( 2013 ) . c. n. varney , k. sun , m. rigol and v. galitski , phys b 82 , 115125 ( 2010 ) ; phys . rev . b 84 , 241105(r ) ( 2011 ) . wang , z .- c . gu , c .- d . gong and d. n. sheng , phys . 107 , 146803 ( 2011 ) . i. vasic , a. petrescu , k. le hur and w. hofstetter , phys . b 91 , 094502 ( 2015 ) . a. amaricci et al . , phys . 114 , 185701 ( 2015 ) ; phys . b , in press ( arxiv:1603.04263 ) . p. goswami and s. chakravarty , phys . 104 , 196803 ( 2011 ) ; arxiv:1603.03763 . k. kobayashi , t. ohtsuki , k .- imura and i. f. herbut , phys . 112 , 016402 ( 2014 ) . b. roy , p. goswami and j. d. sau , phys . b 94 , 041101 ( 2016 ) ; b. roy , y. alavirad and jay d. sau arxiv:1604.01390 . e. g. mishchenko , europhys . 83 , 17005 ( 2008 ) . i. herbut , v. jurii and o. vafek , phys . 100 , 046403 ( 2008 ) . v. jurii , o. vafek and i. herbut , phys . rev . b 82 , 235402 ( 2010 ) . d. e. sheehy and j. schmalian , phys . b 80 , 193411 ( 2009 ) . a. giuliani and v. mastropietro , phys . b 85 , 045420 ( 2012 ) . i. sodemann and m. m. fogler , phys . b 86 , 115408 ( 2012 ) . b. rosenstein , m. lewkowicz , t. maniv , phys . 110 , 066602 ( 2013 ) . d. l. boyda , v. v. braguta , m. i. katsnelson , m. v. ulybyshev , phys . rev . b 94 , 085421 ( 2016 ) . j. m. link , p. p. orth , d. e. sheehy and j. schmalian , phys . rev . b 93 , 235447 . i. herbut , v. jurii , o. vafek and m. j. case , arxiv:0809.0725 . the definition of @xmath433 is meant as the limit as @xmath434 of @xmath435 , which is the anti - periodic extension in @xmath436 and @xmath437 ( of anti - period @xmath75 in both variables ) of @xmath438 , with @xmath439 . at equal times , we let @xmath440 . the same conventions hold for the mixed current - current and current - field correlations introduced below .

a decade of extrasolar planet discoveries has shown that the process of planet formation is more complex than originally anticipated . it leads to a remarkable diversity of planetary configurations , ranging from migrating hot jupiters to eccentric giant planets , as well as our own `` circular '' solar system . following these discoveries , progress has been made in understanding planet formation , but the theory is still incomplete ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the leading scenario for the formation of giant planets is the core accretion mechanism . icy planetesimals located beyond the snow line in their host disk collide repeatedly to grow a core with a modest gaseous atmosphere . if this core succeeds in reaching a critical mass of a few tens of earth masses , runaway accretion of a massive gaseous envelope proceeds and leads , ultimately , to the formation of a gaseous giant planet @xcite . some evidence supporting this scenario has emerged in recent years , in the form of a metallicity trend for stars hosting planets ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , the discovery of a high density hot jupiter @xcite and that of a surprisingly low - mass micro - lensing planet @xcite . however , a long - standing difficulty for the core accretion scenario , which has not yet been fully elucidated , is the fact that the timescale required to build - up critical core masses and thus large gaseous envelopes ( @xmath0-@xmath1 yr ) is comparable to the lifetimes of proto - planetary disks ( e.g. * ? ? ? * ; * ? ? ? orbital migration adds a layer of complication to theories of planet formation . as a result of gravitational interactions with their gaseous disk @xcite , the orbits of planets in the terrestrial mass range are predicted to decay on timescales ( @xmath2 yr ) short compared to disk lifetimes ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? migration is slower for planets of much smaller or much larger masses : in the first case because the torque causing migration is quadratic in planet mass , and in the second case because the planet opens a gap and then migrates on the disk s accretion timescale , which can be comparable to its lifetime . while it is possible or probable that many terrestrial planets form by agglomeration of smaller bodies after the gas is gone , this is not an option for the solid cores of jovian planets since , in the core accretion scenario , the cores must form before the gaseous envelopes can be accreted . the prevalence of jovian planets with orbital periods of only a few days deepens the mystery as it suggests that these planets did migrate but stopped short of merging with their stars at orbital radii where even the accretion timescale would seem to have been very short ( e.g. * ? ? ? * ) . analytic calculations and most hydro - dynamical simulations of migration usually assume a disk that is laminar apart from the waves and shocks excited by the planet itself ( e.g. * ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? ? * ; * ? ? ? * ; * ? ? ? but the effective viscosity of disks probably involves turbulence . @xcite , @xcite , and ( * ? ? ? * hereafter n2005 ) have found in 3d simulations of magneto - rotational turbulence that the instantaneous torque exerted on a planet in the terrestrial mass range is subject to fluctuations many times its mean value , apparently caused by turbulent density fluctuations in the planet s vicinity . in fact , no obvious secular decay manifests itself in the orbits of planets with mass @xmath3 , although because the simulations are limited to @xmath4 planetary orbits compare @xmath5 for jupiter during the lifetime of the proto - solar nebula and the predicted decay in semimajor axis is @xmath6 over this period , the expected secular trend might be difficult to discern amid the fluctuations . the possibility that planetary torques are stochastic raises several issues : 1 . is the time - averaged torque exerted on a planet of a given angular momentum affected ? n2005 finds that turbulence excites planetary eccentricity , @xmath7 . in linear theory , eccentricity allows the planet to interact with the disk at additional resonances that do not couple to circular orbits @xcite . also , at second and higher orders in @xmath7 , the radial positions of the original resonances shift if one compares eccentric and circular orbits at fixed angular momentum ; they do not shift at fixed orbital energy since keplerian energies and mean motions vary only with semi - major axis . thus for eccentric planets , one must distinguish the rate of change of the planetary angular momentum ( torque ) from the rate of change of orbital energy . smoothing of the planetary potential along epicycles tends to reduce the strength of the highest-@xmath8 lindblad resonances that dominate the torque on circular orbits . all of these perturbative effects are expected to become significant when @xmath9 , where @xmath10 is the disk thickness . planets of terrestrial mass achieved such eccentricities in n2005 s simulations . on the other hand , at about the same eccentricity the epicyclic velocities become supersonic , which will probably raise significant shocks , a non - perturbative effect . thus both the sign and magnitude of the changes in the mean torque due to turbulently excited eccentricity are unclear . we are not aware of any systematic investigation of this issue , nor have we attempted one . throughout this paper , we treat the planetary orbit as circular . 2 . even for circular orbits , stochastic changes in semi - major axis on timescales long compared to the period but short compared to the nominal migration time may significantly alter the mean orbital lifetime . this is particularly likely when the local mean torque varies strongly with radius , as it does in the alpha disks studied by @xcite . because of rapid changes in opacity in certain temperature regimes , the disks exhibit sharp peaks in the local migration timescale . under laminar conditions , orbital lifetimes can be dominated by the time required to drift across these peaks . turbulent fluctuations in the torque , however , could allow planets to `` jump over '' the peaks , thus shortening the lifetime . on the other hand , diffusion to large radii where the local migration time is long might prolong the orbital lifetime . 3 . finally , orbital diffusion should allow the occasional planet to survive much longer than the mean lifetime defined by its birthplace . if sufficiently many planets are formed , this might in principle help to reconcile theoretical predictions of rapid _ mean _ migration with the observations at least of the longer - period planets ( though it seems unlikely to explain hot jupiters ) . the methods adopted in this paper allow us to explore the second and third issues above , but not the first . in [ sec : fp ] , we develop a simple advection - diffusion or fokker - planck equation for the probability distribution function of planetary orbital angular momentum in the presence of torque fluctuations . parametrizations of the advection and diffusion coefficients are calibrated against semianalytic formalisms and the simulations of n2005 , respectively . 3 explores analytic steady - state and time - dependent solutions to this equation in steady disks whose properties are power laws in radius . 4 and 5 present numerical , time - dependent solutions in such a disk ( the minimum mass solar nebula ) and in alpha disks with realistic opacities . a summary of our main results and conclusions is given in 6 . planets migrate as a result of torques exerted by the disk . because of turbulence , the torque is a stochastic function with a mean part @xmath11 and a fluctuating part @xmath12 . for a given planetary mass ( @xmath13 ) and a given time - averaged surface density and thickness of the disk [ @xmath14 respectively ] , the mean torque depends only on the semimajor axis , @xmath15 , of the planetary orbit , which is assumed to be circular . the diffusion equation is most conveniently derived in terms of the orbital angular momentum , where @xmath16 is the mass of the central star . ] @xmath17 , rather than @xmath15 , as the spatial independent variable . the fluctuating part of the torque depends on time as well as @xmath18 : @xmath19 . by construction , @xmath20 . the fluctuating torque is taken to be a temporally stationary stochastic process with a finite amplitude and correlation time , so that the integral @xmath21 exists . @xmath22 will play the role of diffusion coefficient . the correlation time can then be defined by @xmath23 ^ 2}\,.\ ] ] temporal stationarity implies that @xmath22 , @xmath24 ^ 2}$ ] , and @xmath25 are independent of @xmath26 . the desired advection - diffusion equation derives from a fokker - planck equation , and its formal derivation is completely standard ( _ e.g. _ * ? ? ? * ) , but it is useful to repeat the derivation here in order to emphasize the underlying assumptions and to ask whether they are justified in this case . two important assumptions have already been introduced : the finiteness of eqs . ( [ ddef ] ) & ( [ tcorr ] ) . let @xmath27 be the probability that a planet of initial angular momentum @xmath18 suffers a change @xmath28 during time interval @xmath29 . moments of the change are @xmath30 of course , except for @xmath31 , these depend implicitly on @xmath18 and @xmath29 . next , let @xmath32 be the probability that the planet has angular momentum @xmath18 at @xmath33 given some initial condition at @xmath34 . the probability at @xmath35 is @xmath36\,d\delta j \nonumber\\ & = & \sum\limits_{n=0}^\infty\frac{(-1)^n}{n ! } \frac{\partial^n}{\partial j^n}\left\{\vphantom{\frac{a}{a } } \overline{(\delta j)^n}f(t , j)\right\}\,.\end{aligned}\ ] ] in the fokker - planck approximation , the changes are presumed small compared to @xmath18 itself but cumulative , so that one can ignore the terms @xmath37 in the sum above . more will be said about the justification for this below . clearly @xmath31 , @xmath38 , and @xmath39 ^ 2}\\[1ex ] & = & \bar\gamma(j)^2(\delta t)^2~+~\int\limits_0^{\delta t}\,d t_m \int\limits_{-2 t_m}^{2 t_m}\,d \tau~ \overline{\delta\gamma ( t_m-\,\tau/2,j)\delta\gamma ( t_m+\,\tau/2,j)}.\end{aligned}\ ] ] the next important assumption is that @xmath25 is short compared to the timescale on which @xmath18 changes by order itself let us write @xmath40 for the latter timescale so that we may take @xmath29 in the range @xmath41 . in that case , the double integral above @xmath42 . at this point , @xmath43\delta t ~+~o\left[(\delta t)^2\right],\ ] ] where the @xmath44 $ ] part includes the @xmath45 contribution to @xmath46 . neglecting these terms on the grounds that @xmath47 , one has the desired advection - diffusion equation : @xmath48 f_j & \equiv & \bar\gamma(j ) f(t , j)~-~\frac{\partial } { \partial j } \left[\vphantom{\frac{1}{2 } } d(j)f(t , j)\right]\,.\end{aligned}\ ] ] an interesting feature of the above result is that if the diffusion coefficient @xmath22 varies with @xmath18 , then it influences the _ mean _ migration rate . that is to say , one can rewrite the flux in eq . ( [ diffusion ] ) as @xmath49 so that @xmath50 contributes to the advection term . we have tried to abstract a value and scaling for @xmath22 from the mhd simulations of n2005 ; the results of @xcite were difficult to translate into our framework . nelson reports that the typical timescale of fluctuations in @xmath51 is of order half an orbital period . hence we take @xmath52 a direct estimate of the correlation function of the data suggests , however , that correlations may persist to much longer than an orbital period , and it is conceivable that the integral ( [ ddef ] ) for @xmath22 does not even exist ( nelson 2005 , private communication ) . if the correlation time does not exist , then the methods of this paper are inapplicable to migration . such long - term correlations imply a `` memory '' in the turbulence , presumably involving persistent structures in addition to the planet of interest : for example , long - lived vortices ( e.g. * ? ? ? * ; * ? ? ? * ) , or of course other planets . pending further numerical evidence , we adopt as a working hypothesis that @xmath25 _ does _ exist but with uncertain magnitude . to complete the diffusion coefficient , we require a parametrization of the variance of the fluctuating torque in terms of time - averaged disk properties . this entails some guesswork , as the simulations of n2005 and @xcite have explored only a limited range of disk models and planetary radii . a natural scale for the gravitational force exerted on the planet by the local gas is @xmath53 , where @xmath54 is the surface density of the disk . this is the force that the planet would feel if suspended just above the disk . the corresponding natural scale for the torque is @xmath55 . therefore , we postulate that the variance of the torque fluctuations is @xmath56 ^ 2}\approx \left(\mathcal{c}_d\times 2\pi g\bar\sigma r_p m_p\right)^2,\ ] ] where @xmath57 is a dimensionless coefficient depending upon the strength of the turbulence . since the parametrization ( [ amptorque ] ) underlies all of our results , it is worth more discussion . the turbulence is accompanied by density perturbations of r.m.s . amplitude @xmath58 on scale @xmath59 . an individual perturbation of this scale exerts a gravitational force per unit mass @xmath60 at distance @xmath61 if @xmath62 , and @xmath63 if @xmath64 . presuming that the forces from different perturbations add in quadrature , the mean - square force is dominated by the closest such perturbation , at @xmath65 , since the number of @xmath59-scale perturbations within distance @xmath61 increases more slowly with @xmath61 than the mean - square contribution from individual perturbations decreases : @xmath66 for @xmath67 and @xmath68 for @xmath69 , versus @xmath70 . thus , the r.m.s . force due to scale @xmath59 is @xmath71 for @xmath62 and @xmath72 for @xmath64 . the contributions to @xmath22 from @xmath62 are probably unimportant because of the weighting by @xmath59 and because smaller scales are likely to have shorter correlation times . if the density power spectrum were as `` blue '' as white noise , @xmath73 , and if the correlation time @xmath74 , then there would be equal contributions from all scales @xmath75 , with the result that the total contribution from these scales would be larger than the contribution from @xmath76 by a logarithmic factor . in reality , the density power spectrum is surely red , since local magnetorotational ( henceforth mri ) simulations find that most of the turbulent magnetic and kinetic energy resides at scales @xmath77 @xcite . thus , the dominant contribution to the force comes from scales @xmath77 . since @xmath78 in this regime , the total r.m.s . force fluctuation is proportional to the total r.m.s . surface - density fluctuation , @xmath79 . it remains to discuss how @xmath80 and hence @xmath81 may depend upon the strength of the turbulence . density fluctuations may arise from fluctuations in gas pressure or in entropy . local mri simulations indicate that @xmath82 @xcite , as might be expected from equipartition , since the internal energy per unit mass associated with a density fluctuation is @xmath83 . in a statistically steady accretion disk , the local thermal time @xmath84 , so if the correlation time of mechanical fluctuations is @xmath85 , entropy fluctuations due to dissipation are expected to be at most @xmath86 . entropy fluctuations may also arise by advection of material in the presence of a background entropy gradient , @xmath87 . the simulations cited above indicate that the turbulent reynolds stress scales with the magnetic stress so @xmath88 , and we assume that the largest displacements have frequencies @xmath89 , so the r.m.s . displacement @xmath90 in any direction . vertical and horizontal entropy gradients are likely to have scale lengths @xmath91 and @xmath92 , respectively , which would seem to make vertical displacements more important . however , the azimuthal force , and hence torque , exerted by a mass element changes only at second order as it is displaced vertically through @xmath93 , so vertical displacements contribute at @xmath94 . radial displacements contribute at first order , @xmath95 . the vertically uniform disk of n2005 has a radial entropy gradient , since the sound speed is constant but the density is not ; however , since n2005 obtains @xmath96 ( in agreement with previous work ) , it happens that @xmath97 in these models , so that we can not distinguish between @xmath98 and @xmath99 . it would be useful to calculate torque fluctuations in a variety of disk models with varying entropy gradients , thicknesses , and ( if possible ) strengths of turbulence . lacking such information , we provisionally assume @xmath100 , where @xmath101 is some function ( probably linear ) . this leads to the parametrization ( [ amptorque ] ) if @xmath102 is a universal constant . as will be seen , diffusion then dominates mean migration , _ i.e. _ inward drift at the rate predicted for a laminar disk , in the outer parts of plausible disk models . if @xmath79 is also proportional to @xmath103 , diffusion becomes even more dominant in the outer parts of a flaring disk . n2005 reports that the r.m.s . fluctuating torque per unit planetary mass is @xmath104 in code units . in the same units , @xmath105 ( nelson 2005 , private communication ) ; the latter is independent of radius because @xmath106 . therefore , @xmath107 is taken as the reference value of the dimensionless factor in equation ( [ amptorque ] ) . following eq . ( [ tcorr ] ) , the diffusion coefficient @xmath22 is the product of the correlation time @xmath25 ( [ tcorrsim ] ) and the torque variance ( [ amptorque ] ) . both factors are uncertain , but within the framework of this advection - diffusion approach , they combine into a single parameter . we hope that future numerical simulations will refine our estimates of this parameter and , more importantly , test the validity of the scalings ( [ tcorrsim ] ) and ( [ amptorque ] ) . for comparison , @xcite s expression for the mean lindblad torque density , @xmath108 leads to the scaling @xmath109 for consistency with our calibration of torque fluctuations , we also calibrate this relation against nelson s simulations of laminar disks , in which the type i torque per unit mass is @xmath110 in code units . the constant of proportionality in eq . ( [ eq : meantorque ] ) is then @xmath111 . using eq . ( [ eq : meantorque ] ) implies that the mean torque is always negative . this is fine for plausible power - law disks but might be misleading for alpha disks , where the surface density and temperature can be highly structured , allowing in principle for positive ( or at least much reduced ) @xmath11 at some radii . thus when modeling alpha disks we use eq . ( [ eq : ward ] ) directly , as described in @xcite . power - law disk models such as the minimum - mass solar nebula ( hereafter mmsn ; @xcite ) allow the advection - diffusion equation to be solved analytically , at least in steady state . some aspects of time - dependent evolution can also be determined analytically for this class of disk models . throughout this section , all disk models are considered to be infinite in radius as well as steady . while unrealistic , this is technically convenient , and we are still able to identify results that are likely to be sensitive to the details of the outer boundary conditions : for example , the mean orbital lifetime . for reference , the mmsn model adopted here has mean surface density @xmath112 and aspect ratio @xmath113 . a steady - state distribution of planets , @xmath114 , satisfies @xmath115 where the right - hand side is a source term representing the rate of formation of planets at radius @xmath15 with angular momentum @xmath17 . analytic solutions to eq . ( [ eq : steadystate ] ) exist when mean torque and the diffusion coefficient are both power - laws in @xmath18 : @xmath116 and the source term is a dirac delta function @xmath117 where @xmath118 is the rate of planet formation at @xmath119 . we see immediately that the flux of planets interior @xmath120 and exterior @xmath121 to the source are both constant and related to each other by @xmath122 first we consider the limiting behavior as @xmath123 . the local timescale for secular inward migration is given by @xmath124 , and the local diffusion timescale is given by @xmath125 . the ratio , @xmath126 indicates that the mean torque becomes asymptotically negligible if @xmath127 . ( this condition holds for the minimum - mass solar nebula , in which @xmath128 and @xmath129 . ) for such disks the solution to eq . ( [ eq : steadystate ] ) at large @xmath18 exterior to the source becomes @xmath130 , where @xmath131 is a constant of integration . we insist that @xmath132 since there should not be an incoming flux of planets from infinity . thus the numerator eventually becomes negative . the diffusion coefficient @xmath22 is unquestionably positive , so @xmath101 becomes negative at sufficiently large @xmath18 . but this is absurd , since @xmath133 is a probability . thus we can not have both a positive flux and a positive @xmath101 at large @xmath18 in steady - state , and it follows that @xmath134 and @xmath135 . all the flux generated at the source is accreted through the inner edge of the disk . we now discuss eq . ( [ eq : steadystate ] ) in full , emphasizing behavior at small and large @xmath18 . let @xmath136 so that @xmath137 $ ] is an integrating factor for eq . ( [ eq : steadystate ] ) . then @xmath138.\ ] ] as before , the upper ( lower ) sign on @xmath139 and @xmath140 applies to @xmath141 ( @xmath142 ) . when @xmath143 and @xmath22 are power laws , @xmath144 can be evaluated explicitly , but we will not write it out . normally the mean torque @xmath145 , so that @xmath146 . then we can choose the constant of integration in eq . ( [ eq : mu ] ) so that @xmath147 for @xmath148 and @xmath149 for @xmath150 . all disks of interest in this paper are dominated by the mean torque at small radii and by diffusion at large @xmath18 . hence @xmath151 $ ] becomes large at small @xmath18 , and so for a well - behaved solution , the contents of the square brackets in eq . ( [ eq : fss ] ) must tend to zero as @xmath152 . this defines a relationship between @xmath153 and @xmath154 . for @xmath141 , @xmath155 , so the integral increases at least as fast as @xmath156 . as before , if the disk is indefinitely extended then we must have @xmath157 to prevent @xmath158 from becoming negative at large @xmath18 . the main conclusion of this discussion is that _ as long as the mean torque is nonpositive in the outer disk , the flux of planets diffusing to infinite radius vanishes . _ we have reached this conclusion by considering steady solutions but will assume that it applies generally . the time - dependent problem is more difficult , but we can determine some general characteristics of the solution by analytic means , most importantly that _ the probability of surviving without accretion in an infinite steady disk , though declining monotonically with time , can have a long power law tail_. for the special case of the mmsn , where @xmath129 and @xmath159 , this probability declines only as @xmath160 . the asymptotic form of the solution at late times is self - similar except near @xmath161 and is described explicitly in dimensionless units by equation ( [ eq : ssim ] ) . the evolution of a planet created with @xmath162 at @xmath34 can be represented by solving eq . ( [ diffusion ] ) with a source term @xmath163 . the laplace transform @xmath164 reduces the time - dependent problem to @xmath165- \frac{\partial}{\partial j}\left[\bar\gamma(j)\hat f(s , j)\right ] -s\hat f(s , j)=-\delta(j - j_s).\ ] ] unfortunately , although eq . ( [ eq : odes ] ) is just an ordinary differential equation in @xmath18 , we can not solve it completely even for power law disks described by eq . ( [ eq : powers ] ) . the difficulty is in part that physically interesting cases have @xmath166 , which is precisely the condition for eq . ( [ eq : odes ] ) to have an irregular singular point at @xmath161 . however , a formal solution can be obtained in powers of @xmath167 , @xmath168 replacing @xmath169 by its taylor series in the laplace transform ( [ eq : laplace ] ) shows that @xmath170 is the @xmath171 moment of @xmath32 with respect to @xmath26 , @xmath172 more interesting than the moments of @xmath101 are the moments of the corresponding flux : @xmath173 the probability that the planet survives longer than @xmath26 before accretion , @xmath174 , is related to the flux at the origin by @xmath175 . therefore , @xmath176 is the @xmath171 moment of the lifetime , and in particular , @xmath177 is the mean lifetime . thus we are motivated to find the @xmath170 . substituting eq . ( [ eq : expans ] ) into eq . ( [ eq : odes ] ) leads to @xmath178''-[\bar\gamma(j)f_0(j)]'&=&-\delta(j - j_s ) , \\ \label{eq : recur } [ d(j)f_n(j)]''-[\bar\gamma(j)f_n(j)]'&=&-n f_{n-1}(j ) \qquad n>0,\end{aligned}\ ] ] where the primes indicate derivatives with respect to @xmath18 . the solution to eq . ( [ eq : f0 ] ) , which is formally identical to the steady - state case ( [ eq : steadystate ] ) , is a green s function for eq . ( [ eq : recur ] ) . it can be constructed from homogeneous solutions @xmath179 and @xmath180 with fluxes @xmath181 and @xmath182 , respectively : @xmath183'~-\bar\gamma(j)y_0(j ) & = & 0,\\ \label{eq : y1 } [ d(j)y_1(j)]'~-\bar\gamma(j)y_1(j ) & = & 1.\end{aligned}\ ] ] the green s function is @xmath184 the formal solution to the system ( [ eq : f0])-([eq : recur ] ) is then @xmath185 the existence of the moments ( [ eq : moment ] ) therefore depends upon the convergence of the integrals in ( [ eq : ssol ] ) . an important special case is the mmsn , where @xmath186 and @xmath187 . it is convenient to choose @xmath188 as the unit of angular momentum and @xmath189 as the unit of time so that @xmath190 and @xmath191 . the homogeneous solutions ( [ eq : y0 ] ) become @xmath192 & \approx & \cases { j^2-(3!!)j^4+(5!!)j^6-\ldots & $ j\to0^+$ \cr 1-\sqrt{\frac{\pi}{2}}\sum\limits_{n=0}^\infty\frac{1}{(2n)!!}j^{-2n-1 } ~+\sum\limits_{n=1}^\infty\frac{1}{(2n-1)!!}j^{-2n } & $ j\to\infty$. } \ ] ] the first expansion is only asymptotic , in keeping with the fact that @xmath161 is an irregular singular point of eq . ( [ eq : y1 ] ) , whereas the second converges for all @xmath193 . armed with these results , we can now determine the convergence of the integrals ( [ eq : ssol ] ) . because @xmath194 as @xmath195 , it follows from ( [ eq : green ] ) that the integrand for @xmath196 is proportional to @xmath197 as @xmath198 . thus the integral over @xmath199 is logarithmically divergent . consequently _ the mean lifetime defined by eq . ( [ eq : tmom ] ) is not finite_. since the divergence is only logarithmic , one may guess ( correctly ) that @xmath200 at late times in the infinite mmsn . the long tail in the distribution of lifetimes results from diffusion to large radii where torques are weak and planets linger . it is natural to seek a self - similar solution to eq . ( [ diffusion ] ) that reflects this behavior . unfortunately @xmath201 defines a preferred scale that prevents self - similarity . omitting @xmath11 from eq . ( [ diffusion ] ) on the grounds that this term is unimportant at @xmath202 removes the obstruction : @xmath203,\ ] ] in the present dimensionless units with @xmath204 as in the mmsn . the tilde distinguishes the solution to the approximate equation ( [ eq : nogamma ] ) from the exact solution @xmath32 of eq . ( [ diffusion ] ) . we require @xmath205 as @xmath195 . on the other hand , @xmath206 should be positive and finite at @xmath161 for the following reason . at late times such that @xmath207 , we expect @xmath32 to match onto a multiple of the constant - flux steady - state solution defined by eq . ( [ eq : y1mmsn ] ) : @xmath208 . we may replace @xmath101 with @xmath209 in this relation if @xmath202 , and in that region @xmath210 . ( [ eq : nogamma ] ) can be solved with these boundary conditions by separation of variables . and @xmath211 , which produces the bessel equation of order one on the righthand side . hankel transforms can then be used . ] for arbitrary positive initial conditions , @xmath212 as promised , this has self - similar form : @xmath18 enters only in the combination @xmath213 [ or @xmath214 in dimensionful units ] . it can be verified that eq . ( [ eq : ssim ] ) is an exact solution of eq . ( [ eq : nogamma ] ) , and its integral over all @xmath18 is @xmath215 , in agreement with our expectation for @xmath174 at late times . for other values of the indices @xmath102 & @xmath216 satisfying @xmath217 , we have the following situation . again , @xmath11 is negligible at large @xmath18 , so eqs . ( [ eq : y0 ] ) & ( [ eq : green ] ) yield @xmath218 at @xmath219 . therefore the integral ( [ eq : ssol ] ) for the first moment @xmath196 diverges if @xmath220 , and consequently , the mean lifetime converges only if @xmath221 . in other words , the mean lifetime exists only if @xmath222 increases more slowly than @xmath223 . the following technical result helps to interpret differences between the late - time behaviors of the planetary distribution in the mmsn and in alpha disks . the large-@xmath26 behavior of @xmath32 is dominated by the singularities of @xmath224 at largest @xmath225 . these correspond to eigenvalues @xmath226 obtained by separating variables in eq . ( [ diffusion ] ) with assumed time dependence @xmath227 . if the eigenvalues are real and discrete , @xmath228 , then @xmath229 at late times , where @xmath230 is the eigenfunction corresponding to @xmath59 . ( we assume that these modes are complete . ) on the other hand , if the eigenvalue spectrum is continuous and extends to arbitrarily small positive @xmath59 , then the late - time decay is slower than exponential , perhaps a power law . defining @xmath231 and setting @xmath232 and @xmath233 , with @xmath234 as in eq . ( [ eq : mu ] ) , one can re - cast the separated version of eq . ( [ diffusion ] ) in self - adjoint sturm - liouville form @xmath235~+\lambda w(j ) g_\lambda(j)=0.\ ] ] in all disks of interest to us , @xmath236 as @xmath152 and the integral of @xmath101 over @xmath18 is finite , whence @xmath237 and @xmath238 . a theorem of @xcite then asserts that if the limit @xmath239^{1/2 } \left[\int_{j}^\infty \frac{dx}{p(x)}\right]^{1/2}\ ] ] vanishes , then the spectrum is discrete and positive . [ the theorem also requires @xmath240 and @xmath241 to be positive and continuous , which is always true for us , and it encompasses `` quasilinear '' generalizations of eq . ( [ eq : sl ] ) . ] it is easy to show that in disks where @xmath242 tends to zero as @xmath152 and to infinity as @xmath195 , the limit ( [ eq : dk ] ) is equivalent to @xmath243^{1/2}.\ ] ] in the mmsn [ @xmath204 ] , this limit diverges , so a continuous spectrum extending to @xmath245 is allowed and indeed must exist , or else we would not have the self - similar behavior discussed above . in the alpha disks studied numerically in [ sec : numerical ] , on the other hand , it appears that @xmath246 as @xmath195 [ fig . ( [ fig:4dprop ] ) ] , so the limit vanishes and a discrete spectrum is implied . this is clearly true of power - law disks with @xmath127 and @xmath247 , so the survival probability @xmath174 should decay exponentially in these disks . we conjecture that the continuum exists and @xmath174 follows a power law for all @xmath248 , because further consideration of eqs . ( [ eq : ssol ] ) indicates that the @xmath171 moment of the lifetime exists only if @xmath249 . the diffusion equation ( [ diffusion ] ) with a source term reads : @xmath250 we find numerical solutions with an implementation of the generic relaxation code described in @xcite . the finite - difference scheme is fully implicit in time . given a distribution @xmath32 at time @xmath26 , we only need to solve the one - dimensional boundary - value problem in @xmath18 to find the solution at time @xmath251 . we thus relax successively from one solution to another over a variable timestep , @xmath252 . controls over the size of the timestep ensure that , at each grid point , the change in each variable over @xmath252 is within an acceptable tolerance . the system is described by two first - order differential equations equivalent to eq . ( [ fpsource ] ) , supplemented with a third differential equation that defines the numerical domain . for simplicity , we do not implement an adaptive mesh but use instead a fixed grid on the numerical domain , chosen to be uniformly spaced in @xmath253 . we take as variables the set @xmath254 . numerical stability of the solutions is achieved with centered - differencing schemes for @xmath101 and @xmath18 , and an upwind - differencing scheme for @xmath255 . three boundary conditions are required . the numerical domain extends from the inner edge of the disk , at @xmath256 , out to @xmath257 . at the inner boundary we trivially set the value of @xmath18 and are free to fix @xmath258 . at the outer boundary we demand the flux in steady - state , as discussed in [ subsec : steady ] , this is a boundary condition that must be satisfied if diffusion dominates at large radii . to avoid potential complications near the disk edge , we extend our numerical domain to sufficiently large radii that it does not influence our results . we have tested our numerical results for power - law models against the analytic steady - state solutions described in [ subsec : steady ] . the numerical solutions converge satisfactorily for a modest grid spacing ( typically 1000 grid points over five decades in radius ) . additional tests at ten times this resolution show clear numerical convergence of the steady - state results . agreement with the late - time behavior expected from the time - dependent self - similar solutions described in [ subsec : timedep ] will also be demonstrated below ( see fig . [ fig : mmsnsv ] ) . we study the combined effects of migration and diffusion on the orbital evolution of low - mass proto - planets for two separate classes of disk models : the minimum mass solar nebula ( mmsn ) and t tauri alpha disks . the mmsn model adopted here has a mean surface density @xmath112 and an aspect ratio @xmath259 , which are normalized to @xmath260 and 0.1 at @xmath261 , respectively . the t tauri alpha disk model adopted is identical to the irradiated disk described by ( * ? ? ? * see their fig . 1 ) , accreting steadily at a rate @xmath262 . we take as our reference model the alpha disk with viscosity parameter @xmath263 . for comparison , we also consider a more massive and cooler alpha disk , with @xmath264 . the inner edge of the disks is located at @xmath265 and @xmath266 in the mmsn and the alpha disks , respectively . the mass of the primary is @xmath267 in the mmsn and @xmath268 in the alpha disks . these slight model differences have very little influence on our results . note that , while the disk models used are assumed to be steady , we are studying the time - dependent orbital evolution of planets embedded in these disks with equation ( [ diffusion ] ) . as a consequence , our method can only apply to proto - planets of small enough mass that they exert a negligible feedback on the disk structure ( broadly speaking , @xmath3 ; e.g. * ? ? ? turbulence is assumed to be present everywhere in our disk models . the magnitude of the corresponding diffusion coefficient is uncertain through both the amplitude and the correlation time of the turbulent torque fluctuations . we quantify these uncertainties with a single parameter , @xmath269 . if we define @xmath270 then @xmath271 ^ 2}=\epsilon\,(2\pi)^3(0.046)^2\frac{\bar{\sigma}^2j^7}{g^2m_*^4m_p^5}.\ ] ] the calibration performed in [ subsec : calib ] suggests that @xmath272 is a reasonable value , which we have assumed to be independent of radius . however , we also investigate the effects of varying this parameter by several orders of magnitude . it is useful to discuss a few general scalings specific to our disk models before we discuss the details of time - dependent numerical solutions . to quantify the relative importance of diffusion _ vs. _ migration in the models , we define a local diffusion time @xmath273 ( for @xmath274 ) and a local migration time , @xmath275 according to eq . ( [ eq : meantorque ] ) , so that @xmath276 in the mmsn . the ratio of these timescales has a particularly simple dependence upon disk and planetary properties : @xmath277 one expects the evolution of planets located in regions of a disk with @xmath278 to be dominated by migration and vice - versa . while the diffusion time is independent of planet mass , the migration time is inversely proportional to @xmath13 . therefore , lower mass proto - planets will be more sensitive to the effects of turbulent torque fluctuations . in addition , the two timescales differ in their dependence on the mean surface density , @xmath279 , which tends to make planets in denser disks more susceptible to the effects of diffusion . as we shall now see , a general consequence of these scalings is that one expects the effects of diffusion to dominate only in the outer regions of proto - planetary disks ( at least for the classes of disk models considered here ) . figure [ fig:4dprop ] shows profiles of various important quantities in the mmsn and our two alpha disk models . panel i shows profiles of mean surface density , @xmath280 . both the slope and normalization of @xmath280 profiles matter for the relative importance of migration _ vs. _ diffusion . panels iii and iv show profiles of local migration and diffusion times in the various disk models . while @xmath281 increases linearly with @xmath18 ( @xmath282 ) in the mmsn , it decreases with increasing radius in the alpha disk models . this is a consequence of the steeper @xmath280 profile in the mmsn . the profiles of @xmath283 , derived from eq . ( [ eq : meantorque ] ) for the mmsn and from the detailed calculations of @xcite for the alpha disk models , show local migration times which decrease toward small radii in both classes of models . finally , panel ii shows the ratio of timescales , @xmath284 , which is important to characterize the behavior of planets embedded in these disks . as mentioned earlier , in all three models , this ratio grows above unity ( indicated by the horizontal dash - dotted line in panel ii ) only at radii @xmath285 if @xmath286 , indicating that diffusion should dominate over migration only in the outer regions of these disks ( and this despite the increasing value of @xmath281 with radius in the mmsn ) . although very useful , the profiles shown in figure [ fig:4dprop ] do not capture all of the complexity of the migration - diffusion problem . first , all the quantities plotted in panels ii - iii - iv were derived for a planet mass @xmath287 and a torque fluctuation normalization @xmath288 . the profiles are easily rescaled for different values of these parameters ( as indicated by labels in fig . [ fig:4dprop ] ) , but it is clear that the exact location of the transition from migration- to diffusion - dominated evolution depends on the proto - planetary disk model adopted , the value of the planet mass , and the normalization @xmath269 of the diffusion coefficient . nevertheless , the dominance of diffusion over migration in the outer regions of these disks is robust because @xmath289 tends to increase with radius in disk models of interest [ eq . ( [ eq : migvsdiff ] ) ] . secondly , as shown in panel iii , in a disk model with realistic opacities , sudden and non - monotonic variations in the value of @xmath283 at radii @xmath290 could complicate the orbital evolution of planets substantially , by stalling their migration as they reach the disk inner regions . this is best explored with global numerical solutions of the time - dependent fokker - planck equation , as seen in [ subsec : burst ] . a third complication arises from the radial dependence of the diffusion driven by turbulent torque fluctuations . as mentioned in [ sec : fp ] , the radial variation of the diffusion coefficient leads to an additional contribution to the advected flux of planets . since @xmath291 for @xmath274 , and since @xmath292 in the mmsn and @xmath293 in the alpha disk models , this additional term tends to advect planets inward in all the disk models considered . we can define a local advection timescale associated with this new contribution , @xmath294 , and compare it to the values of @xmath283 and @xmath281 defined above . we find that this advection timescale is everywhere exactly equal to @xmath281 in the mmsn , while it is shorter than @xmath281 everywhere in our alpha disk models ( as shown explicitly for the @xmath263 case by the short - dashed line in panel iv of fig . [ fig:4dprop ] ) . consequently , even in disk regions where diffusion nominally dominates over migration in our models ( as measured by the ratio @xmath295 ) , this additional contribution due to radially inhomogeneous diffusion should still cause a rather significant effective advection of planets towards the star . we have studied two different types of time - dependent models for the diffusive migration of planets in proto - planetary disks , corresponding approximately to a continuous and a burst scenario for planet formation . in both cases , for the sake of clarity , the source of planets was localized , taking the form of a sharply peaked gaussian approximating the delta source function of eq . ( [ eq : sourceterm ] ) . we have verified that our results do not depend sensitively on the detailed shape of the gaussian source function adopted . as we shall see , the predictions for these rather different scenarios for planet formation are broadly consistent with each other . figure [ fig:4steady ] shows steady - state distributions obtained by numerically solving eq . ( [ fpsource ] ) with a source term centered at @xmath296 , steadily producing earth - mass planets at a rate @xmath297 . the topmost curve in each panel is the steady - state distribution . the enclosed curves , on the other hand , are snapshots of the decaying distributions at selected time intervals after the source is shut off . we have verified that the steady - state distributions for the mmsn in panels i and iii are fully consistent with the analytic solution , eqs . ( [ eq : mu])-([eq : fss ] ) , for power - law disks . a fiducial normalization of @xmath288 for torque fluctuations is assumed in all cases except in panel iii , where @xmath298 has been scaled down to extend the migration - dominated regime out to @xmath299 in the mmsn . thus , at @xmath300 , the disks in panels i and iv are diffusion - dominated and the disks in panels ii and iii are migration - dominated . the distinction between diffusion - dominated and migration - dominated evolutions is most clearly illustrated by comparing panels i and iii . while the initial steady - state peaked distribution is eroded and advected for the most part in panel iii , with little diffusion at large radii , there is significantly more diffusion at large radii in panel i. that distinction is not as clear in the alpha disk models , however . note that , in both panels ii and iv , the initial steady - state distribution is strongly peaked not at the source radius , but at radii where migration stalls ( as seen from panel iii in fig . [ fig:4dprop ] ) . in addition , even though diffusion is more important in the model with @xmath264 in panel iv , the effect appears to be minor in smoothing out only slightly the same sharp distribution features as in panel ii . this is partly explained by the strong contribution of radially inhomogeneous diffusion to inward advection of planets , which is always important in our alpha disk models ( as explained in the previous subsection ) . even though diffusion is equally important at @xmath296 in the models of panel i and iv ( see @xmath301 in fig . [ fig:4dprop ] ) , it is more inhomogeneous in the alpha disk model and thus promotes a significantly more efficient inward advection of planets . our models with a burst of planet formation confirm these general trends . to study this formation scenario , we solve the diffusion equation ( [ diffusion ] ) without source term and impose a specific initial condition for the distribution of planets at @xmath34 instead . the initial distribution , @xmath302 , is modeled as a narrow gaussian centered on @xmath303 . figure [ fig:4spike ] shows the evolution of such distributions of earth - mass planets initially located at @xmath296 in the same four disk models as in fig . [ fig:4steady ] . as before , a comparison between the two mmsn models shows that the effects of diffusion are much more important in the model with @xmath304 ( panel i ) than in the one with @xmath298 ( panel iii ) . in particular , a sub - population of planets diffusing at large radii is clearly visible in panel i. on the contrary , while diffusion still acts to widen the initial distribution in panel iii , the dominant effect is a global advection of the distribution of planets . finally , in agreement with our previous results , figure [ fig:4spike ] shows that the most significant feature of alpha disk models is the effect of rapid variations in the migration rates of planets , which causes them to stall at specific locations in the disk and results in time - varying peaks in the distributions of panels ii and iv . even though diffusion is comparatively as important in panels i and iv , at @xmath300 , its effects appear to be much less important in the alpha disk model of panel iv . we attribute this difference , once again , to the more inhomogeneous diffusion in alpha disk models , which contributes substantially to the inward advection of planets into regions of the disk where diffusion becomes gradually less important and the effects of stalling radii are felt more strongly . two important aspects of the orbital evolution of planets which are not well captured by figure [ fig:4spike ] are the properties of the surviving minority of planets at late times and large radii , and the representative planetary lifetimes in burst - formation models . the fraction of planets that diffuse to orbital radii greater than their formation radius , @xmath305 , can be characterized by the time - dependent quantity @xmath306 where @xmath307 is the angular momentum at a fiducial radius @xmath308>r_s$ ] . hereafter , the integral over the initial distribution appearing in the denominator of eq . ( [ eq : chia ] ) will be normalized to unity . the maximum values of @xmath309 and @xmath310 , reached typically just after the majority of planets were accreted by the star , are listed in table [ tb : chia ] for the four models of figure [ fig:4spike ] . not surprisingly , models with a strong role for diffusion tend to have a fairly large maximum fraction of planets diffusing beyond @xmath311 ( @xmath312 ) and @xmath313 ( @xmath314 ) , while models with little diffusion have much reduced maximum fractions of planets at large radii ( @xmath315 beyond @xmath311 , @xmath316 beyond @xmath313 ) . lccc disk model & torque fluctuations & @xmath317 & @xmath318 + mmsn & @xmath288 & @xmath319 & @xmath320 + alpha - disk ( @xmath263 ) & @xmath288 & @xmath321 & @xmath322 + mmsn & @xmath323 & @xmath324 & @xmath325 + alpha - disk ( @xmath264 ) & @xmath288 & @xmath326 & @xmath327 + we compute three additional time - dependent quantities to help us better characterize the properties of surviving planets in our models . the first is the survival probability , computed as @xmath328 the second is the median orbital radius of surviving planets , @xmath329 , defined by @xmath330 finally , since the mean lifetime is often ill - defined ( [ subsec : timedep ] ) , we calculate the median lifetime @xmath331 defined by @xmath332 . figure [ fig : mmsnsv ] shows how the survival probability and the median radius of surviving planets evolve with time in burst formation models for earth - mass planets in the mmsn . three separate models with burst formation at @xmath333 , @xmath334 and @xmath313 are shown for comparison ( solid lines ) . note that the middle solid line corresponds to the exact same model as the one shown in panel i of figure [ fig:4spike ] . for reference , vertical dotted lines bracket typical lifetimes for gaseous proto - planetary disks ( @xmath335@xmath1 yrs ; e.g. * ? ? ? the median lifetime of the population of planets , which is such that @xmath336 , is indicated by a horizontal dash - dotted line in the upper panel . the lifetimes differ substantially in the three models shown , from much shorter to somewhat longer than the typical disk lifetimes . the self - similar @xmath337 behavior predicted in [ subsec : timedep ] is indeed achieved at late times in the three models . as shown in the lower panel of figure [ fig : mmsnsv ] , the median radius of surviving planets does not evolve much at early times and remains close to the formation radius , as the distribution of planets mostly diffuses out to both small and large radii . once a significant fraction of planets have been accreted by the central star , however , the surviving planets are found at a median radius increasing almost quadratically with time independently of where they were formed . this is just what one would expect from the self - similar solution ( [ eq : ssim ] ) : @xmath338 . figure [ fig : alpsv ] shows the evolution of the same two quantities in burst formation models for earth - mass planets in the alpha disk ( with @xmath263 ) this time . six separate models with burst formation at @xmath333 , @xmath339 , @xmath334 , @xmath340 , @xmath341 and @xmath342 are shown for comparison ( solid lines ) . note that the fourth solid line from the top corresponds to the exact same model as the one shown in panel ii of figure [ fig:4spike ] . the results are qualitatively consistent with the mmsn ones , with a few important differences . there is no self - similar behavior at late times in this case and the more dominant role of advection in the alpha disk model tends to make the median radius of surviving planets move inward initially , until the majority of them are lost to the central star . among the surviving minority , however , @xmath343 asymptotes to @xmath344 at late times , independently of the formation radius , @xmath305 . this behavior is explicable if , as predicted by the theorem cited in [ subsubsec : eig ] , the eigenmodes of eq . ( [ diffusion ] ) are discrete : then @xmath32 converges upon the most slowly decaying mode , and @xmath329 tends to the median radius of this eigenmode . figures [ fig:4dprop ] and [ fig : alpsv ] suggest that this radius is close to that at which @xmath345 [ more accurately , @xmath346 , where @xmath347 is minimized , as might be expected since planets should linger longest where the combined effects of diffusion and migration are weakest . in the mmsn , both timescales increase without limit as @xmath195 , and the spectrum of eigenmodes is apparently continuous . it is also worth emphasizing that the largest peak value of @xmath348 at sub - au distances in alpha disk models ( see panel iii in fig . [ fig:4dprop ] ) effectively limits the migration time of earth - mass planets into their star to values @xmath349 yrs ( for @xmath263 ) , which exceeds typical disk lifetimes . based on this general understanding of diffusive migration in our burst - formation models , let us now focus more systematically on several important model dependencies . figure [ fig : medlrs ] illustrates how the median lifetimes of earth - mass planets formed at different radii in the mmsn and the alpha disk are affected by changes in the normalization of torque fluctuations , @xmath269 . for the nominal @xmath304 value adopted ( indicated by a vertical dash - dotted line in fig . [ fig : medlrs ] ) , diffusion mostly affects the orbital evolution of planets formed at the largest radii , by reducing somewhat their median lifetime . as @xmath269 is increased above this nominal value , the trend becomes clearer . median lifetimes for all formation radii are considerably reduced at larger values of @xmath269 , more so in the mmsn than in the alpha disk . the long - dashed line in figure [ fig : medlrs ] results when we neglect the mean torque @xmath11 in the governing equation ( [ diffusion ] ) . the convergence of the solid curves for the alpha - disk model to a value independent of @xmath305 is related to the value of the diffusion time , @xmath222 , at the disk inner boundary . for sufficiently large @xmath269 , @xmath350 everywhere on the grid . but @xmath222 increases toward small radius in the alpha disks ( fig . [ fig:4dprop ] ) . so the median lifetime in the large-@xmath269 regime is comparable to @xmath222 at the inner boundary , @xmath351 ; we have verified that @xmath352 decreases as @xmath351 is increased . nevertheless , in all cases , the effect of diffusion is to reduce the median lifetime . figure [ fig : medlmp ] illustrates how the median lifetimes of planets of various masses , all formed at @xmath296 in the mmsn and in the alpha disk , are affected by changes in the normalization of torque fluctuations , @xmath269 . for the nominal @xmath304 value , diffusion mostly affects the orbital evolution of the lower mass planets , and reduces somewhat their median lifetimes . this results from the larger values of @xmath348 at lower planet mass , whereas @xmath222 is independent of planet mass . as @xmath269 is increased , diffusion progressively dominates and the distinction among planets of various masses gradually disappears . for large enough values of @xmath269 , the median lifetime of all planets with masses @xmath353 formed at @xmath354 is reduced below typical disk lifetimes . as in the previous figure , however , the lifetimes for @xmath355 in the alpha disk are dominated by the diffusion time at the inner boundary . as anticipated by @xcite , @xcite , and @xcite , our fokker - planck treatment confirms that diffusion driven by turbulent torque fluctuations in proto - planetary disks can greatly influence the orbital evolution of embedded low - mass proto - planets . contrary to some expectations , diffusion does not promote planetary survival but instead systematically reduces the lifetime of most planets in the disk . however , it does help a small fraction of planets diffuse to large radii where they can survive for extended periods . ( we consider that a planet `` survives '' if its orbit remains larger than the inner edge of the disk . ) our results point to a potentially important role for non - deterministic effects in planet formation scenarios . in models with a significant level of diffusion , most proto - planets in the earth - mass range could end up being accreted by their stars , while only a small fraction of them would survive by diffusing to large radii . in scenarios where all proto - planetary disks are efficient at forming planets , this could be interpreted as leading to only a fraction of all potential host stars being successful in keeping a system of surviving planets . even systems starting with very similar initial conditions may end up with very different planetary orbital configurations once their gaseous disks disappear . neglecting the effects of direct gravitational interactions between proto - planets , which are likely to happen from differential rates of migration and diffusion , is one of many model limitations in our work . we have already mentioned that our study is restricted to proto - planets of low enough masses that they do not affect in any way the structure of their host disk . this excludes the late stages of giant planet formation , which are obviously of considerable interest . in particular , it has been suggested that migration and diffusion , each independently , could accelerate the rate of growth of cores until they reach the critical mass at which runaway envelope accretion occurs ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? studying these possibilities within the context of global models such as ours would be interesting . even within the strict regime of applicability of our models , significant uncertainties exist . let us mention a few . we have focused our work on strictly circular orbits . we have assumed that the underlying proto - planetary disk structure does not evolve with time , and that the amplitude @xmath81 of surface - density fluctuations is constant with radius . we have assumed that torque fluctuations can simply be added to a laminar mean torque which is not modified by the turbulence . the laminar torque itself is calculated from a simple extension of the two - dimensional lindblad torque theory , which does not account for the role of corotation torques . despite all these uncertainties , we believe that our results should be broadly valid , provided that turbulence and torque fluctuations in proto - planetary disks satisfy the various stochasticity criteria enunciated in [ sec : fp ] . understanding better the nature of turbulence in proto - planetary disks and in particular indications for long - term correlations in mhd simulations may turn out to be crucial in this context . we would like to issue a plea to the simulation community to study these questions , and in particular , whether the dimensionless amplitude and correlation time of density and torque fluctuations depend only upon the strength of the turbulence ( _ via _ @xmath102 ) or upon the thickness @xmath103 as well ( [ subsec : calib ] ) . the answers will matter both quantitatively ( in determining the relative importance of migration _ vs. _ diffusion in global models such as ours ) and qualitatively ( in justifying a fokker - planck approach or in falsifying it ) . we have seen that details of the protoplanetary disk structure ( mmsn _ vs. _ alpha disks ) do influence our results at the quantitative level . in that respect , the possibility that a magnetically layered structure with quiescent dead zones exists in weakly - ionized regions of proto - planetary disks @xcite may be important . indeed , the results of @xcite suggest that the nature of turbulence in dead zones differs qualitatively from that in mhd turbulent regions , perhaps leading to comparatively reduced torque fluctuations ( see also * ? ? ? * for other suggested effects of dead zones in planet formation scenarios ) . the general tendency for a fraction of planets to diffuse to large radii in proto - planetary disks may have consequences of some observational relevance . we note , for instance , that @xcite have looked into scenarios for outward planet migration in an attempt to explain observed structures in dust disks as resulting from perturbations by planets located at radii beyond those at which they are thought to be able to form . rather than invoking outward migration , this may be expected in scenarios with a prominent role for diffusion , as illustrated by our various models . we thank richard nelson for help with understanding his simulations . this material is based on work supported in part by the national aeronautics and space administration under grant nag5 - 11664 issued through the office of space science . alibert , y. , mordasini , c. , benz , w. & winisdoerffer , c. 2005a , , 434 , 343 alibert , y. , mousis , o. , mordasini , c. & benz , w. 2005b , , 626 , l57 armitage , p. j. & rice , w. k. m. ( 2006 ) , to appear in a decade of extrasolar planets around normal stars ( astro - ph/0507492 ) balbus , s. a. & hawley , j. f. 1998 , rev . phys . , 70 , 1 beaulieu , j .- 2006 , nature , 439 , 437 bender , c. m. & orszag , s. a. 1978 , advanced mathematical methods for scientists and engineers ( new york : mcgraw - hill ) brandenburg , a. , nordlund , a. , stein , r. f. , & torkelsson , u. 1995 , apj , 446 , 741 dangelo , g. , kley , w. & henning , t. 2003 , apj , 586 , 540 drbek , p. & kufner , a. , 2005 , proc . soc . , 134 ( 1 ) , 235 . fleming , t. & stone , j. m. 2003 , apj , 585 , 908 fromang , s. & nelson , r. p. 2005 , , 364 , l81 fromang , s. , terquem , c. & balbus , s. a. 2002 , mnras , 329 , 18 gammie , c. f. 1996 , apj , 457 , 355 gilliland , r. l. et al . 2000 , , 545 , l47 glassgold , a. e. , najita , j. & igea , j. 1997 , apj , 480 , 344 [ erratum : 485 , 920 ] goldreich , p. & tremaine , s. 1980 , , 241 , 425 goldreich , p. & sari , r. 2003 , , 585 , 1024 haisch , k. e. , lada , e. a. & lada , c. j. 2001 , , 553 , l153 hameury , j .- m . , menou , k. , dubus , g. , lasota , j .- , & hur , j .- m . 1998 , , 298 , 1048 hawley , j. f. , gammie , c. f. , & balbus , s. a. 1995 , , 440 , 742 hayashi , c. 1981 , prog . theor . , 70 , 35 ilgner , m. & nelson , r. p. 2006 , , 445 , 205 johansen , a. & klahr , h. 2005 , , 634 , 1353 kley , w. , dangelo , g. & henning , t. 2001 , apj , 547 , 457 korycansky , d. g. & pollack , j. b. 1993 , icarus , 102 , 150 laughlin , g. , steinacker , a. , & adams , f. c. 2004 , , 608 , 489 lin , d. n. c. , bodenheimer , p. & richardson , d. c. 1996 , nature , 380 , 606 lin , d. n. c. & papaloizou , j. 1986 , apj , 309 , 846 lufkin , g. et al . 2004 , mnras , 347 , 421 marcy , g. w. , cochran , w. d. & mayor , m. 2000 , in protostars and planets iv ( tucson : university of arizona press ) , eds . mannings , v. , boss , a.p . , russell , s. s. ) , p. 1285 matsumura , s. & pudritz , r. e. 2003 , , 598 , 645 menou , k. & goodman , j. 2004 , , 606 , 520 mizuno , h. 1980 , prog . of theor . 64 , 544 nelson , r. p. 2005 , , 443 , 1067 ( n2005 ) nelson , a. f. & benz , w. 2003a , apj , 589 , 556 nelson , a. f. & benz , w. 2003b , apj , 589 , 578 nelson , r. p. & papaloizou , j. c. b. 2004 , , 350 , 849 papaloizou , j. c. b. & terquem , c. 1999 , apj , 521 , 823 papaloizou , j. c. b. & terquem , c. 2006 , rep . 69 , 119 pollack , j. b. et al . 1996 , icarus , 124 , 62 rice , w. k. m. & armitage , p. j. 2003 , , 598 , l55 safronov , v. s. 1969 , evoliutsiia doplanetnogo oblaka i obrazovanie zemli i planet ( moscow : nauka ) ( english transl . evolution of the protoplanetary cloud and formation of the earth and planets [ nasa tech . f-677 ] [ jerusalem : israel program sci . [ 1972 ] ) sano , t. , inutsuka , s. , turner , n. , & stone , j.m . 2003 , , 605 , 321 . santos n. c. , israelian g. , mayor m. 2004 , , 415 , 1153 sato , b. et al . 2005 , , 633 , 465 schaefer , c. , speith , r. , hipp , m. , kley , w. 2004 , , 418 , 325 van kampen , n. g. 1981 , stochastic processes in physics and chemistry ( amsterdam : north - holland ) veras , d. & armitage , p. 2004 , , 347 , 613 ward , w. r. 1997a , icarus , 126 , 261 ward , w. r. 1997b , apj , 482 , l211 weldrake , d .t . f. , sackett , p. d. , bridges , t. j. & freeman , k. c. 2005 , , 620 , 1043 wuchterl , g. , guillot , t. & lissauer , j. j. 2000 , in protostars and planets iv ( tucson : university of arizona press ) eds . mannings , v. , boss , a.p . , russell , s. s. , p. 1081

one of the most important aspects of string theory with potential applications to beyond - the - standard - model physics and cosmology is the existence of additional extended objects . a particularly interesting class of these , @xmath2-branes , occur as @xmath3-dimensional surfaces arising as supersymmetric solutions of the low - energy supergravity field equations but have also a high - energy interpretation as surfaces confining the endpoints of open strings @xcite . by now many different configurations of @xmath2-branes have been studied , including parallel and intersecting sets of branes @xcite , the combination of branes with different dimensionalities @xcite and also several possible embeddings in the compactified extra - dimensions @xcite . these models have an enormous potential in the construction of gauge theories with matter and several new extensions of the standard model have been obtained in this way @xcite . branes have also been intensively studied in their application to the problems of early universe cosmology , such as inflation and reheating . in the context of effective field theory treatments of brane - world model building , an immense amount of work has been performed on the astrophysical and cosmological implications of these theories for both flat and warped extra dimensions @xcite , while many studies in string theory have also been undertaken @xcite . one important property of @xmath2-branes is the fact that they interact at large distances via the exchange of closed string modes , which allows one to study them as massive point - like charged objects moving along the dimensions transverse to their world - volumes . an interesting application of this was proposed in @xcite , where it was shown that branes with opposite charges may form bound states similar to electron - positron states , despite the different origin of the central potentials involved . this analogy has motivated the name of _ branonium _ to designate this type of systems . several properties of these bound states have been analyzed in @xcite , including the classical orbits , the associated quantum dynamics and possible compactification schemes . this analysis assumed a probe ( anti)brane moving in the background spacetime created by a fixed central stack of branes , thus neglecting the gravitational backreaction of the probe . in the case of @xmath1-branes , the probe s trajectory follows closed elliptical orbits similar to those found in planetary systems . it was also suggested that the probe s motion would lead to radiation of particles into both the bulk space and the probe s world - volume , making otherwise stable orbits decay . the stability of this system was further studied in @xcite , where it was shown that orbital decay necessarily occurs when geometric moduli of the background spacetime are allowed to vary . in this work , we explore some properties of branonium systems , namely the mechanism leading to particle production in the world - volume of the probe brane . for the particular case of a scalar field living in the world - volume of a probe @xmath1-brane , we show that particle production may occur in a regime of parametric resonance similar to the one behind the phenomenon of preheating after inflation @xcite , leading to an exponential growth of the particle number in the probe s world - volume . we determine how the orbital parameters of the system constrain this mechanism and analyze the possible properties of the resonantly produced particles . also , we study how the probe s trajectory is modified by compactifying three of the dimensions parallel to the branes , so that a 4-dimensional spacetime with a flat friedmann - robertson - walker geometry is obtained . in this case , we analyze the effects of hubble expansion on the orbital parameters and how it modifies the parametric resonance regime . other instability sources such as radiation into bulk modes are discussed as well . we also consider the effects of transverse space compactification on the probe s motion in the case of a compact 3-torus and analyze the associated production of angular momentum . finally , we explore some of the cosmological applications of the resonant particle production mechanism and relate them to possible final configurations of the system , including stable states or brane - antibrane annihilation . in the next section , we begin by introducing some of the general properties of @xmath4 branonia , with particular emphasis on the @xmath5 case , and discuss the effects of hubble expansion on the orbital parameters . in section iii , we study the evolution of scalar particle - modes confined to the probe brane and analyze the properties of the associated parametric resonance mechanism in both the non - expanding and expanding universe cases . we distinguish between the production of massless and massive particles and determine the significance of energy damping into these modes on the probe s motion . we also consider , in this section , the effects of radiation into bulk modes on the resonant production of brane - bound particles . in section iv , we analyze how compactifying the directions transverse to the branes modifies the probe s trajectory and dynamically generates the required angular momentum . possible final states of the system are described in section v , where we also analyze the relevance of the particle production mechanism to cosmology in some particular scenarios . we summarize the main results and conclusions of this work in section vi . branonium corresponds , in its simplest form proposed in @xcite , to a system whose components are a fixed stack of @xmath6 parallel @xmath7-branes and a probe @xmath7-brane , parallel to the stack , whose gravitational backreaction is neglected so that it has no influence on the geometry of the background spacetime . let the coordinates @xmath8 , @xmath9 , correspond to the dimensions parallel to the stack and the coordinates @xmath10 , @xmath11 , correspond to those transverse to the branes . then , the general solution of the supergravity equations of motion in @xmath12 dimensions for @xmath6 parallel @xmath7-branes is given , in the einstein frame , by @xcite : [ eq1 ] ds^2&=&h^-dx_p^2+h^dy^2 , + e^&=&h^ , c_01p=(1-h^-1 ) , which give the @xmath12-dimensional metric , the dilaton @xmath13 and the non - vanishing components of the bulk @xmath3-form field @xmath14}$ ] which couples to the @xmath3-dimensional world - volume of the stack of branes . the harmonic function @xmath15 is a function of the radial coordinate in transverse space , @xmath16 , and has the form : [ eq2 ] h(r)=1+q_pr^ , where @xmath17 and the factor of @xmath18 ensures that the solution is asymptotically the @xmath12-dimensional minkowski space . for the cases we will be interested in studying , the transverse space is at least 2-dimensional , so that @xmath19 . the stack of branes is , thus , located at the origin in the transverse space . the exponents @xmath20 , @xmath21 , @xmath22 and the constant @xmath23 that characterize the solution are given by : [ eq3 ] & = & 4d(d-2 ) , = 4(d-2 ) , + & = & 2 , = 2 , where @xmath24 and [ eq4 ] = _ n^2 + 2dd-2 . the constant @xmath25 defines the coupling of the @xmath26-form field strength to the dilaton in the supergravity action . we will be interested in the particular case of @xmath2-branes in type ii supergravities , which arise as the low energy limit of type ii string theories , so that the @xmath27-form field strength corresponds to a closed string state in the ramond - ramond sector , and @xmath28 . taking @xmath29 , we have @xmath30 , @xmath31 , @xmath32 , @xmath33 , @xmath34 and @xmath35 . it is clear that the condition @xmath19 implies @xmath36 . the constant @xmath37 is related to the charge carried by the stack of @xmath6 branes , and for the case of @xmath2-branes we have [ eq5 ] q_p = c_pg_snl_s^ , where @xmath38 , @xmath39 is the string length and @xmath40 is the string coupling constant . this solution holds in the limit of validity of string perturbation theory and for small curvature , which allows a low energy description of the system in terms of supergravity fields . this approximation is valid as long as the local string coupling , @xmath41 , is everywhere small . also , we need to require the radial coordinate @xmath42 , which will give the distance between the probe and the stack branes , to be large compared to the string length @xmath39 , so that mediation by supergravity bulk fields ( graviton , dilaton and rr - forms ) is the dominant source of brane interactions . before analyzing the motion of the probe brane , we need to consider possible compactifications of the 10 dimensional spacetime that yield our 4-dimensional world at low energies . compactification of dimensions parallel and transverse to the world volume of the branes involve different procedures . in the case of dimensions parallel to the branes , it suffices to consider the kaluza - klein _ ansatz_. if we start with the @xmath12-dimensional spacetime eq . ( [ eq1 ] ) and compactify along a periodic coordinate @xmath43 with radius @xmath44 , so that the branes are wrapped around this dimension , we obtain a @xmath45-dimensional spacetime with metric given by @xcite : [ eq6 ] ds_d^2&=&e^2ds_d^2+e^2(dz+b_dx^)^2 , + ds_d^2&=&h^-dx_p^2+h^dy^2 , + e^&=&h^ , where [ eq7 ] & = & 12(d-1)(d-2 ) , = -(d-2 ) , + & = & 2(d-2 ) , + &=&-2 , =+2 . the dilaton field @xmath13 is not altered by this compactification , while the @xmath46-form rr field can be decomposed as follows : [ eq8 ] c_[n-1]=b_[n-1]+b_[n-2]dz . we take the following truncation of these fields : [ eq9 ] b_[n-1]=0 , ( b_[n-2])_01p-1=(1-h^-1 ) . this truncation leads to a @xmath47-brane supergravity solution in @xmath48 dimensions , characterized by the same harmonic function of the transverse space radial coordinate @xmath15 . applying this _ toroidal compactification _ procedure @xmath49 times we obtain @xmath50-brane solutions living in a @xmath51-dimensional spacetime , with the number of transverse dimensions remaining the same . the process of compactifying the dimensions transverse to the branes is not as straightforward due to the lack of translational invariance which results from the presence of the branes . in particular , to make a given transverse coordinate @xmath52 periodic , so that a similar kaluza - klein mechanism can be applied , one needs to include the appropriate image branes " of the source branes which will ensure the invariance of the configuration under @xmath53 , @xmath54 being the radius of the compact dimension . for example , if a source brane is located at @xmath55 , an image brane has to be considered at @xmath56 . this implies a generalization of the harmonic function in eq . ( [ eq2 ] ) to include branes located at different points in transverse space @xcite : [ eq10 ] h()=1+_iq_p|-|^ , where the vectors @xmath57 denote the positions of the different image branes and @xmath58 . these two compactification schemes allow us to reduce a 10-dimensional solution of the supergravity field equations to a 4-dimensional spacetime which resembles our world . we will be interested in studying a particular case of the general branonium configuration described so far where we have a source stack of d6-branes and also a probe @xmath1 brane . as was shown in @xcite , when the probe carries an antibrane charge , i.e. opposite to the charge of the source branes , the system yields closed elliptical orbits which can be solved analytically , while for other values of @xmath36 the orbits fail to close . this feature has , as we will describe later on , important consequences for particle production in the world - volume of the probe . compactifying 3 of the dimensions parallel to the branes , we obtain effective d3-branes moving in a 3-dimensional transverse space , the shape of the orbits not being affected by this toroidal compactification scheme as mentioned before . if the transverse dimensions are finite but their typical size is much larger than that of the compact parallel dimensions and the interbrane distance , we may consider them to be infinite and neglect the brane images described before . although this would correspond to a 7-dimensional spacetime , from the point of view of fields confined to the world - volume of the source and probe branes it is effectively 4-dimensional . the dynamics in the transverse space will , nevertheless , affect the 4-dimensional dynamics of the fields . in order to make the system more similar to the observable universe , we consider a modification of the solution eq . ( 1 ) so that , after compactification of 3 of the parallel dimensions , the 4-dimensional world - volume of the branes has a flat friedmann - robertson - walker ( frw ) geometry . this is a particular case of the the system studied in @xcite , where the moduli governing the size of both parallel and transverse dimensions were allowed to vary . in this work , we will assume that some dynamical mechanism at the string scale ( see , e.g. @xcite ) or at some other high - energy scale ( such as susy breaking ) fixes all moduli except the scale factor associated with the 3 non - compact dimensions parallel to the branes . hence , in the einstein frame , the complete 7-dimensional line element is given by : [ eq11 ] ds_7 ^ 2=h^-_4[-dt^2+a^2_ijdx^idx^j]+h^_4_mndy^mdy^n , where @xmath59 , @xmath60 and @xmath61 . the harmonic function is in this case @xmath62 . the exponents @xmath63 and @xmath64 can be obtained using eq . ( [ eq3 ] ) and applying eq . ( [ eq7 ] ) for the compactification on a 3-torus , but their precise values will not be necessary in the subsequent calculations . the motion of the probe brane is described by its action , which is divided in two parts . the born - infeld contribution , which concerns the geometry of the world volume and possible gauge fields living on it , is in the string frame given by , for a generic @xmath7-brane probe @xcite : [ eq12 ] s_bi =- t_pd^p+1 e^- , where @xmath65 are world - volume coordinates , @xmath66 gives the @xmath7-brane tension and @xmath67 is the induced metric on the probe . the hatted metric tensor is defined in the string frame , and is related to the einstein frame one via @xmath68 with @xmath69 in the general case . the antisymmetric field strength tensor @xmath70 refers to gauge fields describing open string modes with endpoints on the probe ( in the case of dirichlet branes ) , but in what follows we will set these fields to zero . the wess - zumino part of the action describes the minimal coupling of the ramond - ramond @xmath3-form to the @xmath3-world volume of the probe brane and can be written as @xcite : [ eq13 ] s_wz =- qt_pc_[p+1 ] . the constant @xmath71 gives the sign of the charge of the probe brane , which being a bps saturated state is equal or opposite to the brane tension @xmath66 , so that @xmath72 corresponds to a brane and @xmath73 corresponds to an antibrane @xcite . if we now consider the particular case of d6 branes in 10 dimensions , compactifying 3 of the parallel dimensions and including the scale factor describing the expansion ( or contraction ) of the 3 remaining non - compact dimensions along the brane , and choose the world - volume coordinates such that @xmath74 , we arrive at the following expression for the total action of the probe brane : [ eq14 ] s_4=-t_6v_3d^4x a^3(t ) , where @xmath75 gives the volume of the compact 3-torus around which the d6-branes are wrapped to give the effective 4-dimensional space . we have assumed that the transverse space coordinates are exclusively time - dependent , @xmath76 and defined the velocity of the probe as @xmath77 , with dots denoting time derivatives . as mentioned before , the bound orbits of the probe brane can be exactly solved for the case where the scale factor is constant and it was shown in @xcite that they correspond to closed ellipses . we now wish to analyze the effects of the varying scale factor on these orbits . in this case , a full analytical solution is hard to obtain and it is simpler to consider the evolution of the system when the probe brane moves at larges distances from the stack with small velocities . these two conditions can be expressed as @xmath78 . one must recall that we considered initially that the transverse dimensions can be considered infinite in extent . hence , our analysis will hold as long as the radial coordinate is large compared to the typical length scale of the harmonic function , given by @xmath79 ( assumed larger than the string scale and the compactification scale ) , but still much smaller than the typical size of the transverse dimensions ( assumed much larger than @xmath79 ) . with this hierarchy in mind , we may expand the probe brane s action to obtain , to lowest order , [ eq15 ] s_4-t_6v_3d^4x a^3(t ) . our subsequent analysis will not be affected by the constant term and we may drop it . for a probe antibrane , [ eq16 ] s_4t_6v_3d^4x a^3(t ) . the rotational symmetry of the problem implies that the probe s trajectory will be confined to a plane , which we may choose to be @xmath80 . then , we may define the canonically normalized complex scalar field [ eq17 ] ( y^1+iy^2 ) , so that its action is , in the non - relativistic and large distance limit , [ eq18 ] s_=d^4x a^3(t)(-g^__^*-v ( ) ) , where @xmath81 is the 4-dimensional flat frw metric of the spacetime where the field is defined and the potential is given by : [ eq19 ] v()=2t_6v_3h^-12t_6v_3(1- ) , so that it only depends on @xmath82 . note that in writing the action in this form one must take into account that the spatial derivatives of the field , @xmath83 , @xmath84 , are assumed to vanish , i.e. the probe brane should remain parallel to the source branes during its motion . the analysis of the classical motion of the probe brane is in this way reexpressed as a classical field theory problem of a complex scalar field in the background of a flat frw spacetime . before deriving the equations of motion for the field @xmath85 , it is useful to consider some of the symmetries of the action . first , the rotational symmetry of the problem in the plane transverse to the branes is associated with a conserved angular momentum . in terms of the field theory approach , this corresponds to the global u(1 ) invariance of the action under @xmath86 , where @xmath87 is a constant . the associated noether current induces a conserved charge which gives , in the quantum theory , the particle number operator associated with the field . writing the field as @xmath88 and defining the comoving angular momentum as @xmath89 , the comoving particle number density is given by : [ eq20 ] n = i(^*-^*)=2 ^ 2=2l , while the total particle number , which corresponds to the conserved noether charge , is given by @xmath90 . the energy density and pressure of the field can be obtained from its energy - momentum tensor , yielding : [ eq21 ] & & t_00=||^2+v()=^2+l^2 ^ 2+v ( ) , + p & & t_iia^2=||^2-v()=^2+l^2 ^ 2-v ( ) . the covariant conservation of the energy - momentum tensor can then be expressed as @xmath91 , where the energy of the field is defined as @xmath92 and @xmath93 is the volume of the expanding ( or contracting ) flat frw spacetime . defining the hubble parameter @xmath94 , we can write this in the form : [ eq22 ] ddt=-3h(+p ) . from eqs . ( [ eq21 ] ) and ( [ eq22 ] ) we conclude that @xmath95 , the equality holding only for @xmath96 . thus , in an expanding universe ( @xmath97 ) , the system evolves with a strictly decreasing energy density . the angular momentum evolves according to : [ eq23 ] dldt+3hl=0 , which corresponds to conservation of the total particle number @xmath6 . hence , in an expanding universe , the absolute value of the angular momentum will necessarily decrease , vanishing asymptotically . we , thus , conclude from this simple analysis of conservation laws that the expansion of the universe will make the system reduce both its energy density and its angular momentum . to analyze the details of this evolution , we need to determine the equations of motion for the field @xmath85 . varying the action eq . ( [ eq18 ] ) with respect to @xmath98 , and taking this variation to vanish , we obtain : [ eq24 ] + 3h+v(||)^*=0 . in terms of the transverse space coordinates @xmath10 , @xmath99 , this can be written as : [ eq25 ] y^m+3hy^m+2q_6y^mr^3=0 . this implies that the polar variables @xmath100 and @xmath101 satisfy : [ eq26 ] & + 3h - l^2 ^ 3 + 12 ^ 2=0 , + & + 2 + 3h=0 , where we have defined the constant [ eq27 ] 4q_6(t_6v_32)^3/2 , so that , apart from constant factors , the large distances potential is @xmath102 . it is easy to check that the equation for @xmath101 simply gives the evolution of the angular momentum that we obtained previously in eq . ( [ eq23 ] ) . the evolution of the radial field @xmath100 is determined by the form of the hubble parameter , given by the energy density content of the universe via the friedmann equation in the usual way . this may include not only the energy density of the field @xmath85 but also all other matter , radiation or vacuum energy components . the possibility that this scalar field may drive ( slow - roll ) inflation if , in the early universe , it dominates the energy density has been analyzed in @xcite , for the case where the probe brane moves in a linear trajectory , i.e. @xmath103 . it was shown that the interbrane potential arising from type ii supergravity / string theories is not flat enough to produce the required number of e - foldings , exhibiting an @xmath104-problem " . this is intrinsically related to the assumption that @xmath42 is much smaller than the size of the transverse dimensions . an inflationary period driven by the angular field variable @xmath101 was proposed in @xcite , where it is argued that , despite having a flat potential , this field may provide the constant energy density necessary for inflation if the probe brane moves in a circular orbit with @xmath105 . however , we need to take into account the fact that , during inflation , the angular momentum decays exponentially as @xmath106 , from eq . ( [ eq23 ] ) . the system will hence quickly tend to the @xmath103 case , where sufficient inflation is difficult to obtain . furthermore , the decay of the angular momentum will necessarily alter the evolution of @xmath100 and the value of @xmath107 , so that a slow - roll inflationary mechanism with the probe brane moving in a circular orbit would be hard to construct - branes moving in warped throats @xcite . in particular , the important role of the angular variables in providing accelerated periods of expansion was discussed in @xcite . ] . although we do not wish to completely discard such inflationary mechanisms , we will from now on assume that some other field is responsible for inflation and analyze how the compactified @xmath108 branonium system evolves in the post - inflationary eras . before analyzing in detail the effects of the expansion on the trajectory of the probe , let us recall the general bound orbits of the system in a non - expanding universe . in this case , the system reduces to the well - known problem of a particle in a central @xmath109 attractive potential , admitting closed orbit solutions of the form : [ eq28 ] ( ) = r(1-e^2)1+e . these are closed ellipses with eccentricity and semi - major axis given by : [ eq29 ] e&=&(1 + 4l^2 ^ 2)^12 , + r&=&-2 . the properties of such orbits may be obtained by studying the effective potential @xmath110 . this potential has a minimum at @xmath111 , tends to @xmath112 at the origin and to zero at large distances . the condition @xmath113 then implies that orbits with @xmath114 will be bounded , with @xmath115 . in this case , @xmath100 will oscillate between its minimum and maximum values , @xmath116 and @xmath117 , which can be obtained from eq . ( [ eq28 ] ) by setting @xmath118 and @xmath119 , respectively . at the minimum of @xmath120 , the orbits will be circular with @xmath121 , and @xmath122 . for @xmath123 , the orbits will only be bounded from below with e.g. @xmath124 for @xmath125 , which corresponds to an @xmath126 parabolic orbit . the linear trajectory , with @xmath103 , is a particular case of the latter with no bounds on the value of @xmath100 except for the trivial @xmath127 . for @xmath128 , the orbits are hyperbolic . consider now the expanding case with the scale factor evolving as @xmath129 . this power law behavior is typical of the post - inflationary stages of the universe s evolution , where one may consider a single fluid to give the dominant contribution to the total energy density . for example , in the radiation era we have @xmath130 and in the matter era @xmath131 . we can also take @xmath131 at the end of inflation , when the oscillations of the inflaton about the minimum of its potential , with an equation of state corresponding to that of non - relativistic matter , dominate the energy density . this model only holds away from the transitions between these periods as in these cases at least two of the components give similar contributions to the energy density . we will consider the evolution of the branonium system for arbitrary @xmath87 , starting at some instant @xmath132 when the scale factor has a value @xmath133 . the hubble parameter is then of the form @xmath134 , which implies from eq . ( [ eq23 ] ) that the angular momentum evolves according to : [ eq30 ] l = l_0(tt_0)^-3 , with @xmath135 being its initial value . it is then clear that @xmath136 as @xmath137 , as we have mentioned earlier . the equation for the radial field @xmath100 can now be written as : [ eq31 ] + 3t - l_0 ^ 2 ^ 3(tt_0)^-6 + 12 ^ 2=0 . let us start by analyzing how the system evolves when placed initially in a would - be circular orbit , so that [ eq32 ] ( t_0)=2l_0 ^ 2_c0 , as @xmath138 . it is easy to see that @xmath100 can not remain constant for @xmath139 due to the decay of the angular momentum . we may , however , look for solutions where the condition for circular orbits is maintained during the motion of the probe brane , i.e. [ eq33 ] ( t)=2l^2(t)=_c0(tt_0)^-6_c(t ) . substituting into eq . ( [ eq31 ] ) we obtain [ eq34 ] _ c0t_0 ^ 2(tt_0)^-6 - 26(3 + 1)=0 , which is satisfied only in the non - expanding case , @xmath140 , and for @xmath141 . as we are interested in cases where @xmath142 , we conclude that @xmath143 is not an exact solution of the equations of motion . it is , however , an approximate solution at late times and , as we will check later , gives the global evolution of the system in an expanding universe , so that it is useful to consider some of its properties . in particular , the angular frequency evolves as : [ eq35 ] _ c = l_c^2(tt_0)^9 , so that the orbital period decreases as @xmath144 . such a solution eventually fails to satisfy the non - relativistic and large distance approximation as the radius of the orbit decreases and its angular velocity grows . in particular , the parameter @xmath145 which controls this approximation grows , for @xmath146 , as [ eq36 ] hv^2(t_6v_3)^4(q_6l_0)^4(tt_0)^12 . hence , given the values of @xmath147 and @xmath75 , it is the ratio @xmath148 that controls how long the approximations remain valid . consider now trajectories with a non - vanishing initial eccentricity . in an expanding universe , the effective potential becomes time - dependent , although its asymptotic properties at the origin and at infinity remain the same . as the angular momentum redshifts away , the minimum of the effective potential decreases , the same happening with the value of @xmath100 at this minimum , according to the circular solution defined in eq . ( [ eq33 ] ) . as discussed earlier , the energy density of the field always decreases during the motion of the probe brane in an expanding universe . hence , if the system is initially in one of the closed orbits with @xmath149 , the latter condition will be satisfied for all @xmath139 . this means that , at all times , the probe brane will be in one of the closed elliptical orbits defined earlier . from eq . ( [ eq29 ] ) , we also conclude that the semi - major axis of the orbit will decrease as the energy density evolves to more negative values . the eccentricity of the orbits may , however , not remain constant , as it depends on the variation of both the energy density and the angular momentum . in fact , using eqs . ( [ eq22 ] ) and ( [ eq23 ] ) , one can show that the eccentricity varies according to [ eq37 ] de^2dt=-12hl^2 ^ 2(3+p)=-24hl^2 ^ 2(2t+v ) . where we defined the kinetic energy of the field as @xmath150 , so that @xmath151 and @xmath152 . thus , in an expanding universe , for finite @xmath135 , one expects the eccentricity of the orbit to vary unless @xmath153 . this last condition is satisfied on average in the non - expanding case , corresponding to the virial theorem . it does not necessarily hold , however , in an expanding universe , with the semi - major axis of the orbits decaying as discussed above . it is nevertheless clear that , at late times , as @xmath107 and @xmath154 decrease , the variation of the eccentricity of the orbits should be smaller . the sign of @xmath155 is also not definite , so that the eccentricity may either increase or decrease during the motion of the probe brane . from eqs . ( [ eq29 ] ) and ( [ eq33 ] ) , we obtain : [ eq38 ] r(t)=_c(t)1-e^2(t ) . hence , the variation of the semi - major axis follows the decrease of the circular solution obtained earlier , being also affected by the variation of the eccentricity . in order to have a better understanding of the evolution of the probe brane s motion in an expanding universe , we have solved the equations of motion numerically . measuring all quantities in terms of the string length , i.e. setting @xmath156 , we choose , as an example , the values @xmath157 and @xmath158 we follow the example analyzed in @xcite . ] . we set the initial conditions at @xmath159 to be those of a non - expanding elliptical orbit with angular momentum @xmath135 and eccentricity @xmath160 , which we take to be the only free orbital parameters . the field is initially at its maximum value @xmath161 , with @xmath162 and @xmath119 . after determining the numerical solution for @xmath163 , we computed the corresponding energy density and eccentricity evolution , according to eqs . ( [ eq21 ] ) and ( [ eq29 ] ) , respectively . we used the numerical solution for the eccentricity to compute @xmath164 , according to eq . ( [ eq38 ] ) , and also the maximum and minimum values of @xmath163 at each orbit , given by : [ eq39 ] _ min(t)&=&r(t)(1-e(t ) ) , + _ max(t)&=&r(t)(1+e(t ) ) . we have also determined the evolution of the angular field variable @xmath101 using @xmath165 . in figure 1 we have plotted the results obtained for the values @xmath166 and @xmath167 in a matter - dominated universe , @xmath131 , with @xmath168 . seems quite large , one must take into account that it is measured in units of the string time , which is of order @xmath169 seconds if the string scale is of the order of the planck scale . thus , we are considering the probe brane to be moving quite early in the history of the universe . ] and @xmath167 in a matter - dominated universe , @xmath131 , with @xmath168 . the plots show ( a ) the radial field and associated quantities , ( b ) the energy density , ( c ) the eccentricity and ( d ) the motion of the probe brane in the complex plane of the field @xmath85.,title="fig : " ] [ fig : figure_1 ] observing the plots shown in this figure , we conclude that , as expected , the probe brane evolves continually through elliptical orbits of decreasing semi - major axis , so that they fail to close . the radial field then oscillates between minimum and maximum values which decrease in time . the global decrease follows that of the circular solution , eq . ( [ eq33 ] ) , but exhibits an oscillating behavior associated with the variation of the energy density . although the latter strictly decreases , as expected , it also oscillates with decreasing period and amplitude , so that at later stages its evolution tends to be smooth . consequently , one observes oscillations in the evolution of the eccentricity of the orbit . in the example shown in figure 1 , these oscillations have a small amplitude . however , at earlier times , when the effects of hubble expansion are more significant , numerical simulations show that this amplitude can be quite large . all simulations show that @xmath170 starts increasing and , as the period and amplitude of the oscillations decay , it tends to a constant value . a constant eccentricity at late times implies that , in this limit , the energy density should , on average , vary as [ eq40 ] l^-2(t)(tt_0)^6 , tending smoothly to more negative values as observed in figure 1 . then , from eq . ( [ eq22 ] ) , [ eq41 ] = -3h=6h , so that @xmath171 or , equivalently , @xmath172 . thus , although at early times the effect of the expansion makes the eccentricity vary significantly , at late times the system _ virializes _ and starts evolving smoothly between orbits of constant eccentricity and decaying radius . this result will be useful later for studying particle production in the probe brane at late times . another interesting consequence of the expansion of the universe is a rotation of the axis of the elliptical orbits with time , although this is a small effect in the example shown in figure 1 . this is simply a result of the growing eccentricity of the orbits which deviates the maximum and minimum values of @xmath100 from @xmath119 and @xmath118 , respectively . the orbital axis will , however , stabilize at late times , as @xmath170 tends to a constant value . numerically , one also observes a decrease in the amplitude of the oscillations of @xmath163 . this is given by [ eq42 ] ( t)_max(t)-_min(t)=r(t)e(t ) , so that the observed decay of the amplitude is mainly due to the decrease of @xmath164 as discussed above . the asymptotic value of the eccentricity depends not only on the initial time @xmath132 at which the probe starts its motion but also on the initial values of the orbital angular momentum and eccentricity . numerical simulations show that larger values of @xmath135 lead to a more pronounced growth of the eccentricity , in agreement with eq . ( [ eq37 ] ) . the dependence on @xmath160 is less trivial and numerically one finds a larger eccentricity variation for @xmath160 close to 0 and to @xmath173 . this is , however , very small for @xmath160 close to 1 , which is expected , as the energy density of the field can not increase to produce hyperbolic orbits with @xmath174 if initially @xmath149 . the eccentricity always grows if initially @xmath105 , so that @xmath175 and @xmath176 , as most of the eccentricity variation occurs initially when hubble expansion is more significant . the opposite behavior should be observed if initially @xmath177 . although the example we have shown refers to a matter - dominated universe , @xmath131 , our qualitative discussion and analytical results hold for all @xmath142 , in particular to the @xmath130 radiation - dominated universe . to summarize the results of this section , we conclude that , in an expanding universe , the probe brane follows elliptical orbits with decreasing radius and increasing frequency and whose eccentricity exhibits an oscillating behavior and asymptotically tends to a constant value larger than the initial one . if the motion of the probe brane begins at late times with a small angular momentum , the latter effect is negligible . the probe brane will asymptotically collide with the branes in the stack , as @xmath178 , but our approximations will break down before this happens , so that we expect the motion of the probe brane to deviate significantly from our previous results as its angular velocity grows and it spirals towards the central stack . the collision will , however , occur if nothing else prevents the probe from losing energy and angular momentum as the universe expands , and may lead to the annihilation of the probe antibrane with one of the source branes . for now , we will use the results obtained in this section to study the production of particles in the probe brane and we will return to this issue in section v of this work . in the original discussion of the branonium system @xcite , it was argued that the fields confined to the probe brane , such as the gauge fields mentioned earlier , become time - dependent due to the motion of the probe through the background spacetime created by the central stack . consequently , an observer bound to the probe interprets this variation as corresponding to the production of particles associated with these fields . such radiation of energy into brane particle - modes arises only if the distance between the probe and the stack varies in time , thus creating a time - dependent background from the point of view of brane - bound observers . although the power radiated into these modes was estimated in @xcite , many aspects of this particle production mechanism remain unclear . in this section , we will analyze this mechanism in more detail , revealing some new properties of particle production in branonium , and discuss how the expansion of the universe modifies this process . the fields confined to the brane arise , from the string theory point of view , as states associated with open strings whose endpoints are attached to the brane , as discussed earlier . for a single @xmath7-brane , which is a @xmath179 bps supersymmetric state , these include bosonic fields such as gauge bosons and scalar fields , as well as their fermionic superpartners . if the probe corresponds itself to a stack of @xmath180 parallel @xmath7-branes , other degrees of freedom associated with a @xmath181 vector supermultiplet may arise @xcite . here , we will study the simplest case of a real scalar field @xmath104 confined to the probe brane and follow an effective field theory approach . for the @xmath108 system described earlier , we write the effective action for the scalar field @xmath104 in the form : [ eq43 ] s_=-t_6d^7 e^-(-12^__-12m^2 ^ 2 ) , where @xmath182 denotes the mass of the scalar field , which includes its bare mass as well as quantum corrections , and all other quantities are those defined earlier in this work . when compactifying 3 of the dimensions parallel to the brane as before , the scalar field will be decomposed in its kaluza - klein ( kk ) modes , leading to a tower of 4-dimensional massive modes , or kk states . we will focus on the evolution of the zero mode of the field @xmath104 , whose mass will simply be given by @xmath182 . to simplify the notation , we will denote this mode as @xmath104 , although one must bear in mind that it is not the original 7-dimensional field . the global factor of @xmath147 will not affect our discussion and we may absorb it into the definition of the field . first , let us consider the case @xmath183 . then , the classical equations of motion arising from the action eq . ( [ eq43 ] ) are given by : [ eq44 ] _ ( e^-^_)=0 . using the results obtained in section ii , and defining the function @xmath184^{1\over2}$ ] , we can write this as [ eq45 ] + ( 3h - f)-1a^2(1-hv^2)^2=0 , where @xmath185 and @xmath186 is the flat 3-dimensional laplacian of the field . to construct the associated quantum description , we follow the semi - classical approach to the quantization of scalar fields in curved space @xcite . in a curved background spacetime with a time - varying geometry , as is the case of the world - volume of the probe brane , the induced time - variation of the fields modifies the usual canonical quantization procedure , as the quantum operators become themselves time - dependent . in particular , the creation and annihilation operators associated with the field will now evolve in time as the background changes , the same happening with the associated multi - particle states . this is the main reason behind the production of particles in a dynamical background . let us start by expanding the field in fourier modes of the form : [ eq46 ] ( , t)=(a__(t)e^i+ a^_^*_(t)e^-i ) . in the quantum theory , @xmath187 and @xmath188 become the annihilation and creation operators associated with the fourier mode @xmath189 of the field . expanding the field in this way , we include all the time dependence of the field in the functions @xmath190 and @xmath191 , while the operators remain time - independent . the conjugate momentum to the field @xmath104 , obtained by computing @xmath192 , can then be written as [ eq47 ] ( , t)=a^3f(a__e^i+ a^_^*_e^-i ) . if the creation and annihilation operators satisfy the canonical commutation relations [ eq48 ] [ a_,a_]=[a^_,a^_]=0 , [ a_,a^_]=^3(- ) , then the canonical commutators for the field and its conjugate momentum , @xmath193=i\delta^3(\mathbf{x}-\mathbf{y})$ ] , can only be obtained if the mode functions satisfy the following wronskian normalization condition : [ eq49 ] _ ^*_-^*__=ifa^3 . this condition is essential for consistency of the quantization procedure . we now follow a semi - classical approach and consider the evolution of the field modes to be given by the classical equations of motion , assuming that any quantum corrections to their propagation can be neglected . substituting eq . ( [ eq46 ] ) into eq . ( [ eq45 ] ) , we see that each of the fourier modes evolves independently according to [ eq50 ] _ + ( 3h - f)_+k^2a^2(1-hv^2)_=0 . and similarly for the complex conjugate modes @xmath194 . notice that this equation depends only on @xmath195 but not the direction of the momenta @xmath196 . this is a consequence of the isotropy of our metric _ ansatz _ , and we may write without loss of generality @xmath197 . it is also clear , from eq . ( [ eq50 ] ) , that the physical momentum of the modes is related to their comoving momentum @xmath198 via [ eq51 ] k_phys = ka . this expression includes both the expected redshift of the modes in an expanding universe and the effect of the probe s motion through the dynamical background . let us now rescale the field modes by defining the mode functions @xmath199 . these satisfy [ eq52 ] _ k+_k^2x_k=0 , which corresponds to the equation for a harmonic oscillator with a variable frequency given by : [ eq53 ] _ k^2&=&k^2a^2(1-hv^2)-14(3h - f)^2 - 12(3h - f)= + & = & k_phys^2+^2 , where we have defined [ eq54 ] ^2(t)-14(3h - f)^2 - 12(3h - f ) . as each mode behaves as a harmonic oscillator , we can define the associated particle number via [ eq55 ] e_k&=&12|x_k|^2 + 12_k^2|x_k|^2=_k(n_k+12 ) + n_k&=&_k2(|x_k|^2_k^2+|x_k|^2)-12 . it is well known that this quantity is an adiabatic invariant for a harmonic oscillator with variable frequency , as happens , for example , in an oscillating pendulum whose length is decreased infinitely slowly @xcite . thus , the number of _ quanta _ in a given fourier mode @xmath198 can only change if its frequency varies in a non - adiabatic way , which can be expressed by the following condition @xcite : [ eq56 ] |d_kdt|_k^2 . for our particular case , this gives , in the limit @xmath78 , [ eq57 ] |hk_phys^2-|(k_phys^2+^2)^32 . this condition will then constrain the number of particles produced in each physical momentum mode by the motion of the probe brane . to quantify this number of particles , we need to look for solutions of the mode equation , eq . ( [ eq52 ] ) . we will start by discussing what happens in a non - expanding universe and then we will analyze the effects of hubble expansion on particle production . consider the motion of the probe brane in a non - expanding universe , @xmath96 , and assume that the non - relativistic and large distance approximations discussed earlier are valid . as we have seen in section ii , the probe follows closed elliptical trajectories if its energy ( density ) is negative , @xmath149 . if @xmath78 , we have to leading order : [ eq58 ] 1-hv^21 , k_physk , fh^1/2 . then , we may take the following approximations : [ eq59 ] f&=&ff , + f&&12 , so that we can write : [ eq60 ] ^2=-14f^2 + 12f-516(hh)^2 + 14hh . an exact analytical expression for the time variation of the radial field @xmath163 can be obtained in a non - expanding universe , being given implicitly by [ eq61 ] ( t)=r . the angular variable @xmath200 satisfies [ eq62 ] t - t_0=^-1 , where @xmath201 and @xmath202 is the angular frequency of the orbit . the angular variables @xmath203 and @xmath101 are related by : [ eq63 ] = . for small eccentricity orbits , @xmath204 , eq . ( [ eq62 ] ) gives , to leading order in @xmath205 : [ eq64 ] _ 0+(t - t_0 ) . choosing @xmath206 we have , in this approximation , [ eq65 ] ( t)r . this allows us to determine the non - relativistic and large distance expression for @xmath207 , in the case of small eccentricity orbits . to lowest order , we then obtain : [ eq66 ] ^2e^2((t - t_0 ) ) . defining [ eq67 ] ^2e^2 , the variable frequency of the mode with comoving momentum @xmath198 can , hence , be approximated by : [ eq68 ] _ k^2(t)k^2+^2((t - t_0 ) ) . if we now rewrite eq . ( [ eq52 ] ) in terms of the rescaled time variable @xmath208 , we obtain : [ eq69 ] x^_k+(a_k-2q(2z))x_k=0 , where [ eq70 ] a_k , q-2 ^ 2 ^ 2 . ( [ eq69 ] ) has the form of the well - known mathieu equation @xcite with parameters @xmath209 and @xmath71 determined by the orbital parameters of the probe brane and by the value of the comoving momentum @xmath198 of each mode . in the parameter space @xmath210 , the mathieu equation exhibits both stable and unstable solutions , the latter being closely associated with the phenomenon of parametric resonance @xcite . as we are working under the assumption that the probe brane s orbit has a large radius and a small eccentricity , it is easy to check that @xmath211 ( the analysis is independent of the sign of @xmath71 ) . in this region of parameter space , the mathieu equation exhibits instabilities in a series of narrow resonance bands near @xmath212 , with @xmath27 being a positive integer , and with a width in comoving momentum space approximately given by @xmath213 . in these resonance bands , the solution evolves as @xmath214 , with a real exponent @xmath215 , according to floquet s theorem . the most important of these is the first resonance band , which occurs when [ eq71 ] 1-|q|-18q^2a_k1+|q|-18q^2 . the exponent @xmath216 can then be approximately written as [ eq72 ] _ k . this is , as claimed , real for @xmath217 , having a maximum value of @xmath218 for @xmath219 . this implies that particle - modes lying inside this resonance band will be exponentially amplified , leading to a resonant production of particles . thus , from eq . ( [ eq70 ] ) , the center of the resonance band occurs for comoving momentum @xmath220 , while its lower and upper bounds are given by [ eq73 ] k_min&&k_c , + k_max&&k_c , so that the resonance band has a width in comoving momentum space of @xmath221 . this is quite small compared to the value of the center of the resonance , as @xmath222 , which means that particle production occurs in a regime of narrow parametric resonance . the particles produced by this mechanism have typical energies of order @xmath223 , i.e. with half the typical energy of the field @xmath85 . this agrees with the discussion in @xcite , where it is stated that particles are produced in pairs with opposite momenta , as if resulting directly from the decay of the interbrane distance field - particles ensures momentum conservation . ] . we have found in this work , however , that these particles are produced resonantly , the associated field modes being exponentially amplified . after the modes inside the resonance band have been amplified sufficiently , we may write , from eq . ( [ eq55 ] ) , [ eq74 ] n_ke^2_kz = e^_k(t - t_0 ) , where we used that @xmath224 , so that the term involving @xmath225 in eq . ( [ eq55 ] ) can be neglected . this also implies that , although the particle number in each mode grows exponentially , the resonance takes a long time to develop . in particular , the typical resonance time , @xmath226 , is much larger than the orbital period of the probe brane , which is of order @xmath227 . this is characteristic of the narrow resonance regime and , hence , a significant particle number is only produced after the probe brane has completed a large number of orbits , given by @xmath228 . to have an idea of the order of magnitude of these quantities , consider the scenario studied in section ii , where @xmath229 with @xmath158 . then , for an orbit with eccentricity @xmath230 and an angular momentum @xmath231 , we have @xmath232 , which satisfies the large distances approximation , and @xmath233 . this implies that the probe brane will have to complete approximately @xmath234 orbits for the particle number to be amplified by one e - folding . to illustrate the results obtained earlier , we have solved eq . ( [ eq52 ] ) for the example described above . an initial vacuum state with zero particle number corresponds to the following initial conditions : [ eq75 ] x_k(t_0)=1 , _ k(t_0)=-i . in figure 2 we have plotted the numerical results obtained for the comoving momentum at the center of the resonance band , @xmath235 . the linear evolution observed for the logarithm of the particle number shows that , as expected , this quantity is being amplified exponentially with time , but that a significant number of particles is produced only after the probe brane has completed a few thousands of orbits . and @xmath231 , giving @xmath236 . the time coordinate @xmath237 is given in units of the orbital period @xmath238.,title="fig : " ] [ fig : figure_2 ] we note that , although this solution was obtained numerically , a formal analytical solution of the mathieu equation can be written in terms of the mathieu sine and cosine functions . in figure 3 we also show the evolution of the particle number for a mode outside the limits of the resonance band . in this case , @xmath239 takes imaginary values and @xmath240 exhibits an oscillating behavior with a small amplitude , confirming that only for comoving momenta @xmath241 a significant particle number is produced by the elliptical motion of the probe brane . it is also worth mentioning that , in the limit of circular orbits , @xmath242 , the parameter @xmath71 vanishes and there is no amplification of any of the particle momentum modes . this confirms our early assumption that the interbrane brane distance needs to vary for particle production to occur . and @xmath231 , giving @xmath236 . the comoving momentum of the mode is @xmath243 . the time coordinate @xmath237 is given in units of the orbital period @xmath238.,title="fig : " ] [ fig : figure_3 ] so far we have not included the backreaction effects of the produced particles on the probe s motion . we will return to this later in section iii c , where we estimate the energy density damped into brane - modes by the resonance and discuss how this affects the probe brane s orbital decay . consider now the case where the field has a non - vanishing mass , @xmath244 . in this case , the classical equations of motion eq . ( [ eq44 ] ) can be generalized to [ eq76 ] _ ( e^-^_)-e^-m^2=0 , or , explicitly , [ eq77 ] + ( 3h - f)-1a^2(1-hv^2)^2+(1-hv^2)h^12m^2=0 . each of the fourier modes then evolves independently according to [ eq78 ] _ k+(3h - f)_k+ . this implies that all modes have , in this case , an effective physical mass which is different from the mass parameter @xmath182 due to the motion of the probe brane : [ eq79 ] m_eff^2=(1-hv^2)h^12m^2 . notice that the mass is not , however , affected by the expansion of the universe . when writing the equations of motion in terms of the rescaled mode function @xmath245 defined earlier , we again obtain the equation for a harmonic oscillator with a varying frequency , which is now given by : [ eq80 ] _ k^2=k_phys^2+m_eff^2+^2 , where all the quantities are time - dependent in the general case . let us now set @xmath96 and compare these results with the ones obtained for the particle production mechanism in the massless case . first , notice that , in the non - relativistic and large distance approximation , we have @xmath246 . thus , the leading order modification introduced by the mass of the field can be obtained by simply replacing @xmath247 . following the same procedure as before , one can reduce the particle - mode equation to the mathieu equation , with the parameter @xmath209 being now given by : [ eq81 ] a_k=4(k^2+m^2)^2 . the parameter @xmath71 , which quantifies the strength and width of the resonance phenomenon , is not modified by the introduction of the mass parameter , so that we expect the particle production mechanism to occur as before . in this case , however , the center of the resonance band is shifted to a lower momentum , @xmath248 , which gives @xmath249 . the upper and lower limits of the resonance band are also modified , being now given by : [ eq82 ] k_min= , + k_max= , which in turn alters the width of the resonance band in the obvious way . if , as in the massless case , we have @xmath250 , we may write @xmath251 , which is larger than the corresponding value in the massless case , satisfying nevertheless the narrow resonance condition . notice that this implies an upper bound for the mass of the particles which can be produced by this mechanism , [ eq83 ] m . in the particle description discussed earlier , this simply means that each particle " in the interbrane distance field @xmath85 needs to have enough energy to decay into two @xmath104-particles at rest . a residual kinetic energy is also necessary if all the modes inside the resonance band are to be excited . in the non - relativistic and large distance approximation , the orbital frequency of the probe brane is assumed to be small , so that only particles with small masses may be excited during its motion . one must recall , however , that we are measuring all quantities in terms of the typical string parameters . in particular , the orbital frequency should be small compared to the string energy scale . if the latter corresponds to the planck scale , then the produced particles can still be quite massive . for the example studied earlier in this section , with orbital parameters @xmath252 , particles up to masses of order @xmath253 gev can be resonantly created . recalling the definition of @xmath85 in terms of the physical interbrane distance , eq . ( [ eq17 ] ) , we may write the upper bound on the mass of the produced particles in terms of the physical semi - major axis of the orbit , @xmath254 . inserting back the missing @xmath39 factors , we obtain : [ eq84 ] m_max = m_s , where @xmath255 is the string energy scale . as we assumed in the beginning that @xmath256 , it is clear from this expression that @xmath257 , which allows nevertheless for the production of very massive particles . in general , particles living in the world - volume of the probe brane will be massive , so that we need to use the results obtained above to describe the parametric resonance . for masses which are small compared to the typical orbital energies of the probe , the simpler treatment of the @xmath183 case will be sufficient . let us now consider the case of an expanding universe , with @xmath258 . the inclusion of the scale factor will modify our previous analysis of particle production in several ways . first , it will change the frequency of each harmonic oscillator particle - mode , namely modifying the factor @xmath259 , given in eq . ( [ eq54 ] ) . also , the physical momenta of the particle - modes will be redshifted as the universe expands , so that , in the non - relativistic and large distance approximation , we have @xmath260 . finally , the motion of the probe brane will be altered , as we have seen in section ii , so that the semi - major axis of elliptical orbits will decrease as @xmath261 , making the angular frequency consequently increase as @xmath262 . furthermore , if the hubble parameter is sufficiently large , the orbital eccentricity may exhibit a significant growth . all the modifications make a complete analytical study of particle production in the probe brane rather complex . in order to determine the leading effects of the expansion , we will consider the motion of the probe brane at sufficiently late times , so that some of the effects of the expansion can be discarded . as we have concluded in section ii , the variation of the eccentricity of the orbits is , in this regime , very small , so that it may be neglected . the explicit effects of the expansion in modifying the factor @xmath207 may also be discarded in this limit , provided that @xmath263 and @xmath264 , or explicitly [ eq85 ] t(r_physq_6 ) , t^2(r_physq_6 ) . the first of these conditions will , _ a priori _ , be more constraining . for the example we have been following in this section , the first condition gives @xmath265 while the second only implies the constraint @xmath266 . with these constraints in mind , we now have to analyze how hubble expansion modifies the factors @xmath267 and @xmath268 in the variable harmonic oscillator frequency . the radial interbrane distance field @xmath100 can now be written , in the low eccentricity orbit approximation , as : [ eq86 ] ( t)r(t ) , with [ eq87 ] r(t)=r_0(tt_0)^-6 , ( t)=_0(tt_0)^9 . computing its time derivatives , we find that they involve terms arising from hubble expansion which are suppressed by at least one power of @xmath237 , being negligible at late times . there are also terms involving the quantity @xmath269 . if the time necessary for particle production to occur is smaller than the age of the universe at that time , we may also discard these terms . after some algebra , the leading order modification to @xmath207 is then given by [ eq88 ] ^2(t)^2(t)((t)(t - t_0 ) ) . the quantity @xmath270 is now time - dependent and can be obtained trivially by including the appropriate time variation of the orbital parameters in its non - expanding expression , eq . ( [ eq67 ] ) . this gives [ eq89 ] ^2(t)=_0 ^ 2(tt_0)^24 , _ 0 ^ 2e_0 ^ 2 . then , taking into account the momentum redshift , we can write the particle - mode equation , to lowest order , as : [ eq90 ] _ k+x_k=0 . following the same reasoning as in the non - expanding case , we rescale the time coordinate by defining the variable @xmath271 . as the angular frequency of the orbit is now time - dependent , this change of variables will introduce new terms in the equation . these will , however , be suppressed by powers of @xmath237 and involve first and second derivatives of the mode function @xmath272 . in the non - expanding case , these were quite small compared to @xmath272 , and one expects them to be negligible in this case as well . using that , to lowest order , @xmath273 , we may write the leading order equation for the mode function in terms of the variable @xmath274 as [ eq91 ] x^_k+[a_k(z)-2q(z)(2z)]x_k=0 , with [ eq92 ] a_k(z)a_k0(1-z ) , q(z)q_0(1+z ) , where @xmath275 and @xmath276 correspond to the initial values of the parameters of the mathieu equation , given by their non - expanding case expressions , and the coefficients @xmath20 and @xmath277 are given by [ eq93 ] , = 310 . we , hence , conclude that the leading order effect of the expansion is to make the coefficients of the mathieu equation vary in time . in particular , their variation is , to a first approximation , linear in the variable @xmath274 and controlled by the single parameter @xmath20 , which will be small at late times . one must take into account , however , that eq . ( [ eq91 ] ) is not the mathieu equation and , namely , does not satisfy the conditions of floquet s theorem which allowed us to define the coefficient @xmath239 . nevertheless , we can use the results obtained in the non - expanding case from the properties of the mathieu equation as a guide to describe the behavior of the modes when @xmath97 . first , notice that , as @xmath209 decreases with @xmath274 , the position of the center of the resonance band , at @xmath249 , will be shifted to higher momentum modes : [ eq94 ] k_c(z)=k_c01-z , k_c0 . thus , we expect modes with comoving momentum larger than @xmath278 to be excited as the probe s orbit decays . to better understand the evolution of each mode , let us analyze the evolution of the exponent @xmath279 , which is given by : [ eq95 ] _ the zeros of @xmath279 occur for [ eq96 ] z_1=-|q_0|+a_k0 - 1a_k0+|q_0| , z_2=|q_0|+a_k0 - 1a_k0-|q_0| , with @xmath280 . this means that @xmath279 will be real for @xmath281 and pure imaginary for @xmath282 and @xmath283 . figure 4 illustrates the typical evolution of @xmath279 for a generic fourier mode . for a generic mode . the shaded area corresponds to the period the mode spends inside the resonance band.,title="fig : " ] [ fig : figure_4 ] thus , each mode may only experience the resonant regime during a finite amount of time , when @xmath279 is real and the mode is inside the resonance band . during the periods where @xmath279 is pure imaginary , we expect the particle number to exhibit an oscillating behavior . in the non - expanding case , the frequency and amplitude of these oscillations for modes outside the resonance band is determined by the value of @xmath284 , so that the larger the latter the smaller the period and the amplitude of the oscillations . as , in an expanding universe , @xmath285 changes in time , we expect the frequency and amplitude of the oscillations of the particle number to vary as the probe brane s orbit decays . similarly , as @xmath286 is not constant when the mode is inside the resonance band , the strength of the resonance is also expected to vary . particle modes will then have different behaviours according to their position relative to the initial resonance band , which can be parametrized by : [ eq97 ] a_k01+|q_0| . recalling that the centre of the original resonance band corresponds to @xmath287 and that the probe s motion begins at @xmath288 , we need to distinguish three different types of modes : 1 . @xmath289 + as both @xmath290 and @xmath291 are negative in this case , these modes will always be outside the resonance band , with an imaginary exponent @xmath279 for all @xmath292 . 2 . @xmath293 + this case corresponds to the particle - modes inside the resonance band in a non - expanding universe , with @xmath294 but @xmath295 . hence , they will start inside the resonance band but stop experiencing the resonant regime for @xmath283 . 3 . @xmath296 + although these modes are outside the initial resonance band , they will experience the resonant behavior at some later stage , as @xmath297 . for both type ii and type iii modes , the maximum value of @xmath298 occurs for @xmath299 , giving , for @xmath300 , [ eq98 ] _ k(z_max)(1 + 310|q_0| ) . this gives the same result as in the non - expanding limit for @xmath287 . it is easy to check that @xmath301 as @xmath302 , the same happening for @xmath290 and @xmath291 . this implies that only modes for which @xmath294 ( @xmath303 ) and @xmath295 ( @xmath304 ) will be excited in this limit , in agreement with the fact that only type ii modes are excited in the non - expanding case ( @xmath305 ) and that these experience the parametric resonance during an infinite period of time . it is also clear from eq . ( [ eq98 ] ) that higher momentum modes are excited by the resonance at later times , as expected from the evolution of the center of the resonance band . the total time each type ii or type iii mode spends inside the resonance band is given by : [ eq99 ] z_bandz_2-z_1(1 - 710|q_0| ) . it is clear that this quantity decreases with @xmath20 , so that the smaller the hubble parameter the more time each mode is excited by the resonance . in the limit @xmath302 , @xmath306 , but as previously discussed this limit only applies to type ii modes , as all the others are outside the resonance band in this limit . one also concludes that higher momentum modes will spend less time inside the resonance band , which suggests these will be less excited . the analysis of the properties of the exponent @xmath279 thus provides a very clear insight on the qualitative evolution of all momentum modes . we would like , however , to be able to quantify the particle number produced by the resonance in each mode . in the non - expanding case , @xmath307 , which implied @xmath308 in the limit of large particle number for type ii modes . despite the variation of @xmath239 in the @xmath97 case , we may take this exponent to be approximately constant during an infinitesimal interval between @xmath274 and @xmath309 . then , during this interval , @xmath272 will be amplified by a factor @xmath310 if the mode is inside the resonance band . integrating this result , we expect that , after a significant number of particles has been produced , [ eq100 ] n_k(z)(2_z_i^z_k(z)dz ) . the initial time @xmath311 refers to the time the mode enters the resonance band , i.e. @xmath312 and @xmath313 for type ii and type iii modes , respectively . one can use eq . ( [ eq100 ] ) to estimate the total particle number produced in each mode by the parametric resonance . the details of this calculation are given in appendix a. for type ii and type iii modes , one obtains : [ eq101 ] n_k^res ( ii)(z_2)&=&q_0 ^ 2211+|q_0|(+2 + + & + & ) , + n_k^res ( iii)(z_2)&=&q_0 ^ 221+|q_0| . ( [ eq101 ] ) constitutes the main result of this section , giving the leading order expressions for the particle number density produced by the resonance in an expanding universe . from these approximated expressions , we can see that the parameter @xmath314 controls the strength of the resonance , a significant number of particles being produced only if @xmath315 . recall that , in the non - expanding case , the typical resonance time was given by @xmath316 . the typical time a mode spends inside the resonance band in the expanding case is , from eq . ( [ eq99 ] ) , of order @xmath317 . hence , apart from numerical factors , the condition for a significant number of particles to be produced is simply stating that the time a given mode spends inside the resonance band should be greater than the typical time the resonance takes to develop , @xmath318 . from the results in eq . ( [ eq101 ] ) , we can also conclude that the total particle number produced increases with @xmath319 for @xmath320 . on the other hand , for @xmath296 , it exhibits a slow decrease with @xmath319 , showing that the production of high momentum modes is suppressed . this agrees with the fact that high momentum modes spend less time inside the resonance band , as obtained above . to check the results obtained so far for the resonant particle production in an expanding universe , we have solved eq . ( [ eq91 ] ) numerically . in order to simplify the computation , we have rescaled the mode functions via @xmath321 , so that initially we have approximately [ eq102 ] _ k(0)1 , ^_k(0)-i . we then computed the function [ eq103 ] _ k(z)=|_k(z)|^2+|^_k(z)|^2 . this is related to the physical particle number density of each mode , given by eq . ( [ eq55 ] ) , approximately by @xmath322 , taking the limit of large particle number . note that @xmath323 initially , as we neglected the @xmath179 factor in eq . ( [ eq55 ] ) , so that the effects of this normalization have to be taken into account before the resonance produces a significant number of particles . in figure 5 , we illustrate the results obtained for modes of types i , ii and iii , including also the analytical prediction obtained from eq . ( [ eq100 ] ) . these simulations correspond to an original resonance strength @xmath324 , the effects of the expansion being quantified by @xmath325 ( recall eq . ( [ eq93 ] ) ) . these values ensure the validity of the approximations made and also that a significant particle number is produced for modes of types ii and iii . and @xmath325 . the plots correspond to ( a ) a type i mode , with @xmath326 , ( b ) a type ii mode , with @xmath287 and ( c ) a type iii mode with @xmath327 . the solid line corresponds to the numerical solution in all three cases , while in ( b ) and ( c ) the dashed line gives the corresponding analytical prediction . the shaded area corresponds to the period that each mode spends inside the resonance band . ] observing the plots shown in this figure , we see that all three particle - modes follow the expected behavior . the type i mode exhibits an oscillating particle number , with oscillations of decreasing amplitude and period , confirming that , in fact , no net particle number is produced reflects the normalization chosen in eq . ( [ eq103 ] ) and does not correspond to a physical particle number . ] . both the type ii and the type iii modes are exponentially amplified during a finite period , after which their particle number oscillates with decreasing amplitude and period , tending to an adiabatically constant value . also , as expected , the type iii mode exhibits an oscillating behavior before entering the resonance band . we can also conclude from figure 5 that the particle number follows the predicted evolution inside the resonance band , the main differences between the numerical and the analytical solutions occurring near the endpoints of the resonance band , where the transitions between resonant and oscillating regimes take place . the discrepancies for low particle number are also due to the normalization of @xmath328 , as discussed earlier . it is nevertheless clear that eq . ( [ eq100 ] ) gives a quite good description of the resonant particle production regime in an expanding universe , although it underestimates the particle number density , and that the contribution of the non - resonant periods are subdominant , as expected . the total particle number produced by the end of the resonant regime , @xmath329 , was computed for several values of @xmath319 , keeping @xmath330 and @xmath20 fixed at the values chosen above . the results we have obtained are shown in figure 6 . . plot ( a ) shows the results for small values of @xmath319 , including type ii and iii modes , and plot ( b ) shows the results for large values of @xmath319 , where only type iii modes are present . the solid line corresponds to the numerical results while the dashed line gives the corresponding analytical prediction . all results correspond to @xmath324 and @xmath325.,title="fig : " ] [ fig : figure_6 ] one observes that @xmath329 increases with @xmath319 for type ii modes and slowly decreases with @xmath319 for type iii modes , in agreement with eq . ( [ eq101 ] ) . the main discrepancies between the numerical results and the analytical prediction are again due to the smooth transition between the resonant and non - resonant regimes at @xmath291 , with the latter giving a subdominant contribution to the total particle number , which is more significant for type iii modes . the different number of oscillations the modes undergo before entering the resonance band give the oscillations that can be observed in figure 6 . these are suppressed for large @xmath319 as the pre - resonance oscillations of the particle number give a negligible contribution for high momentum modes . numerical simulations also show that , as expected , the particle number in all modes decreases with @xmath20 , so that hubble expansion may inhibit the resonant amplification of particle modes at early times . thus , we conclude that our analysis of the resonance regime gives a good description of the mechanism of particle production in branonium for an expanding universe and that the corrections arising from the non - resonant regimes are subdominant . extrapolation of this discussion to high momentum modes is , however , difficult , as these will only be excited at late times , where the approximations we have considered may no longer hold . the production of high momentum particles is nevertheless suppressed , as we have concluded earlier . furthermore , at each given time @xmath331 , only the modes with @xmath332 have entered the resonance band , which means that only those modes for which [ eq104 ] a_0k < a_0k^max(z ) have started being excited . this gives the following momentum cut - off for modes which have already been amplified at @xmath139 : [ eq105 ] k_max(t)=12 ^ 12(1 - 20)^-12 . notice that this gives the expected endpoint of the resonance band in the limit @xmath333 , given by eq . ( [ eq73 ] ) . the generalization of these results for massive particles follows the same reasoning used in the non - expanding case , with the substitution @xmath334 within our approximations . note that , in the expanding case , the physical momentum of the particles is redshifted while their mass remains the same as the universe expands . this modification will then produce a more complex evolution of the parameter @xmath335 in the modified mathieu equation , eq . ( [ eq91 ] ) , given to lowest order by : [ eq106 ] a_k(z)=a_k0(1-z)+4m^2_0 ^ 2(1-z ) , where @xmath336 and all other quantities are defined as before . although this will alter the way the resonance band moves into higher momentum modes as the probe s orbit decays , the evolution of the modes is expected to follow the same qualitative behavior as in the massless case , so that we will not analyze this in more detail . we , thus , conclude that , to lowest order , the parametric resonance survives in an expanding universe , although modes are excited only during a finite period . these results hold only at late times but we note once again that all quantities are measured with respect to the typical string values . we need , however , to bear in mind that eq . ( [ eq91 ] ) is not the mathieu equation and that several other terms in eq . ( [ eq52 ] ) may become important at early times . these may alter the excitation time and the amplification of each mode , or even prevent any particle production . it is nevertheless clear that the effects of the expansion are suppressed at the typical energies involved in the probe s motion , so that one still expects a significant number of particles to be produced . as the resonance develops , the probe brane loses energy to excited @xmath104-particle modes and one may wonder whether the energy radiated by this mechanism is sufficient to affect the probe brane s motion . let us start by computing the energy density of massless particles produced by the resonance in a non - expanding universe . we have seen that each rescaled mode function @xmath272 has an associated harmonic oscillator energy function @xmath337 . for the mode functions @xmath338 , the associated energy function is approximately the same in the non - relativistic and large distance limit , where @xmath339 . hence , the total energy density radiated into brane particle - modes is given by : [ eq107 ] _ p = e_k=12 ^ 2_0^+dk k^2_k(n_k+12 ) , where we have used the momentum space isotropy discussed earlier in this work . after a significant number of particles has been produced , we may take @xmath340 and neglect the @xmath179 factor in eq . ( [ eq107 ] ) . also , we know that particles are produced in a narrow resonance band centered at @xmath341 with a band width @xmath342 . the particle number distribution in momentum space can then be well approximated by a gaussian distribution of the form : [ eq108 ] n_k(t)=n_k_c(t)(-12(k - k_c)^2k(t)^2 ) . the gaussian distribution width @xmath343 should correspond to a fraction of the resonance band width , so that we may write @xmath344 . the coefficient @xmath345 is expected to be time - dependent as the central mode @xmath235 is more amplified than all the other modes in the resonance band , so that the distribution should become more sharply peaked about this value with time . the advantage of writing the particle number distribution in this form is that it allows us to further approximate the result by a @xmath346-function distribution , taking into account that @xmath347 , as discussed earlier . introducing the correct normalization , we may then write : [ eq109 ] n_k(t)n_k_c(t)k(t)(k - k_c ) . using @xmath348 , we obtain : [ eq110 ] _ p & & ( 2)^3k_resn_k_c + & & |q|(2)^4 n_k_c , where the time dependence of @xmath349 , @xmath345 and consequently @xmath350 is implicit . notice that , when @xmath351 , we have @xmath352 , as the resonance vanishes in this limit . to have an estimate of the order of magnitude of this quantity , take the example we have considered before in the discussion of particle production in a non - expanding universe , for which @xmath353 and @xmath354 , in units of the string length . by solving the equations of motion numerically and fitting the obtained particle number distribution to the gaussian distribution given in eq . ( [ eq108 ] ) , we may determine the values of @xmath345 and @xmath349 . for an initial time @xmath355 , we obtained the following results : [ eq111 ] n_k_c(5t_0)&&7.510 ^ 2 , ( 5t_0)0.20 , + n_k_c(10t_0)&&1.610 ^ 7 , ( 10t_0)0.12 . these two examples confirm that , indeed , the resonance width decreases in time . from eq . ( [ eq110 ] ) , we obtain for these two cases @xmath356 and @xmath357 , in units of the string energy scale . the energy density radiated into brane particle - modes should then be compared with the energy density of the interbrane distance field @xmath85 , which is given by @xmath358 . for the example considered above , @xmath359 , so that the value obtained for the energy density radiated into particles up to @xmath360 is negligible when compared to the probe brane s energy density . in general , we obtain : [ eq112 ] _ p||(2)^103 n_k_c , which gives the main result of this subsection . this means that , in the previous example , if we take @xmath361 , up to a @xmath104-particle number density of @xmath362 can be produced without affecting significantly the motion of the probe brane . these results can be easily generalized for the massive case , by taking into account the changes in @xmath235 and @xmath363 discussed earlier . for @xmath250 , we obtain : [ eq113 ] _ p||k_c(2)^73 n_k_c , with @xmath364 and which , as expected , reduces to eq . ( [ eq112 ] ) in the limit @xmath365 . as the momentum of the produced particles decreases with @xmath182 , it is easy to conclude that the energy damped into brane particle - modes is smaller for more massive particles ( recall that the strength of the resonance is not affected by the mass of the particles ) . the analysis of the energy radiated into brane particle - modes is more difficult to perform in the expanding universe case , as more effects have to be taken into account . first , we need to recall the approximate relation between @xmath328 and the physical particle number . next , we need to notice that the mode functions @xmath366 are redshifted by a factor of @xmath367 with respect to the harmonic oscillator mode functions @xmath272 . finally , for the purposes of determining the energy density of the produced particles , we may take @xmath368 . then , we have [ eq114 ] _ p(t)_k . we see that the energy density of the produced particles is redshifted by the usual power of @xmath369 , but is further reduced due to the redshift of the physical momentum and energy of the modes . we can use the expressions obtained for the particle number produced by the resonance in eq . ( [ eq101 ] ) to determine @xmath328 , as we have concluded that the effects of the non - resonant periods are subdominant . writing them in terms of the comoving momentum of the modes , we conclude that , for type iii modes , [ eq115 ] _ p(t)dk k^3(q_0 ^ 2_0 ^ 281k^2 ) . to determine the limits of this integral , one must recall that , at a given time , only a finite number of modes has entered the resonance band . in particular , one may use the value computed in eq . ( [ eq105 ] ) as the upper limit of this integral , although some of these modes have not yet been completely excited by the resonance . the lower limit of the integral will be the first type iii mode , with @xmath370 . this gives a finite energy density radiated into brane modes at each finite time @xmath139 . this energy density is redshifted by hubble expansion , but at the same time modes with increasing energy enter the resonance band . this process will , however , stop at some point , as our approximations will break down when the motion of the probe becomes relativistic and gets too close to the central stack . even if the resonance mechanism persists in this limit , although certainly in a different form , the probe will eventually modify its trajectory , either by colliding with the central stack or via some other mechanism that stabilizes its motion . the contribution of type ii modes is obviously finite and can be computed by integrating eq . ( [ eq114 ] ) using the corresponding expression for @xmath328 given in eq . ( [ eq101 ] ) , with a lower limit @xmath278 and an upper limit @xmath371 for the integral . no particular insight is , however , gained by computing the exact expressions for the energy density for both type ii and type iii modes . instead , we note that the dominant contribution to the energy density will be given by high momentum modes which enter the resonance band at late times . for these , the effects of the expansion of the universe will be less significant and one may use the results obtained in the non - expanding case to compute the energy density produced by the resonance . in this case one should use the values of @xmath71 and @xmath372 at the end of the resonance mechanism . as discussed in @xcite , the probe s trajectory in the background created by the central stack may induce not only the production of brane - bound particles but also radiation into bulk modes , namely gravitational , rr - form and dilatonic fields . this effect is due to the accelerated motion of the probe brane and most of the power is radiated into lower - spin fields , in particular scalar fields which couple to the brane s orbital monopole moment . the probe will thus lose energy through this process , contributing to the decay of its orbit . this will certainly modify the mechanism of resonant production of particle - modes in the brane , namely by varying the orbital frequency , which determines the comoving momentum of the modes that are excited by the parametric resonance . one then expects the effects of radiation into bulk modes to be quite similar to those induced by the universe s expansion , so that we will not analyze this process in detail . nevertheless , it is important to estimate its contribution to the decay of the probe s orbit . the power radiated into bulk scalar modes was estimated in @xcite , where it was shown that the number of orbits the probe can complete before the interbrane distance becomes of order @xmath39 is approximately given by : [ eq116 ] ( r_il_s)^321(g_s^3n)^12 , where @xmath373 is the initial value of the physical radius of the orbit , assuming it is circular . ( this gives a good estimate for the decay time , even though we are interested in small eccentricity orbits for the resonance process . @xmath374 corresponds to the ratio of the decay time to the initial orbital period . ) this expression shows that , for the probe to complete a large number of orbits before decay , it needs to be at a large distance from the central stack , in units of the string length . also , one needs @xmath375 to obtain a sufficiently large value of @xmath374 . this value should be compared to the number of orbits of the probe necessary for the resonance to be effective , which in the non - expanding case is of order @xmath376 . hence , the production of a significant number of particles requires @xmath377 , which gives for the initial radius of the orbit : [ eq117 ] r_il_s(32|q|)^23n^13g_s . for example , if @xmath378 and there are 10 branes in the central stack , we need the initial interbrane distance to be larger than about @xmath379 for the resonance to be effective , which is not too large a number taking into account that @xmath40 is parametrically small . as , in the analysis of the parametric resonance mechanism , we have assumed the interbrane distance to be large compared to the string length , we expect the effects of radiation into bulk modes to be initially negligible . this process should , however , become more important as the orbit decays and increases its acceleration . it is even possible that , at late times , it overcomes hubble expansion as the main energy loss process . nevertheless , the end of the resonance should still be determined by the break down of the non - relativistic and large distance approximation , as this also controls the amount of energy damped into bulk modes . so far we have considered the probe to move at distances from the central stack which are small compared to the typical size of the transverse space directions , so that we may neglect the effects of compactification on the probe s motion . in this section , we will analyze the leading order effects introduced by the finite size of these directions , and show that they may lead to the creation of orbital angular momentum , a necessary condition for resonant particle production to take place . we will consider the simplest case of compactifying the three transverse directions on an isotropic 3-torus of size @xmath380 , which is defined by the identifications @xmath381 . as mentioned earlier , the harmonic function associated with the central stack configuration needs to be modified in this case by including the appropriate brane images " , according to eq . ( [ eq10 ] ) . these images are placed at points with coordinates @xmath382 , with integer @xmath383 , defining a hypercubic lattice corresponding to the covering space of the 3-torus , @xmath384 . although toroidal compactifications are too simple to give realistic particle physics phenomenology , this will be sufficient to illustrate the main effects of compactification on the probe brane s trajectory . the interbrane potential associated with the generalized harmonic function eq . ( [ eq10 ] ) is obtained by computing the propagator for a massless field on a torus , which satisfies : [ eq118 ] ^2g(,)=(-)-1v _ , where @xmath385 is the volume of the transverse 3-torus . the term @xmath386 is included for consistency , so that the integral over the compact manifold of both sides of eq . ( [ eq118 ] ) vanishes . if one expands the green s function @xmath387 in terms of eigenfunctions of the laplacian operator @xmath388 , one concludes that the @xmath386 term removes the unphysical zero - mode which makes the green s function diverge . this term arises naturally from the curvature of the non - compact 4-dimensional spacetime @xcite . it has been shown that one can write the massless propagator on the 3-torus in terms of an integral involving the jacobi - theta function @xmath389 @xcite : [ eq119 ] g()g(,)=1r__0^s . although this expression gives the full compact space propagator , we are mainly interested in the leading order modifications introduced by compactification when the probe is moving closer to the central stack than to any of its images ( orbits around any of the brane images are equivalent due to the symmetries of the hypercubic lattice ) . in @xcite , an expansion of this propagator about the origin was computed using ewald s method for calculating potentials in hypercubic lattices in the context of solid state physics . the leading corrections to the central @xmath390 potential are given by : [ eq120 ] g()=-14r - r^26v_-c_sv_^13-a_4h_4()-a_6h_6()- the values for the coefficients of the various terms were found numerically to be @xmath391 , @xmath392 and @xmath393 . the functions @xmath394 and @xmath395 are harmonic functions of order @xmath396 and @xmath397 , respectively , and the series continues with harmonic terms of higher ( even ) orders . the leading harmonic correction to the propagator is given by : [ eq121 ] h_4()=1r_^5 . recalling that the function @xmath398 contributes to the interbrane potential through the graviton - dilaton and rr - form interactions , we may write this potential for @xmath399 , to order @xmath400 and discarding constant terms , as [ eq122 ] v()&=&-2q_6 , where we defined the function [ eq123 ] f(y^ir ) . we conclude that , to this order , the interbrane potential is modified by two terms . both terms correspond to repulsive contributions to the potential arising from the overall attraction of the image branes and their coefficients are suppressed by a factor @xmath401 , so that their effects are negligible for large transverse volume . jellium " term , of order @xmath402 and whose name arises in the context of solid state physics , gives an isotropic contribution while the asymmetry " term , of order @xmath400 , gives an anisotropic contribution to the potential . we will be more interested in the effects of the asymmetry term , as its breaks the rotational symmetry of the transverse space and leads to the generation of orbital angular momentum . the effects of the jellium term will , however , be more significant close to the central stack . it is easy to show that , for large transverse volume , this term will , to leading order , make the probe brane s orbits precess . this is due to its repulsive nature and , for small eccentricity orbits , the deficit angle is approximately given by @xmath403 . this effect will , however , be suppressed at late times as the probe s angular momentum is redshifted away by hubble expansion . the inclusion of the asymmetry term makes the analysis quite difficult , as the potential is no longer central and depends on both angular coordinates in the transverse space . this may lead , in particular , to non - planar orbits of the probe brane . however , if we set @xmath404 initially , the probe will feel no force along this direction and its trajectory will be confined to the @xmath405 plane as before . we will focus on this particular case , bearing in mind that in general non - planar trajectories may arise . within these assumptions , the problem reduces , as before , to the evolution of the canonically normalized complex scalar field @xmath85 , defined in eq . ( [ eq17 ] ) , in an expanding universe . its potential can be written as [ eq124 ] v()=-||-_j||^2-_a(_r^4+_i^4 - 6_r^2_i^2 ) , where @xmath406 and @xmath407 are , respectively , the real and imaginary parts of the field , @xmath408 was defined in eq . ( [ eq27 ] ) and [ eq125 ] _ j , _ a=32a_4t_6v_3q_6r_^5 give the strength of the jellium and asymmetry terms . this potential is no longer invariant under the global u(1 ) symmetry of the non - compact case , signaling that the probe s angular momentum is no longer conserved . this is quite similar to the u(1)-violating potential governing the evolution of the scalar field in the affleck - dine mechanism @xcite , where non - conservation of angular momentum plays a crucial role in generating the u(1 ) baryon number asymmetry in our universe . the potential remains , however , invariant under the hypercubic group symmetries @xmath409 and @xmath410 . in terms of the polar angle @xmath101 , this means that @xmath411 rotations as well as reflections about the @xmath412 axis are preserved by the asymmetry term , as one can easily conclude by writing eq . ( [ eq123 ] ) for @xmath80 as [ eq126 ] f()=1 - 2 ^ 2(2 ) . hence , we only need to consider initial conditions such that @xmath413 or equivalently @xmath414 . the u(1 ) symmetry violation precludes a complete analytical description of the orbits . however , it is not difficult to obtain a qualitative insight on the main features of the probe s trajectories . if the probe is placed significantly far from the central stack ( but still closer to it than to any of the image branes ) , the asymmetry term will not be negligible and some angular momentum will be created or destroyed . if the probe has no angular momentum initially , as we would expect immediately after inflation , then it will necessarily acquire some and be placed in an orbit around the central stack , instead of just falling towards it along the radial direction . as both the jellium and the asymmetry interactions are repulsive , it is possible that this trajectory does not remain bound to the central stack , becoming connected to one or more branes in the hypercubic lattice . hubble expansion will , however , redshift the probe s energy and angular momentum , so that one expects the probe to become bound to only one brane stack at late times . as the orbital radius decreases , the u(1)-violating term becomes less significant and the particle number associated with @xmath85 should asymptotically become constant , i.e. the angular momentum should vary only due to the universe s expansion . one also expects the jellium term , as well as the asymmetry term , to induce some precession of the orbital axis , which should stop at late times when these terms become negligible . due to the hypercubic symmetries , along @xmath415 the force acting on the probe will be in the radial direction , as @xmath416 , and no angular momentum will be created in these particular directions . for @xmath417 , angular momentum creation should , however , be a generic feature . to have a better understanding of how the initial conditions affect the amount of angular momentum created or destroyed , we have computed the force acting on the probe brane and the associated torque , given by : [ eq127 ] = -v =- v_r - v_i , + = ||=-_rv_i+_iv_r , where we have considered the field @xmath85 as a vector in the @xmath418 plane . recall that the torque gives the variation of the angular momentum , which in terms of the particle number density @xmath27 defined in eq . ( [ eq20 ] ) can be written as : [ eq128 ] dndt+3hn= . we have plotted the isotorque contours for some particular parameter values in figure 7 . we have restricted the position of the probe to the region @xmath419 , where the expansion of the potential about @xmath420 is sufficiently accurate and the probe s motion is bound to the central stack ( near the limits of this region , the orbits may not be bound to the central stack , as the actual region where the force points towards the central stack has a more complicated shape ) . ) plane . coordinates are normalized to the size of the transverse torus , with the value @xmath421 in this example , in units of the string length . we considered @xmath158 and @xmath422 in these units . note that the torque changes sign when crossing one of the @xmath423 axis , with integer @xmath27 , being negative for @xmath417 . also , the absolute value of the torque increases as one moves away from the origin.,title="fig : " ] [ fig : figure_7 ] observing this figure we see that the torque is maximized close to the boundary of the region mentioned above ( darker areas for @xmath414 ) . also , the longer the probe s trajectory remains close to this boundary the more angular momentum it is likely to gain . thus , maximum angular momentum creation should occur if the probe is placed below the centre of the hypercubic cell . the initial value of the hubble parameter @xmath107 will also affect the amount of angular momentum created . if it is of the same order of magnitude as the orbital frequency , one expects the probe to be driven into lower torque regions within a few periods . lower values of the hubble parameter should allow the probe s angular momentum to oscillate significantly as the probe moves through alternate regions of positive and negative torque , but no net angular momentum will be gained after completing the first few orbits . the asymptotic value of the comoving particle number @xmath424 will , hence , be determined by the initial path of the probe . we have simulated the evolution of the probe brane numerically to illustrate this discussion . the initial values for the real and imaginary parts of the field were defined as follows : [ eq129 ] _ r , i(t_0)=_r , ir _ , _ r , i=0 , with @xmath425 , giving @xmath417 , and @xmath426 , which ensures the probe is placed initially within the region where it is bound to the central stack . we have also computed the eccentricity of the orbit , as defined in eq . ( [ eq29 ] ) , with the probe s energy density including both the jellium and asymmetry terms . this quantity should only become meaningful at late times , when the latter terms become negligible , but its evolution tracks the creation of angular momentum along the probe s trajectory . figure 8 shows an example of the results obtained for the probe s orbit . in this case , the probe is placed below the centre of the hypercubic cell and , as expected , a significant angular momentum is produced during its motion . one observes , as expected , a small precession of the orbital axis and the decay of the orbit into regions of lower torque , which drives the orbital eccentricity towards a constant value . initial conditions were chosen in this example so that the inverse hubble parameter is of the same order of magnitude as the orbital period and , hence , the orbit is quickly stabilized . the eccentricity tends in this example to a constant value of @xmath427 . notice the similarities with the example illustrated in figure 1 , the main difference residing in the fact that , in this case , the probe has initially no angular momentum . , @xmath421 and @xmath428 . the probe is initially stationary at the point @xmath429 for @xmath430 and we consider a matter - dominated universe with @xmath131 . the asymptotic value obtained for the eccentricity is @xmath427.,title="fig : " ] [ fig : figure_8 ] in figure 9 we plot the asymptotic values of the eccentricity obtained numerically for different initial positions in the plane @xmath418 , considering two distinct values of the initial hubble parameter in a matter - dominated universe . in both cases , one observes that smaller eccentricity orbits are obtained near the larger torque regions below the centre of the hypercubic cell , as expected . we see that , in general , some angular momentum is produced , although the majority of initial conditions leads to highly eccentric orbits , with @xmath431 . we also conclude that the probe s orbit is generically less eccentric if its motion starts at later times , so that the initial hubble damping of the asymmetry term is smaller . as one may observe in figure 9 , in particular in plot ( b ) , the combination of the asymmetry term and hubble expansion makes the final eccentricity vary in a non - trivial way with the initial position of the probe , namely near @xmath432 , where the orbits may in some cases become unbound . note that these results correspond to particular values of the parameters of the problem and that the values of the asymptotic eccentricity are highly dependent on these parameters , namely the relevant distance scales @xmath79 and @xmath380 . , with @xmath433 ( with similar results for @xmath434 ) , assuming @xmath158 , @xmath421 and @xmath428 . both plots were obtained for a matter - dominated universe with @xmath131 and the motion of the probe starts at ( a ) @xmath435 and ( b ) @xmath436 . all parameters are given in units of the string length . contours were plotted for @xmath437 , with @xmath438 , and darker regions correspond to lower eccentricities.,title="fig : " ] [ fig : figure_9 ] these results show that a significant amount of angular momentum can be created by u(1)-violating terms in the probe s potential arising from compactification of the transverse dimensions , which is a key ingredient for the development of the parametric resonance we have studied in this work . the simple 3-torus compactification thus ilustrates how low eccentricity orbits can be created for a range of initial conditions , even if the probe has initially no angular momentum . potential applications of the resonant particle production mechanism to cosmology will be largely determined by the system s final configuration . as we have discussed before , the @xmath108 branonium we have analyzed in this work is unstable , as the probe ( anti)brane loses energy and angular momentum due to hubble expansion . at some stage , the distance between the probe and the central stack of branes will consequently become of the order of the string length ( although our approximations will break down long before this happens ) . when this occurs , the scalar mode associated with open strings stretching between the probe and the source branes becomes tachyonic , signaling an instability of the system @xcite . condensation of this tachyonic mode will then lead to the annihilation of the probe brane with one of the branes in the central stack . the non - supersymmetric nature of the @xmath108 configuration , which is a general property of all systems mixing brane and antibrane states , is behind this instability . namely , it makes the system evolve into a @xmath179 bps state with @xmath439 parallel @xmath1 branes , with the same total charge as the original configuration but with a lower energy . interesting scenarios may arise if some additional mechanism stabilizes the probe brane at a finite distance from the central stack , in which case the produced particles may survive after the resonance mechanism ends . an attractive possibility for stabilizing multiple brane systems was suggested in @xcite . when supersymmetry breaking occurs , one or more fields within the supermultiplets of the theory typically acquire masses , their value being set by the energy scale at which susy is broken . massive gauge potentials will then induce short - range interactions which , if repulsive , may balance the gravitational atraction between different branes . notice that the presence of a probe antibrane itself breaks susy but that this does not induce bulk field masses , so that other sources of susy breaking need to be considered in this case . suppose , for example , that the @xmath440 rr - form and consequently its magnetic dual form @xmath441 become massive after susy breaking , the same happening to the dilaton field , @xmath13 . both the rr - form and dilaton - mediated interactions will then be described by yukawa potentials whose range is determined by the masses @xmath442 and @xmath443 , respectively . if @xmath444 , the dilaton - mediated interaction will be exponentially suppressed for interbrane distances of order @xmath445 . assuming @xmath446 , we may write the relevant terms in the potential for @xmath447 as [ eq130 ] v(r)=m^4(-1r ) , where @xmath448 gives the overall constant factor . the constant @xmath345 parametrizes the effective strength of the rr - interaction compared to the gravitational atraction and one expects @xmath449 as the dilaton - mediated interaction is negligible . if , as we have assumed so far , the probe is a @xmath0-brane , the rr - interaction , although short - ranged , is attractive and can not balance the @xmath390 gravitational part . if , however , the probe is a @xmath1-brane , we have @xmath450 and stabilization of the system may be possible . in this case , for @xmath451 , the rr - interaction is exponentially suppressed and the probe s motion is governed only by the gravitational term . the evolution of the probe brane will , in this case , be very similar to that of a probe antibrane in a massless rr - form background which we have studied in this work , the potential having a smaller strength than in the latter case . this difference does not modify the qualitative features of the probe s trajectory and all the results we have derived in this work remain valid using an effective central stack charge @xmath452 . in particular , one expects the parametric resonance to develop in the probe s world - volume under the same conditions as before and the effects of transverse space compactification to provide the necessary source of angular momentum for similar initial conditions . the fate of a probe @xmath1-brane will , however , be quite different than that of a probe @xmath0-brane . as the probe s orbit decays due to the universe s expansion , the interbrane distance will eventually become of order @xmath445 . at these distances , the repulsive rr - interaction becomes relevant to the probe s motion and stabilization may be possible . the potential given in eq . ( [ eq130 ] ) has a local minimum formally given by : [ eq131 ] r_0=-m_rr^-1 , where @xmath453 is the lambert w - function , defined as the inverse of @xmath454 . after its angular momentum has been redshifted away by hubble expansion , one expects the probe to settle at this local minimum . this may not be the absolute minimum of the full potential , which includes the dilaton - mediated interaction and other supergravity and string theory corrections closer to the central stack . hence , absolute stabilization of the probe can not be guaranteed , but it is reasonable to expect at least a long - lived metastable state . from the properties of the lambert w - function , one concludes that @xmath455 for @xmath456 , being positive for @xmath457 , and that @xmath458 strictly grows with @xmath345 . hence , as expected , stabilization at large distances from the central stack is only possible if the rr - repulsion is stronger than the gravitational attraction . in this case , one expects the probe brane to stabilize away from the central stack at a distance @xmath459 . the scale of supersymmetry breaking will then determine the cosmological implications of the particle production mechanism analyzed in this work . if susy is broken at very high energies , inducing a large mass for the rr - potential ( but still smaller than the dilaton s mass ) , the probe will stabilize very close to the central stack and no tachyonic modes are present to induce annihilation . the stable probe and the central stack will then be characterized by a general broken gauge symmetry group @xmath460 , where @xmath461 is the number of branes in the probe , with an exponentially large number of particles charged under the @xmath181 gauge group if the resonance develops before stabilization occurs . this may be relevant , for example , to the generation of the baryon asymmetry in our universe . on the other hand , a parametrically small soft susy breaking mass will stabilize the probe at large distances . the parametric resonance may then be relevant for dark matter production , as particles in the probe s world - volume will necessarily interact weakly with the visible sector if the latter is embedded in the central stack . as we have shown , these particles may be supermassive ( although parametrically small compared to the string scale ) , so that this scenario could provide an effective mechanism for a non - thermal production of wimpzillas @xcite . one also expects the parametric resonance to amplify other bosonic modes living in the probe s world - volume , such as yang - mills fields if the probe has more than one brane . if these mediate baryon number - violating forces , their interactions with central stack fields could then provide a mechanism for baryogenesis as the one discussed in @xcite , in which case the probe would need to be stabilized before nucleosynthesis . it is also possible for fermionic particles living in the probe s world - volume to be produced in resonance , although pauli blocking makes this process significantly different from the bosonic case , according to the discussion given in @xcite . thus , if the probe can be stabilized , a resonant production of fermionic dark matter particles may also be achieved through this process . one also expects the interbrane distance field to oscillate about the local minimum of the potential , possibly generating a second stage of resonant amplification . this may be induced from coupling to other fields , as in the standard preheating mechanism , or gravitationally , as in the case described in this work . although this second stage may contribute significantly to the final particle number density in the probe brane s world - volume , we will not discuss them in further detail but rather refer the reader to the discussions given in this work and in the literature @xcite . we emphasize that all these scenarios depend on the particular embedding of the standard model fields in this setup , an issue for which there is still no complete answer and that is closely related to the nature of susy breaking and its communication to the standard model . in this work we have analyzed the mechanism of particle production in @xmath108 branonium systems , as well as in @xmath462 systems with susy breaking . we have shown that , in the limit of large distances and small velocities and for small eccentricity orbits , a parametric resonance will develop , producing scalar particles confined to the probe brane s world - volume . massless particles are produced in a narrow resonance band centered at the comoving momentum corresponding to one half of the probe s orbital frequency @xmath372 , as if resulting from the direct decay of the particles associated with the interbrane distance field . the strength of the resonance is given by the exponent @xmath239 defined in eq . ( [ eq72 ] ) . massive particles are produced in the same way , with energies of order @xmath463 . these energies are small compared to the string energy scale but , as the latter should be close to the planck scale , very massive particles can be produced by the resonance mechanism , according to eq . ( [ eq84 ] ) . for both massless and massive particles , the associated particle number grows exponentially in time and a large number of particles can be produced by this mechanism , just as in the case of preheating after inflation @xcite . however , the resonance surely requires the probe to complete a large number of orbits before the particle - modes can be significantly amplified , due to the small value of the exponent @xmath464 . hubble expansion of the 4-dimensional effective flat frw spacetime makes the probe s orbit decay , redshifting its energy and angular momentum . this alters the development of the parametric resonance significantly if the probe s motion does not occur at sufficiently late times . if , however , its motion begins late in the history of the universe , at least compared to typical string times , the resonance will still develop with the associated resonance band being shifted towards higher momentum modes . each mode will be excited during the finite period it spends inside the resonance band , after which the associated particle number becomes an adiabatic invariant . a significant number of particles , given appproximately by eq . ( [ eq101 ] ) , can be produced if this period is sufficiently long for the resonance to develop , according to the condition @xmath465 , where @xmath20 quantifies the effects of the expansion and was defined in eq . ( [ eq93 ] ) . we have also concluded that a large number of brane - bound particles can be produced without affecting the probe s orbital motion , as can be seen in eq . ( [ eq112 ] ) . radiation into bulk closed string modes will also have a negligible effect if the probe moves at sufficiently large distances , as assumed for the validity of our study . a realistic implementation of this mechanism requires particular initial configurations with large interbrane distances , non - relativistic velocities and almost circular orbits . we have shown that angular momentum creation may result from the effects of compactification of the directions transverse to the branes , illustrating this for the case of a compact 3-torus , where low eccentricity orbits can be produced for some range of initial conditions . in this work , we have also discussed how , for a probe @xmath1-brane , an interbrane potential may arise from supersymmetry breaking , if e.g. the relevant rr - form and the dilaton , @xmath13 , gain a mass in the process . stabilization of the probe is then possible at late times and at distances of order @xmath445 , if @xmath466 . such a stabilization avoids annihilation with the central stack , as necessarily happens for a probe antibrane , making the parametric resonance regime potentially relevant for baryon number or dark matter generation , for example . in our analysis , the parametric resonance arises only for @xmath108 or @xmath462 systems , where the @xmath390 potential allows the probe to move in closed elliptical orbits in a non - expanding universe . it is crucial for the development of the resonance that the interbrane distance field @xmath100 exhibits an oscillating behavior , although the orbits do not need to close , as we have seen for the case of an expanding universe . thus , other @xmath4 or @xmath467 branonium systems with @xmath468 , where the probe trajectories are necessarily unbounded from below , can not exhibit such resonant particle production . additional stabilization potentials may , however , provide the required oscillatory behavior , so that it may be possible to find a resonant particle production mechanism in other cases . it is nevertheless clear that resonant particle production in branonium systems may play an important role in our universe s evolution and we hope with this work to motivate future exploitation of their properties and applications . 0.2 cm * acknowledgements * 0.2 cm the authors would like to thank graham ross , martin schvellinger , andre lukas and yang - hui he for useful discussions . jgr is supported by fct ( portugal ) under the grant sfrh / bd/23036/2005 . jmr is partially supported by the ec network @xmath469 framework programme research and training network quest for unification " ( mrtn - ct-2004 - 503369 ) . this work was partially supported by the eu fp6 marie curie research and training network universenet " ( mrtn - ct-2006 - 035863 ) . 0.5 cm the particle number produced by the parametric resonance can be estimated by computing the integral in eq . ( [ eq100 ] ) . this gives the physical particle number apart from constant factors , subdominant non - resonant contributions and also an @xmath369 redshift factor due to hubble expansion . we have , in order to compute this function , let us write the exponent @xmath279 in the form : [ eqa2 ] _ k(z)=12 , where [ eqa3 ] a&=&q_0 ^ 2 ^ 2-a_k0 ^ 2 ^ 2 , + b&=&2(q_0 ^ 2+a_k0(a_k0 - 1 ) ) , + c&=&q_0 ^ 2-(a_k0 - 1)^2 . for all modes of interest , @xmath470 , while we have @xmath471 for type ii modes and @xmath472 for type iii modes . the sign of @xmath473 will not affect our results . it is also useful to define @xmath474 , which is explicitly given by [ eqa4 ] d^2=4q_0 ^ 2(2a_k0(a_k0 - 1)+(a_k0 - 1)^2 ^ 2+a_k0 ^ 2 ^ 2 ) , and is positive for all type ii and type iii momentum modes . with these considerations , we find , for @xmath475 , [ eqa5 ] n_k^res(z)&=12a_z_i^z . taking into account that @xmath476 , @xmath477 and also that @xmath478 , we can write the total particle number produced by the resonance for type ii modes as : [ eqa6 ] n_k^res(ii)(z_2)=-14a . similarly , for type iii modes , we obtain [ eqa7 ] n_k^res(iii)(z_2)=8d^2(-a)^32 . it is easy to check that eqs . ( [ eqa6 ] ) and ( [ eqa7 ] ) give the same result for the mode in the transition between types ii and iii , at @xmath479 , as expected . recalling that @xmath300 , these expressions can be approximated by those given in eq . ( [ eq101 ] ) . j. dai , r. g. leigh and j. polchinski , mod . lett . a * 4 * ( 1989 ) 2073 ; j. polchinski , phys . lett . * 75 * , 4724 ( 1995 ) [ arxiv : hep - th/9510017 ] . m. berkooz , m. r. douglas and r. g. leigh , nucl . b * 480 * , 265 ( 1996 ) [ arxiv : hep - th/9606139 ] . g. aldazabal _ et al . _ , j. math . phys . * 42 * , 3103 ( 2001 ) [ arxiv : hep - th/0011073 ] ; l. e. ibanez , f. marchesano and r. rabadan , jhep * 0111 * , 002 ( 2001 ) [ arxiv : hep - th/0105155 ] ; d. bailin , g. v. kraniotis and a. love , phys . b * 530 * , 202 ( 2002 ) [ arxiv : hep - th/0108131 ] ; d. cremades , l. e. ibanez and f. marchesano , jhep * 0207 * , 022 ( 2002 ) [ arxiv : hep - th/0203160 ] ; a. m. uranga , class . quant . grav . * 20 * , s373 ( 2003 ) [ arxiv : hep - th/0301032 ] ; c. kokorelis , jhep * 0209 * ( 2002 ) 029 [ arxiv : hep - th/0205147 ] ; c. kokorelis , jhep * 0208 * , 036 ( 2002 ) [ arxiv : hep - th/0206108 ] ; n. ohta and p. k. townsend , phys . b * 418 * , 77 ( 1998 ) [ arxiv : hep - th/9710129 ] . + for recent reviews see , e.g. , r. blumenhagen , m. cvetic , p. langacker and g. shiu , ann . rev . nucl . part . * 55 * , 71 ( 2005 ) [ arxiv : hep - th/0502005 ] ; r. blumenhagen , b. kors , d. lust and s. stieberger , phys . * 445 * , 1 ( 2007 ) [ arxiv : hep - th/0610327 ] . m. r. douglas , in _ cargese 1997 , strings , branes and dualities _ [ arxiv : hep - th/9512077 ] ; a. a. tseytlin , nucl . b * 475 * , 149 ( 1996 ) [ arxiv : hep - th/9604035 ] ; a. a. tseytlin , class . * 14 * , 2085 ( 1997 ) [ arxiv : hep - th/9702163 ] ; m. s. costa , nucl . b * 490 * , 202 ( 1997 ) [ arxiv : hep - th/9609181 ] ; z. guralnik and s. ramgoolam , nucl . b * 499 * , 241 ( 1997 ) [ arxiv : hep - th/9702099 ] . n. arkani - hamed , s. dimopoulos and g. r. dvali , phys . d * 59 * , 086004 ( 1999 ) [ arxiv : hep - ph/9807344 ] ; g. r. dvali and s. h. h. tye , phys . b * 450 * , 72 ( 1999 ) [ arxiv : hep - ph/9812483 ] ; n. arkani - hamed , _ et al . _ , b * 567 * , 189 ( 2000 ) [ arxiv : hep - ph/9903224 ] ; p. binetruy , c. deffayet and d. langlois , nucl . b * 565 * , 269 ( 2000 ) [ arxiv : hep - th/9905012 ] ; j. m. cline , c. grojean and g. servant , phys . rev . lett . * 83 * , 4245 ( 1999 ) [ arxiv : hep - ph/9906523 ] ; c. csaki , _ et al . _ , phys . lett . b * 462 * ( 1999 ) 34 [ arxiv : hep - ph/9906513 ] ; s. s. gubser , phys . d * 63 * , 084017 ( 2001 ) [ arxiv : hep - th/9912001 ] ; a. hebecker and j. march - russell , nucl . b * 608 * , 375 ( 2001 ) [ arxiv : hep - ph/0103214 ] ; h. a. bridgman , k. a. malik and d. wands , phys . d * 65 * , 043502 ( 2002 ) [ arxiv : astro - ph/0107245 ] ; a. kehagias and e. kiritsis , jhep * 9911 * , 022 ( 1999 ) [ arxiv : hep - th/9910174 ] ; p. kraus , jhep * 9912 * ( 1999 ) 011 [ arxiv : hep - th/9910149 ] . + for reviews of brane cosmology from the effective field theory viewpoint see , eg , d. langlois , prog . suppl . * 148 * , 181 ( 2003 ) [ arxiv : hep - th/0209261 ] ; r. maartens , living rev . * 7 * , 7 ( 2004 ) [ arxiv : gr - qc/0312059 ] . c. p. burgess _ et al . _ , jhep * 0107 * , 047 ( 2001 ) [ arxiv : hep - th/0105204 ] . c. p. burgess _ et al . _ , jhep * 0203 * , 052 ( 2002 ) [ arxiv : hep - th/0111025 ] . g. shiu and s. h. h. tye , phys . lett . b * 516 * , 421 ( 2001 ) [ arxiv : hep - th/0106274 ] ; n. t. jones , h. stoica and s. h. h. tye , jhep * 0207 * , 051 ( 2002 ) [ arxiv : hep - th/0203163 ] ; j. h. brodie and d. a. easson , jcap * 0312 * , 004 ( 2003 ) [ arxiv : hep - th/0301138 ] ; h. firouzjahi and s. h. h. tye , phys . b * 584 * , 147 ( 2004 ) [ arxiv : hep - th/0312020 ] . + for reviews of brane inflation in string theory see , eg , e. kiritsis , fortsch . phys . * 52 * , 200 ( 2004 ) [ phys . * 421 * , 105 ( 2005 errat,429,121 - 122.2006 ) ] [ arxiv : hep - th/0310001 ] ; s. h. henry tye , arxiv : hep - th/0610221 . j. h. traschen and r. h. brandenberger , phys . d * 42 * , 2491 ( 1990 ) ; l. kofman , a. d. linde and a. a. starobinsky , phys . lett . * 73 * , 3195 ( 1994 ) [ arxiv : hep - th/9405187 ] ; y. shtanov , j. h. traschen and r. h. brandenberger , phys . d * 51 * , 5438 ( 1995 ) [ arxiv : hep - ph/9407247 ] ; d. boyanovsky , h. j. de vega , r. holman , d. s. lee and a. singh , phys . d * 51 * , 4419 ( 1995 ) [ arxiv : hep - ph/9408214 ] ; m. yoshimura , prog . phys . * 94 * , 873 ( 1995 ) [ arxiv : hep - th/9506176 ] ; d. i. kaiser , phys . d * 53 * , 1776 ( 1996 ) [ arxiv : astro - ph/9507108 ] ; s. y. khlebnikov and i. i. tkachev , phys . rev . lett . * 77 * , 219 ( 1996 ) [ arxiv : hep - ph/9603378 ] . l. kofman , a. d. linde and a. a. starobinsky , phys . d * 56 * , 3258 ( 1997 ) [ arxiv : hep - ph/9704452 ] ; p. b. greene , l. kofman , a. d. linde and a. a. starobinsky , phys . d * 56 * , 6175 ( 1997 ) [ arxiv : hep - ph/9705347 ] ; g. n. felder , l. kofman and a. d. linde , phys . d * 59 * ( 1999 ) 123523 [ arxiv : hep - ph/9812289 ] . s. kachru , r. kallosh , a. linde , j. m. maldacena , l. p. mcallister and s. p. trivedi , jcap * 0310 * , 013 ( 2003 ) [ arxiv : hep - th/0308055 ] . s. shandera , b. shlaer , h. stoica and s. h. h. tye , jcap * 0402 * , 013 ( 2004 ) [ arxiv : hep - th/0311207 ] . a. sen , jhep * 9808 * , 012 ( 1998 ) [ arxiv : hep - th/9805170 ] ; a. sen , in _ cargese 1999 , progress in string theory and m - theory _ , arxiv : hep - th/9904207 . e. kolb , d. j. h. chung and a. riotto , in _ buenos aires 1998 , trends in theoretical physics 2 _ , arxiv : hep - ph/9810361 . e. w. kolb , a. linde and a. riotto , phys . lett . * 77 * , 4290 ( 1996 ) [ arxiv : hep - ph/9606260 ] . p. b. greene and l. kofman , phys . rev . d * 62 * , 123516 ( 2000 ) [ arxiv : hep - ph/0003018 ] .

theories with extra spacetime dimensions have drawn considerable attention ever since the original proposal by kaluza and klein . there has been renewed interest in such theories since the emergence of string theory . several new ideas in this context have been proposed and have interesting consequences for particle phenomenology and cosmology @xcite . in these higher - dimensional models , spacetime is usually taken to be a product of a four - dimensional spacetime and a compact manifold of dimension @xmath0 . while gravity can propagate freely through the extra dimensions , standard model particles are confined to the four dimensional spacetime . observers in this three spatial - dimensional wall ( a 3-brane " ) will measure an effective planck scale @xmath1 , where @xmath2 is the volume of the compact space . if @xmath2 is large enough it could make planck scale of the order of tev , thus removing the hierarchy between the planck and the weak scale . subsequently , randall _ et al . _ @xcite proposed a higher - dimensional scenario that is based on nonfactorizable geometry and accounts for the hierarchy without introducing large extra dimensions . however , the braneworld model itself is not stable and it was shown in ref . @xcite that by introducing a scalar field in the bulk , the modulus - namely the brane separation in the rs model - can be stabilized without any fine - tuning . assumption of negligibly small scalar backreaction on the metric in the gw approach prompted further work in this direction , where the modification of the rs metric due to backreaction of the bulk fields has been derived ( see @xcite ) . the stability issues in such cases have been reexamined for time - dependent cases @xcite ; also the effect of gauge fields or higher form fields have been studied in several works ( see @xcite ) . in an effort to search for the signatures of extra dimensions , the roles of the kaluza - klein modes of different bulk fields on the phenomenology at the standard model brane are of crucial importance . for the five - dimensional rs model , @xcite determined the bulk scalar kk modes and their self - interactions . it is found that due to the exponential redshift factor in the rs model , kk scalar modes in this spacetime have tev scale mass splitting and inverse tev couplings ( see @xcite ) . this is in sharp contrast to the kk decomposition in product spacetimes , which for large compactified dimensions , give rise to a large number of light kk modes ( see @xcite ) with a very small coupling with brane fields . due to this very distinct feature , the rs model has interesting consequences @xcite . motivated by string theory and other extra - dimensional models where one can have several extra dimensions , in this paper we extend the results of the bulk scalar model in five - dimensional rs space - time to arbitrary number of warped dimensions and have obtained the kk decomposition of the scalar kk masses . we have shown that in these multiply warped models we have much larger number of kk modes than the five - dimensional rs counterpart with effective couplings in the inverse tev range . we have also discussed possible numerical values for various parameters in our theory and have used them to get possible numerical values of low - lying kk mode masses in our multiply warped model , showing explicit localization in tev brane . our results also establish a general formula for determining these kk masses and couplings in the presence of any arbitrary number of extra dimensions . the paper is organized as follows : we give a brief explanation for six - dimensional doubly warped spacetime in sec . [ bmsix ] , the calculation for bulk scalar field has been done in this six - dimensional spacetime in sec . [ bmbulk ] . the same calculations have been finally extended to higher - dimensional spacetime with arbitrary number of extra dimensions in secs . [ bmseven ] and [ bmbulk7 ] . the paper ends with a short discussion of our results . in this section we shall discuss doubly compactified six - dimensional spacetime with @xmath3 orbifolding along each compactified direction . for a detailed discussion we refer the reader to @xcite . the manifold under consideration is given by , @xmath4\times s^{1}/z_{2}$ ] @xcite . we let the compactified dimensions to @xmath5 and @xmath6 , respectively . the noncompactified dimensions are taken to be , @xmath7 . the moduli along the compact dimensions are given by @xmath8 and @xmath9 , respectively . the corresponding metric ansatz is taken as @xmath10+r_{z}^{2}dz^{2},\ ] ] with @xmath11 . thus we have four orbifold fixed points , which are given by @xmath12 , respectively . the total bulk - brane action could be given by @xmath13 + \int d^{4}xdydz \left[v_{3}\delta(z)+v_{4}\delta(z-\pi)\right ] \\ & & s_{4}=\sum _ { i=1}^{2}\sum _ { j=1}^{2 } \int d^{4}xdydz \sqrt{-g_{vis}}\left(l - v\right)\delta ( y - y_{i})\delta ( z - z_{j}).\end{aligned}\ ] ] here the brane potentials in general have the particular functional dependence @xmath14 and @xmath15 . finally the full six - dimensional einstein s equation is given by , @xmath16 in the above expression @xmath17 , @xmath18 are bulk indices , @xmath19 , @xmath20 run over the usual four spacetime coordinates given by @xmath21 . the quantities @xmath22 and @xmath23 are the metric in @xmath24 and @xmath25 hypersurfaces , respectively . the line element derived from the above einstein s equation turns out to be @xcite @xmath26+r_{z}^{2}dz^{2}.\ ] ] in the above line element we have the following identification for the constants @xmath27 and @xmath28 given by @xmath29 the boundary terms lead to the brane tensions and using the einstein s equation across the two boundaries at @xmath30 , @xmath31 , respectively , thus we readily obtain @xmath32 thus the two 4-branes situated at @xmath30 and @xmath31 would have a @xmath6-dependent brane tension . the fact that the two tensions are equal and opposite is reminiscent of the original rs - form . similarly we get the brane tensions for other two 4-branes as @xmath33 here @xmath34 were introduced to account orbifolding along the @xmath6-direction and with @xmath35 being a constant , the resulting hypersurface should have only a constant energy density . the fact that @xmath36 is dependent on the coordinate @xmath6 makes the two hypersurfaces for y orbifolding to have a z - dependent energy density . the 3-brane located at @xmath37 suffers no warping and can be identified with the planck brane . the other three can be valid choices for standard model ( visible ) brane . however , if we assume that there is no brane having lower energy scale than ours , we are forced to identify the sm brane to be located at @xmath38 . the suppression factor on the tev brane can be given by @xmath39 the desired suppression of @xmath40 on the tev brane can be obtained by choosing different combinations of the parameters @xmath28 and @xmath27 . however we also have an extra relation as presented in eq . ( [ bm5 ] ) , which shows that in order to avoid large hierarchy between the two moduli @xmath8 and @xmath9 , either of the two parameters @xmath28 and @xmath27 must be large and the other should be small . for example , we can easily assume @xmath41 and @xmath42 . however a small hierarchy also exists in the original rs model , where there exists an one order of magnitude hierarchy between @xmath43 and @xmath27 , satisfying planck - to - tev scale warping by @xmath44 . a natural question that arises with this discussion is whether stabilization of these moduli to the desired values is possible . for the five - dimensional rs model this has been shown in @xcite by introducing a bulk stabilizing scalar field and tuning the vev of the scalar field at the boundaries . the modulus in the theory is stabilized near planck length without any fine - tuning . in this case as well we can adopt a similar procedure by introducing a bulk stabilizing scalar field . again choosing appropriate vev at the boundary , we can stabilize @xmath8 and @xmath9 to desired values @xcite . in our six - dimensional braneworld scenario with @xmath5 and @xmath6 dependence , the action for the scalar field can be expressed such that @xmath45 where the coupling parameters , @xmath46 tend to infinity as the scalar field approach to the following values , @xmath47 , @xmath48 , @xmath49 and @xmath50 . we take @xmath51 . now following ref . @xcite we can obtain the equation of motion in the separable form as @xmath52= \left(r_{y}^{2}b^{2}m^{2}-p\right)\chi ( z),\end{aligned}\ ] ] where we have made the following decomposition , @xmath53 and @xmath54 is the separability constant . also in the above expression prime denote differentiation with respect to @xmath5 and dot denotes differentiation with respect to z. finally , the above equations with appropriate boundary condition @xcite can be solved , which when substituted into the action leads to an effective potential for the moduli as @xmath55,\end{aligned}\ ] ] where we have used the following shorthand notations , @xmath56 , @xmath57 and @xmath58 . then solving the equations , @xmath59 and @xmath60 and then through the second derivatives with respect to @xmath61 and @xmath27 we can arrive at the minimum values of @xmath28 and @xmath27 . as an illustrative example we can start with @xmath62 and @xmath63 , leading to @xmath64 and @xmath65 . note that these values can resolve the gauge hierarchy problem . thus along this line any higher - dimensional braneworld models can have their moduli stabilized . from now on we shall assume that such a stabilization has been performed and all the moduli hence forth will have those stabilized values . in this analysis , following the stabilizing bulk scalar model , we have assumed that the backreaction of the bulk stabilizing field is negligibly small . moreover from the action of the bulk stabilizing scalar it may be observed that at the boundaries the stabilizing scalar tends to their vevs @xmath66 when the coupling @xmath67 tends to infinity . this is exactly similar to the five - dimensional counter part of the goldberger - wise calculation of modulus stabilization . as a result at the boundaries , the stabilizing scalar is frozen to different values @xmath66 and hence does not contribute to the dynamics of the model . in this section we carry out the kaluza - klein decomposition of a nongravitational bulk scalar field propagating in the spacetime described by eq . ( @xmath68 ) in the spirit of the work @xcite with bulk scalar field . we find that in these multiply warped spacetime the sm brane contains larger number of tev scale scalar kk modes than the five - dimensional rs model . this has significant phenomenological consequences @xcite . we consider a free scalar field in the bulk for which the action is given by @xmath69 where @xmath70 with @xmath71 is given by eq . ( @xmath68 ) , and m is of order of @xmath72 . after an integration by parts , this can be written as @xmath73,\end{aligned}\ ] ] where @xmath74 . now we make the following substitution for kk decomposition , @xmath75 the following normalization conditions are imposed on the fields @xmath19 and @xmath20 , @xmath76 @xmath77 the differential equation satisfied by the function @xmath78 is @xmath79 where @xmath80 stands for kk mode mass eigenvalue . the above differential equation can be further simplified and cast to the following form , @xmath81 the above equation can be solved in terms of bessel functions of first and second order as @xmath82,\ ] ] with @xmath83 representing an overall normalization . with extra - dimension parameter @xmath5 . the vertical line represents the @xmath31 line showing the fact that the quantity @xmath84 is maximum at @xmath31 , the position of tev brane.,width=384,height=288 ] now we can proceed much further . the mass modes determined by @xmath80 must be real . this reality condition imposed on the mass modes requires the differential operator on the left hand side of eq . ( [ bm14a ] ) to be self - adjoint . this self - adjointness imply that derivatives of @xmath85 should be continuous at the orbifold fixed points . these gives two conditions on the parameters @xmath80 and @xmath86 , expressed as , @xmath87 \left[2y_{2}(x_{m}e^{-c\pi})+x_{m}e^{-c\pi}y'_{2}\left(x_{m}e^{-c\pi}\right)\right ] \nonumber \\ & -&\left[2y_{2}\left(x_{m}\right)+x_{m}y'_{2}\left(x_{m}\right)\right ] \left[2j_{2}(x_{m}e^{-c\pi})+x_{m}e^{-c\pi}j'_{2}\left(x_{m}e^{-c\pi}\right)\right]\end{aligned}\ ] ] where @xmath88 . since to make planck scale down to tev scale we should have @xmath89 . then the above equation reduces to the following form , @xmath90 . then for light mode masses we have @xmath91 to be order of unity @xcite . this keeps @xmath92 also order of unity . then from eq . ( [ bm14a ] ) as well as from fig.[fig1 ] we observe that the modes @xmath85 are larger near the 3-brane at @xmath31 , which makes these low - mass kaluza - klein modes to be found preferentially near the @xmath31 region ( see fig.[fig1 ] ) . thus , with the tev brane being situated at @xmath31 , we observe that the low - mass kk modes are exponentially suppressed and hence confined to the tev brane . also for these low - lying kk mass modes the coefficient @xmath86 is of the order of @xmath93 , which shows that we can ignore the @xmath94 part compared to @xmath95 , while performing integrals involving @xmath84 . similar analysis for @xmath96 yields @xmath97 the solution , apart from an overall normalization , can be expressed as @xmath98 h_{\sqrt{5/2}kz}\left(\frac{-10k^{2}+b_{m}^{2}r_{z}^{2}(1+\cosh(2k\pi))}{10k^{2}}\right ) \nonumber \\ & + & e_{m}exp\left[-\frac{5}{2}k^{2}z^{2}\right]~_{1}f_{1 } \left(-\frac{-10k^{2}+b_{m}^{2}r_{z}^{2}(1+\cosh(2k\pi))}{10k^{2}},\frac{1}{2},\frac{5k^{2}z^{2}}{2}\right),\end{aligned}\ ] ] where @xmath99 is the kummer confluent hypergeometric function and @xmath100 is the hermite polynomial of degree n. then from fig.[fig2 ] we observe that this function is also maximum at @xmath101 . hence the low - lying kk mass modes are confined to the tev brane located at @xmath101 . from the behavior of both @xmath84 and @xmath102 we find that all the low - lying kk mass modes are confined to @xmath103 brane , i.e. , the tev brane . a possible experimental signature of the bulk scalar kk modes can originate via coupling of the bulk scalar to dihiggs in the form @xmath104 . for @xmath105 the dominant decay channels are @xmath106 and @xmath107 which leads to multijets as final product which though may be difficult to differentiate from the qcd back ground @xcite . also when the mass of bulk scalar is in the range of @xmath108 to @xmath109 gev then enhanced production of @xmath110 occurs . moreover for bulk scalar mass in the range @xmath111 to @xmath108 gev we have a relatively larger cross sections for the diphoton channel . in this region due to small mixing the branching ratio is dominated by @xmath106 and @xmath107 . the diphoton channel is a very promising search channel as branching ratio remains more or less at constant level even up to @xmath112 threshold @xcite . this might become possible if the lhc runs extends the diphoton searches for invariant masses above existing @xmath113gev . + to determine the parameters of the solution we proceed as follows : we want @xmath114 to be real , as it appears in the mass modes . thus self - adjointness also applies in this case . this implies that derivatives of @xmath96 to be continuous around the orbifold fixed points . at @xmath101 , this is trivially satisfied irrespective of the quantity @xmath115 . however at @xmath116 all the terms are suppressed by @xmath117 , thus the self - adjointness there leads to @xmath118 at large values of @xmath6 , confluent hypergeometric series have a large value . being in the denominator the term can be neglected for practical purposes . with extra - dimension parameter @xmath6 . the graph clearly shows the fact that the quantity @xmath102 is maximum at @xmath101 , the position of tev brane.,width=384,height=288 ] using the above equations we readily obtain the following action for the field @xmath119 as , @xmath120 where we have the following expression for the element @xmath121 , @xmath122 and @xmath123 as , @xmath124 now from the previous discussion we have the solution for these two sets of functions @xmath78 and @xmath125 , which can be used to determine @xmath121 and @xmath126 in order to obtain the masses of the kk modes by evaluating the quantity @xmath127 . in contrast to the five - dimensional situation ( see @xcite ) where the masses of the bulk fields appear as a diagonalized mass matrix , in this case the bulk field @xmath128 manifests itself to some four - dimensional observer as an infinite kk tower with mass being determined by the the quantity @xmath127 such that a scalar @xmath129 has a mass @xmath130 after an appropriate diagonalization procedure . the solution for @xmath78 is presented in equation ( [ bm14a ] ) . a similar solution was obtained by wise et . al ( see @xcite ) except for the fact that we have bessel functions of second order . following their discussion we can argue in a similar manner that the lightest kk modes have mass parameter @xmath80 suppressed exponentially with respecanct to the the scale @xmath131 appearing in eq . ( [ bm8 ] ) . since we have taken @xmath131 to be order of planck scale and @xmath28 to be around @xmath132 , by stabilization these mass modes @xmath80 are in the tev range . also from the solution we could observe that the modes @xmath78 are larger in the region @xmath31 . this has been explained earlier through graphical presentation of the function @xmath84 . the solution for @xmath125 has been presented in eq . ( [ bm15a ] ) . though we have argued following the graphical presentation of the function @xmath102 , we now provide a theoretical motivation for our above - mentioned results . the solution has an overall factor of @xmath133 $ ] and we see that the solution has maximum value around @xmath101 . hence the bulk field being a product of these two functions @xmath85 and @xmath134 as shown in eq . ( [ bm9 ] ) has mass parameter in the tev range and has maximum value around @xmath38 . now from sec . [ bmsix ] this is precisely the sm brane . hence , the bulk field has a maximum in the sm brane ; i.e. , the kk modes are most likely to be found in that region where @xmath80 and @xmath114 are in the tev range . this sets the stage for kk excitations to have tev scale mass splitting on the sm brane . now we would like to compute some low - lying kk mode masses numerically . for that we need to fix some parameters , @xmath27 , @xmath28 , @xmath9 and the bulk mass of the scalar field @xmath131 . we shall take the bulk mass to be in planck scale . then we can determine the remaining parameters , by making the following demands : ( a ) if we have a gauge boson field in this multiply warped scenario , its lowest massive kk modes should lead to @xmath135 and @xmath136 boson masses , ( b ) the suppression @xmath137 as presented in eq . ( [ branem01 ] ) should be @xmath138 , and finally ( c ) the hierarchy between @xmath8 and @xmath9 should be small . the kk mode of the gauge boson in this multiply warped spacetime can be obtained from ref . @xcite . this desired mass for @xmath135 and @xmath136 boson @xmath139gev can be obtained with @xmath140 and @xmath141gev , about 14 times smaller compared to planck scale . the other parameters @xmath27 and @xmath28 can be determined using small hierarchy between @xmath8 and @xmath9 along with desired warping of @xmath140 . this finally leads to , the following estimation : @xmath142 , @xmath143 and the ratio between moduli being @xmath144 . the suppression factor turns out to be @xmath145 . thus we will calculate the low - lying kk masses for our bulk scalar field with these sets of parameters ( see table.[tab1 ] ) . we now present the self - interactions of the bulk scalar field . from the four - dimensional point of view these self - interactions can induce couplings between the kk modes . in this case self - couplings of the light modes are suppressed by the warp factor and hence if the planck scale set the six - dimensional couplings , the low - lying kk modes have tev range self - interactions . we present the interaction term in the action with coupling parameter @xmath146 such that @xmath147 where the coupling @xmath146 is of the order of unity . then we can expand in modes and the self - interactions of light kk states become @xmath148 thus the effective four dimensional coupling constants are @xmath149 which reduces to , @xmath150^{2m}\int _ { 0}^{\pi}(\beta _ { q})^{2m}dz\ ] ] in the large @xmath151 and @xmath152 limit . hence we observe that the relevant scale for four - dimensional physics is not the scale set by planck scale , i.e. , @xmath17 , but this is @xmath153 . hence the kk reduction has lead the couplings from planck scale to the tev scale by the warp factor on the sm brane located at @xmath38 . .the masses of the kk modes of the scalar field are given in gev units . we have chosen the following values , @xmath141 gev , @xmath142 , @xmath143 . some representative masses of low - lying kk modes are given . [ cols="^,^,^,^ " , ] from the above discussion we now try to obtain some bounds on the parameters in our model , e.g. , @xmath8 , @xmath9 from the requirement of precision electroweak test . for that purpose we can use the same setup and put a bulk gauge boson whose kk modes can be detected in precision electroweak tests . we define a quantity denoted by @xmath154 where @xmath155 is the mass of @xmath135 gauge boson and @xmath156 is the mass of higher kk modes of the bulk gauge boson and @xmath157 is the effective four - dimensional gauge coupling along with @xmath158 to be the gauge couplings for higher kk modes . then from ref . @xcite we could argue that for precision electroweak test we should have @xmath159 with @xmath160 percent confidence level . from this result we can get the following bounds on the parameters of this model , @xmath161gev . this leads to a bound on @xmath162 as well by assuming a small hierarchy between the two moduli as , @xmath163gev . from ref . @xcite it can be easily verified that this bound is respected by gauge couplings and kk mode masses . thus these multiply warped models indeed satisfy precision electroweak tests . with an aim to arrive at a generic result we shall now try to extend our analysis with one more extra dimension . for that purpose we start with a seven - dimensional spacetime where the spacelike dimensions are successively warped . in other words the manifold of interest could be given by @xmath164\right\rbrace \times s^{1}/z_{2}\right]\times s^{1}/z_{2}$ ] . then the total brane - bulk action can be given by @xmath165 \nonumber \\ & + & \int d^{4}xdydzdw \left[v_{3}\delta(z)+v_{4}\delta(z-\pi)\right ] \nonumber \\ & + & \int d^{4}xdydzdw \left[v_{5}\delta(y)+v_{6}\delta(y-\pi)\right],\end{aligned}\ ] ] with appropriate actions ( @xmath166 ) for 12 possible 4-branes at the edges @xmath167 , @xmath168 and @xmath169 . we also have eight possible 3-branes at the corners @xmath170 . by natural extension of the method as illustrated in the previous section we get the line element and other parameters such that @xcite , @xmath171+r_{z}^{2}dz^{2}\right\rbrace + \re _ { w}^{2}dw^{2 } \nonumber \\ \ell ^{2}&=&\frac{-\lambda _ { 7}\re _ { w}^{2}}{15 } \nonumber \\ k&= & \frac{\ell r_{z}}{\re _ { w}\cosh(\ell \pi ) } \nonumber \\ c&=&\frac{\ell r_{y}}{\re _ { w}\cosh(k\pi)cosh(\ell \pi)}=\frac{kr_{y}}{r_{z}\cosh(k\pi)}\end{aligned}\ ] ] it may be of interest that the 5-brane at @xmath172 does not represent a flat metric ( y and z dependencies ) . in order to obtain substantial warping along the @xmath173 direction ( from @xmath172 to @xmath174 ) , one need to make @xmath175 substantial ( same order of magnitude as rs scenario ) . the seven - dimensional or triply warped model has a structure analogous to that of six - dimensional one , not only in the , form of functional dependence but also on the nature of warping . this method can easily be extended to even higher dimensions . also note that the orbifolding requires branes situated at edges of n - dimensional hypercube with 3-branes at the corners . if one of the direction suffers a large warping any other direction should have small warping so that there is no large hierarchy coming from the moduli . in this case also we have several candidates for our sm brane . however applying the fact that no brane should have less energy than ours , leads to @xmath176 to be sm brane . following the methods of previous sections , we shall carry out the kaluza - klein decomposition of a bulk scalar field propagating in the spacetime given by eq . ( @xmath177 ) . as in the previous section in this case as well we can write the bulk scalar field in terms of product of four functions . by making kk decomposition we again end up with kk mass modes having tev scale masses and splittings . the action for the bulk scalar field in this seven - dimensional spacetime can be given as @xmath178.\ ] ] from the line element as given by eq . ( @xmath179 ) , we readily obtain the following form for the action @xmath180,\end{aligned}\ ] ] where @xmath181 . we make the following substitution for the bulk field : @xmath182 we also impose the following normalization for the functions @xmath183 , @xmath184 and @xmath185 , @xmath186 now applying integration by parts to the integral as presented in eq . ( @xmath187 ) we readily obtain @xmath188 \nonumber \\ & -&\frac{1}{2}\frac{1}{r_{y}^{2}}\left[\sum_{pqrabc } \frac{\cosh^{4}(\ell w)}{\cosh^{4}(\ell \pi)}\frac{\cosh^{3}(kz)}{\cosh^{3}(k\pi ) } \phi _ { pqr}\phi _ { abc}\beta _ { q}\beta _ { b}\gamma _ { r}\gamma _ { c}\alpha _ { p } \partial _ { y}\left(e^{-4\sigma}\partial _ { y}\alpha _ { a}\right)\right ] \nonumber \\ & -&\frac{1}{2}\frac{1}{r_{z}^{2}}\left[\sum_{pqrabc } \frac{\cosh^{4}(\ell w)}{\cosh^{4}(\ell \pi)}e^{-4\sigma } \phi _ { pqr}\phi _ { abc}\alpha _ { p}\alpha _ { a}\gamma _ { r}\gamma _ { c}\beta _ { q } \partial _ { z}\left(\frac{\cosh^{5}(kz)}{\cosh^{5}(k\pi)}\partial _ { z}\beta _ { b}\right)\right ] \nonumber \\ & -&\frac{1}{2}\frac{1}{\re _ { w}^{2}}\left[\sum \frac{\cosh^{5}(kz)}{\cosh^{5}(k\pi)}e^{-4\sigma } \phi _ { pqr}\psi _ { abc}\alpha _ { p}\alpha _ { a}\beta _ { q}\beta _ { b}\gamma _ { r } \partial _ { w}\left(\frac{\cosh^{6}(\ell w)}{\cosh^{6}(\ell \pi ) } \partial _ { w}\gamma _ { c}\right)\right ] \nonumber \\ & + & \frac{1}{2}m^{2}\left[\sum \frac{\cosh^{6}(\ell w)}{\cosh^{6}(\ell \pi)}\frac{\cosh^{5}(kz)}{\cosh^{5}(k\pi)}e^{-4\sigma } \phi _ { pqr}\phi _ { abc}\alpha _ { p}\alpha _ { a}\beta _ { q}\beta _ { b}\gamma _ { r}\gamma _ { c}\right ] \big \rbrace.\end{aligned}\ ] ] then we make the following choice for the differential equations satisfied by the functions @xmath189 , @xmath102 and @xmath190 , @xmath191 the first equation as presented in ( [ bmm26a ] ) can be solved and has an identical solution as that obtained in the previous section . however , for convenience we rewrite the solution , @xmath192\ ] ] the second equation as given by eq . ( [ bmm26b ] ) has the same solution as presented in eq . ( [ bm15a ] ) but we rewrite it here , @xmath193 h_{\sqrt{5/2}kz}\left(\frac{-10k^{2}+b_{q}^{2}r_{z}^{2}(1+\cosh(2k\pi))}{10k^{2}}\right ) \nonumber \\ & + & e_{q}exp\left[-\frac{5}{2}k^{2}z^{2}\right]~_{1}f_{1 } \left(-\frac{-10k^{2}+b_{q}^{2}r_{z}^{2}(1+\cosh(2k\pi))}{10k^{2}},\frac{1}{2},\frac{5k^{2}z^{2}}{2}\right)\end{aligned}\ ] ] the third eq . ( [ bmm26c ] ) has the following solution along with an overall normalization , @xmath194 h_{\sqrt{3}\ell w } \left(\frac{-12\ell ^{2}+c_{r}^{2}\re_{w}^{2}(1+\cosh(2\ell \pi))}{12\ell ^{2}}\right ) \nonumber \\ & + & f_{r}exp\left[-3\ell ^{2}w^{2}\right]~_{1}f_{1 } \left(-\frac{-12\ell ^{2}+c_{r}^{2}\re_{w}^{2}(1+\cosh(2\ell \pi))}{24\ell ^{2}},\frac{1}{2},3\ell ^{2}w^{2}\right)\end{aligned}\ ] ] here also we have @xmath195 and @xmath196 to be bessel functions of first and second order respectively . along with these @xmath100 represents hermite polynomials and @xmath99 is the kummer confluent hypergeometric series . the arbitrary constants @xmath197 , @xmath198 and @xmath199 can be determined by the self - adjoint criteria and have the following expressions @xmath200 hence our final expression for the action is given by @xmath201 \label{bm27a } \\ m_{pqrabc}&=&\left\lbrace a_{p}^{2}\delta _ { pa}\delta _ { qb}\delta _ { rc}+b_{p}^{2}p_{rc}\delta _ { pa}\delta _ { qb } + c_{p}^{2}p_{qb}q_{rc}\delta _ { pa}+~m^{2}p_{pa}q_{qb}r_{rc}\right\rbrace , \label{bm27b}\end{aligned}\ ] ] where we have defined the following quantities , @xmath202 now from the previous discussion we can find the solution for the three sets of functions @xmath78 , @xmath125 and @xmath203 which in turn determine @xmath123 , @xmath204 and @xmath205 . therefore from these three functions the explicit expression for kk mass modes can be determined from the quantity @xmath206 as given by eq . ( [ bm27b ] ) . in this case as well the bulk field @xmath128 manifests itself to some four - dimensional observer as a scalar @xmath207 whose mass is determined by eqs . ( [ bm27a ] ) and ( [ bm27b ] ) . the solution for @xmath78 as presented in eq . ( [ bm26a ] ) has the same nature as obtained by wise _ ( see @xcite ) . in this case as well lightest kk modes have mass modes determined by @xmath80 , suppressed exponentially with respect to the the scaler mass @xmath131 which we have taken to be order of planck scale . thus these mass modes @xmath80 are in the tev range whereas m is of order of @xmath72 . the solution for @xmath125 and @xmath203 has been presented in eqs . ( [ bm26b ] ) and ( [ bm26c ] ) . the solution can be seen to include exponential factors such as @xmath133 $ ] , @xmath208 $ ] and we see that when mass parameter @xmath114 is of the order of tev , solutions have maximum value around @xmath101 as obtained earlier in sec . [ bmbulk ] as well . from the solution of @xmath203 it is evident that the solution has maximum value around @xmath174 . hence the bulk field being a product of these three functions @xmath209 , @xmath210 and @xmath211 as shown in eq . ( [ bmm4 ] ) , has mass parameter in the tev range and has maximum value to find the modes around @xmath212 which is the location of the sm brane . also the bulk field is maximum in the sm brane ; i.e. , the kk modes are most likely to be found in the tev region as the @xmath80 , @xmath114 and @xmath213 are in the tev range . along with the above line of arguments we could in principle have plotted all the functions @xmath209 , @xmath210 and @xmath211 and for all of them we have the functions to take maximum value at @xmath31 , @xmath101 and @xmath174 , precisely at the location of the tev brane . for completeness we present the self - interactions of the bulk scalar field in this seven - dimensional spacetime . from the four - dimensional point of view these self - interactions induce couplings between the kk modes . in this case also the effective self - couplings are suppressed by the warp factor and if the planck scale sets the six - dimensional couplings and the low - lying kk modes have tev range self - interactions . we present the interaction term in the action with coupling parameter @xmath146 such that @xmath214 where the coupling @xmath146 is of the order of unity . then we could expand in modes and hence the self - interactions of light kk states are given by @xmath215 the effective four - dimensional coupling constants therefore are being given by @xmath216 which in the large @xmath151 , @xmath152 and @xmath217 limit reduces to @xmath218^{2 m } \int _ { 0}^{\pi } dz(\beta _ { q})^{2 m } \int _ { 0}^{\pi } dw(\gamma _ { r})^{2m}.\end{aligned}\ ] ] hence we observe that the relevant scale for four - dimensional physics is not the scale set by planck scale but @xmath219 . hence the kk reduction lead to the tev scale couplings by the warp factor on the sm brane located at @xmath212 . now this result can easily be extended to any higher dimension spacetime . for n extra dimensions we can write the action for the bulk field as , @xmath220\ ] ] where @xmath70 with @xmath221 is given by a generalization of eq . ( @xmath177 ) , and m is of order of @xmath72 . then the kk splitting for the bulk field can be expressed as the following decomposition , @xmath222 thus among these n extra dimensions , one will have the solution given by eq . ( [ bm26a ] ) , and then the other @xmath223 solutions are being given by generalization of eq . ( [ bm26b ] ) such that the numerical values will be different but form of the solution remains unaltered . for nth extra dimension @xmath224 the solution for the mode can therefore be expressed as @xmath225 exp\left[-\frac{1}{2}n k^{2}w^{2}\right ] h_{\frac{\sqrt{3+n}}{2}kw } \left(\frac{-6k^{2}-2nk^{2}+m_{r}^{2}r^{2}(1+\cosh(2k \pi))}{2k^{2}(3+n)}\right ) \nonumber \\ & + & f_{r}exp\left[-\frac{3}{2}k^{2}w^{2}\right]\exp \left[-\frac{1}{2}n k^{2}w^{2}\right ] \nonumber \\ & \times & _ { 1}f_{1 } \left(-\frac{-6k^{2}-2nk^{2}+m_{r}^{2}r^{2}(1+\cosh(2k \pi))}{4k^{2}(3+n)},\frac{1}{2},\frac{1}{2}(3+n)k^{2}w^{2}\right)\end{aligned}\ ] ] hence the bulk field as viewed by a four - dimensional observer leads to a mass matrix whose components can be obtained by solving the eigenvalue problem as presented by each separable functions in the expansion given by eq . ( [ bmm6 ] ) . also all these eigenvalues have tev scale masses and the bulk field also has maximum value at @xmath226 , which is the sm brane . hence the standard model particles can be taken as low - lying kaluza - klein modes of a bulk field propagating in any number of extra - dimensional spacetime . the effective self - coupling is this case turns out to be @xmath227 @xmath228^{2 m } \int _ { 0}^{\pi } dz(\beta _ { q})^{2 m } \int _ { 0}^{\pi } dw(\gamma _ { r})^{2m}\cdots.\ ] ] thus finally we have obtained the kk mass modes and their self - interactions for n extra dimensions . we have also observed that in all these cases the kk mass modes and self - interactions are suppressed by the warp factor near the sm brane and hence all are in tev scale . hence this properties can be used to search for the tev range kk mass modes and self - interactions in next generation colliders . in this paper we generalize the work presented in ref . @xcite on the bulk scalar field to determine its kk modes and the effective self - interaction in a multiple warped spacetime . for arbitrary number of extra dimensions , we have derived the expressions for the kk mode masses and their self - interactions . various components determining the masses are in the tev range because of the warp factor suppression . moreover the bulk scalar field has been shown to have maximum value at the sm brane . hence the low - lying kk modes for the bulk scalar fields lie in the tev range with inverse tev self - coupling . thus the appearance of kk mode masses and couplings at tev scale are generic features of warped dimensional models with any number of extra warped dimensions as long as we want to resolve the gauge hierarchy problem without introducing any hierarchical moduli . we have also introduced the moduli stabilization mechanism in these multiply warped models and have obtained the stabilized values for the moduli . then we have presented a compact and generic formula to determine all the mass modes and their couplings for models with any arbitrary number of warped extra dimensions . this work now can be extended to other forms of bulk fields , which in turn may lead to the possibility of identifying various standard model particles as the low - lying kk excitation of various bulk fields , where the small warping in multiple directions can explain mass splitting in standard model particles as discussed in @xcite and @xcite . the close spacing of the low - lying kk modes along with enhanced coupling makes it likely for them to be seen as a series of close - lying resonances . in order to investigate the role of kk mass modes through collider - based experiments , we consider the interaction of various modes with thermselves , i.e. , self - interactions , and we have obtained that all of them are suppressed to the tev scale by the warp factor . also from the numerical values of masses for low - lying kk modes , we readily observe that the masses in the standard rs model get split into infinite number of mass modes , with very close spacings , which is a very interesting feature of these multiply warped models and can be probed in future runs of lhc . the other things to be noted are that with stronger coupling of the kk modes of the bulk scalar field , one expects the decay widths to be larger , and thus the peaks to be broader as we go to higher and higher dimensions . the nature of the line shapes , therefore , will be an interesting benchmark to distinguish between higher - dimensional and lower - dimensional kk signals if such excitations appear during the high - luminosity runs of the lhc . s.c . is funded by a spm fellowship from csir , government of india . i. antoniadis , phys . b * 246 * , 377 ( 1990 ) ; + j.d . lykken , phys . d * 54 * , r3693 ( 1996 ) ; + r. sundrum , _ ibid . _ * 59 * , 085009 ( 1999 ) ; + k.r . dienes , e. dudas and t. gherghetta , phys . b * 436 * , 55 ( 1998 ) . o. dewolfe , d.z . freedman , s.s . gubser and a. karch , phys . d * 62 * , 046008 ( 2000 ) ; + g.f . giudice , r. ratazzi and j.d . wells , nucl . b544 * , 3 ( 1999 ) ; + e.a . mirabelli , m. perelstein and m.e . peskin , phys . lett . * 82 * , 2236 ( 1999 ) . w.d . goldberger and m.b . wise , phys . d * 60 * , 107505 ( 1999 ) ( hep - ph/9907218 ) ; + s. kachru , m.b . schulz and e. silverstein , phys . d * 62 * , 045021 ( 2000 ) ; + e. dudas and m. quiros , nucl . phys . * b721 * , 309 ( 2005 ) . giudice , r. ratazzi and j.d . wells , nucl . * b544 * 3 ( 1999 ) ; + e.a . mirabelli , m. perelstein and m.e . peskin , phys . lett . * 82 * 2236 ( 1999 ) ; + t. han , j.d . lykken and r.j.zhang , phys . rev . * d59 * 105006 ( 1999 ) ; + j.l . hewett , phys . 82 * 4765 ( 1999 ) .

quasielastic nucleon knockout reactions have been extensively investigated with the aim of obtaining precise information on nuclear structure . the present work focuses on exclusive proton - induced @xmath8 and @xmath9 processes , whereby the residual @xmath10 nucleus is left in the discrete part of its energy spectrum . a sketch of the @xmath8 reaction is given in fig . [ fig : schem_repres ] . for a historical overview of the research into proton - induced nucleon emission off nuclear targets the reader is referred to refs . @xcite . reaction . the incoming proton undergoes `` soft '' initial - state interactions with the target before knocking out a bound proton through the primary high - momentum - transfer @xmath11 scattering . both the scattered and the ejected proton suffer final - state interactions while leaving the nucleus . the scattered and the ejected proton are detected in coincidence , while the residual nucleus remains unobserved.,scaledwidth=50.0% ] in quasielastic @xmath0 scattering the projectile is elastically scattered from a single bound nucleon in the target nucleus , resulting in the struck nucleon being knocked out of the target nucleus . this relatively simple reaction mechanism of one `` hard '' nucleon - nucleon collision is obscured by the `` soft '' initial- and final - state interactions ( ifsi ) of the incident and two outgoing nucleons with the nuclear medium . consequently , every model for @xmath8 and @xmath9 reactions has two issues to address : first , the description of the `` hard '' wide - angle scattering that leads to the ejection of the struck nucleon and second , the distorting mechanisms of the `` soft '' small - angle ifsi . concerning the treatment of the `` hard '' @xmath12 scattering part , essentially two methods exist . a so - called `` cross - section factorized '' approximation @xcite can be adopted so that the nucleon - nucleon scattering cross section enters as a multiplicative factor in the differential @xmath0 cross section . some results of exclusive @xmath8 measurements interpreted with this cross - section factorized form can be found in refs . the inclusion of spin - dependence in the description of the ifsi , however , breaks this factorization scheme . in that situation an alternative technique can be used : the amplitude factorized form of the cross section @xcite . in this approach , the two - body @xmath12 interaction can be approximated by the interpolation of phase shifts @xcite from free elastic @xmath12 scattering . various phenomenological forms to fit the amplitudes have also been used in the past . traditionally , the nucleon - nucleon scattering matrix has been parametrized in terms of five lorentz invariants @xcite , a method usually dubbed as the ia1 model or the spvat ( scalar , pseudoscalar , vector , axial vector , tensor ) form of the @xmath12 scattering matrix . differential cross section calculations adopting these five - term representations have been reported in refs @xcite . it should be noted , however , that although the spvat form gives reasonable predictions of @xmath0 observables , it is , in principle , not correct , as a five - term parametrization of the relativistic @xmath12 scattering matrix is inherently ambiguous @xcite . tjon and wallace @xcite have shown that a complete expansion of the @xmath12 scattering matrix ( commonly called the ia2 model ) contains @xmath13 independent invariant amplitudes . to date , the only calculations employing this general lorentz invariant representation have been performed in the context of the relativistic plane wave impulse approximation ( rpwia ) , i.e. , a model which ignores all ifsi mechanisms @xcite . the ifsi effects are typically computed by means of the distorted wave impulse approximation ( dwia ) theoretical framework @xcite . generally , in a dwia approach the scattering wave functions of the incoming and two outgoing nucleons are generated by solving the schrdinger or dirac equation with complex optical potentials . parametrizations for these optical potentials are usually not gained from basic grounds , but are obtained by fitting elastic nucleon - nucleus scattering data . several optical - potential parameter sets @xcite have been used in the description of quasifree proton scattering off nuclei . in the past , both nonrelativistic and relativistic dwia versions have proven successful in predicting @xmath0 cross sections over a wide energy range ( @xmath14@xmath15 mev ) and for a whole scope of target nuclei . most of the available calculations for the exclusive @xmath0 process addressed incident proton kinetic energies of a few hundred mev . in this work we aim at extending the formalism to scattering in the gev energy regime . as a matter of fact , the majority of dwia frameworks rely on partial - wave expansions of the exact solution to the scattering problem , an approach which becomes increasingly cumbersome at higher energies . in this energy range the eikonal approximation @xcite , which belongs to the class of high - energy semiclassical methods , offers a valid alternative for describing the ifsi . nonrelativistic eikonal studies of the @xmath8 reaction used in combination with optical potentials can be found in refs . @xcite . in this paper we propose a relativistic and cross - section factorized formalism based on the eikonal approximation for computing exclusive @xmath0 cross sections at incident proton energies in the few hundred mev to gev range . the eikonal formalism is implemented relativistically in combination with optical potentials @xcite , as well as with glauber theory @xcite , which is a multiple - scattering extension of the eikonal approximation . the two frameworks only differ in the way they treat the ifsi . the main focus will be on the optical - potential approach , as this method turns out to be the most suitable for the description of the ifsi for the kinematical settings discussed in this work . the paper is organized as follows . in secs . [ sec : cross_sect_matr_elt ] and [ sec : rpwia ] the factorized cross section is derived in the rpwia formalism . thereafter , the different methods to deal with the ifsi are developed in sec . [ sec : ifsi ] . [ sec : ifsi_factor ] is devoted to a presentation of optical - potential and glauber results for the ifsi factor . this is a function which accounts for all ifsi effects when computing the @xmath0 observables . the optical - potential predictions of our model are compared with cross section data that have been collected at pnpi and triumf in sec . [ sec : cross_sect_results ] . first , we present our calculations for the @xmath2c , @xmath3o , and @xmath4ca@xmath6 and @xmath16 pnpi data for @xmath1 gev incoming proton energies @xcite . second , the cross sections for @xmath5he@xmath6 scattering at @xmath7 mev are compared to the triumf data by van oers _ et al . finally , sec . [ sec : concl ] states our conclusions . in this section the formalism for the description of @xmath8 reactions is outlined . the generalization to @xmath9 reactions is straightforward . we conform to the conventions of bjorken and drell @xcite for the @xmath17 matrices and the dirac spinors , and take @xmath18 . the four - momenta of the incident and scattered proton are denoted as @xmath19 and @xmath20 . the proton momenta @xmath21 and @xmath22 define the scattering plane . the four - momentum transfer is given by @xmath23 , where @xmath24 , @xmath25 , and @xmath26 are the four - momenta of the target nucleus , residual nucleus , and the ejected proton . the standard convention @xmath27 is followed for the four - momentum transfer . in the laboratory frame , the fivefold differential cross section can be written as @xmath28 here , @xmath29 is the invariant matrix element which reflects the transition between the initial and final states . the hadronic recoil factor is given by @xmath30 with the energy transfer @xmath31 , the three - momentum transfer @xmath32 , and @xmath33 the angle between @xmath34 and @xmath35 . the @xmath8 matrix element is given by @xmath36 where @xmath37 is the unknown two - body operator describing the high - momentum transfer `` hard '' @xmath11 scattering , @xmath38 the ground state of the even - even target nucleus and @xmath39 the discrete state in which the residual nucleus is left . in coordinate space the matrix element takes on the form @xmath40 for the sake of brevity of the notations , only the spatial coordinates are explicitly written . in this section , the @xmath8 matrix element of eq . ( [ eq : matrix_elt ] ) will be analyzed in the rpwia . in this approach , only one hard collision between the projectile and a bound nucleon is assumed to occur , knocking the bound nucleon out of the target nucleus . the modelling of the `` soft '' ifsi processes , which affect both the incoming and outgoing protons , will be considered in sec . [ sec : ifsi ] . in evaluating the @xmath8 matrix element of eq . ( [ eq : matrix_elt ] ) , a mean - field approximation for the nuclear wave functions is adopted . we also assume factorization between the `` hard '' @xmath12 coupling and the nuclear dynamics . for reasons of conciseness , the forthcoming derivations are explained for the @xmath41 case . the generalization to arbitrary mass number @xmath42 is rather straightforward . the antisymmetrized @xmath43-body wave function in the initial state is of the slater determinant form @xmath44 details on the bound - state single - particle wave functions @xmath45 entering this mean - field @xmath43-body wave function can be found in appendix [ app : bound ] . the wave function of the incoming proton is given by a relativistic plane wave @xmath46 e^{i \vec{k } \cdot \vec{r } } \chi_{\frac{1}{2}m_{s } } = e^{i \vec{k } \cdot \vec{r } } \ : u ( \vec{k } , m_{s } ) \ ; . \label{eq : relat_plane_wave}\end{aligned}\ ] ] the @xmath43-body wave function in the final state reads @xmath47 relative to the target nucleus ground state written in eq . ( [ eq : ini_wave_fctn_slater ] ) , the wave function of eq . ( [ eq : fin_wave_fctn_slater ] ) refers to the situation whereby the struck proton resides in a state `` @xmath48 '' , leaving the residual @xmath10 nucleus as a hole state in that particular single - particle level . the outgoing protons are represented by relativistic plane waves . since both the initial and the final wave functions are fully antisymmetrized , one can choose the operator @xmath49 to act on two particular coordinates ( @xmath50 and @xmath51 ) . without any loss of generality , the @xmath8 matrix element of eq . ( [ eq : matrix_elt ] ) can be written as @xmath52 with @xmath53 the levi - civita tensor . in the rpwia , @xmath54 inserting this expression in eq . ( [ eq : matrix_elt_part_coord ] ) one obtains @xmath55 there are @xmath56 possible choices ( permutations ) for the indices @xmath57 , @xmath58 , , all giving the same contribution to the matrix element . accordingly , the above expression can be rewritten as @xmath59 because the bound - state wave functions are normalized to unity ( @xmath60 ) and @xmath61 , the matrix element can be further simplified to @xmath62 including a direct and an exchange term . substitution of the general form of the scattering operator @xmath63 , with @xmath64 the @xmath12 scattering amplitude in momentum space , in the above expression ( [ eq : rpwia_dir_exch_matrix_elt ] ) leads to @xmath65 for the direct term and a similar expression for the exchange term . here @xmath66 is the relativistic wave function for the bound nucleon in momentum space ( for details see appendix [ app : bound ] ) and @xmath67 is the missing momentum . in order to arrive at a cross - section factorized expression for eq . ( [ eq : fivefold_diff_cross_section ] ) , the quasielastic off - shell proton - proton scattering matrix element will be related to the free on - shell proton - proton cross section . for this purpose , we insert the completeness relation @xmath68 = 1 \label{eq : completeness}\ ] ] in @xmath69 and obtain the following expression for the matrix element @xmath70 here , @xmath71 is the matrix element for free @xmath11 scattering @xmath72 factorization breaks down , even when ifsi are disregarded , owing to the negative - energy projection term . to recover factorization , the negative - energy projection term is neglected in the remainder of this work @xmath73 using the expression of the relativistic bound - nucleon wave function in momentum space given in appendix [ app : bound ] , the @xmath74 contraction in eq . ( [ eq : pos_energy_proj_matr_elt ] ) reduces to @xcite @xmath75 where @xmath76 , @xmath77 indicates the spin projection of the spin spherical harmonic @xmath78 on a spin state @xmath79 , and the radial function in momentum space @xmath80 is given by @xmath81 with @xmath82 and @xmath83 the bessel transforms of the standard upper and lower radial functions of the bound - nucleon wave function in coordinate space ( see appendix [ app : bound ] for details ) and @xmath84 . upon squaring eq . ( [ eq : pos_energy_proj_matr_elt ] ) , the @xmath11 and nuclear bound - state parts get coupled by the summation over the intermediate spins @xmath85 and @xmath86 : @xmath87 after summation over @xmath57 , the struck nucleon s generalized angular momentum quantum number , the square of @xmath88 yields a @xmath89 , i.e. , becomes diagonal in @xmath85 . thereby , use is made of the following identity @xmath90 this leads to the decoupling between the @xmath11 scattering and the bound - state part in the matrix element : @xmath91 with @xmath92 the last factor in eq . ( [ eq : decoupled_squared_matrix_elt ] ) can be related to the free @xmath11 scattering center - of - mass cross section @xmath93 so that the rpwia differential @xmath8 cross section of eq . ( [ eq : fivefold_diff_cross_section ] ) can be written in the cross - section factorized form @xmath94 here , @xmath85 is the mandelstam variable for the @xmath11 scattering , not to be confused with the intermediate spin from the preceding eqs . ( [ eq : completeness])([eq : decoupled_squared_matrix_elt ] ) . in the numerical calculations which will be presented in the forthcoming sections , the free proton - proton cross section @xmath95 is obtained from the said code @xcite . it is well known that the factorized rpwia result of eq . ( [ eq : factorized_rpwia_cross_sect ] ) adopts an oversimplified description of the reaction mechanism . the momentum distribution @xmath96 , which represents the probability of finding a proton in the target nucleus with missing momentum @xmath97 , will be modified by the scatterings of the incoming and outgoing protons in the nucleus . therefore it is necessary to incorporate the effects of these ifsi in the model . first , in sec . [ sec : dist_factoriz ] , the differential @xmath8 cross section is written in a factorized form taking ifsi effects into account . next , the relativistic eikonal methods used for dealing with the ifsi effects in this work , are discussed in depth . two methods will be used . the relativistic optical model eikonal approximation ( romea ) is the subject of sec . [ sec : romea ] , whereas the relativistic multiple - scattering glauber approximation ( rmsga ) is discussed in sec . [ sec : rmsga ] . in both versions of the relativistic eikonal framework for @xmath0 reactions presented here ( romea and rmsga ) , the antisymmetrized initial- and final - state @xmath43-body wave functions , @xmath98 \label{eq : dist_ini_wave_fctn}\ ] ] and @xmath99 \ ; , \label{eq : dist_fin_wave_fctn}\end{aligned}\ ] ] differ from their respective rpwia expressions of eqs . ( [ eq : ini_wave_fctn_slater ] ) and ( [ eq : fin_wave_fctn_slater ] ) through the presence of the operators @xmath100 , @xmath101 , and @xmath102 . these define the accumulated effect of all interactions that the incoming and emerging protons undergo in their way into and out of the target nucleus . since the ifsi violate factorization , some additional approximations are in order . first , only central ifsi are considered , i.e. , spin - orbit contributions are omitted . further , the zero - range approximation is adopted for the `` hard '' @xmath12 interaction , allowing one to replace the coordinates of the two interacting protons ( @xmath50 and @xmath51 ) by one single collision point in the distorting functions @xmath100 , @xmath101 , and @xmath102 . this leads to the distorted momentum - space wave function @xmath103 similar to eq . ( [ eq : fourier_bound_wave ] ) , but with the additional ifsi factor @xmath104 accounting for the soft ifsi effects . now , along the lines of @xcite , it is natural to define a distorted wave amplitude @xmath105 so that the distorted momentum distribution is given by the square of this amplitude , @xmath106 this distorted momentum distribution has the following properties . first , it takes into account the distortions for the incoming and outgoing protons . second , it reduces to the plane wave momentum distribution @xmath96 in the plane wave limit when assuming that @xmath107 satisfies the relation @xmath108 between the upper and lower components . using the ansatz ( [ eq : dist_mom_dist ] ) for the distorted momentum distribution , the differential @xmath8 cross section can be cast in the form @xmath109 it differs from the rpwia expression ( [ eq : factorized_rpwia_cross_sect ] ) through the introduction of a `` distorted '' momentum distribution @xmath110 . as shown for example in refs . @xcite , in the relativistic eikonal limit the scattering wave function of a nucleon with energy @xmath111 and spin state @xmath112 subject to a scalar ( @xmath113 ) and a vector potential ( @xmath114 ) takes on the form @xmath115 e^{i \vec{k } \cdot \vec{r } } e^{i s(\vec{r } ) } \chi_{\frac{1}{2 } m_{s } } \ ; , \label{eq : eikonal_scatt_wave}\ ] ] where the eikonal phase @xmath116 reads @xmath117 \biggr ] \ ; , \label{eq : eikonal_phase}\ ] ] with @xmath118 and the average momentum @xmath119 pointing along the @xmath120-axis . the central and spin - orbit potentials @xmath121 and @xmath122 in the above expression are determined by @xmath113 and @xmath114 and their derivatives . in general , a fraction of the strength from the incident beam is removed from the elastic channel into the inelastic ones . these inelasticities are commonly implemented by means of the imaginary part of the optical potential . in evaluating the ifsi effects , three approximations are introduced . first , the dynamical enhancement of the lower component of the scattering wave function ( [ eq : eikonal_scatt_wave ] ) , which is due to the combination of the scalar and vector potentials , is neglected . second , the impulse operator @xmath123 is replaced by the asymptotic momentum @xmath124 of the nucleon . as mentioned before , the spin - orbit potential @xmath122 is also omitted . as a result , the effects of the interactions of the incoming and outgoing protons with the residual nucleus are implemented in the distorted momentum - space wave function of eq . ( [ eq : dist_mom_space_wave_fctn ] ) through the following phase factors @xmath125 @xmath126 @xmath127 [ eq : eikonal_ifsi_operators ] with the @xmath120-axes of the different coordinate systems lying along the trajectories of the respective particles ( @xmath120 along the direction of the incoming proton @xmath21 , @xmath128 along the trajectory of the scattered proton @xmath22 , and @xmath129 along the path of the ejected nucleon @xmath34 ) and @xmath130 , @xmath131 , and @xmath132 the coordinates of the collision point @xmath133 in the respective coordinate systems . the geometry of the scattering process is illustrated in fig . [ fig : geometry ] . the integration limits guarantee that the incoming proton only undergoes isi up to the point where the `` hard '' @xmath12 collision occurs and the outgoing protons are only subject to fsi after this `` hard '' collision . , @xmath134 , and @xmath135 are the impact parameters for each of the three paths for a collision occurring at @xmath133 . @xmath136 , @xmath137 , and @xmath138 are the @xmath120 coordinates of the collision point in the respective coordinate systems . @xmath139 and @xmath140 are the angles of the outgoing nucleons relative to the incoming proton direction.,scaledwidth=50.0% ] it is worth remarking that the eikonal ifsi operators of eq . ( [ eq : eikonal_ifsi_operators ] ) are one - body operators , i.e. , they do not depend on the coordinates @xmath141 of the residual nucleons . the normalization of the bound - state wave functions simplifies the ifsi factor ( [ eq : ifsi_factor ] ) considerably to @xmath142 in the romea case . in the numerical calculations , we have employed the global @xmath143 parametrizations of cooper _ et al . _ @xcite and the optical potential of van oers _ et al . _ @xcite to describe the pnpi and triumf data , respectively . hereafter , the @xmath8 calculations which adopt eq . ( [ eq : eikonal_ifsi_operators ] ) as a starting basis are labeled the relativistic optical model eikonal approximation ( romea ) . in the romea approach , all the ifsi effects are parametrized in terms of mean - field like optical potentials , i.e. , the ifsi are seen as a scattering of the nucleon with the residual nucleus as a whole . as the energy increases , shorter distances are probed and the scattering with the individual nucleons becomes more relevant . for proton kinetic energies @xmath144 gev , the highly inelastic and diffractive character of the underlying elementary proton - nucleon scattering cross sections makes the glauber approach @xcite more natural . this method reestablishes the link between proton - nucleus interactions and the elementary proton - proton and proton - neutron scattering . it essentially relies on the eikonal , or , equivalently , the small - angle approximation and the assumption of consecutive cumulative scatterings of a fast nucleon on a composite target containing `` frozen '' point scatterers ( nucleons ) . in relativistic glauber theory , the scattering wave function of a nucleon with energy @xmath111 and spin state @xmath112 reads @xcite @xmath145 e^{i \vec{k } \cdot \vec{r } } \chi_{\frac{1}{2}m_{s } } \ ; . \label{eq : glauber_scatt_wave}\ ] ] the operator @xmath146 implements the subsequent elastic or `` mildly inelastic '' collisions of the fast nucleon with the `` frozen '' spectator nucleons @xmath147 \ ; , \label{eq : glauber_operator}\ ] ] where @xmath148 ensures that the nucleon only interacts with other nucleons if they are localized in its forward propagation path . given the diffractive nature of @xmath149 collisions at gev energies , the profile function @xmath150 for central elastic @xmath149 scattering is parametrized in a functional form of the type @xmath151 at lower energies that part of the profile function proportional to @xmath152 is non - gaussian and makes significant contributions to nuclear scattering . rather than eq . ( [ eq : profile_fctn ] ) a parametrization in terms of the arndt @xmath12 phases @xcite is appropriate at lower energies . for the calculations presented here , which address higher energies , the gaussian - like real part of @xmath150 is the dominant contributor and the use of eq . ( [ eq : profile_fctn ] ) is justified . the parameters in eq . ( [ eq : profile_fctn ] ) can be determined directly from elementary nucleon - nucleon scattering experiments and include the total @xmath149 cross sections @xmath153 , the slope parameters @xmath154 , and the ratios of the real to the imaginary part of the scattering amplitude @xmath152 . we obtained these glauber parameters through interpolation of the data base available from the particle data group @xcite ( for more details , see ref . @xcite ) . as in the romea framework , only the central spin - independent contribution is retained and the impulse operator is replaced by the nucleon momentum . the glauber operators in eq . ( [ eq : ifsi_factor ] ) take the following form @xmath155 \ ; , \ ] ] @xmath156 \ ; , \ ] ] @xmath157 \ ; , \ ] ] [ eq : glauber_ifsi_operators ] where @xmath133 denotes the collision point and @xmath141 are the positions of the frozen spectator protons and neutrons in the target . the @xmath158 , @xmath159 , and @xmath160 coordinate systems are defined as in the previous section . the step functions make sure that the incoming proton can only interact with the spectator nucleons which it encounters before the `` hard '' collision and the outgoing protons can only interact with the spectator nucleons which they find in their forward propagation paths . contrary to romea , the glauber ifsi operators of eq . ( [ eq : glauber_ifsi_operators ] ) are genuine @xmath42-body operators , so the integration over the coordinates of the spectator nucleons in eq . ( [ eq : ifsi_factor ] ) has to be carried out explicitly . this makes the numerical evaluation of the glauber ifsi factor very challenging . standard numerical integration techniques were adopted to evaluate the ifsi factor and no additional approximations , such as the commonly used thickness - function approximation , were introduced . henceforth , we refer to calculations based on eq . ( [ eq : glauber_ifsi_operators ] ) as the relativistic multiple - scattering glauber approximation ( rmsga ) . in this section , results for the ifsi factor ( [ eq : ifsi_factor ] ) are given for the knockout of nucleons from the fermi level in @xmath2c , @xmath3o , and @xmath4ca , at an incident energy @xmath161 gev and a scattered proton kinetic energy @xmath162 mev . thereby , we adopt coplanar scattering angles @xmath163 on opposite sides of the incident beam , i.e. , kinematics corresponding with the pnpi experiment of ref . all ifsi effects are included in the ifsi factor @xmath164 . note that in the absence of initial- and final - state interactions the real part of the ifsi factor equals one , whereas the imaginary part vanishes identically . the @xmath8 ifsi factor is a function of three independent variables @xmath165 . the @xmath120-axis is chosen along the direction of the incoming beam @xmath21 , the @xmath166-axis lies along @xmath167 and the @xmath168-axis lies in the scattering plane defined by the proton momenta @xmath21 and @xmath22 . @xmath169 and @xmath170 denote the polar and azimuthal angles with respect to the @xmath120-axis and the @xmath168-axis , respectively . the radial coordinate @xmath171 represents the distance relative to the center of the target nucleus . to gain a better insight into the dependence of the ifsi factor on @xmath171 , @xmath169 , and @xmath170 , we calculated the contribution of the three distorting functions @xmath100 , @xmath101 , and @xmath102 to the ifsi factor . in figs . [ fig : re.12c.p.edad.phi0.3d ] and [ fig : im.12c.p.edad.phi0.3d ] results are displayed for the computed real and imaginary part of @xmath172 , for proton emission from the fermi level in @xmath2c . the results were computed within the romea framework , using the edad1 optical - potential parametrization of @xcite . the @xmath169 dependence can be interpreted as follows . for a given @xmath171 , the distance that the incoming proton travels through the target nucleus before colliding `` hard '' with a target nucleon , decreases with increasing angle @xmath169 . as a consequence , small values of @xmath169 induce the largest isi . for the fsi of the scattered proton , the opposite holds true and @xmath173 for a large @xmath171 value corresponds to an event whereby the `` hard '' collision transpires at the outskirts of the nucleus and the scattered proton has to travel through the whole nucleus before it becomes asymptotically free , thus giving rise to the smallest ( largest ) values for the real ( imaginary ) part of the ifsi factor . unlike the scattered proton , which moves almost collinear to the @xmath120-axis , the ejected nucleon leaves the nucleus under a large scattering angle @xmath140 . hence , the fsi are minimal for @xmath169 close to @xmath174 or @xmath175 and maximal for @xmath169 around @xmath176 . finally , the @xmath169 dependence of the complete ifsi factor is the result of the interplay between the three distorting effects , with the strongest scattering and absorption observed at @xmath169 close to @xmath174 , @xmath176 , and @xmath175 . figure [ fig : re.40ca.n.edai.phi0.radial ] displays the real part of the ifsi factor as a function of @xmath171 at various values of @xmath169 . the romea calculations were performed for the same kinematics as in figs . [ fig : re.12c.p.edad.phi0.3d ] and [ fig : im.12c.p.edad.phi0.3d ] , and use the edai optical - potential fit of @xcite . turning first to the upper left panel , it can be inferred from the picture that the isi effects increase with growing @xmath171 for @xmath177 . for @xmath178 and increasing @xmath171 , initially , the growing distance the proton has to travel through the nucleus leads to a decrease of the real part of @xmath100 . this is followed by an increase for larger @xmath171 up to @xmath179 . this reduction in isi effects with increasing @xmath171 is brought about by the incoming proton s path through the nucleus moving away from the nuclear interior and closer to the less dense nuclear surface . the other curves of the upper left figure illustrate a general trend for @xmath180 : as @xmath171 increases , the real part of the ifsi factor grows correspondingly . as can be appreciated from fig . [ fig : re.40ca.n.edai.phi0.radial ] as well as from the previous figures , the global behavior of the @xmath101 factor describing the scattered proton s fsi can be related to that of the isi factor @xmath100 through the substitution @xmath181 . this approximate symmetry can be attributed to the small scattering angle @xmath139 , i.e. , the scattered proton leaves the nucleus almost parallel to the incoming proton s direction . in the bottom left panel , the additional curve ( @xmath182 , i.e. , close to @xmath176 ) represents the situation of maximal fsi of the ejected nucleon . for this @xmath169 value , the path of the ejected nucleon passes through the center of the nucleus and the distance travelled through the nucleus increases with @xmath171 . accordingly , the real part of @xmath102 is a monotonously decreasing function of @xmath171 . the other extreme is the @xmath183 case where increasing @xmath171 means less fsi . for the other @xmath169 values , the absorption reaches its maximum for some intermediate @xmath171 value . again , the combination of @xmath100 , @xmath101 , and @xmath102 determines the total ifsi factor with the strongest attenuation predicted in the nuclear interior . the dependence of the ifsi factor on the azimuthal angle of the collision point is quite straightforward . one representative result is displayed in fig . [ fig : re.16o.p.edad2.r3fm.3d ] . here , @xmath184 ( @xmath185 ) refers to a situation where the `` hard '' @xmath12 collision occurs in the upper ( lower ) hemisphere with respect to the @xmath186-plane ( see fig . [ fig : geometry ] for a collision point located in the upper hemisphere ) . due to the cylindrical symmetry about the @xmath120-axis , the factor describing the isi of the incoming proton is independent of @xmath170 . regarding the scattered proton , we observe the least fsi in the upper hemisphere , since the proton then avoids passing through the highly absorbing nuclear interior . for the ejected nucleon , the contrary applies and the strongest fsi effects are found for @xmath187 . as the @xmath188-plane is defined as the scattering plane , the ifsi factor possesses the symmetry @xmath189 for coplanar scattering . results for the emission from levels other than the fermi level have not been plotted here , as it turns out that the ifsi factors hardly depend on the single - particle level in which the struck nucleon resides . the peculiar spatial characteristics of the different single - particle orbits have an impact on the observables , though . indeed , the distorted momentum - space wave function @xmath190 of eq . ( [ eq : dist_mom_space_wave_fctn ] ) is determined by the values of the ifsi factor folded with a relativistic bound - state wave function @xmath191 . as the particles experience less ifsi close to the nuclear surface , one obtains a stronger reduction of the quasifree cross section for nucleon knockout from a level which has a larger fraction of its density in the nuclear interior . this will become apparent in fig . [ fig : delta_r ] , but even more so in sec . [ sec : pnpi ] . figure [ fig : delta_r ] shows a function @xmath192 which represents the contribution of the nuclear region with radial coordinate @xmath171 to the differential cross section . the procedure to calculate this function is similar to the method exposed in ref . @xcite and is developed in appendix [ app : radial_contrib ] . comparison of the upper and lower panels illustrates that ifsi mechanisms make the @xmath8 cross sections reflect surface mechanisms , unlike the @xmath193 reaction where the weakly interacting electron probes the entire nuclear volume and only the outgoing proton interacts with the residual system . apart from the shift to higher @xmath171 , the ifsi brings about a strong reduction in the magnitude of the cross sections , whereby the fermi level is least affected . even though @xmath192 is concentrated in the surface region , the average density seen through this reaction still amounts to @xmath194 @xmath195 ( @xmath196 @xmath195 ) or @xmath197 ( @xmath198 ) of the central density in the case of @xmath199 knockout from @xmath2c ( @xmath4ca ) . in the case of emission from the fermi level , on the other hand , the average density is only @xmath200 ( @xmath201 ) of the central density for a @xmath2c ( @xmath4ca ) target . also , the ifsi factors for neutron emission are almost identical to the corresponding ifsi factors for proton knockout and , as expected , the overall effect of ifsi is more pronounced for heavier target nuclei . in this subsection , we investigate the sensitivity of the computed ifsi factors to the adopted parametrizations for the optical potentials and compare the romea results with the rmsga predictions . as can be seen in fig . [ fig : re.12c.n.romea_compar.radial_polar_azimuth ] , the ifsi factor depends on whether @xmath42-dependent ( edad1/edad2 ) or @xmath42-independent ( edai ) fits for the potentials are selected , but the global features are comparable . figure [ fig : re.12c.p.rmsga.phi0.3d ] , as contrasted to fig . [ fig : re.12c.p.edad.phi0.3d ] , demonstrates that the rmsga method adequately describes the isi of the incoming proton and the fsi of the scattered proton . however , the discrepancies between romea and rmsga become significant in the calculation of the fsi of the ejected nucleon ( note the different scales in the bottom left panels of figs . [ fig : re.12c.p.edad.phi0.3d ] and [ fig : re.12c.p.rmsga.phi0.3d ] ) , and therefore also in the complete ifsi factor . the noted difference is attributed to the low ejectile kinetic energy ( @xmath202 mev for the specific case of fig . [ fig : re.12c.p.rmsga.phi0.3d ] , and comparable values for knockout from other levels and other nuclei ) . at such low energies , the rmsga predictions are not realistic due to the underlying approximations , mostly the postulation of linear trajectories and frozen spectator nucleons . so , for the kinematics discussed here , the romea method is to be preferred over the rmsga one , as the latter overestimates the distortion for the low - energetic ejectile nucleon . c to the adopted choice for the parametrization of the optical potentials . results of romea calculations with the edad1 ( solid curve ) , edad2 ( dashed curve ) , and edai ( dot - dashed curve ) optical potentials are shown . kinematics as in fig . [ fig : re.12c.p.edad.phi0.3d].,title="fig:",scaledwidth=50.0% ] c to the adopted choice for the parametrization of the optical potentials . results of romea calculations with the edad1 ( solid curve ) , edad2 ( dashed curve ) , and edai ( dot - dashed curve ) optical potentials are shown . kinematics as in fig . [ fig : re.12c.p.edad.phi0.3d].,title="fig:",scaledwidth=50.0% ] c to the adopted choice for the parametrization of the optical potentials . results of romea calculations with the edad1 ( solid curve ) , edad2 ( dashed curve ) , and edai ( dot - dashed curve ) optical potentials are shown . kinematics as in fig . [ fig : re.12c.p.edad.phi0.3d].,title="fig:",scaledwidth=50.0% ] the pnpi experiment @xcite was carried out with an incident proton beam of energy @xmath1 gev . the scattered proton was detected at @xmath203 with a kinetic energy between @xmath204 and @xmath205 mev , while the knocked - out nucleon was observed at @xmath206 having a kinetic energy below @xmath207 mev . figures [ fig : diffcross.12c.p.romea.edai][fig : diffcross.40ca.n.romea.edai ] display a selection of differential cross section results as a function of the kinetic energy of the most energetic nucleon in the final state . the edai optical potential @xcite was used for the romea calculations . the other parametrizations of ref . @xcite produce similar predictions , whereas the rmsga approach fails to give an adequate description of the data because of the low kinetic energy of the ejected nucleon . since the experiment of ref . @xcite only measured relative cross sections , the romea results were normalized to the experimental data . the romea calculations reproduce the shapes of the measured differential cross sections . furthermore , comparison of the rpwia and romea calculations shows that the effect of the ifsi is twofold . first , ifsi result in a reduction of the rpwia cross section that is both level and @xmath42 dependent . from the figures it is clear that ejection of a nucleon from a deeper lying level leads to stronger initial- and final - state distortions . this reflects the fact that the incoming and outgoing nucleons encounter more obstacles when a deeper lying bound nucleon is probed . the @xmath42 dependence also conforms with our expectations , i.e. , the ifsi effects are larger for heavier nuclei . besides the attenuation , the ifsi also make the measured missing momentum different from the initial momentum of the struck nucleon . as can be inferred from fig . [ fig : momdistr.12c.p.1p32.romea.edai ] , this momentum shift leads to an asymmetry between the positive and negative missing - momentum side of the momentum distribution . note that positive missing momentum corresponds to @xmath208 . c@xmath6 momentum distribution for the @xmath209 state as a function of the missing momentum . the solid ( dashed ) curve are romea ( rpwia ) calculations.,scaledwidth=50.0% ] finally , we present some results for the @xmath5he@xmath6 reaction at an incident proton energy of @xmath7 mev . figure [ fig : diffcross.4he.p.vanoers.250 ] compares the data from the triumf experiment @xcite with romea calculations using the optical potential of van oers _ et al . _ @xcite . the typical shape for knockout of an @xmath85-state proton is reproduced by the romea predictions . this fair agreement of the romea results with the data demonstrates that our romea model also works satisfactorily at lower incident energies . a relativistic and cross - section factorized framework to describe the ifsi in quasielastic @xmath0 reactions has been outlined . the model , which relies on the eikonal approximation , can use either optical potentials or glauber multiple - scattering theory to deal with ifsi . thanks to the freedom of choice between these two substantially different techniques , our model is expected to be applicable at both intermediate and high incident energies . the properties of the ifsi factor have been investigated for an incident proton energy of @xmath1 gev . not surprisingly , the strongest attenuation occurs in the nuclear interior and heavier target nuclei are found to induce larger ifsi effects . also , the surface - peaked character of the @xmath0 reaction was clearly established to be a consequence of the ifsi . whereas the different types of optical - potential sets contained in ref . @xcite yield comparable ifsi factors , the rmsga calculations exhibit an unrealistic behavior for these kinematics . the romea model has been used to calculate cross sections for the kinematics of two different experiments : quasielastic proton scattering from @xmath2c , @xmath3o , and @xmath4ca at @xmath1 gev , and @xmath5he@xmath6 scattering at @xmath7 mev . the predictions are shown to reproduce the shape of the data reasonably well at both incoming energies , thereby providing support for the wide applicability range of our model . although the rmsga approach was deemed unsuitable for the kinematics discussed here , it should prove useful when trying to describe nuclear transparency data . work in this direction is in progress . this work was supported by the fund for scientific research , flanders ( fwo ) and the research council of ghent university . for spherically symmetric potentials , the solutions @xmath210 to a single - particle dirac equation have the form @xcite @xmath211 \ ; , \label{eq : bound_state}\ ] ] where @xmath58 denotes the principal , @xmath212 and @xmath57 the generalized angular momentum quantum numbers . the @xmath213 are the spin spherical harmonics and determine the angular and spin parts of the wave function , @xmath214 with @xmath215 the fourier transform of the relativistic bound - nucleon wave function is given by @xmath216 \ ; , \label{eq : fourier_bound_wave}\ ] ] with @xmath84 . the radial functions @xmath82 and @xmath83 in momentum space are obtained from their counterparts in coordinate space : @xmath217 @xmath218 with @xmath219 the spherical bessel function of the first kind . the differential @xmath0 cross section ( [ eq : factorized_ifsi_cross_sect ] ) is proportional to the distorted momentum distribution @xmath110 of eq . ( [ eq : dist_mom_dist ] ) . when approximating the completeness relation ( [ eq : completeness ] ) as @xmath220 i.e. , neglecting the negative - energy term as in sec . [ sec : rpwia ] , this amounts to @xmath221 thus , with @xmath222 defined as @xmath223 the function @xmath224 \label{eq : delta_r } \\ & = & \sum_{m } \bigl ( \overline{d } \left ( r_{1 } \right ) \ : \phi_{\alpha_1}^{d } + \overline{\phi_{\alpha_1}^{d } } \ : d \left ( r_{1 } \right ) \bigr ) \nonumber\end{aligned}\ ] ] represents the contribution of an infinitesimal interval in @xmath171 around @xmath225 to the @xmath0 cross section . this procedure also enables us to estimate the average density seen through this reaction as @xmath226 g. jacob and th . j. maris , rev . mod . phys . * 38 * , 121 ( 1966 ) . g. jacob and th . j. maris , rev . * 45 * , 6 ( 1973 ) . p. kitching , w. j. mcdonald , th . j. maris , and c. a. z. vasconcellos , adv . . phys . * 15 * , 43 ( 1985 ) . n. s. chant and p. g. roos , phys . c * 15 * , 57 ( 1977 ) . s. kullander , f. lemeilleur , p. u. renberg , g. landaud , j. yonnet , b. fagerstrm , a. johansson , and g. tibell , nucl . phys . * a173 * , 357 ( 1971 ) . r. k. bhowmik , c. c. chang , p. g. roos , and h. d. holmgren , nucl . a226 * , 365 ( 1974 ) . r. bengtsson , t. berggren , and ch . gustafsson , phys . rep . * 41 * , 191 ( 1978 ) . p. kitching , c. a. miller , w. c. olsen , d. a. hutcheon , w. j. mcdonald , and a. w. stetz , nucl . phys . * a340 * , 423 ( 1980 ) . w. t. h. van oers , b. t. murdoch , b. k. s. koene , d. k. hasell , r. abegg , d. j. margaziotis , m. b. epstein , g. a. moss , l. g. greeniaus , j. m. greben , j. m. cameron , j. g. rogers , and a. w. stetz , phys . rev . c * 25 * , 390 ( 1982 ) . d. s. carman , l. c. bland , n. s. chant , t. gu , g. m. huber , j. huffman , a. klyachko , b. c. markham , p. g. roos , p. schwandt , and k. solberg , phys . b452 * , 8 ( 1999 ) . n. s. chant and p. g. roos , phys . c * 27 * , 1060 ( 1983 ) . r. a. arndt , i. i. strakovsky , and r. l. workman , phys . c * 62 * , 034005 ( 2000 ) ; scattering analysis interactive dial - in program ( said ) , http://gwdac.phys.gwu.edu/. j. a. mcneil , l. ray , and s. j. wallace , phys . c * 27 * , 2123 ( 1983 ) ; j. a. mcneil , j. r. shepard , and s. j. wallace , phys . lett . * 50 * , 1439 ( 1983 ) ; j. r. shepard , j. a. mcneil , and s. j. wallace , _ ibid . _ * 50 * , 1443 ( 1983 ) . c. j. horowitz , phys . c * 31 * , 1340 ( 1985 ) ; c. j. horowitz and m. j. iqbal , _ ibid . _ * 33 * , 2059 ( 1986 ) ; d. p. murdock and c. j. horowitz , _ ibid . _ * 35 * , 1442 ( 1987 ) ; c. j. horowitz and d. p. murdock , _ ibid . _ * 37 * , 2032 ( 1988 ) . o. v. maxwell , nucl . a600 * , 509 ( 1996 ) ; * a638 * , 747 ( 1998 ) ; o. v. maxwell and e. d. cooper , _ ibid . _ * a656 * , 231 ( 1999 ) . e. d. cooper and o. v. maxwell , nucl . a493 * , 468 ( 1989 ) ; o. v. maxwell and e. d. cooper , _ ibid . _ * a513 * , 584 ( 1990 ) ; * a565 * , 740 ( 1993 ) ; * a574 * , 819 ( 1994 ) ; * a603 * , 441 ( 1996 ) . y. ikebata , phys . c * 52 * , 890 ( 1995 ) . j. mano and y. kudo , prog . theor . phys . * 100 * , 91 ( 1998 ) . c. a. miller , k. h. hicks , r. abegg , m. ahmad , n. s. chant , d. frekers , p. w. green , l. g. greeniaus , d. a. hutcheon , p. kitching , d. j. mack , w. j. mcdonald , w. c. olsen , r. schubank , p. g. roos , and y. ye , phys . c * 57 * , 1756 ( 1998 ) . r. neveling , a. a. cowley , g. f. steyn , s. v. frtsch , g. c. hillhouse , j. mano , and s. m. wyngaardt , phys . c * 66 * , 034602 ( 2002 ) . d. l. adams and m. bleszynski , phys . * b136 * , 10 ( 1984 ) . j. a. tjon and s. j. wallace , phys . c * 32 * , 1667 ( 1985 ) . b. i. s. van der ventel and g. c. hillhouse , phys . c * 69 * , 024618 ( 2004 ) . d. f. jackson and t. berggren , nucl . 62 * , 353 ( 1965 ) . a. nadasen , p. schwandt , p. p. singh , w. w. jacobs , a. d. bacher , p. t. debevec , m. d. kaitchuck , and j. t. meek , phys . c * 23 * , 1023 ( 1981 ) . p. schwandt , h. o. meyer , w. w. jacobs , a. d. bacher , s. e. vigdor , m. d. kaitchuck , and t. r. donoghue , phys . c * 26 * , 55 ( 1982 ) . d. g. madland , los alamos national laboratory report la - ur-87 - 3382 ( unpublished ) . e. d. cooper , b. c. clark , r. kozack , s. shim , s. hama , j. i. johansson , h. s. sherif , r. l. mercer , and b. d. serot , phys . rev . c * 36 * , r2170 ( 1987 ) . e. d. cooper , b. c. clark , s. hama , and r. l. mercer , phys . b206 * , 588 ( 1988 ) . s. hama , b. c. clark , e. d. cooper , h. s. sherif , and r. l. mercer , phys . c * 41 * , 2737 ( 1990 ) . e. d. cooper , s. hama , b. c. clark , and r. l. mercer , phys . c * 47 * , 297 ( 1993 ) . l. antonuk , p. kitching , c. a. miller , d. a. hutcheon , w. j. mcdonald , g. c. neilson , and w. c. olsen , nucl . phys . * a370 * , 389 ( 1981 ) . c. samanta , n. s. chant , p. g. roos , a. nadasen , j. wesick , and a. a. cowley , phys . rev . c * 34 * , 1610 ( 1986 ) . a. a. cowley , j. j. lawrie , g. c. hillhouse , d. m. whittal , s. v. frtsch , j. v. pilcher , f. d. smit , and p. g. roos , phys . c * 44 * , 329 ( 1991 ) . a. a. cowley , g. j. arendse , j. a. stander , and w. a. richter , phys . lett . * b359 * , 300 ( 1995 ) . a. a. cowley , g. j. arendse , r. f. visser , g. f. steyn , s. v. frtsch , j. j. lawrie , j. v. pilcher , t. noro , t. baba , k. hatanaka , m. kawabata , n. matsuoka , y. mizuno , m. nomachi , k. takahisa , k. tamura , y. yuasa , h. sakaguchi , t. itoh , h. takeda , and y. watanabe , phys . c * 57 * , 3185 ( 1998 ) . g. p. mccauley and g. e. brown , proc . london * 71 * , 893 ( 1958 ) . c. j. joachain , _ quantum collision theory _ ( elsevier , amsterdam , 1975 ) . r. j. glauber and g. matthiae , nucl . b21 * , 135 ( 1970 ) . s. j. wallace , phys . c * 12 * , 179 ( 1975 ) . d. r. yennie , in _ hadronic interactions of electrons and photons _ , edited by j. cummings and d. osborn ( academic press , new york , 1971 ) , p. 321 . s. l. belostotsky , yu . v. dotsenko , n. p. kuropatkin , o. v. miklukho , v. n. nikulin , o. e. prokofiev , yu . a. scheglov , v. e. starodubsky , a. yu . tsaregorodtsev , a. a. vorobyov , and m. b. zhalov , in _ proceedings of the international symposium on modern developments in nuclear physics _ ( novosibirsk , 1987 ) , p. 191 . j. d. bjorken and s. d. drell , _ relativistic quantum mechanics _ ( mcgraw - hill , new york , 1964 ) . j. a. caballero , t. w. donnelly , e. moya de guerra , and j. m. udas , nucl . phys . * a632 * , 323 ( 1998 ) . j. r. vignote , m. c. martnez , j. a. caballero , e. moya de guerra , and j. m. udas , phys . c * 70 * , 044608 ( 2004 ) . r. d. amado , j. piekarewicz , d. a. sparrow , and j. a. mcneil , phys . c * 28 * , 1663 ( 1983 ) . d. debruyne , j. ryckebusch , w. van nespen , and s. janssen , phys . c * 62 * , 024611 ( 2000 ) . d. debruyne , j. ryckebusch , s. janssen , and t. van cauteren , phys . lett . * b527 * , 62 ( 2002 ) . j. ryckebusch , d. debruyne , p. lava , s. janssen , b. van overmeire , and t. van cauteren , nucl . phys . * a728 * , 226 ( 2003 ) . k. hagiwara _ et al . _ , d * 66 * , 010001 ( 2002 ) , http://pdg.lbl.gov . k. hatanaka , m. kawabata , n. matsuoka , y. mizuno , s. morinobu , m. nakamura , t. noro , a. okihana , k. sagara , k. takahisa , h. takeda , k. tamura , m. tanaka , s. toyama , h. yamazaki , and y. yuasa , phys . 78 * , 1014 ( 1997 ) . j. d. walecka , _ electron scattering for nuclear and nucleon structure _ ( cambridge university press , cambridge , 2001 ) .

many models have been developed for behavioral changes , but only a few are formulated in terms of mathematical relations . for example , * _ game theory _ ( von neumann and morgenstern , 1944 ) , based on the concept of success of meeting strategies , is used for the description of cooperation and competition processes among individuals , * _ decision theories _ ( domencich and mcfadden , 1975 , ortzar and willumsen , 1990 ) , assuming the maximization of utility , successfully model the choice behavior among several alternatives , * _ diffusion models _ ( coleman , 1964 , bartholomew , 1967 , granovetter , 1983 , kennedy , 1983 ) mathematically describe the spread of behaviors or opinions , rumors , innovations , etc . all these models are related to a more general behavioral model discussed in the following . this model is based on boltzmann-_like equations _ and includes _ spontaneous _ ( or _ externally induced _ ) behavioral changes and behavioral changes by _ pair interactions _ of individuals ( sect . [ s2 ] ) . these changes are described by _ transition rates_. they reflect the results of mental and psychical processes , which could be simulated with the help of osgood and tannenbaum s ( 1955 ) _ congruity principle _ , heider s ( 1946 ) _ balance theory _ or festinger s ( 1957 ) _ dissonance theory_. however , it is sufficient for our model to determine the transition rates empirically ( sect . the _ ansatz _ used for the transition rates distinguishes _ imitative _ and _ avoidance processes _ , and assumes _ utility maximization _ by a variant of the _ multinomial logit model _ ( domencich and mcfadden , 1975 , ortzar and willumsen , 1990 ) ( sect . [ s2.1 ] ) . in section [ s3 ] a consequent mathematical formulation related to an idea of lewin ( 1951 ) is developed , according to which the behavior of individuals is guided by a _ social field_. this formulation is achieved by a second order taylor _ approximation _ of the boltzmann - like equations leading to _ diffusion equations_. because of their relation with the boltzmann equation ( boltzmann , 1964 ) and the fokker - planck equation ( fokker , 1914 , planck , 1917 ) they will be called the boltzmann - fokker - planck _ equations_. according to these new equations the most probable behavioral change is given by a vectorial quantity that can be interpreted as _ social force _ ( sect . [ s3.1 ] ) . the social force results from external influences ( the environment ) as well as from individual interactions . in special cases the social force is the derivative ( gradient ) of a potential . this potential reflects public opinion , social norms and trends , and will be called the _ social field_. by _ diffusion coefficients _ individual variation of the behavior ( the `` freedom of will '' ) is taken into account . in section [ s4 ] representative cases are illustrated by computer simulations . the boltzmann - fokker - planck model for the behavior of individuals under the influence of a social field shows some analogies with the physical model for the behavior of electrons in an electric field ( e.g. of an atomic nucleus ) ( helbing , 1992a , c ) . in particular , individuals and electrons influence the concrete form of the _ effective _ social , respectively , electric field . however , the behavior of electrons is governed by a _ different _ equation : the schrdinger _ equation _ ( schrdinger , 1926 , davydov , 1976 ) . let us consider a population consisting of a great number @xmath0 of individuals . concerning a special topic of interest , these individuals show a behavior @xmath1 out of several possible behaviors in the set @xmath2 . due to `` freedom of the will '' one can not expect a deterministic theory for the temporal change @xmath3 of the individual behavior @xmath4 to be realistic . however , one can construct a model for the change of the _ probability distribution _ @xmath5 of behaviors @xmath4 within the given population ( @xmath6 , @xmath7 ) . a theory of this kind is , of course , stochastic . in order to take into account several _ types _ @xmath8 of behavior , we may distinguish @xmath9 _ subpopulations _ @xmath8 consisting of @xmath10 individuals ( @xmath11 ) . then , the following relation holds : @xmath12 our goal is now to find a suitable equation for the probability distribution @xmath13 of behaviors within subpopulation @xmath8 ( @xmath14 , @xmath15 ) . if we neglect memory effects ( cf . sect . [ s6.1 ] ) , the desired equation is of the form @xmath16 whereas the _ inflow _ into @xmath1 is given as the sum over all absolute transition rates describing changes from an _ arbitrary _ behavior @xmath17 to @xmath1 , the _ outflow _ from @xmath1 is given as the sum over all absolute transition rates describing changes from @xmath1 to _ another _ behavior @xmath17 . since the absolute transition rate of changes from @xmath1 to @xmath17 is the product @xmath18 of the _ relative transition rate _ @xmath19 for a change to behavior @xmath17 _ given _ @xmath1 , and the probability @xmath13 of behavior @xmath1 , we arrive at the explicit equation @xmath20 \ , . \label{boltz}\ ] ] @xmath19 has the meaning of a transition probablility from @xmath1 to @xmath17 per unit time and takes into account the behavioral variations between the individuals ( occurring even within the same type @xmath8 of behavior ! ) . in the following we have to specify the relative transition rates @xmath19 , which will turn out to be _ effective _ transition rates . if we restrict the model to spontaneous ( or externally induced ) behavioral changes and behavioral changes due to pair interactions , we have ( helbing , 1992a , c ) : @xmath21 @xmath22 describes the rate of spontaneous ( resp . externally induced ) transitions from @xmath1 to @xmath17 for individuals of subpopulation @xmath8 . @xmath23 is the transition rate for two individuals of types @xmath8 and @xmath24 to change their behaviors from @xmath1 and @xmath25 to @xmath17 and @xmath26 , respectively , due to pair interactions . the total frequency of these interactions is proportional to the probability @xmath27 of behavior @xmath25 within subpopulation @xmath24 and the number @xmath28 of individuals of type @xmath24 . we have to sum up over @xmath24 , @xmath25 , and @xmath26 since all specifications of these variables are effectively connected with transitions from @xmath1 to @xmath17 of individuals of subpopulation @xmath8 . inserting ( [ effrate ] ) into ( [ boltz ] ) , we now obtain the socalled boltzmann-_like equations _ ( helbing , 1992a , c ) [ boltzlike ] @xmath29 \\ & + & \sum_{b=1}^a \sum_{{x}'\in \omega } \sum_{{y}\in \omega } \sum_{{y}'\in \omega } w_{ab}({\mbox{\boldmath $ x$}},{\mbox{\boldmath $ y$}}'|{\mbox{\boldmath $ x$}}',{\mbox{\boldmath $ y$}};t ) p_b({\mbox{\boldmath $ y$}},t)p_a({\mbox{\boldmath $ x$}}',t ) \nonumber \\ & - & \sum_{b=1}^a \sum_{{x}'\in \omega } \sum_{{y}\in \omega } \sum_{{y}'\in \omega } w_{ab}({\mbox{\boldmath $ x$}}',{\mbox{\boldmath $ y$}}'|{\mbox{\boldmath $ x$}},{\mbox{\boldmath $ y$}};t ) p_b({\mbox{\boldmath $ y$}},t)p_a({\mbox{\boldmath $ x$}},t)\end{aligned}\ ] ] with @xmath30 obviously , ( [ boltzlike]b ) depends nonlinearly ( quadratically ) on the probability distributions @xmath13 ( resp . @xmath27 ) which is due to the pair interactions . the boltzmann - like equations originally had been developed for the description of the kinetics of gases ( boltzmann , 1964 ) . however , they have also been applied to attitude formation ( helbing , 1992b , c ) and the avoidance behavior of pedestrians ( helbing , 1992c , d ) . it is possible to generalize the model to simultaneous interactions of an arbitrary number of individuals ( i.e. , higher order interactions ) ( helbing , 1992a , c ) . however , in most cases behavioral changes are dominated by pair interactions ( _ dyadic _ interactions ) . many of the phenomena occurring in social interaction processes can already be understood in terms of pair interactions . for models of behavioral changes the following special form of the effective transition rates ( [ effrate ] ) has been found to be suitable ( helbing , 1992b , c , e ) : @xmath31 r^a({\mbox{\boldmath $ x$}}'|{\mbox{\boldmath $ x$}};t ) \ , . \label{concrates}\ ] ] here , * @xmath32 is a measure of the rate of spontaneous ( or externally induced ) behavioral changes within subpopulation @xmath8 . * @xmath33 [ resp . @xmath34 is the _ readiness _ of an individual of subpopulation @xmath8 to change behavior from @xmath1 to @xmath17 spontaneously [ resp . in pair interactions ] . * @xmath35 is the _ interaction rate _ of an individual of subpopulation @xmath8 with individuals of subpopulation @xmath24 . * @xmath36 is a measure for the frequency of _ imitative processes _ @xmath37 where an individual of subpopulation @xmath8 takes over the behavior @xmath17 of an individual of subpopulation @xmath24 . the total frequency of imitative processes is proportional to the probablility @xmath38 of behavior @xmath17 within subpopulation @xmath24 . * @xmath39 is a measure for the frequency of _ avoidance processes _ @xmath40 where an individual of subpopulation @xmath8 changes the behavior @xmath1 to another behavior @xmath17 if meeting an individual of subpopulation @xmath24 with the same behavior ( defiant behavior , snob effect ) . the total frequency of avoidance processes is proportional to the probablility @xmath41 of behavior @xmath1 within subpopulation @xmath24 . a more detailled discussion of the different kinds of interaction processes and of _ ansatz _ ( [ concrates ] ) is given in publications of helbing ( 1992b , c , e ) . for @xmath42 we take the quite general form [ util ] @xmath43 with @xmath44 ( cf . weidlich and haag , 1988 , helbing , 1992c ) . then , the readiness @xmath42 for an individual of subpopulation @xmath8 to change behavior from @xmath1 to @xmath17 will be greater , * the greater the _ difference _ in the _ utilities _ @xmath45 of behaviors @xmath17 and @xmath1 , * the smaller the _ incompatibility ( `` distance '' ) _ @xmath46 between the behaviors @xmath1 and @xmath17 . similar to ( [ util ] ) we use @xmath47 and , therefore , allow the utility function @xmath48 for spontaneous ( or externally induced ) behavioral changes to differ from the utility function @xmath49 for behavioral changes in pair interactions . _ ansatz _ ( [ util ] ) is related to the _ multinomial logit model _ ( domencich and mcfadden , 1975 , ortzar and willumsen , 1990 ) . it assumes _ utility maximization _ with incomplete information about the exact utility of a behavioral change from @xmath1 to @xmath17 , which is , therefore , estimated and stochastically varying ( cf . helbing , 1992c ) . computer simulations of the boltzmann - like equations ( [ boltz ] ) , ( [ concrates ] ) , ( [ util ] ) are discussed and illustrated in helbing ( 1992b , c , e ) ( cf . also sect . [ s4 ] ) . the boltzmann - like equations ( [ boltz ] ) , ( [ concrates ] ) include a variety of special cases , which have become very important in the social sciences : * the _ logistic equation _ ( pearl , 1924 , verhulst , 1845 ) describes limited growth processes . let us consider the situation of two behaviors @xmath50 ( i.e. , @xmath51 ) and one subpopulation ( @xmath52 ) . @xmath53 may , for example , have the meaning to apply a certain strategy , and @xmath54 not to do so . if only imitative processes @xmath55 and processes of spontaneous replacement @xmath56 are considered , one arrives at the _ logistic equation _ @xmath57 * the _ gravity model _ ( zipf , 1946 , ravenstein , 1876 ) describes processes of exchange between different places @xmath1 . its dynamical version results for @xmath58 , @xmath59 , @xmath60 , and @xmath52 : @xmath61 p({\mbox{\boldmath $ x$}},t)p({\mbox{\boldmath $ x$}}',t ) \ , .\ ] ] here , we have dropped the index @xmath8 because of @xmath62 . @xmath5 is the probability of being at place @xmath1 . the absolute rate of exchange from @xmath1 to @xmath17 is proportional to the probabilities @xmath5 and @xmath63 at the places @xmath1 and @xmath17 . @xmath64 is often chosen as a function of the metric distance @xmath65 between @xmath1 and @xmath17 : @xmath66 . * the _ behavioral model _ of weidlich and haag ( 1983 , 1988 , weidlich , 1991 , 1994 ) is based on spontaneous transitions . we obtain this model for @xmath67 and @xmath68 because of the dependence of the utilities @xmath48 on the behavioral distributions @xmath41 the model assumes _ indirect interactions _ , which are , for example , mediated by the newspapers , tv or radio . @xmath69 is the _ preference _ of subpopulation @xmath8 for behavior @xmath1 . @xmath70 are _ coupling parameters _ describing the influence of the behaviorial distribution within subpopulation @xmath24 on the behavior of subpoplation @xmath8 . for @xmath71 , @xmath70 reflects the _ social pressure _ of behavioral majorities . * the _ game dynamical equations _ ( hofbauer and sigmund , 1988 , schuster _ et . al . _ , 1981 , helbing , 1992c , e , 1993 ) result for @xmath72 , @xmath60 , and @xmath73 where @xmath74 for a detailled interpretation of these relations see helbing ( 1992c , e ) . + the explicit form of the game dynamical equations is + [ game ] + @xmath75 \\ & + & \nu_{aa}(t ) p_a({\mbox{\boldmath $ x$}},t ) \big [ e_a({\mbox{\boldmath $ x$}},t ) - \langle e_a \rangle \big ] \ , .\end{aligned}\ ] ] + whereas ( [ game]a ) again describes spontaneous behavioral changes ( _ `` mutations '' _ , innovations ) , ( [ game]b ) reflects competition processes leading to a _ `` selection '' _ of behaviors with a _ success _ @xmath76 that exceeds the _ average success _ @xmath77 the success @xmath76 is connected with the socalled _ payoff matrices _ @xmath78 by @xmath79 ( helbing , 1992c , e ) . @xmath80 means the success of behavior @xmath1 with respect to the environment . + since the game dynamical equations ( [ game ] ) agree with the _ selection mutation equations _ ( hofbauer and sigmund , 1988 ) they are not only a powerful tool in social sciences and economy ( axelrod , 1984 , von neumann and morgenstern , 1944 , schuster _ et . al . _ , 1981 , helbing , 1992c , e , 1993 ) , but also in evolutionary biology ( fisher , 1930 , eigen , 1971 , eigen and schuster , 1979 , feistel and ebeling , 1989 ) . we shall now assume the set @xmath2 of possible behaviors forms a _ continuous _ space . the @xmath81 dimensions of this space correspond to different characteristic _ aspects _ of the considered behaviors . in the continuous formulation , the sums in ( [ boltz ] ) , ( [ effrate ] ) have to be replaced by integrals : [ cont ] @xmath82 \ , , \ ] ] @xmath83 a reformulation of the boltzmann - like equations ( [ cont ] ) via a second order taylor _ approximation _ ( kramers - moyal expansion ( kramers , 1940 , moyal , 1949 ) ) leads to _ diffusion equations _ ( helbing , 1992a , c ) : [ bfp ] @xmath84 + \frac{1}{2 } \sum_{i , j=1}^n \frac{\partial}{\partial x_{i}}\frac{\partial } { \partial x_{j } } \big[q_{a i j}({\mbox{\boldmath $ x$}},t ) p_a({\mbox{\boldmath $ x$}},t)\big ] \label{bfp}\ ] ] with the effective _ drift coefficients _ @xmath85 and the effective _ diffusion coefficients _ contains additional terms due to another derivation of ( [ bfp ] ) . however , they make no contributions , since they result in vanishing surface integrals ( helbing , 1992c ) . ] @xmath86 because of their relation with the boltzmann equation ( boltzmann , 1964 ) and the fokker - planck equation ( fokker , 1914 , planck , 1917 ) equations ( [ bfp ] ) will be called the boltzmann - fokker - planck _ equations _ in the following ( cf . helbing 1992a , c ) . in the boltzmann - fokker - planck equations the drift coefficients @xmath87 govern the systematic change ( `` drift '' , motion ) of the distribution @xmath13 , whereas the diffusion coefficients @xmath88 describe the spread of the distribution @xmath13 due to fluctuations resulting from the individual variation of behavioral changes . for _ ansatz _ ( [ concrates ] ) , the effective drift and diffusion coefficients can be split into contributions due to spontaneous ( or externally induced ) transitions ( @xmath89 ) , imitative processes ( @xmath90 ) , and avoidance processes ( @xmath91 ) : [ sum ] @xmath92 where @xmath93 and @xmath94 the behavioral changes induced by the _ environment _ are included in @xmath95 and @xmath96 . the boltzmann - fokker - planck equations ( [ bfp ] ) are equivalent to the stochastic equations ( langevin equations , 1908 ) [ langevin ] @xmath97 with @xmath98 g_{ajk}({\mbox{\boldmath $ x$}},t ) \label{drift}\ ] ] and @xmath99 ( cf . stratonovich , 1963 , helbing , 1992c ) . for an individual of subpopulation @xmath8 the vector @xmath100 with the components @xmath101 describes the contribution to the change of behavior @xmath1 that is caused by behavioral fluctuations @xmath102 ( which are assumed to be delta - correlated and gaussian ( helbing , 1992c ) ) . since the diffusion coefficients @xmath88 and the coefficients @xmath103 are usually small quantities , we have @xmath104 ( cf . ( [ drift ] ) ) , and ( [ langevin ] ) can be put into the form @xmath105 whereas the fluctuation term describes individual behavioral variations , the vectorial quantity @xmath106 drives the systematic change of the behavior @xmath4 of individuals of subpopulation @xmath8 . therefore , it is justified to denote @xmath107 as _ social force _ acting on individuals of subpopulation @xmath8 . with that we have attained a very intuitive formulation of social processes , according to which behavioral changes are caused by social forces . on the one hand , social forces influence the behavior of the individuals , but on the other hand , due to interactions , the behavior of the individuals also influences the social forces via the behavioral distributions @xmath13 ( cf . ( [ cont]b ) , ( [ effdr ] ) ) . that means , @xmath108 is a function of the social processes within the given population . under the integrability conditions @xmath109 there exists a time - dependent _ potential _ @xmath110 so that the social force is given by its derivative ( by its gradient @xmath111 ) : @xmath112 the potential @xmath113 can be understood as _ social field_. it reflects the social influences and interactions relevant for behavioral changes : the public opinion , trends , social norms , etc . clearly , the social force is not a force obeying the newtonian ( 1687 ) laws of mechanics ( cf . greenwood , 1988 ) . instead , the social force @xmath107 is a vectorial quantity with the following properties : * @xmath107 drives the temporal change @xmath3 of another vectorial quantity : the behavior @xmath4 of an individual of subpopulation @xmath8 . * the component @xmath114 r^a({\mbox{\boldmath $ x$}}'|{\mbox{\boldmath $ x$}};t)\ ] ] of the social force @xmath107 describes the reaction of subpopulation @xmath8 on the behavioral distribution within subpopulation @xmath24 and usually differs from @xmath115 , which describes the influence of subpopulation @xmath8 on subpopulation @xmath24 . * neglecting fluctuations , the behavior @xmath4 does not change if @xmath107 vanishes . @xmath116 corresponds to an _ extremum _ of the social field @xmath113 , because it means @xmath117 we will now compare our results with lewin s ( 1951 ) _ social field theory_. from social psychology it is well - known that the behavior of an individual is determined by the totality of _ environmental influences _ and his or her _ personality_. inspired by electro - magnetic field theory , social field theory claims that environmental influences can be considered as a _ dynamical force field _ , which should be mathematically representable . a temporal change of this field will evoke a _ psychical tension _ which , then , induces a _ ( behavioral ) compensation_. in the following it will turn out that our model allows a fully mathematical specification of the field theoretical ideas , which was still to be found : * let us assume that an individual s objective is to behave in an optimal way with respect to the social field @xmath113 , that means , he or she tends to a behavior corresponding to a _ minimum _ of the social field . * if the behavior @xmath1 does not agree with a minimum of the social field @xmath113 this evokes a force @xmath118 that is given by the gradient of the social field @xmath113 ( pulling into the direction of steepest descent of @xmath113 ) . the force @xmath107 plays the role of the _ psychical tension_. it induces a behavioral change according to @xmath119 * the behavioral change @xmath3 drives the behavior @xmath4 towards a minimum @xmath120 of the social field @xmath113 . when the minimum @xmath120 is reached , then @xmath121 holds and , therefore , @xmath116 . as a consequence , the psychical tension vanishes , that means , it is _ compensated _ by the previous behavioral changes . + when the psychical tension @xmath107 vanishes then , except for fluctuations , no behavioral changes take place in accordance with ( [ accor ] ) . the individual has reached an _ equilibrium _ within the social field , then . * note , that the social fields @xmath113 of different subpopulations @xmath8 usually have different minima @xmath120 . this means that individuals of different types @xmath8 of behavior will normally feel different psychical tensions @xmath107 . in other words , index @xmath8 distinguishes different _ the boltzmann - fokker - planck equations are able to describe a broad spectrum of social phenomena . in the following , some of the results shall be illustrated by computer simulations . we shall examine the case of @xmath122 subpopulations , and a situation for which the interesting aspect of the individual behavior can be described by a certain _ position _ @xmath123 on a one - dimensional continuous scale ( i.e. , @xmath124 , @xmath125 ) . a concrete example for this situation would be the case of _ opinion formation_. here , conservative and progressive thinking individuals could be distinguished by different subpopulations . the position @xmath123 would describe the grade of approval or disapproval with respect to a certain political option ( for example , sdi , power stations , golf war , etc . ) . in the one - dimensional case , the integrability conditions ( [ intcond ] ) are automatically fulfilled , and the social field @xmath126 is well - defined . the parameter @xmath127 can be chosen arbitrarily . we will take for @xmath127 the value that shifts the absolute minimum of @xmath128 to zero , that means , @xmath129 * since we will restrict the simulations to the case of imitative or avoidance processes , the shape of the social field @xmath128 changes with time only due to changes of the probability distributions @xmath130 ( cf . ( [ sum ] ) ) , that means , due to behavioral changes of the individuals ( see figures [ fi1 ] to [ fi6 ] ) . in the one - dimensional case one can find the formal _ stationary solution _ @xmath131 which we expect to be approached in the limit of large times @xmath132 . due to the dependence of @xmath133 and @xmath134 on @xmath135 , equations ( [ abh ] ) are only _ implicit _ equations . however , from ( [ abh ] ) we can derive the following conclusions : * if the diffusion coefficients are constant ( @xmath136 ) , ( [ abh ] ) simplifies to @xmath137 \right ) \ , , \ ] ] that means , the stationary solution @xmath135 is completely determined by the social field @xmath138 . especially , @xmath135 has its maxima at the positions @xmath139 , where the social field @xmath138 has its minima ( see fig . [ fi1 ] ) . the diffusion constant @xmath140 regulates the width of the behavioral distribution @xmath135 . for @xmath141 there were no individual behavioral variations , and the behavioral distribution @xmath135 were sharply peaked at the deepest minimum @xmath139 of @xmath138 . * if the diffusion coefficients @xmath134 are varying functions of the position @xmath123 , the behavioral distribution @xmath135 is also influenced by the concrete form of @xmath134 . from ( [ abh ] ) one expects high behavioral probabilities @xmath135 where the diffusion coefficients @xmath134 are small ( see fig . [ fi2 ] , where the probability distribution @xmath142 can not be explained solely by the social field @xmath143 ) . * since the stationary solution @xmath135 depends on both , @xmath133 and @xmath134 , different combinations of @xmath133 and @xmath134 can lead to the same probability distribution @xmath135 ( see fig . [ fi4 ] in the limit of large times ) . for the following simulations , we shall assume @xmath144 $ ] and use the _ ansatz _ [ ans ] @xmath145 for the readiness @xmath146 to change from @xmath123 to @xmath147 ( cf . ( [ util ] ) ) . with the utility function @xmath148 subpopulation @xmath8 prefers behavior @xmath149 . @xmath150 means the _ indifference _ of subpopulation @xmath8 with respect to variations of the position @xmath123 . moreover , we take @xmath151 where @xmath152 can be interpreted as measure for the _ range of interaction_. according to ( [ dist ] ) , the rate of behavioral changes is the smaller the greater they are . only small changes of the position ( i.e. , between neighboring positions ) contribute with an appreciable rate . figure [ fi1 ] to [ fi6 ] show the respective values of @xmath152 , @xmath153 , and @xmath154 used in the simulations . note , that @xmath152 and the readiness @xmath155 are different quantities . for very small values of the range @xmath152 of interaction the diffusion coefficients @xmath134 can be neglected and the fluctuations play a neglible role , that means , behavioral changes are mainly given by the social field ( see fig . 1 ) . for greater but still small values of @xmath152 , the diffusion coefficients have to be taken into account in order to fully understand the temporal development of the behavioral distribution @xmath130 ( see fig . if @xmath152 exceeds a certain value , the taylor approximation is invalid , and the boltzmann - like equations should be applied instead of the boltzmann - fokker - planck equations . let @xmath156 be the degree of _ sympathy _ which individuals of subpopulation @xmath8 feel towards individuals of subpopulation @xmath24 . then , one expects the following : whereas the frequency @xmath36 of imitative processes will be increasing with @xmath156 , the frequency @xmath39 of avoidance processes will be decreasing with @xmath156 . this functional relationship can , for example , be described by @xmath157 with @xmath158 @xmath159 is a measure for the frequency of imitative processes within subpopulation @xmath8 , @xmath160 a measure for the frequency of avoidance processes . if we assume the sympathy between individuals of the same subpopulation to be be maximal , we have @xmath161 . in the following simulations of imitative processes we assume the preferred positions to be @xmath164 and @xmath165 . with @xmath166 the individuals of subpopulation @xmath62 like the individuals of subpopulation @xmath167 , but not the other way round . that means , subpopulation 2 influences subpopulation 1 , but not vice versa . as expected , in both behavioral distributions @xmath130 there appears a maximum around the preferred behavior @xmath149 . in addition , due to imitative processes of subpopulation 1 , a second maximum of @xmath168 develops around the preferred behavior @xmath169 of subpopulation 2 . this second maximum is small , if the indifference @xmath153 of subpopulation 1 with respect to variations of the position @xmath123 is low ( see fig . [ fi1 ] ) . for high values of the indifference @xmath153 even the _ majority _ of individuals of subpopulation 1 imitates the behavior of subpopulation 2 ( see fig . [ fi2 ] ) ! one could say , the individuals of subpopulation 2 act as _ trendsetters_. this phenomenon is typical for _ we shall now consider the case @xmath170 for which the subpopulations influence each other mutually with equal strengths . if the indifference @xmath150 with respect to changes of the position @xmath123 is small in both subpopulations @xmath8 , _ each _ probability distribution @xmath130 has _ two _ maxima . the higher maximum is located around the preferred position @xmath149 . a second maximum can be found around the position preferred in the _ other _ subpopulation . it is the higher , the greater the indifference @xmath150 is ( see fig . [ fi3 ] ) . however , if @xmath150 exceeds a certain value in at least one subpopulation , a socalled _ phase transition _ ( that means , a qualitative different situation ) occurs , since only _ one _ maximum develops in each behavioral distribution @xmath130 ! despite the fact that the social fields @xmath128 and diffusion coefficients @xmath171 of the subpopulations @xmath8 are different because of their different preferred positions @xmath149 ( and different utility functions @xmath172 ) , the behavioral distributions @xmath130 agree after some time ! especially , the maxima @xmath139 of the distributions @xmath130 are located at the _ same _ position @xmath173 in both subpopulations . one could say , the two subpopulations made a _ compromise_. the compromise @xmath173 is nearer to the position @xmath149 of the subpopulation @xmath8 with the lower indifference @xmath150 ( see fig . [ fi4 ] ) . for the simulation of avoidance processes we assume with @xmath176 and @xmath177 that both subpopulations nearly prefer the same behavior . figure [ fi5 ] shows the case , where the individuals of different subpopulations dislike each other : @xmath178 this corresponds to a mutual influence of each subpopulation on the other . the computational results indicate that * individuals avoid behaviors which are found in the other subpopulation . * the subpopulation @xmath62 with the lower indifference @xmath179 is distributed around the preferred behavior @xmath180 and pushes away the other subpopulation ! despite the fact that the initial behavioral distribution @xmath181 agrees in both subpopulations , there is nearly no overlapping of @xmath168 and @xmath182 after some time . this is typical of _ polarization phenomena _ in the society . well - known examples are the development of _ ghettos _ or the formation of _ extremist groups_. in figure [ fi6 ] , we assume that the individuals of subpopulation 2 like the individuals of subpopulation 1 and , therefore , do not react to the behaviors in subpopulation 1 with avoidance processes : @xmath183 as a consequence , @xmath182 remains unchanged with time , whereas @xmath168 drifts away from the preferred behavior @xmath180 due to avoidance processes . surprisingly , the polarization effect is much smaller than in figure [ fi5 ] ! the distributions @xmath168 and @xmath182 overlap considerably . this is , because the slope of @xmath182 is smaller than in figure [ fi5 ] ( and remains constant ) . as a consequence , the probability for an individual of subpopulation 1 to meet a disliked individual of subpopulation 2 with the same behavior @xmath123 can hardly be decreased by a small behavioral change . one may conclude , that polarization effects ( which often lead to an escalation ) can be reduced , if individuals do not return dislike with dislike . for practical purposes one has , of course , to determine the model parameters from empirical data . therefore , let us assume we know empirically the distribution functions @xmath184 , [ the interaction rates @xmath185 , ] and the effective transition rates @xmath186 ( @xmath187 ) for a couple of times @xmath188 . the corresponding effective social fields @xmath189 and diffusion coefficients @xmath190 are , then , easily obtained as [ empi ] @xmath191 with @xmath192 and @xmath193 much more difficult is the determination of the utility functions @xmath194 , @xmath195 , the distance functions @xmath196 , and the rates @xmath197 , @xmath198 , @xmath199 . this can be done by numerical minimization of the _ error function _ @xmath200 p_a^{\rm e}({\mbox{\boldmath $ x$}},t_l ) \right\}^2 \ , , \label{error}\ ] ] for example with the method of _ steepest descent _ ( cf . al . _ , 1977 ) . in ( [ error ] ) , we have introduced the abbreviation @xmath201 \mbox{e}^{u^a({x}',t_l ) - u^a({x},t_l ) } \ , . \label{assum}\ ] ] it turns out ( cf . helbing , 1992c ) , that the rates @xmath202 have to be taken constant during the minimization process ( e.g. , @xmath203 ) , whereas the parameters @xmath204 , @xmath205 , @xmath206 and @xmath207 are to be varied . for @xmath208 one inserts [ dista ] @xmath209 with @xmath210 ^ 2 \ ! + w_{\rm e}^a({\mbox{\boldmath $ x$}}|{\mbox{\boldmath $ x$}}';t_l ) g_a({\mbox{\boldmath $ x$}},{\mbox{\boldmath $ x$}}';t_l ) \big [ p_a^{\rm e}({\mbox{\boldmath $ x$}}',t_l ) \big]^2\ ] ] and @xmath211 ^ 2 \ ! + \big [ g_a({\mbox{\boldmath $ x$}},{\mbox{\boldmath $ x$}}';t_l ) p_a^{\rm e}({\mbox{\boldmath $ x$}}',t_l ) \big]^2 \ , .\ ] ] ( [ dista ] ) follows from the minimum condition for @xmath212 ( cf . helbing , 1992c ) . since @xmath213 may have a couple of minima due to its nonlinearity , suitable start parameters have to be taken . especially , the numerically determined rates @xmath214 and @xmath215 have to be non - negative . if @xmath213 is minimal for the parameters @xmath204 , @xmath205 , @xmath212 , @xmath202 , @xmath214 and @xmath215 , this is ( as can easily be checked ) also true for the scaled parameters @xmath216 in order to obtain unique results we put @xmath217 and @xmath218 which leads to @xmath219 and @xmath220 @xmath221 and @xmath222 are _ mean utilities _ , whereas @xmath223 is a kind of _ unit of distance_. the distances @xmath224 are suitable quantities for _ multidimensional scaling _ ( kruskal and wish , 1978 , young and hamer , 1987 ) . they reflect the `` psychical structure '' ( psychical topology ) of individuals of subpopulation @xmath8 , since they determine which behaviors are more or less related ( compatible ) ( comp . to osgood _ et . al . _ , 1957 ) . by the dependence on @xmath8 , @xmath224 distinguishes different psychical structures resulting in different types @xmath8 of behavior and , therefore , different `` characters '' ( personalities ) . it is not easy to find suitable data for the determination of the model parameters , since usually there only exist data for the temporal development of a certain behavioral distribution @xmath225 , but not for the corresponding effective transition rates @xmath226 . however , for a few countries all necessary data are known about _ migration _ between different regions ( see weidlich and haag , 1988 ) . in this case , the behavior @xmath125 means _ to live in region _ @xmath227 , where @xmath228 is the number of distinguished regions . in the following , the results for migration in west germany shall be presented . west germany is divided into 10 federal states and the region of west berlin ( see figure [ fi7 ] and table [ ta1 ] ) . the data for these @xmath229 regions can be found in the _ statistische jahrbcher _ of the years 1960 to 1985 on an annual basis ( @xmath230 , @xmath231 , @xmath232 , @xmath233 ) . our data analysis will assume with respect to migration one more or less homogeneous population , that means , we have @xmath52 , and the index @xmath8 can be dropped . figure [ fi8 ] shows some examples for the temporal variation of the effective transition rates @xmath234 . using the method of steepest descent , the minimization of the error function ( [ error ] ) , ( [ assum ] ) gives the following results : @xmath235 that means , avoidance processes are negligible . the rate @xmath236 of spontaneous behavioral changes and the rate @xmath237 of imitative processes are depicted in figure [ fi9 ] . the utility functions @xmath238 for spontaneous changes and the utility functions @xmath239 for imitative processes are illustrated in figures [ fi10 ] and [ fi11 ] . the irregulatity of the utility functions @xmath239 indicates that they probably fit random fluctuations of the migration data . indeed , a mathematical analysis proves that the term @xmath240 only explains 5.2 percent of the variance of the effective transition rates @xmath241 . therefore , it makes _ no significant contribution _ to their mathematical description , and the migration rates @xmath241 of west germany can already be represented by the model @xmath242 this result agrees with the model of weidlich and haag ( 1988 ) ! figure [ fi12 ] shows the corresponding rate @xmath236 of spontaneous changes , and figure [ fi13 ] depicts the utility functions @xmath238 . the distances @xmath243 can be calculated from the formulas ( [ dista ] ) , ( [ scal ] ) , and ( [ undi ] ) . they are not only a measure for the mean _ geographical distances _ , but also for _ transaction costs _ ( e.g. , removal costs ) and _ psychical differences _ ( of the language , mentality , etc . ) . the replacement of the time dependent distances @xmath243 with the time independent values @xmath244 defined by @xmath245 ( see table [ ta2 ] ) allows for a further _ model reduction_. the optimal value of the rate of spontaneous behavioral changes is , then , given by @xmath246 although the reduced model @xmath247 only needs @xmath248 variables for the description of @xmath249 effective transition rates @xmath250 , it attains a very good correlation of 0.984 with the empirical data . for a more detailled discussion of this model , see weidlich and haag ( 1988 ) . in this article , a behavioral model has been proposed that incorporates in a consistent way many models of social theory : the diffusion models , the multinomial logit model , lewin s field theory , the logistic equation , the gravity model , the weidlich - haag model , and the game dynamical equations . this very general model opens new perspectives concerning a theoretical description and understanding of behavioral changes , since it is formulated fully mathematically . it takes into account spontaneous ( or externally induced ) behavioral changes and behavioral changes due to pair interactions . two important kinds of pair interactions have been distinguished : imitative processes and avoidance processes . the model turns out to be suitable for computational simulations , but it can also be applied to concrete empirical data . the formulation of the model in the previous sections has neglected _ memory effects _ that may also influence behavioral changes . however , memory effects can be easily included by generalizing the boltzmann - like equations to @xmath253 \\ & & \qquad \quad \displaystyle + \frac{1}{2 } \sum_{i , j=1}^n \frac{\partial}{\partial x_{i}}\frac{\partial } { \partial x_{j } } \big[q_{a i j}^{t - t'}({\mbox{\boldmath $ x$}},t ' ) p_a({\mbox{\boldmath $ x$}},t')\big ] \bigg\ } \end{array } \label{membfp}\ ] ] obviously , in these formulas there only appears an additional integration over past times @xmath257 ( helbing , 1992c ) . the influence of the past results in a dependence of @xmath258 , @xmath259 , and @xmath260 on @xmath261 . the boltzmann - like equations ( [ boltzlike ] ) resp . the boltzmann - fokker - planck equations ( [ bfp ] ) used in the previous sections result from ( [ memb ] ) resp . ( [ membfp ] ) in the limit of short memory . the force model described in section [ s3 ] can serve as a new mathematical modelling concept . for example , it has successfully been applied to the simulation of _ pedestrian behavior _ ( cf . helbing , 1991 ) . another interesting field is an application of the force model to _ group dynamics _ and _ group formation_. in this case we take @xmath262 , that means , each subpopulation consists of one individual only ( @xmath263 ) . since @xmath10 is violated , then , the temporal change @xmath264 of @xmath13 , which describes the probability of individual @xmath8 to show the behavior @xmath1 at time @xmath265 , is additionally subject to fluctuations ( cf . helbing , 1992a , c ) . the interaction rates @xmath266 are related to the _ adjacency matrix _ and describe the _ social ( interpersonal ) network _ ( burt , 1982 ) . moreover , the sympathy matrix @xmath267 is affected by the social processes . the crucial task for simulating group dynamics is , therefore , to set up equations for the temporal change of @xmath156 . the topic of group dynamics will be treated in a forthcoming paper . + this work has been financially supported by the _ volkswagen stiftung _ and the _ deutsche forschungsgemeinschaft _ ( sfb 230 ) . the author is grateful to prof . w. weidlich and dr . r. reiner for valuable discussions and commenting on the manuscript . + helbing , d. ( 1992e ) a mathematical model for behavioral changes by pair interactions . in : g. haag , u. mueller , and k. g. troitzsch ( eds . ) economic evolution and demographic change , pp . 330348 . springer , berlin . + osgood , ch . e. , and tannenbaum , p. h. ( 1955 ) the principle of congruity in the prediction of attitude change . _ psychological review _ * 62 * : 4255 . + pearl , r. ( 1924 ) _ studies in human biology_. williams & wilkins , baltimore .

recent developments in the fabrication and control of nanostructures and molecular devices have stimulated a growing interest in the study and research of heat conductance in such devices . @xcite as efforts are being made to better utilize and control these devices , the understandings of the mechanisms for accumulation and dissipation of heat are of importance . the literature on the topic considers different physical setups , differing in the process by which heat is mediated , the devices under investigation and the nature of the heat baths involved in the heat transfer process . @xcite we consider heat transfer between two baths held at different temperatures through a steady - state current of energy . the linear response regime , which relies on the equilibrium properties of the system , arises in the limit of small gradient of temperatures . a linear response study of a system similar to the one that is presented in this paper was done recently by entin - wohlman and others and the thermopower properties were calculated @xcite . in the general case , however , the system is far from equilibrium , as the temperature gradient is finite . the description of such systems is a challenging problem of great interest in current research , as many of the concepts and techniques used to describe equilibrium setups are inadequate . exact solutions of systems far from equilibrium are particularly desirable as they may offer both a benchmark and an unbiased understanding of the underlying physics . in this paper we present an asymptotically - exact calculation of the heat current through a molecular junction , under explicit nonequilibrium conditions manifested by a finite temperature gradient between the two baths to which the molecule is coupled the electrons in a conduction band and a bosonic bath . in a typical molecular bridge , molecular orbitals are coupled simultaneously to the lead electrons and to the vibrational modes of the molecule , with the former degrees of freedom reduced to a single effective band in the absence of a bias voltage . @xcite this coupling to the vibrational modes of the molecule is believed to have an essential role in heat transfer processes . @xcite in the continuum limit , a minimal model for an unbiased molecular bridge therefore consists of a vibrational mode that is coupled by displacement to the conduction electrons at the origin , and is also coupled to a bosonic bath of vibrational modes , as described by the hamiltonian of eq . ( [ h ] ) . the solution is asymptotically exact in the sense that it is governed by the smallness of the electron - phonon coupling @xmath0 with respect to the effective energy band of the conductance electrons . the thermal properties of the bosonic hamiltonian of eq . ( [ h - bosonic ] ) , and similar hamiltonians describing heat current between harmonic baths , were studied before in different physical setups than presented here @xcite . the results we derive in this paper corroborate those studies , and adjust them to describe the electronic systems which we shall present . the paper is organized as follows . we begin in sec . [ sec : model ] by presenting the model and the physical systems which it may represent , under suitable mappings . in sec . [ sec : mapping ] we then introduce the nonequilibrium condition and map the model onto a form quadratic in bosons . this quadratic nature of the hamiltonian is then exploited , in sec . [ sec : resonance_heat_current ] , in order to calculate exactly the heat current in the system . in sec . [ sec : master_equation ] we then turn to a master equation approach , which is perturbative in nature , in order to calculate again the heat current . this approach allows us to gain some useful physical understanding of the processes involved . section [ sec : off - resonance ] addresses the case where the localized fermionic level is taken off - resonance , which breaks the particle - hole symmetry of the problem and adds a linear term to our quadratic hamiltonian . we then turn , in sec . [ sec : conclusions ] to present our conclusions . the model we consider consists of a single molecule coupled simultaneously to a conduction band and to a bosonic bath , which may represent , depending on the context , either the substrate phonons or any other continuum of bosonic degrees of freedom . a finite temperature gradient is applied between the electronic and bosonic baths , generating a nonzero heat flow from the hotter bath to the colder one . physically this model can describe any of a number of systems , three of which are portrayed schematically in fig . [ fig : fig1 ] : a single molecule adsorbed on a metallic surface ( upper panel ) , a molecular bridge ( middle panel ) , and a single - molecule transistor embedded in one of the arms of an aharonov - bohm interferometer ( lower panel ) . under suitable conditions , to be specified below , each of these setups can be described by the generic continuum - limit hamiltonian @xcite @xmath1 which is the focus of the present work . in the above hamiltonbian , the one - dimensional fermionic field @xmath2 represents the conduction - electron degrees of freedom , @xmath3 denotes the molecular vibrational mode , and @xmath4 are the modes of the bosonic bath . the fermionic field obeys canonical anticommutation relations @xmath5 subject to the regularization @xmath6 , where @xmath7 is a suitable short - distance cutoff . in the above model we have restricted ourselves to the case where the coulomb repulsion between the lead and the localized fermionic level is negligible , and omitted that term . the bosonic bath is characterized by the coupling function @xmath8 which is assumed to have the standard power - law form @xmath9 here , @xmath10 is a high - energy cutoff and @xmath11 is a dimensionless coupling constant parameterizing the coupling strength to the bosonic bath . the power @xmath12 is of particular interest as it corresponds to an ohmic bath.@xcite the parameter @xmath0 describes the displacement coupling between the electrons and the vibrational mode , while @xmath13 represents ordinary potential scattering . ) . top to bottom : ( a ) a molecule adsorbed on a surface , with the bulk phonons serving as a bosonic thermal bath ; ( b ) a single - molecule transistor , where the phonons of the substrate are coupled to local vibrational mode ; ( c ) an aharonov - bohm interferometer with a molecular device embedded in one of its arms.,title="fig:",scaledwidth=44.0% ] ) . top to bottom : ( a ) a molecule adsorbed on a surface , with the bulk phonons serving as a bosonic thermal bath ; ( b ) a single - molecule transistor , where the phonons of the substrate are coupled to local vibrational mode ; ( c ) an aharonov - bohm interferometer with a molecular device embedded in one of its arms.,title="fig:",scaledwidth=44.0% ] ) . top to bottom : ( a ) a molecule adsorbed on a surface , with the bulk phonons serving as a bosonic thermal bath ; ( b ) a single - molecule transistor , where the phonons of the substrate are coupled to local vibrational mode ; ( c ) an aharonov - bohm interferometer with a molecular device embedded in one of its arms.,title="fig:",scaledwidth=44.0% ] the generic model described by the hamiltonian of eq . ( [ h ] ) is found in discussions of different physical systems consisting of a single molecule which is coupled to large environments . it has a long history that dates back to the 1970s , when it was proposed as a model for the electron - phonon coupling in mixed - valence compounds . @xcite in the modern context of nanostructures it is expected to properly describe the physics of single - molecule devices away from coulomb - blockade valleys where a single unpaired spin resides on the molecule . we shall now turn to present several of these setups . the most direct realization of the model given by eq . ( [ h ] ) is that of a single molecule adsorbed on a metallic surface . adsorbed molecules on surfaces have been intensely studied over the years , both theoretically and experimentally ( for a brief review see ref . [ ] ) . in such molecules , the vibrational modes of adsorbed molecules relax by interaction with the surface conductance electrons as well as by coupling to the bulk phonons , and the interaction between the electrons and the vibrational mode plays an important role . @xcite one of the common models used to describe such interactions @xcite is the hamiltonian of eq . ( [ h ] ) , prior to the reduction of the relevant conductance electrons degrees of freedom to one - dimensional fields . adopting that description , the bosonic bath described by the operators @xmath14 and @xmath4 in our hamiltonian represent the bulk phonons . the molecule is then brought into contact with an stm - tip or other electronic lead , and the conductance electrons couple to the electronic level on the molecule . focusing on the electronic mode that couples to the molecule , we can map them onto the one - dimensional field that is represented by the electronic field operators @xmath15 and @xmath2 in the hamiltonian . a detail of the process can be found in ref . [ ] . in such a system , generally one would take @xmath16 , as the conductions electrons are held in resonance with the electronic level on the molecule . another system that can be described by the model of eq . ( [ h ] ) is that of a molecular bridge a single molecule sandwiched between two electronic leads . in such single - molecule devices the electron - phonon interaction plays an important role , as the molecular orbitals are coupled simultaneously to the lead electrons and to the vibrational modes of the molecule itself . the molecular bridge is typically placed on an insulating substrate which provides an additional phononic bath that couples to the molecular vibrational modes . @xcite focusing on a molecular junction held between leads with no bias voltage , one can choose a symmetric and anti - symmetric basis for the lead electrons , thus the molecule interacts with an effective single electronic band . the molecule itself is modeled by a single spinless electronic level @xmath17 with energy @xmath13 which is coupled by displacement to a local vibrational mode @xmath3 with frequency @xmath18 . this vibrational mode is further coupled by displacement to a bath of phonons @xmath4 . the level is then coupled to the single band of electrons via a hopping matrix element @xmath19 . the resulting hamiltonian is given by @xmath20 with @xmath21 and @xmath22 typically the hamiltonian is treated either in the weak - coupling limit using perturbation theory in @xmath0 , or using the lang - firsov transformation @xcite and the polaronic approximation in the limit where @xmath19 is small . in the limit of a broad level close to resonance , however , this hamiltonian can be reduced @xcite to the continuum limit hamiltonian of eq . ( [ h ] ) as described in detail in ref . this mapping process give rise to an effective high - energy cutoff which is determined by the hybridization width of the level @xmath23 . the effective bandwidth @xmath24 of the continuous field @xmath15 is related to the hybridization width by @xmath25 . the final system considered is an aharonov - bohm interferometer with a molecular device embedded in one of its arms . this setup is a more complicated variant of the molecular bridge described above , as there is an additional transmission channel between the two leads , and a magnetic flux @xmath26 threading the ring . lately , this system has attracted a considerable interest , following the work of entin - wohlman and others @xcite who computed the transport coefficients relating the linear - response electric and heat currents to the voltage bias and temperature gradient . restricting our attention to the case where the two electronic leads are held in identical temperature and chemical potential , the hamiltonian describing such a device is given @xcite by @xmath27 , with @xmath28 here , @xmath29 describes the left ( l ) and right ( r ) lead electrons , @xmath30 describes the arm of the interferometer without the device , and @xmath31 describes the device itself and its coupling to the leads . the general term @xmath32 may contain any local interactions at the dot that do not involve the lead electrons . in our case , it will include the local vibrational mode , the electron - phonon coupling and the bosonic bath . however , for the purpose of the mapping shown here , no further assumptions are needed on the form of @xmath32 beyond it not involving the lead electrons . we begin by transferring to the symmetric and anti - symmetric basis , defining the new symmetric field @xmath33\ ] ] where @xmath34 . the green function @xmath35 pertaining to this field at @xmath36 can now be calculated from the green functions of the left and right fields at the origin , given by matrix form as @xmath37^{-1 } & -te^{i\varphi } \\ -te^{-i\varphi } & \left[g_r(z)\right]^{-1}\end{array } \right)^{-1}.\end{aligned}\ ] ] using this we write the dot s green function as @xmath38^{-1},\ ] ] where @xmath39 is the self - energy contribution of the @xmath32 . close to resonance , the role of the symmetric electronic green function @xmath40 will be to renormalize @xmath13 and the level width . therefore , our effective hamiltonian of eq . ( [ h ] ) generates the same correlation functions pertaining to the dot operators as the original hamiltonian , ensuring the validity of the calculations of the heat current . to this end , we have to adjust the parameters of our effective hamiltonian such that the effective electronic band - width is @xmath41 with @xmath42 , where @xmath43 the renormalized level energy is effected by the flux threading the ring @xmath44 where @xmath45 . as the flux @xmath26 contributes only to the effective @xmath13 , which will be shown to effect the heat current only at quartic orders or through the renormalization of the coupling coefficient @xmath0 , one concludes that the magnetic flux effects the heat current similarly . moreover , its effect is an even function of the flux . the nonequilibrium conditions that will give rise to the heat current will be manifested by assuming to hold the electronic and bosonic baths at different temperatures @xmath46 and @xmath47 , respectively . this temperature gradient may be intentional and well controlled , or can be the by - product of some other dynamics that inevitably causes an imbalance between the two heat baths . in either case we assume that all other energy relaxation mechanisms between the electronic and the bosonic bath are sufficiently inefficient such that all local relaxation can be regarded as mediated by the molecule . in the keldysh spirit,@xcite we account for the temperature difference by taking the initial density operator in the distant past to have the form @xmath48 where @xmath49 and @xmath50 are the reciprocal temperatures and @xmath51 represent the two decoupled baths . the system is then evolved in time according to the full hamiltonian @xmath52 until steady state is reached . the electronic hamiltonian @xmath53 contains a natural high - energy cutoff or bandwidth @xmath54 , which , depending on the context , may represent either the actual conduction - electron bandwidth or the hybridization width of a certain molecular orbital . our subsequent solution of the nonequilibrium state is confined to the weak - coupling regime , @xmath55 , which serves as a prerequisite for some of the realizations of hamiltonian of eq . ( [ h ] ) depicted in fig . [ fig : fig1 ] . in this limit one can apply abelian bosonization @xcite to convert the hamiltonian and the initial density operator to a form quadratic in bosonic operators . specifically , the hamiltonian of eq . ( [ h ] ) takes the form @xmath56 \sum_{k > 0 } \xi_k ( a^{\dagger}_k + a_k ) \nonumber \\ & + ( b^{\dagger } + b ) \sum_n \lambda_n ( \gamma^{\dagger}_n + \gamma_n ) , \label{h - bosonic}\end{aligned}\ ] ] where the first two terms on the right - hand side correspond to the free hamiltonian terms @xmath53 and @xmath57 that appear in @xmath58 ; @xmath59 and @xmath60 with @xmath61 ( @xmath62 being the size of the system ) are canonical bosonic creation and annihilation operators corresponding to the fourier modes of the fermionic density ; @xmath63 equals @xmath64 ; and the coefficients @xmath65 , which have the dimension of one over length , are given by @xmath66 note that we have omitted in eq . ( [ h - bosonic ] ) the contribution of the @xmath67 mode of the fermionic density as it has no effect on our problem of interest . as for the coupling constants @xmath68 and @xmath69 , these have the dimension of energy times length , and are given to linear order in @xmath0 and @xmath13 by @xcite @xmath70 as we shall argue below , one must include higher orders in @xmath13 to account for its effect on the heat current . our interest is in the steady - state heat current flowing between the two baths . formally there are several heat - current operators one can define , e.g. , the heat current flowing into the bosonic bath or the heat current flowing out of the electronic bath , all of which must coincide in steady state . for convenience we shall focus on the heat current flowing into the bosonic bath , whose corresponding operator @xmath71\ ] ] reads @xmath72 our goal is to evaluate the steady - state expectation value @xmath73 . first we consider the case where @xmath74 , which corresponds for a molecular bridge to electronic resonance conditions . technically , this limit is somewhat easier to address as the bosonic hamiltonian is purely quadratic . a nonzero @xmath69 introduces a term linear in bosonic operators to the hamiltonian of eq . ( [ h - bosonic ] ) , whose treatment requires some care . we defer discussion of the off - resonance case to sec . [ sec : off - resonance ] below . to compute the heat current we begin by writing it in the form @xmath75 where @xmath76 in steady state all two - time correlation functions reduce to a function of the time difference only . this allows one to convert to the energy domain by fourier transforming with respect to the time difference . we apply this procedure to @xmath77 , and aiming at evaluating @xmath78 we resort to ordinary diagrammatic perturbation expansion to write an expression for @xmath79 . to this end , it is useful to define the free ( i.e. , @xmath80 ) green functions of the bath bosons in matrix form as @xmath81,\ ] ] for the retarded and advanced functions , and @xmath82,\ ] ] for the lesser and greater functions , where @xmath83 is the bose - einstein distribution function corresponding to the temperature @xmath84 . similarly , we shall denote the fully dressed green functions of the molecular vibrational mode @xmath3 by @xmath85 and @xmath86 , for which the @xmath87 matrix notation @xmath88\ ] ] is used . [ the same @xmath87 matrix notation applies to @xmath89 above . ] having laid out the building blocks for the perturbation expansion , we rely on langreth theorem @xcite to have the identity , at steady - state @xmath90_{1 p } \left\ { \left [ g^a_{\gamma , n}(\epsilon ) \right]_{22 } - \left [ g^a_{\gamma , n}(\epsilon ) \right]_{11 } \right\ } + \left [ g^{r}(\epsilon ) \right]_{1 p } \left\ { \left [ g^>_{\gamma , n}(\epsilon ) \right]_{22 } - \left [ g^>_{\gamma , n}(\epsilon ) \right]_{11 } \right\ } \right ] , \label{g^>_b , n}\ ] ] which we shall now turn to evaluate . focusing initially on the retarded and advanced green functions @xmath85 , these acquire the form @xmath91^{-1 } , \label{g^ra}\ ] ] where @xmath92 \nonumber \\ & + \sum_n \lambda_n^2 \left [ \frac{1 } { \epsilon - \omega_n \pm i\eta } - \frac{1 } { \epsilon + \omega_n \pm i\eta } \right ] , \label{s^ra}\end{aligned}\ ] ] are the corresponding self - energy functions . there are two contributions to @xmath93 : one due to the coupling to the electronic bath [ the first term on the right - hand side of eq . ( [ s^ra ] ) ] , and another due to the coupling to the bosonic bath [ the second term on the right - hand side of eq . ( [ s^ra ] ) ] . denoting these two terms by @xmath94 and @xmath95 , respectively , the former can be expressed in a closed analytical form @xcite in terms of the exponential integral function : @xcite @xmath96 . \label{s^ra_e}\end{aligned}\ ] ] here , @xmath97 is the density of states per unit length , and @xmath98 equals @xmath99 . as for the second contribution @xmath95 , it can be conveniently expressed in terms of the coupling function @xmath100 , @xmath101 d\epsilon ' . \label{s^ra_b}\ ] ] for an ohmic bath with @xmath12 , eq . ( [ s^ra_b ] ) can be evaluated in closed analytic form to obtain @xmath102 . \label{s^ra_b - ohmic}\ ] ] proceeding to the lesser and greater green functions @xmath86 , these read @xmath103 \sigma^ { < , > } ( \epsilon ) g^{a}(\epsilon ) , \label{g^<>}\ ] ] with the lesser and greater self - energy functions @xmath104 . \label{s^<>}\end{aligned}\ ] ] here , @xmath105 is the bose - einstein distribution function corresponding to the temperature @xmath106 . note that eq . ( [ s^ < > ] ) can be conveniently related to the two components of the retarded self - energy through @xmath107 which generalizes the standard equilibrium relation @xmath108 . to evaluate @xmath109 of eq . ( [ g^>_b , n ] ) , it is useful to utilize the identities @xmath110_{1 p } & = |g(\epsilon)|^2 2 \omega_0 ( \epsilon + \omega_0 ) \sigma^{>}(\epsilon ) , \nonumber \\ \sum_{p = 1 , 2 } \left [ g^{r}(\epsilon ) \right]_{1 p } & = |g(\epsilon)|^2 ( \epsilon + \omega_0 ) \left [ \epsilon^2 - \omega_0 ^ 2 - 2\omega_0 \sigma^{a}(\epsilon ) \right ] , \end{aligned}\ ] ] which follow directly from eqs . ( [ g^ra ] ) and ( [ g^ < > ] ) . here we have introduced the auxiliary function @xmath111 and made use of the fact that @xmath112^{\ast}$ ] . inserting these identities into eq . ( [ g^>_b , n ] ) one obtains @xmath113 \left [ \delta(\epsilon - \omega_n ) + \delta(\epsilon + \omega_n ) \right ] , \ ] ] which reduces by virtue of eq . ( [ s^<>-via - s^r ] ) to @xmath114 \left [ \delta(\epsilon - \omega_n ) + \delta(\epsilon + \omega_n ) \right ] .\ ] ] plugging this result into eq . ( [ j_q - via - g ] ) , employing the relations @xmath115 and @xmath116 = - [ n_e(\epsilon ) - n_b ( \epsilon ) ] $ ] , and exploiting the fact that @xmath117 is restricted to positive energies , we finally arrive at @xmath118 , \label{j_q - general}\ ] ] where we have explicitly written out the function @xmath119 that appears in the integrand [ see eq . ( [ g - def ] ) ] . equation ( [ j_q - general ] ) is the central result of this paper . it provides an exact expression for the heat current corresponding to the bosonic hamiltonian of eq . ( [ h - bosonic ] ) , for a general coupling function @xmath117 . since @xmath120 $ ] with @xmath121 is positive definite for @xmath122 ( negative definite for @xmath123 ) , the heat current flows , as it physically should , from the hotter bath to the colder one . below we analyze in detail the characteristics of @xmath124 in different temperature and coupling regimes . the form of expression for the heat current bears a similarity to landauer formula . other works studying thermal conductance through local contacts have arrived at a similar expressions or used a landauer - type expression as a starting point . @xcite in the context of heat current between bosonic reservoirs this expression was derived in earlier works discussing hamiltonians similar to the one in eq . ( [ h - bosonic ] ) @xcite . this type of expression survives even when accounting perturbatively for the interaction with vibrational modes . @xcite we begin with the low - temperature limit , @xmath125 ( throughout this paper we assume that @xmath126 ) . in this limit , the bose - einstein distribution functions @xmath127 and @xmath128 that enter the integrand of eq . ( [ j_q - general ] ) have decayed long before @xmath129 and @xmath119 have changed in any significant manner from their @xmath130 values . thus , to a good approximation one can ( i ) set @xmath131 in the integrand of eq . ( [ j_q - general ] ) , and ( ii ) extend the upper integration limit to infinity . taking the coupling function @xmath117 to have the power - law form of eq . ( [ j - power - law ] ) this yields @xmath132 d \epsilon .\ ] ] the resulting integral can now be carried out analytically using @xmath133 where @xmath134 is the riemann zeta function . @xcite this in turn gives @xmath135 with @xmath136 . lastly , @xmath137 has the explicit expression @xmath138 } , \ ] ] allowing one to express the coefficient @xmath139 entirely in terms of the basic model parameters entering the bosonic hamiltonian of eq . ( [ h - bosonic ] ) . as can be seen from eq . ( [ j_q - low - t ] ) , the low - temperature heat current shows a rather strong temperature dependence . in particular , the linear - response heat conductance @xmath140 varies as @xmath141 , corresponding to @xmath142 for an ohmic bath . this should be contrasted with the heat conductance of a generic noninteracting electronic tunnel junction , which varies linearly with @xmath143 at sufficiently low temperature . @xcite for the heat conductance of a tunnel junction to display a superlinear temperature dependence of the form found here , its transmission coefficient must vanish in a power - law fashion at the fermi energy . this power - law behavior of the heat current at low temperatures was also observed for systems in which the two bosonic reservoirs are connected by a system with few degrees of freedom such as in the spin - boson model @xcite . the similarity between these systems and the system under consideration here , at low - temperatures with respect to @xmath18 , stems from the fact that at this temperature regime only the lowest lying levels of the vibrational mode of the molecule are available for transferring energy between the baths . as commented above , an ohmic bath with @xmath12 is of particular interest . focusing on this case and on the hierarchy @xmath144 , we devise below an analytical expression for the nonequilibrium heat current , encompassing the crossover from the low - temperature regime , @xmath145 , to the intermediate - temperature one , @xmath146 . this expression is approximate as it employs a power series expansion of the self - energies , but yields accurate results within the regime where @xmath24 and @xmath10 are the largest energy scales in the system . to this end , consider the function @xmath147 which enters the integrand of eq . ( [ j_q - general ] ) . since the bose - einstein distribution functions @xmath127 and @xmath128 decay on a scale far smaller than @xmath148 , it suffices to accurately represent @xmath147 for @xmath149 . this allows one to expand the exact expressions for @xmath150 and @xmath151 [ see eqs . ( [ s^ra_e ] ) and ( [ s^ra_b - ohmic ] ) above ] in @xmath152 and @xmath153 to obtain @xmath154 , \\ \sigma^{r}_{b}(\epsilon ) & = - 2 \pi \alpha \omega_c \left [ i\pi y + 2 + { \cal o } \left ( y^2 \right ) \right ] .\end{aligned}\ ] ] settling with linear orders in @xmath155 and @xmath156 , the function @xmath147 is well approximated for @xmath157 by @xmath158 where @xmath159 equals @xmath160 with the softened frequency @xmath161 ^ 2 } \label{omega}\ ] ] and the relaxation rate @xmath162 . \label{tau}\ ] ] the softened frequency @xmath163 characterizes the dressed excitations of the phonon , and thus serves as the energy scale which determines the cross over from the high - temperature to the low - temperature behavior . within this approximation for @xmath147 , the heat current for an ohmic bath becomes @xmath164 } { | ( \epsilon - z_{+ } ) ( \epsilon - z_{- } ) |^2 } , \label{j_q - ohmic - int}\ ] ] where we have set in addition @xmath165 and extended the upper integration limit to infinity ( both approximations being well justified by the hierarchy @xmath166 ) . the resulting integral in eq . ( [ j_q - ohmic - int ] ) can be performed in a closed analytic form in terms of the digamma function @xcite @xmath167 . skipping details of the algebra we quote here only the end result : @xmath168 + i\pi\tau \left ( t_{e } - t_{b } \right ) \right\ } . \label{j_q - ohmic}\ ] ] it is straightforward to confirm using the asymptotic expansion @xcite @xmath169 that eq . ( [ j_q - ohmic ] ) properly reduces for @xmath170 to eq . ( [ j_q - low - t ] ) with @xmath12 , including the precise value of the prefactor @xmath139 . more interesting is the limit @xmath171 , when the leading contribution to eq . ( [ j_q - ohmic ] ) crosses over to a linear dependence on the temperature difference @xmath172 . thus , as the larger of the two temperatures exceeds the vibrational resonance energy , the heat current continues to increase linearly with @xmath173 . this behavior markedly differs from that of a resonant electronic tunnel junction , whose heat current depends logarithmically on @xmath173 above the resonance energy . the physical difference stems , as we show below , from the bosonic nature of the vibrational mode , which can be excited to exceedingly high energies by creating ever more phonons . this should be contrasted with a resonant electronic level , which can only be empty or occupied . having analyzed analytically certain limits , we now proceed to a complete numerical evaluation of the exact heat current of eq . ( [ j_q - general ] ) at arbitrary temperatures @xmath106 and @xmath84 . as the temperatures @xmath106 and @xmath84 enter the expression for the heat current only through the term @xmath174 $ ] in the integrand , it is anti - symmetric under replacing them . as such , we may restrict ourselves to calculations in which @xmath84 is held constant and the heat current is calculated for different values of @xmath106 , and generalize the results for opposite values by inverting the direction of the current . in fig . [ fig : j_q - low - t ] we have addressed the case where one bath ( the bosonic one ) is held at a constant low temperature @xmath175 , while scanning different values of @xmath106 higher than that temperature . we have plotted the heat current in the case of a sub - ohmic bath ( with @xmath176 ) , an ohmic bath and a super - ohmic bath ( with @xmath177 ) . the graph shows a clear crossover from a power - law behavior to a linear dependence as @xmath173 increases to values of the order of @xmath18 . in the inset of the graph we have plotted , on a log - log scale , the heat current for small values of @xmath173 for each type of bath , and a dashed line following the expected power - law behaviors at low temperatures given at eq . ( [ j_q - low - t ] ) . the agreement between the calculated heat current and the expected one is excellent while @xmath178 , breaking at about @xmath179 . figure [ fig : j_q - high - t ] describes the heat current as well , but focuses on the case where the temperature of the bosonic bath is held at the high value of @xmath180 . in this case the heat current about @xmath181 displays a linear dependence on @xmath173 , with different slope for every type of bosonic bath . in fig . [ fig : diff_conductance ] we have plotted the linear - response heat conductance @xmath182 , as defined in eq . ( [ conductance_defined ] ) , for different temperatures and types of baths . as the temperatures rises , the conductance increases until it saturates at about @xmath183 . the different exponents relating to the type of bosonic baths are evident in the low - temperature regime . between the electronic and bosonic baths , as a function of the temperature difference between the baths . here the temperature of the bosonic bath was held constant at a low value @xmath175 and the temperature of the electronic bath @xmath106 was changed . the heat current was calculated for different types of bosonic baths , where we considered a sub - ohmic case ( @xmath176 , green ) , an ohmic case ( red ) and a super - ohmic case ( @xmath184 , blue ) . inset : a log - log plot of the low - temperature regime . the dashed lines are following the appropriate power - laws predicted in that regime by eq . ( [ j_q - low - t ] ) . here @xmath124 is measured in units of the basic heat current @xmath185 , where @xmath186 , and @xmath187 are in the weak coupling regime . the cutoffs used are @xmath188 . , scaledwidth=44.0% ] and we calculated the current for different values of @xmath106 . the heat current was calculated for different types of bosonic baths , where we have considered a sub - ohmic case ( @xmath176 , green ) , an ohmic case ( red ) and a super - ohmic case ( @xmath184 , blue ) . inset : a zoom over the regime about the point @xmath189 where the heat current displays a linear dependence on @xmath173 . here @xmath124 is measured in units of the basic heat current @xmath185 , where @xmath186 , and @xmath187 are in the weak coupling regime . the cutoffs used are @xmath188 . , scaledwidth=44.0% ] between the baths as a function of the temperature @xmath143 in which both bath are helds . the conductance is calculated for different types of bosonic baths , where we have considered a sub - ohmic case ( @xmath176 , green ) , an ohmic case ( red ) and a super - ohmic case ( @xmath184 , blue ) . here @xmath182 is measured in units of a basic heat conductance @xmath190 , where @xmath186 , and @xmath187 are in the weak coupling regime . the cutoffs used are @xmath188 . , scaledwidth=44.0% ] although our solution for @xmath124 is formally exact within the bosonic hamiltonian of eq . ( [ h - bosonic ] ) , it is instructive to develop a more transparent physical picture that would , in particular , elucidate the source of distinction between the bosonic system under consideration and a conventional resonant tunnel junction . to this end , we devise below a master equation approach , applicable at weak coupling . this approach to address heat flow was previously applied by segal @xcite , considering a different setup . leijnse and others @xcite have also used this method to calculate the thermopower properties in a similar setup . the basic components of the theory are the probabilities @xmath191 to find the local boson @xmath3 at time @xmath19 in the state where @xmath192 . these probabilities are connected at the golden - rule approximation by rate equations of the form @xmath193 ( terms with @xmath194 should be omitted for @xmath195 ) , with the transition rates @xcite @xmath196 , \\ w_{n \to n - 1 } & = 2\pi n \left\ { f(\omega_0 ) [ 1 + n_e(\omega_0 ) ] \right . \nonumber \\ & \;\;\;\;\;\;\;\;\;\;\;\ ; \left . + \lambda(\omega_0 ) [ 1 + n_b(\omega_0 ) ] \right\ } .\end{aligned}\ ] ] here , @xmath197 equals @xmath198 . the golden - rule approximation used here corresponds to lowest order perturbation theory in both the coupling to the phononic bath [ represented by the coupling function @xmath199 and the coupling to the bosonic bath [ represented by the coupling function @xmath117 ] . it is useful to define at this point the reduced transition rates @xmath201 , \\ w_{\downarrow } & = 2\pi \left\ { f(\omega_0 ) [ 1 + n_e(\omega_0 ) ] + \lambda(\omega_0 ) [ 1 + n_b(\omega_0 ) ] \right\ } , \end{aligned}\ ] ] such that @xmath202 and @xmath203 . focusing on steady state , when @xmath204 , eq . ( [ master - equations ] ) can be recast as an infinite set of coupled linear equations , @xmath205 whose solution is @xmath206 with @xmath207 . here , @xmath208 is a normalization factor which comes to ensure that @xmath209 . thus , the probabilities @xmath210 obey a boltzmann - like form with an effective temperature @xmath211 defined by @xmath212 consider first the case of thermal equilibrium , when @xmath213 and @xmath214 . under these circumstances @xmath215 hence @xmath211 equals @xmath143 irrespective of details of the two baths . once a temperature gradient is applied between the two reservoirs , i.e. , @xmath216 , then @xmath211 falls in between @xmath217 and @xmath218 , as follows from the equality @xmath219 with @xmath220 } { f(\omega_0 ) [ 1 + n_e(\omega_0 ) ] + \lambda(\omega_0 ) [ 1 + n_b(\omega_0 ) ] } . \label{q - def}\ ] ] equation ( [ t_eff - via - exponents ] ) can be solved analytically in the high - temperature limit , @xmath221 , where fermi s golden rule ( and thus our master equation approach ) is expected to apply . specifically , expanding each of the exponents to linear order in @xmath18 one obtains @xmath222 this expression can further be simplified by noting that @xmath223 depends implicitly on @xmath106 and @xmath84 through the bose - einstein distribution functions @xmath224 and @xmath225 . plugging these relations into eq . ( [ q - def ] ) and inserting the resulting expression for @xmath223 into eq . ( [ t_eff - via - q ] ) , one finally arrives at @xmath226 where we have made repeated usage of the fact that @xmath221 . it should be stressed that this result equally applies to all forms of the bosonic bath , be it ohmic , sub - ohmic or super - ohmic . the steady - state solution to the probablities @xmath210 can be used in turn to calculate the heat current . focusing again on the heat current flowing between the bosonic bath and the local phonon , the latter involves the transition rates @xmath227 to and from bosonic bath . explicitly , the heat current takes the form @xmath228 where @xmath229.\end{aligned}\ ] ] using these rates the expression for the heat current gains the compact form @xmath230 , \label{j_q - master - eq}\ ] ] where @xmath231 is the average occupancy of the localized phonon . recalling that @xmath211 lies between the temperatures of the bosonic and electronic baths , it is clear that @xmath124 is positive ( negative ) when @xmath122 ( @xmath232 ) , which gives the correct direction of the heat flow . the above result provides a transparent picture for the linear dependence of the heat current on the temperature gradient in the high - temperature regime . in this regime , the expression for @xmath124 can be further approximated using eq . ( [ t_eff_high_t ] ) for the effective temperature , which results in @xmath233 hence , the linear dependence on the temperature gradient stems from the large occupancy of the localized phonon , which does not saturate with increasing temperature . segal @xcite , using the master equation approach , described a similar linear dependence of the heat flow on the temperature difference . this should be contrasted with a spinless electronic resonant level , that can only be empty or singly occupied . while eq . ( [ j_q - master - eq ] ) properly captures the physics of the high - temperature regime , it fails to produce the required power - law behavior in the low - temperature regime , @xmath234 . indeed , replacing the bose - einstein functions with simple exponents and using eq . ( [ t_eff - via - exponents ] ) for the effective temperature , the heat current becomes @xmath235 where we have omitted the exponentially small bose - einstein distribution functions in the expression for @xmath223 [ see eq . ( [ q - def ] ) ] . thus , the master equation approach predicts activated low - temperature behavior in place of the correct power - law form . the fact that the master equation approach well describes the high - temperature regime but fails at low temperatures is by itself not surprising . here we can exploit the exact solution to carefully examine the range of validity of the approach and the simple physical picture that it lends . generally speaking , the quality of the master equation approach depends on two parameters : ( 1 ) the temperatures involved , and ( 2 ) the strength of the coupling constants . naturally , the master equation approach loses quantitative accuracy as the coupling constants are increased since fermi s golden - rule corresponds to lowest - order perturbation theory . nevertheless , as we shall show , the qualitative picture can remain quite accurate even when the coupling constants are increased . to critically examine the master equation approach we shall focus on the three quantities : ( 1 ) the effective temperature @xmath211 defined by the average occupancy of the localized phonon [ see eq . ( [ t_eff_from_exact ] ) below ] ; ( 2 ) the boltzmann - like distribution of the localized phonon , and ( 3 ) the heat current @xmath124 . as shown above , the master equation approach predicts an effective temperature @xmath211 which is manifest in the average phonon occupancy @xmath236 . similarly , one can define an effective temperature from the exact phononic occupancy @xmath237 according to @xmath238 where @xmath163 is the softened phonon frequency , approximated by eq . ( [ omega ] ) . usage of the softened frequency , rather than the bare one , comes to account for higher - order correction not included in the golden - rule approximation . we emphasis that eq . ( [ t_eff_from_exact ] ) serves as an ad - hoc definition of an effective temperature , which does not , by itself , imply a boltzmann - like distribution . we shall examine this latter point in the following subsection . figure [ fig : t_eff ] compares the effective temperature extracted from the exact solution with the master equation result of eq . ( [ t_eff - via - exponents ] ) , scanning different coupling strengths . there is an excellent agreement at weak coupling which gradually deteriorates as the combined coupling to the two baths is increased . the agreement is controlled by the decay time @xmath239 of eq . ( [ tau ] ) , which depends quadratically on the coupling @xmath0 to the electronic bath and linearly on the coupling @xmath11 to the bosonic one . for the strongest coupling considered ( right - hand side of the lower - right panel ) the deviation between the two curves is of the order of @xmath240 , which is still quite moderate . and for different values of the coupling to the bosonic reservoir @xmath11 . the red line represents the effective temperature calculated based on the master equation approach while the black line is the effective temperature extracted from the phononic occupancy as given in eq . ( [ t_eff_from_exact ] ) . in this graph @xmath241 , the bosonic bath was taken to be ohmic and the temperatures used are @xmath242 , @xmath243 , which lie well within the applicatbility temperature range of the master equation approximation . , scaledwidth=44.0% ] the boltzmann - like distribution of the probablities @xmath210 predicts all the moments of the phononic occupancy @xmath244 . the ratio between the moments can be expressed using the moments themselves , giving the ratio between the first two moments @xmath245 which does not directly depend on the effective temperature , and is true for every boltzmann - like distribution , regardless of the value of the temperature . we shall use this ratio as a benchmark for examining how close the phononic distrubtion is to a boltzmann - like one , as the exact ratio is calculated using the phononic green functions at steady state @xmath246 + \left|g^{<}_{bb}(t , t)\right|^2 \right\}. \nonumber\end{aligned}\ ] ] the calculated ratio deviates from the boltzmann - like one due to the presence of the term @xmath247 , which is identically zero for thermal distributions . in fig . [ fig : is_thermal ] we have plotted the exact ratio between the second and first moments of the phononic occupation and compared it to the expected ratio were the distribution was boltzmann - like . similarly to the comparison done for the effective temperature , the comparison was done as a function of the coupling strength . for small couplings the exact ratio matches excellently the one expected from a boltzmann - like distribution , thus confirming the approximation . as the coupling is increased , the exact ratio deviates from the botlzmann - like prediction , and the non - thermal nature of the distribution becomes prominent . we feel it is important to stress , however , that the distribution is not thermal for any coupling strength . as a third quantity to which we look in order to examine the validity of the master equation approximation is the heat current between the two baths @xmath124 . the exact expression for the heat current is given in its integral form in eq . ( [ j_q - general ] ) and the expression derived from the master equation approach given in eq . ( [ j_q - master - eq ] ) . the comparison of the results of these two expressions is plotted in fig . [ fig : j_q_master_eq ] where we have scanned different temperatures while holding the temperature difference between the baths constant . in accordance with our previous analysis of the results , the master equation reproduces quite well the exact result only at high temperatures , and deviating from it significantly as the temperatures are lowered . as a function of the electron - phonon coupling @xmath0 and for different values of the coupling to the bosonic reservoir @xmath11 . the red line represents the ratio as predicted by the master equation while the black line is the actual ratio calculated according to the exact solution . in this graph @xmath241 , the bosonic bath was taken to be ohmic and the temperatures used are @xmath242 , @xmath243 , which lie well within the applicability temperature range of the master equation approximation . , scaledwidth=44.0% ] and the approximate calculation relying on the master equation approach as given in eq . ( [ j_q - master - eq ] ) . the temperature difference between the reservoirs is held fixed with @xmath248 and both temperatures are changed . inset : the ratio of the approximated result from the exact one , showing a larger difference as the temperature is lowered . here @xmath124 is measured in units of the basic heat current @xmath185 , where @xmath186 , and @xmath249 are in the weak coupling regime . the cutoffs used are @xmath188 . , scaledwidth=44.0% ] so far only the resonant case where @xmath16 was considered . since @xmath13 can be controlled by using suitable gate voltages , studying its effects on the heat current is of particular interest . in this section we address the case where @xmath13 is given some nonzero value , breaking the particle - hole symmetry of the model . staying in accordance with the rest of our discussion , where we have assumed that the bandwidth @xmath24 is the largest energy scale of the system , we are interested in the regime where @xmath250 in addition to @xmath251 . we shall demonstrate that in this regime the leading contribution of @xmath13 to the heat current is either @xmath252 or @xmath253 . before getting to the task of explicitly calculating the heat current for the off - resonant case , we first consider the symmetries of the hamiltonian of eq . ( [ h ] ) . if @xmath16 then the system holds a particle - hole symmetry , which breaks for nonzero values of @xmath13 . writing the hamiltonian as a function of @xmath13 and carrying out an particle - hole transformation @xmath254 one finds out that @xmath255 , while the expression for @xmath124 remains unchanged . thus we conclude that @xmath124 is an even function of @xmath13 . next we turn to the calculation done in sec . [ sec : resonance_heat_current ] under resonance conditions and aim at adjusting it to the case where @xmath256 . since the @xmath13 term in the hamiltonian of eq . ( [ h - bosonic ] ) is linear in the operators @xmath257 and @xmath258 , one may express its effects on the heat current by appropriate corrections to the bare green functions pertaining to these operators . we will then be able to use the functions , dressed by the @xmath13 term , to re - calculate the self - energies @xmath259 of eq . ( [ s^ra ] ) and @xmath260 of eq . ( [ s^ < > ] ) . to this end , we consider a free hamiltonian to which we add an @xmath13 term @xmath261 where @xmath262 . we will designate by @xmath263 the unperturbed green functions given , in energy space , by @xmath264 , \label{g0_ra_epsilon_d}\ ] ] @xmath265 , \label{g_les_grea_epsilon_d}\ ] ] and by @xmath266 the dressed functions with respect to the hamiltonian . the hamiltonian can be diagonalized exactly by introducing new bosonic creation and annihilation operators @xmath267 , by which it takes the form @xmath268 this is a free hamiltonian with respect to the @xmath269 s . the correlation functions between the original bosonic operators are related to the correlation functions of the @xmath269 s by the fact that @xmath270 we can thus adjust the lesser and greater green functions , in the energy domain , by adding the appropriate term proportional to @xmath271 , which reflects the fact that the extra term added by @xmath13 on the right - hand - side of eq . ( [ nonresonant_new_bosonic ] ) is independent of time @xmath272 we continue to note that the retarded and advanced green functions remain unchanged by this addition of non - zero @xmath13 , as the extra term is constant and drops out when the commutation relations are taken . having arrived at the conclusion that the corrections due to the @xmath13 term exist only at @xmath273 , we point to the fact that the integrand in the expression for the heat current given in eq . ( [ j_q - general ] ) depends directly on @xmath274 . therefore , all such terms do not contribute to the heat current , leaving it independent of the value of @xmath13 . as @xmath13 reflects the energy associated with the charging of the level , one would expect that its value will directly affect the heat current . we understand the independence of the latter on the value of @xmath13 as a result of the weak - coupling regime @xmath275 , and expect that as either of these values is increased , @xmath13 will play a role in determining the heat current . taking into account that the heat current is an even function of @xmath13 , we conclude our discussion in noting that the leading contribution in this regime is not lower than @xmath276 or @xmath277 . in this paper we have presented an asymptotically exact calculation of the heat current between a bosonic bath and a fermionic bath , that is mediated by a single molecule . the calculation is based on a mapping of continuous model given by the hamiltonian of eq . ( [ h ] ) onto a form quadratic in bosonic operators @xcite . this mapping is valid in the weak coupling regime , where @xmath278 , and @xmath24 is the effective electronic bandwidth . this model may describe , under suitable mappings , several physically and experimentally relevant setups , the most relevant being a molecule adsorbed on a surface , a molecular junction and an aharonov - bohm interferometer with a molecular device embedded in one of its arms . the exact calculation yields a landauer - type expression for the heat current , given in eq . ( [ j_q - general ] ) . such an expression stands in accordance with previous works on thermal currents in confined nanostructures . @xcite at low temperatures , the heat current strongly depends on the nature of the bosonic bath . assuming that the bosonic bath has a power - law form and is characterized by the power @xmath279 , the low - temperature linear - response heat conductance varies as @xmath280 . at high temperatures , however , the heat current depends linearly on the temperature difference between the two baths , regardless of the nature of the power - law governing the bosonic bath . the crossover between the low and high - temperatures regimes is at the scale of the softened vibrational mode frequency @xmath163 given in eq . ( [ omega ] ) . the high - temperature behavior , which is markedly different than the transmission through a purely electronic system , is explained by the bosonic nature of vibrational mode , which can be excited to high energies by creating more phonons . this is illustrated by a master equation analysis of the system , which is perturbative in nature but is justified in the high - temperature and weak coupling regime . in that regime the master equation approach reproduces the heat current that was calculated exactly previously , and also offers an effective temperature that we assign to the local vibrational mode . it should be stressed that even at that regime it does not have a thermal distribution , and the effective temperature is a useful illustrative approximation . as our solution is exact only at weak electron - phonon coupling , one would expect different features to appear in the heat current as the interaction strength is increased . it would be interesting to compare our results with such an analysis , and see how our calculation persists into the strong - coupling regime . we leave that to future work . 99 for recent reviews see , e.g. , _ introducing molecular electronics _ , edited by g. cuniberti , g. fagas , and k. richter , lecture notes in physics vol . 680 ( springer , new york , 2005 ) ; m. galperin , m. a. ratner , and a. nitzan , j. phys . : condens . matter * 19 * , 103201 ( 2007 ) . wang , j. wang , and j. t. l , eur . j. b * 62 * , 381 - 404 ( 2008 ) . d. g. cahill , w. k. ford , k. e. goodson , g. d. mahan , a. majumdar , h. j. maris , r. merlin and s. r. phillpot , j. appl . phys . * 93 * , 793 ( 2003 ) . d. r. ward , d. a. corley , j. m. tour and d. natelson , nature nanotechnology * 6 * , 33 - 38 ( 2011 ) m. galperin , a. nitzan , and m. a. ratner , phys . b * 75 * , 155312 ( 2007 ) . o. entin - wohlman , y. imry and a. aharony , phys . b * 82 * , 115314 ( 2010 ) . o. entin - wohlman and a. aharony , phys . rev . b * 85 * , 085401 ( 2012 ) . a. ozpineci and s. ciraci , phys . b * 63 * , 125415 ( 2001 ) . d. segal , a. nitzan , and p. hnggi , j. chem . * 119 * , 6840 ( 2003 ) . l. a. wu and d. segal , phys . e * 83 * , 051114 ( 2011 ) . b. sothmann , r. sanchez , a. n. jordan and m. bttiker , phys . b. * 85 * , 205301 ( 2012 ) . b. sothmann and m. bttiker , epl * 99 * , 27001 ( 2012 ) . l. i. glazman and m. raikh , jetp lett . * 47 * , 452 ( 1988 ) . g. schulze , k. j. franke , a. gagliardi , g. romano , c. s. lin , a. l. rosa , t. a. niehaus , th . frauenheim , a. di carlo , a. pecchia and j. i. pascual , phys . * 100 * , 136801 ( 2008 ) ; g. schulze , k. j. franke and j. i. pascual , new j. phys . * 10 * , 065005 ( 2008 ) ; k. j. franke and j. i. pascual , j. phys . : condens . matter * 24 * , 394002 ( 2012 ) . m. tsutsui , m. taniguchi , k. yokota and t. kawai , appl . lett . * 96 * , 103110 ( 2010 ) . d. segal , phys . lett . * 100 * , 105901 ( 2008 ) . t. ojanen and a. p. jauho , phys . 100 * , 155902 ( 2008 ) . t. ruokola and t. ojanen , phys . b. * 83 * , 045417 ( 2011 ) . a. dhar and d. roy , j. stat . * 125 * 801 ( 2006 ) . a. dhar , adv . phys . * 57 * , 457 ( 2008 ) . k. saito and t. kato , arxiv:0706.1234 we employ units in which @xmath281 . a. j. legget , s. chakravarty , a. t. dorsey , m. p. a. fisher , a. garg , and w. zwerger , rev . phys . * 59 * , 1 ( 1987 ) . d. sherrington and s. von molnr , solid st . comm . * 16 * , 1347 ( 1975 ) . r. r. cavanagh , e. j. heilweil and j. c. stephenson , surface sciente * 299 * , 643 ( 1994 ) . j. w. gadzuk , phys . b * 24 * , 1651 ( 1981 ) ; e. blaisten - barojas and j. w. gadzuk , j. chem . phys . * 97 * , 862 ( 1992 ) . y. vinkler , a. schiller , and n. andrei , phys . b * 85 * , 035411 ( 2012 ) . h. l. edwards , q. niu and a. l. de lozanne , appl . lett . * 63 * , 1815 ( 1993 ) ; h. l. edwards , q. niu , g. a. georgakis and a. l. delozanne , phys . b * 52 * , 5714 ( 1995 ) . j. r. prance , c. g. smith , j. p. griffiths , s. j. chorley , d. anderson , g. a. c. jones , i. farrer and d. a. ritchie , phys . lett . * 102 * , 146602 ( 2009 ) . i. g. lang and yu . a. firsov , zh . eksp . * 43 * , 1843 ( 1962 ) [ sov . jetp * 16 * , 1301 ( 1963 ) ] . b. dra and a. halbritter , phys . b. * 80 * , 155402 ( 2009 ) . b. dra , phys . b. * 75 * , 245113 ( 2007 ) . b. dra and m. gulcsi , phys . b. * 78 * , 165111 ( 2008 ) . w. hofstetter , j. knig and h. schoeller , phys . 87 * , 156803 ( 2001 ) . l. v. keldysh , sov . jetp * 20 * , 1018 ( 1965 ) . f. d. m. haldane , j. phys . c * 14 * , 2585 ( 1981 ) . d. c. langreth , in _ linear and nonlinear electron transport in solids _ ( plenum pres , new york , 1976 ) , vol . 17 of _ nato advanced study institute , series b : physics _ , eds . j. t. devreese and v. e. van doren . see , e.g. , _ handbook of mathematical functions _ , eds . m. abramowitz and i. a. stegun ( dover , new york , 1972 ) , chapter 5 . see , e.g. , _ handbook of mathematical functions _ , eds . m. abramowitz and i. a. stegun ( dover , new york , 1972 ) , chapter 23 . f. zhan , s. denisov and p. hnggi , phys . b * 84 * , 195117 ( 2011 ) . see , e.g. , _ handbook of mathematical functions _ , eds . m. abramowitz and i. a. stegun ( dover , new york , 1972 ) , chapter 6 . d. segal , phys . b * 73 * , 205415 ( 2006 ) . m. leijnse , m. r. wegewijs and k. flensberg , phys . b * 82 * , 045412 ( 2010 ) .

in numerical analysis there exist numerous examples of objects forming a _ group _ , i.e. objects that compose in an associative manner , have an inverse and identity element . typical examples are the group of orthogonal matrices or the group of runge kutta methods . _ semigroups _ , sets of objects close under composition but not inversion , like for instance the set of all matrices and explicit runge kutta methods , are also well studied in literature . however , there are important examples of objects that are neither a group nor a semigroup . one important case is the class of objects closed under a ` sandwich - type ' product , @xmath0 . for example , the collection of all symmetric positive definite matrices and all selfadjoint runge kutta methods . the sandwich - type composition @xmath1 for numerical integrators was studied at length in @xcite and references therein . however , if inverses are well defined , we may wish to replace the sandwich product with the algebraically nicer _ symmetric product _ @xmath2 . spaces closed under such products are called symmetric spaces and are the objects of study in this paper . there is a parallel between the theory of lie groups and that of symmetric spaces . for lie groups , fundamental tools are the lie algebra ( tangent space at the identity , closed under commutation ) and the exponential map from the lie algebra to the lie group . in the theory of symmetric spaces there is a similar notion of tangent space . the resulting object is called a lie triple system ( lts ) , and is closed under _ double commutators _ , @xmath3 $ ] . also in this case , the exponential maps the lts into the symmetric space . an important decomposition theorem is associated with symmetric spaces and lie triple systems : lie algebras can be decomposed into a direct sum of a lts and a subalgebra . the well known splitting of a matrix as a sum of a symmetric and a skew symmetric matrix is an example of such a decomposition , the skew symmetric matrices are closed under commutators , while the symmetric matrices are closed under double commutators . similarly , at the group level , there are decompositions of lie groups into a product of a symmetric space and a lie subgroup . the matrix polar decomposition , where a matrix is written as the product of a symmetric positive definite matrix and an orthogonal matrix is one example . in this paper , we are concerned with the application of such structures to the numerical analysis of differential equations of evolution . the paper is organised as follows : in 2 we review some general theory of symmetric spaces and lie triple systems . applications of this theory in numerical analysis of differential equations are discussed in 3 , which , in turn , can be divided into two parts . in the first ( 3.13.3 ) , we review and discuss the case of differential equations on matrix groups . the properties of these decompositions and numerical algorithms based on them are studied in a series of papers . polar - type decompositions are studied in detail in @xcite , with special emphasis on optimal approximation results . the paper @xcite is concerned with important recurrence relations for polar - type decompositions , similar to the baker - campbell - hausdorff formula for lie groups , while @xcite employ this theory to reduce the implementation costs of numerical methods on lie groups @xcite . we mention that polar - type decompositions are also closely related to the more special root - space decomposition employed in numerical integrators for differential equations on lie groups in @xcite . in @xcite it is shown that the generalized polar decompositions can be employed in cases where the theory of @xcite can not be used . in the second part of 3 ( [ sec:3.3 ] and beyond ) , we will consider the application of this theory to numerical methods for the solution of differential equations with symmetries and reversing symmetries . by backward error analysis , numerical methods can be thought of as exact flows of nearby vector fields . the main goal is then to remove from the numerical method the _ undesired _ part of the error ( either the one destroying the symmetry or the reversing symmetry ) . these error terms in the modified vector field generally consist of complicated derivatives of the vector field itself and are not explicitly calculated , they are just used formally for analysis of the method . in this context , the main tools are compositions at the group level , using the flow of the numerical method and the symmetries / reversing symmetries , together with their inverses . there is a substantial difference between preserving reversing symmetries and symmetries for a numerical method : the first can be always be attained by a finite ( 2 steps ) composition ( scovel projection @xcite ) , the second requires in general an infinite composition . thus symmetries are generally more difficult to retain than reversing symmetries . for the retention of symmetry , we review the thue morse symmetrization technique for arbitrary methods and present a new yoshida - like symmetry retention technique for self - adjoint methods . the latter has always positive intermediate step sizes and can be of interest in the context of stiff problems , which typically require step size restrictions . we illustrate the use of these symmetrisation methods by some numerical experiments . finally , section 4 is devoted to some concluding remarks . in this section we present some background theory for symmetric spaces and lie triple systems . we expect the reader to be familiar with some basic concepts of differential geometry , like manifolds , vector fields , etc . for a more detailed treatment of symmetric spaces we refer the reader to @xcite and @xcite which also constitute the main reference of the material presented in this section . we shall also follow ( unless otherwise mentioned ) the notational convention of @xcite : in particular , @xmath4 is a set ( manifold ) , the letter @xmath5 is reserved for groups and lie groups , gothic letters denote lie algebras and lie triple systems , latin lowercase letters denote lie - group elements and latin uppercase letters denote lie - algebra elements . the identity element of a group will usually be denoted by @xmath6 and the identity mapping by @xmath7 . [ def:2.1 ] a _ symmetric space _ is a manifold @xmath4 with a differentiable _ symmetric product _ @xmath8 obeying the following conditions : * @xmath9 * @xmath10 * @xmath11 and moreover * every @xmath12 has a neighbourhood @xmath13 such that for all @xmath14 in @xmath13 @xmath15 implies @xmath16 . the latter condition is relevant in the case of manifolds @xmath4 with open set topology ( as in the case of lie groups ) and can be disregarded for sets @xmath4 with discrete topology : a discrete set @xmath4 endowed with a multiplication obeying ( i)(iii ) will be also called a symmetric space . a _ pointed symmetric space _ is a pair @xmath17 consisting of a symmetric space @xmath4 and a point @xmath18 called _ base point . _ note that when @xmath4 is a lie group , it is usual to set @xmath19 . moreover , if the group is a _ matrix _ group with the usual matrix multiplication , it is usual to set @xmath20 ( identity matrix ) . the left multiplication with an element @xmath21 is denoted by @xmath22 , @xmath23 and is called _ symmetry around _ @xmath12 . note that @xmath24 because of ( i ) , hence @xmath12 is fixed point of @xmath22 and it is isolated because of ( iv ) . furthermore , ( ii ) and ( iii ) imply @xmath22 is an involutive automorphism of @xmath4 , i.e. @xmath25 . symmetric spaces can be constructed in several different ways , the following are important examples : 1 . [ item:1 ] manifolds with an intrinsically defined symmetric product . as an example , consider the @xmath26-sphere as the set of unit vectors in @xmath27 . the product @xmath28 turns this into a symmetric space . the above operation is the reflection of points on a sphere . this can be generalized to @xmath29-dimensional subspaces in @xmath30 ( @xmath31 ) in the following manner : assume that @xmath32 $ ] is a full rank matrix . define @xmath33 the orthogonal projection operator onto the range of @xmath12 , @xmath34 . consider the reflection @xmath35 . define @xmath36 . this operation obeys the conditions ( i)(vi ) whenever @xmath37 are @xmath38 full rank matrices . in particular , note that ( i ) is equivalent to @xmath39 , i.e. the reflection is an involutive matrix . [ item:2 ] subsets of a continuous ( or discrete ) group @xmath5 that are closed under the composition @xmath40 , where @xmath41 is the usual multiplication in @xmath5 . groups themselves , continuous , as in the case of lie groups , or discrete , are thus particular instances of symmetric spaces . as another example , consider the set of all symmetric positive definite matrices as a subset of all nonsingular matrices , which forms a symmetric space with the product @xmath42 3 . [ item:3 ] symmetric elements of automorphisms on a group . an automorphism on a group @xmath5 is a map @xmath43 satisfying @xmath44 . symmetric elements _ are defined as @xmath45 it is easily verified that @xmath46 obeys ( i)(iv ) when endowed with the multiplication @xmath40 , hence it is a symmetric space . as an example , symmetric matrices are symmetric elements under the matrix automorphism @xmath47 . [ item:4 ] homogeneous manifolds . given a lie group @xmath5 and a subgroup @xmath48 , a homogeneous manifold @xmath49 is the set of all left cosets of @xmath48 in @xmath5 . not every homogeneous manifold possesses a product turning it into a symmetric space , however , we will see in theorem [ th:2.1 ] that any connected symmetric space arises in a natural manner as a homogeneous manifold . [ item:5 ] jordan algebras . let @xmath50 be a finite - dimensional vector space with a bilinear multiplication @xmath51 such that @xmath52 ( powers defined in the usual way , @xmath53 ) , with unit element @xmath6 . define @xmath54 and set @xmath55 . then , the set of _ invertible _ elements of @xmath56 is a symmetric space with the product @xmath57 in the context of symmetric matrices , take @xmath58 , where @xmath41 denotes the usual matrix multiplication . after some algebraic manipulations , one can verify that the product @xmath59 as in example [ item:2 ] . let @xmath5 be a connected lie group and let @xmath60 be an analytic involutive automorphism , i.e. @xmath61 and @xmath62 . let @xmath63 denote @xmath64 , @xmath65 its connected component including the base point , in this case the identity element @xmath6 and finally let @xmath66 be a closed subgroup such that @xmath67 . set @xmath68 . [ th:2.1 ] the homogeneous space @xmath69 is a symmetric space with the product @xmath70 and @xmath71 is a symmetric space with the product @xmath72 . the space of symmetric elements @xmath71 is isomorphic to the homogeneous space @xmath73 . moreover , every connected symmetric space is of the type @xmath74 and also of the type @xmath71 . the interesting consequence of the above theorem is that _ every connected symmetric space _ is also a homogeneous space , which implies a factorization : as coset representatives for @xmath73 one may choose elements of @xmath71 , thus any @xmath75 can be decomposed in a product @xmath76 , where @xmath77 . in other words , @xmath78 the matrix polar decomposition is a particular example , discussed in [ sec:3.1 ] . the automorphism @xmath60 on @xmath5 induces an automorphism on the lie algebra @xmath79 and also a canonical decomposition of @xmath79 . let @xmath80 and @xmath81 denote the lie algebras of @xmath5 and @xmath66 respectively and denote by @xmath82 the differential of @xmath60 at @xmath6 , @xmath83 note that @xmath82 is an involutive automorphism of @xmath79 and has eigenvalues @xmath84 . moreover , @xmath85 implies @xmath86 . set @xmath87 . then , @xmath88 it is easily verified that @xmath89 \subset \gg{k } , \quad { [ \gg{k } , \gg{p } ] } \subset \gg{p } , \quad { [ \gg{p } , \gg{p } ] } \subset \gg{k},\ ] ] that is , @xmath81 is a subalgebra of @xmath79 while @xmath90 is an ideal in @xmath81 . given @xmath91 , its canonical decomposition @xmath92 is @xmath93 , with @xmath94 and @xmath95 , @xmath96 we have already observed that there is close connection between projection matrices , reflections ( involutive matrices ) and hence symmetric spaces . in a linear algebra context , this statement can be formalized as follows . recall that a matrix @xmath97 is a _ projection _ if @xmath98 . [ th : proj ] to any projection matrix @xmath97 there corresponds an involutive matrix @xmath99 . conversely , to any involutive matrix @xmath100 there correspond two projection matrices @xmath101 and @xmath102 . these projections satisfy @xmath103 and @xmath104 , moreover @xmath105 , i.e. the projection @xmath106 projects onto the @xmath84 eigenspace of @xmath100 . note that if @xmath100 is involutive , so is @xmath107 , which corresponds to the opposite identification of the @xmath84 eigenspaces . a matrix @xmath66 , whose columns are in the span of the @xmath108 eigenspace is said to be _ block - diagonal _ with respect to the automorphism , while a matrix @xmath109 , whose columns are in the span of the @xmath110 eigenspace , is said to be _ 2-cyclic_. in the context of lemma [ th : proj ] , we recognize that @xmath94 is the 2-cyclic part and @xmath111 is the block - diagonal part . namely , if @xmath112 is represented by a matrix , then @xmath113 where @xmath114 restricted to the appropriate subspaces . then , @xmath115 and @xmath116 corresponds to the block - diagonal part @xmath66 and @xmath117 corresponding to the 2-cyclic part @xmath109 . in passing , we mention that the decomposition ( [ eq:2.3 ] ) is called a _ cartan decomposition _ whenever the cartan killing form @xmath118 is nondegenerate , hence it can be used to introduce a positive bilinear form @xmath119 . if this is the case , the linear subspaces @xmath120 are orthogonal . the involutive automorphism @xmath60 need not be defined at the group level @xmath5 and thereafter lifted to the algebra by ( [ eq:2.1 ] ) . it is possible to proceed the other way around : an involutive algebra automorphism @xmath82 on @xmath79 , which automatically produces a decomposition ( [ eq:2.2])-([eq:2.3 ] ) , can be used to induce a group automorphism @xmath60 by the relation @xmath121 and a corresponding group factorization ( [ eq : polardecomp ] ) . thus , we have an `` upstairs - downstairs '' viewpoint : the group involutive automorphisms generate corresponding algebra automorphisms and vice versa . this `` upstairs - downstairs '' view is useful : in some cases , the group factorization ( [ eq : polardecomp ] ) is difficult to compute starting from @xmath12 and @xmath60 , while the splitting at the algebra level might be easy to compute from @xmath112 and @xmath82 . in other cases , it might be the other way around . in lie group theory lie algebras are important since they describe infinitesimally the structure of the tangent space at the identity . similarly , lie triple systems give the structure of the tangent space of a symmetric space . [ sec:2.2 ] ( @xcite ) [ def:2.2 ] a vector space with a trilinear composition @xmath122 $ ] is called a _ lie triple system _ ( lts ) if the following identities are satisfied : * @xmath123=0 $ ] , * @xmath122+[y , z , x ] + [ z , x , y ] = 0 $ ] , * @xmath124 = [ [ x , y , u ] , v , w ] + [ u , [ x , y , v],w ] + [ u , v,[x , y , w]]$ ] . a typical way to construct a lts is by means of an involutive automorphism of a lie algebra @xmath79 . with the same notation as above , the set @xmath90 is a lts with the composition @xmath125 = [ [ x , y ] , z].\ ] ] vice versa , for every lts there exists a lie algebra @xmath126 and an involutive automorphism @xmath60 such that the given lts corresponds to @xmath90 . the algebra @xmath126 is called _ standard embedding _ of the lts . in general , any subset of @xmath79 that is closed under the operator @xmath127 \label{eq : t_x}\ ] ] is a lie triple system . it can be shown that being closed under @xmath128 guarantees being closed under the triple commutator . let @xmath129 be the group of @xmath130 invertible matrices . consider the map @xmath131 it is clear that @xmath60 is an involutive automorphism of @xmath129 . then , according to theorem [ th:2.1 ] , the set of symmetric elements @xmath132 is a symmetric space . we observe that @xmath71 is the set of invertible symmetric matrices . the symmetric space @xmath71 is disconnected and particular mention deserves its connected component containing the identity matrix @xmath133 , since it reduces to the set of symmetric positive definite matrices . the subgroup @xmath63 consists of all orthogonal matrices . the decomposition ( [ eq : polardecomp ] ) is the classical polar decomposition , any nonsingular matrix can be written as a product of a symmetric matrix and an orthogonal matrix . if we restrict the symmetric matrix to the symmetric positive definite ( spd ) matrices , then the decomposition is unique . in standard notation , @xmath134 is denoted by @xmath135 ( spd matrix ) , while @xmath136 is denoted by @xmath137 ( orthogonal matrix ) . at the algebra level , the corresponding splitting is @xmath138 , where @xmath139 is the classical algebra of skew - symmetric matrices , while @xmath140 is the classical set of symmetric matrices . the latter is not a subalgebra of @xmath141 but is closed under @xmath128 , hence is a lie triple system . the decomposition ( [ eq : algebrasplitting ] ) is nothing else than the canonical decomposition of a matrix into its skew - symmetric and symmetric part , @xmath142 it is well known that the polar decomposition @xmath143 can be characterized in terms of best approximation properties . the orthogonal part @xmath137 is the best orthogonal approximation of @xmath12 in any orthogonally invariant norm ( e.g. 2-norm and frobenius norm ) . other classical polar decompositions @xmath144 , with @xmath135 hermitian and @xmath137 unitary , or with @xmath135 real and @xmath137 coinvolutory ( i.e. @xmath145 ) , can also be fitted in this framework @xcite with the choice of automorphisms @xmath146 ( hermitian adjoint ) , and @xmath147 respectively ( where @xmath148 denotes the complex conjugate of @xmath12 ) . the group decomposition @xmath144 can also be studied via the algebra decomposition . in @xcite such decompositions are generalized to arbitrary involutive automorphisms , and best approximation properties are established for the general case . clearly , also other types of automorphisms can be considered , generalizing the group factorization ( [ eq : polardecomp ] ) and the algebra splitting ( [ eq : algebrasplitting ] ) . for instance , there is a large source of involutive automorphisms in the set of involutive _ inner automorphisms _ @xmath154 that can be applied to subgroups of @xmath155 to obtain a number of interesting factorizations . the matrix @xmath156 has to be involutive , @xmath157 , but it need not be in the group : the factorization makes sense as long as @xmath158 is in the group ( resp . @xmath159 is in the algebra ) . as an example , let @xmath160 be the group of orthogonal matrices , and let @xmath161 = i - 2 e_1 e_1^t$ ] , where @xmath162 denotes the @xmath163th canonical unit vector in @xmath30 . obviously , @xmath164 , as @xmath165 ; nevertheless , we have @xmath166 , as long as @xmath167 , since @xmath168 and @xmath169 , thus @xmath170 . it is straightforward to verify that the subgroup @xmath63 of theorem [ th:2.1 ] consists of all orthogonal @xmath171 matrices of the form @xmath172 where @xmath173 . thus the corresponding symmetric space is @xmath174 . matrices belong to the same coset if their first column coincide , thus the symmetric space can be identified with the @xmath175-sphere @xmath176 . the corresponding splitting of a skew - symmetric matrix @xmath177 is @xmath178 thus any orthogonal matrix can be expressed as the product of the exponential of a matrix in @xmath90 and one in @xmath81 . the space @xmath90 can be identified with the tangent space to the sphere in the point @xmath179 . different choices of @xmath156 give different interesting algebra splittings and corresponding group factorizations . for instance , by choosing @xmath156 to be the anti - identity , @xmath180 $ ] , one obtains an algebra splitting in _ persymmetric _ and _ perskew - symmetric _ matrices . the choice @xmath181 $ ] for symplectic matrices , gives the splitting in lower - dimensional symplectic matrices forming a sub - algebra and a lie - triple system , and so on . in @xcite , such splittings are used for the efficient approximation of the exponential of skew - symmetric , symplectic and zero - trace matrices . in @xcite similar ideas are used to construct computationally effective numerical integrators for differential equations on stiefel and grassman manifolds . a similar framework can be used to obtain coordinates on lie groups @xcite , which are of interest when solving differential equations of the type @xmath182 by reducing the problem recursively to spaces of smaller dimension . recall that @xmath183 where @xmath184 obeys the differential equation @xmath185 , where @xmath186 , @xmath187 being the @xmath188th bernoulli number , see @xcite . decomposing @xmath189 , where @xmath190 , the solution can be factorized as @xmath191 where @xmath192 + \sum_{j=1}^\infty 2^{2j } c_{2j } \mathrm{ad}^{2j}_p \pi^{-}x , \quad c_{2j}=\frac{b_{2j}}{(2j ) ! } , \label{eq : pdot } \\ \dot k & = & \mathrm{dexp}^{-1}_k ( \pi^{+}x - 2 \sum_{j=1}^\infty ( 2^{2j}-1 ) c_{2j } \mathrm{ad}^{2j-1}_p \pi^{-}x ) . \label{eq : kdot}\end{aligned}\ ] ] note that ( [ eq : kdot ] ) depends on @xmath109 , however , it is possible to formulate it solely in terms of @xmath66 , @xmath193 and @xmath194 , but the expression becomes less neat . in block form : @xmath195 where @xmath196 . the above formula paves the road for a recursive decomposition , by recognizing that @xmath197 is the @xmath198 function on the restricted sub - algebra @xmath199 . by introducing a sequence of involutive automorphisms @xmath200 , one induces a sequence of subalgebras , @xmath201 , of decreasing dimention , @xmath202 . note also that the functions appearing in the above formulation are all analytic functions of the @xmath203-operator , and are either odd or even functions , therefore they can be expressed as functions of @xmath204 on @xmath205 . in particular , this means that , as long as we can compute analytic functions of the @xmath204 operator , the above decomposition is computable . thus , the problem is reduced to the computation of analytic functions of the 2-cyclic part @xmath109 as well as analytic functions of @xmath206 ( trivialized tangent maps and their inverse ) . the following theorem addresses the computation of such functions using the same framework of lemma [ th : proj ] . let @xmath109 be the 2-cyclic part of @xmath207 with respect to the involution @xmath100 , i.e. @xmath208 . let @xmath209 . for any analytic function @xmath210 , we have @xmath211 where @xmath212 and @xmath213 . a similar result holds for @xmath206 , see @xcite . it is interesting to remark that if @xmath214 then @xmath215 where @xmath216 as above . in particular , the problem is reduced to computing analytic functions of the principal square root of a matrix @xcite : numerical methods that compute these quantities accurately and efficiently are very important for competitive numerical algorithms . typically , @xmath217 is a low - rank matrix , hence computations can be done using eigenvalues and eigenfunctions of the @xmath203 operator restricted to the appropriate space , see @xcite . in particular , if @xmath218 are vectors , then @xmath219 is a scalar and the formulas become particularly simple . these coordinates have interesting applications in control theory . some early use of these generalized cartan decompositions ( [ eq:2.2])([eq:2.3 ] ) ( which the author calls @xmath220-_grading _ ) to problems with nonholonomic constraints can be found in @xcite . in @xcite , the authors embrace the formalism of symmetric spaces and use orthogonal ( cartan ) decompositions with applications to nmr spectroscopy and quantum computing , using adjoint orbits as main tool . generally , these decompositions can be found in the literature , but have been applied mostly to the cases when @xmath221=\{0\ } $ ] or @xmath222=\{0\ } $ ] , or both , as ( [ eq : pdot])-([eq : kdot ] ) become very simple and the infinite sums reduce to one or two terms . the main contribution of @xcite is the derivation of such differential equations _ and _ the evidence that such equations can be solved efficiently using linear algebra tools . see also @xcite for some applications to control theory . let @xmath223 be the group of diffeomorphisms of a manifold @xmath4 onto itself . we say that a map @xmath224 has a symmetry @xmath225 if @xmath226 ( the multiplication indicating the usual composition of maps , i.e. @xmath227 ) , while if @xmath228 we say that @xmath229 is a reversing symmetry of @xmath230 @xcite . without further ado , we restrict ourselves to involutory symmetries , the main subject of this paper . symmetries and reversing symmetries are very important in the context of dynamical systems and their numerical integration . for instance , nongeneric bifurcations can become generic in the presence of symmetries and vice versa . thus , when using the integration time - step as a bifurcation parameter , it is vitally important to remain within the smallest possible class of systems . reversing symmetries , on the other hand , give rise to the existence of invariant tori and invariant cylinders @xcite . it is a classical result that the set of symmetries possess the structure of a group they behave like automorphisms and fixed sets of automorphisms . the group structure , however , does not extend to reversing symmetries and fixed points of anti - automorphisms , and in the last few years the set of reversing symmetries has received the attention of numerous numerical analysts . in @xcite it was observed that the set of fixed points of an involutive anti - automorphism @xmath231 was closed under the operation @xmath232 that mclachlan et al . called `` sandwich product . '' pseudogroup_. ] indeed , our initial goal was to understand such structures and investigate how they could be used to devise new numerical integrators for differential equations with some special geometric properties . we recognise , cfr . [ sec:2.1 ] , that the set of fixed points of an anti - automorphism is a symmetric space . conversely , any connected space of invertible elements closed under the `` sandwich product '' , is the set of the fixed points of an involutive automorphism ( cfr . theorem [ th:2.1 ] ) and has associated to it a lts . to show this , consider the well known symmetric bch formula , @xmath233 @xcite , which is used extensively in the context of _ splitting methods _ @xcite . because of the sandwich - type composition ( symmetric space structure ) , the corresponding @xmath234 must be in the lts space , and this explains why it can be written as powers of the double commutator operators @xmath235 applied to @xmath112 and @xmath236 . a natural question to ask is : what is the automorphism @xmath60 having such sandwich - type composition as anti - fixed points ? as @xmath237 , we see that , by writing @xmath238 , we have @xmath239 . in the context of numerical integrators , the automorphism @xmath60 consists in changing the time @xmath240 to @xmath241 . this will be proved in [ sec:3.4 ] . if @xmath4 is a finite dimensional smooth compact manifold , it is well known that the infinite dimensional group of @xmath223 of all smooth diffeomorphisms @xmath242 is a lie group , with lie algebra @xmath243 of all smooth vector fields on @xmath4 , with the usual bracket and exponential map . it should be noted , however , that the exponential map is not a one - to - one map , not even in the neighbourhood the identity element , since there exist diffeomorphisms arbitrary close to the identity which are not on any one - parameter subgroup and others which are on many . however , the regions where the exponential map is not surjective become smaller and smaller the closer we approach the identity @xcite , and , for our purpose , we can disregard these regions and assume that our results are formally true . there are two different settings that we can consider in this context . the first is to analyze the set of differentiable maps that possess a certain symmetry ( or a discrete set of symmetries ) . the second is to consider the structure of the set of symmetries of a fixed diffeomorphism . the first has a continuous - type structure while the second is more often a discrete type symmetric space . [ th:3.4 ] the set of diffeomorphisms @xmath230 that possess @xmath229 as an ( involutive ) reversing symmetry is a symmetric space of the type @xmath71 . denote @xmath244 it is clear that @xmath60 acts as an automorphism , @xmath245 moreover , if @xmath229 is an involution then so is also @xmath60 . note that the set of diffeomorphisms @xmath230 that possess @xmath229 as a reversing symmetry is the space of symmetric elements @xmath246 defined by the automorphism @xmath60 ( cfr . [ sec:2 ] ) . hence the result follows from theorem [ th:2.1 ] . [ th:3.4b ] the set of reversing symmetries acting on a diffeomorphism @xmath230 is a symmetric space with the composition @xmath247 . if @xmath248 is a symmetry of @xmath230 then so is also @xmath249 , since @xmath250 and the assertion follows by taking the inverse on both sides of the equality . in particular , if @xmath248 is a symmetry of @xmath230 it is also true that @xmath251 is a reversing symmetry of @xmath252 . next , we observe that if @xmath248 and @xmath253 are two reversing symmetries of @xmath230 then so is also @xmath254 , since @xmath255 it follows that the composition @xmath256 is an internal operation on the set of reversing symmetries of a diffeomorphism @xmath230 . with the above multiplication , the conditions i)iii ) of definition [ def:2.1 ] are easily verified . this proves the assertion in the case when @xmath257 has a discrete set of reversing symmetries . in what follows , we assume that @xmath258 is differentiable and involutory ( @xmath259 ) and @xmath260 . acting on @xmath261 , we have @xmath262 where @xmath263 is the tangent map of @xmath264 . the pullback is natural with respect to the jacobi bracket , @xmath265 = \mathcal{t } _ * [ x , y ] \mathcal{t},\ ] ] for all vector fields @xmath266 . hence the map @xmath82 is an involutory algebra automorphism . let @xmath267 and @xmath90 be the eigenspaces of @xmath82 in @xmath268 . then @xmath269 where @xmath270 is the lie algebra of vector fields that have @xmath264 as a symmetry and @xmath271 is the lie triple system , vector fields corresponding to maps that have @xmath264 as a reversing symmetry . thus , as is the case for matrices , every vector field @xmath112 can be split into two parts , @xmath272 having @xmath264 as a symmetry and reversing symmetry respectively . in the context of ordinary differential equation , let us consider @xmath273 given an arbitrary involutive function @xmath264 , the vector field @xmath274 can always be canonically split into two components , having @xmath275 as a symmetry and reversing symmetry respectively . however , if one of these components equals zero , then the system ( [ eq:3.3 ] ) has @xmath264 as a symmetry or a reversing symmetry . let us consider the ode ( [ eq:3.3 ] ) , whose exact flow will be denoted as @xmath276 . backward error analysis for odes implies that a ( consistent ) numerical method for the integration of ( [ eq:3.3 ] ) can be interpreted as the sampling at @xmath277 of the flow @xmath278 of a vector field @xmath279 ( the so called _ modified vector field _ ) which is close to @xmath274 , @xmath280 where @xmath134 is the order of the method ( note that setting @xmath277 , the local truncation error is of order @xmath281 ) . consider next the map @xmath60 on the set of flows depending on the parameter @xmath282 defined as @xmath283 where @xmath284 , with @xmath285 . the map @xmath60 is involutive , since @xmath286 , and it is easily verified by means of the bch formula that @xmath287 , hence @xmath60 is an automorphism . consider next @xmath288 then @xmath289 if and only if @xmath290 , namely the method @xmath291 is _ selfadjoint_. the set of one - parameter , consistent , selfadjoint numerical schemes is a symmetric space in the sense of theorem [ th:2.1 ] , generated by @xmath60 as in ( [ eq:3.4 ] ) . next , we perform the decomposition ( [ eq:2.2 ] ) . we deduce from ( [ eq:3.4 ] ) that @xmath292 hence , @xmath293 is the subalgebra of vector fields that are odd in @xmath282 , and @xmath294 is the lts of vector fields that possess only even powers of @xmath282 . thus , if @xmath279 is the modified vector field of a numerical integrator @xmath291 , its canonical decomposition in @xmath295 is @xmath296 the first term containing only odd powers of @xmath282 and the second only even powers . then , if the numerical method @xmath297 is selfadjoint , it contains only odd powers of @xmath282 locally ( in perfect agreement with classical results on selfadjoint methods @xcite ) . in @xcite it has been shown that it is possible to generalize the polar decomposition of matrices to lie groups endowed with an involutive automorphism : every lie group element @xmath298 sufficiently close to the identity can be decomposed as @xmath299 where @xmath300 , the space of symmetric elements of @xmath60 , and @xmath301 , the subgroup of @xmath5 of elements fixed under @xmath60 . furthermore , setting @xmath302 and @xmath303 , one has that @xmath304 is an odd function of @xmath240 and it is a best approximant to @xmath298 in @xmath63 in @xmath63 right - invariant norms constructed by means of the cartan killing form , provided that @xmath5 is semisimple and that the decomposition @xmath305 is a cartan decomposition . assume that @xmath230 , the exact flow of the differential equation ( [ eq:3.3 ] ) , has @xmath229 as a reversing symmetry ( i.e. @xmath306 , where @xmath307 ) , while its approximation @xmath308 has not . we perform the polar decomposition @xmath309 i.e. @xmath310 has @xmath311 as a reversing symmetry , while @xmath312 has @xmath229 as a symmetry . since the original flow has @xmath311 as a reversing symmetry ( and not symmetry ) , @xmath313 is the factor that we wish to eliminate . we have @xmath314 . hence the method obtained composing @xmath308 with @xmath315 has the reversing symmetry @xmath229 every other step . to obtain @xmath310 we need to extract the square root of the flow @xmath316 . now , if @xmath317 is a flow , then its square root is simply @xmath318 . however , if @xmath319 is the flow of a consistent numerical method ( @xmath320 ) , namely the numerical integrator corresponds to @xmath321 , it is not possible to evaluate the square root @xmath322 by simple means as it is not the same as the numerical method with half the stepsize , @xmath323 . the latter , however , offers an approximation to the square root : note that @xmath324 an expansion which , compared with @xmath321 , reveals that the error in approximating the square root with the numerical method with half the stepsize is of the order of @xmath325 a term that is subsumed in the local truncation error . the choice @xmath326 as an approximation to @xmath310 ( we stress that each flow is now evaluated at @xmath327 ) , yields a map that has the reversing symmetry @xmath275 at each step by design , since @xmath328 note that @xmath329 , where @xmath330 is the inverse ( or adjoint ) method of @xmath331 . if @xmath60 is given by ( [ eq:3.4 ] ) , then @xmath332 and this algorithm is precisely the _ scovel projection _ @xcite originally proposed to to generate selfadjoint numerical schemes from an arbitrary integrator , and then generalized to the context of reversing symmetries @xcite . [ th:3.5 ] the generalized scovel projection is equivalent to choosing the @xmath246-factor in the polar decomposition of a flow @xmath291 under the involutive automorphism @xmath333 , whereby square roots of flows are approximated by numerical methods with half the stepsize . another algorithm that can be related to the generalized polar decomposition of flows is the application of the thue morse sequence to improve the preservation of symmetries by means of a numerical integrator @xcite . given an involutive automorphism @xmath334 and a numerical method @xmath291 in a group @xmath5 of numerical integrators , iserles et al . @xcite construct the sequence of methods @xmath335 } : = \varphi_h , \quad \varphi^{[k+1 ] } : = \varphi^{[k ] } \sigma ( \varphi^{[k ] } ) , \qquad k=0,1 , 2,\ldots . \label{eq : thue}\ ] ] since @xmath336 } = \sigma^0 \varphi^{[k]}$ ] , it is easily observed that the @xmath136-th method corresponds to composing @xmath337}$ ] and @xmath338}$ ] according to the @xmath136-th thue morse sequence @xmath339 , as displayed below in table [ tab:3.1 ] ( see @xcite ) . .thue morse iterations for the method @xmath336}$ ] . [ cols="^,<,<",options="header " , ] iserles et al . ( @xcite ) showed , by a recursive use of the bch formula , that each iteration improves by one order the preservation of the symmetry @xmath340 by a consistent numerical method , where @xmath340 is the involutive automorphism such that @xmath341 . the main argument of the proof is that if the method @xmath342}$ ] has a given symmetry error , the symmetry error of @xmath343})$ ] has the opposite sign . hence the two leading symmetry errors cancel and @xmath344}$ ] has a symmetry error of one order higher . in other words , if the method @xmath291 preserves @xmath345 to order @xmath134 , then @xmath336}$ ] preserves the symmetry @xmath345 to order @xmath346 every @xmath347 steps . as @xmath60 changes the sign of the symmetry error only , if a method @xmath291 has a given approximation order @xmath134 , so does @xmath348 . thus @xmath349 } = \sigma ( \varphi_h)$ ] can be used as initial condition in ( [ eq : thue ] ) , obtaining the conjugate thue morse sequence @xmath350 . by a similar argument , also @xmath351 } : = \varphi_h \hbox { or } \chi^{[0 ] } : = \sigma(\varphi_h ) , \quad \chi^{[k+1 ] } : = \sigma ( \chi^{[k ] } ) \chi^{[k ] } , \qquad k=0,1 , 2,\ldots , \label{eq : thue_compl}\ ] ] generate sequences with increasing order of preservation of symmetry . as an illustration of the technique , consider the spatial pde @xmath352 , where @xmath353 is analytic in @xmath354 . let the pde be defined over the domain @xmath355\times[-l , l]$ ] with periodic boundary conditions and initial value @xmath356 . if @xmath356 is symmetric on the domain , i.e. @xmath357 , so is the solution for all @xmath240 . the symmetry is @xmath358 . now , assume that we solve the equation by the method of alternating directions , where the method @xmath230 corresponds to solving with respect to the @xmath12 variable keeping @xmath14 fixed , while @xmath359 with respect to the @xmath14 variable keeping @xmath12 fixed . the symmetry will typically be broken at the first step . nevertheless , we can get a much more symmetric solution if the sequence of the directions obeys the thue morse sequence ( iteration @xmath336}$ ] ) or the equivalent sequence given by iteration @xmath360}$ ] . this example is illustrated in figures [ fig : alt_dir]-[fig : alt_dir2 ] . for the method of alternating directions for the pde @xmath361 , on a square , with initial condition @xmath362 and periodic boundary conditions . the integration is performed using the first order method @xmath363 ( forward euler ) , where @xmath364 is a circulant divided difference discretization matrix of the differential operators @xmath365 ( second order central differences ) . the experiments are performed with constant stepsize @xmath366 . the order of the overall approximation is the same as the original method ( first order only ) , the only difference is the order of the directions , chosen according to the thue morse sequence . the different symbols correspond to different sampling rates : every step , every second step , fourth , eight , . from top left to bottom right : sequences 0 , 01 , 0110 , 01101001 , etc . ] versus step size @xmath282 for the method of alternating directions for the pde @xmath367 , as above . the basic method @xmath230 ( method 0 " ) is now a symmetric composition with the heun method : half step in the @xmath12 direction , full step in the @xmath14 direction , half step in the @xmath12 direction , all of them performed with the heun method ( improved euler ) . it is immediately observed that the symmetry error is improved by choosing the directions according to the thue morse sequence . the order of the method ( global error ) remains unchanged ( order two ) . ] in a famous paper that appeared in 1990 ( @xcite ) yoshida showed how to construct high order time - symmetric integrators starting from lower order time - symmetric symplectic ones . yoshida showed that , if @xmath230 is a selfadjoint numerical integrator of order @xmath368 , then @xmath369 is a selfadjoint numerical method of order @xmath370 provided that the coefficients @xmath371 and @xmath372 satisfy the condition @xmath373 whose only real solution is @xmath374 in the formalism of this paper , time - symmetric methods correspond to @xmath71-type elements with @xmath60 as in ( [ eq:3.4 ] ) and it is clearly seen that the yoshida technique can be used in general to improve the order of approximation of @xmath246-type elements . a similar procedure can be applied to improve the order of the retention of symmetries and not just reversing symmetries . to be more specific , let @xmath375 be a symmetry of the given differential equation , namely @xmath376 , with @xmath377 , @xmath378 , @xmath379 denoting the pullback of @xmath340 to @xmath380 ( see [ sec:3.3 ] ) . here , the involutive automorphism is given by @xmath381 so that @xmath382 [ prop : yoshida_self ] assume that @xmath383 is the flow of a self - adjoint numerical method of order @xmath368 , @xmath384 , @xmath385 , where @xmath386 , @xmath387 and @xmath274 has @xmath388 as a symmetry . the composition @xmath389}_{h}(t ) = \varphi_{ah}(at ) \sigma(\varphi_{bh}(bt ) ) \varphi_{ah}(at ) , \label{eq : yoshida.2}\ ] ] with @xmath390 has symmetry error @xmath370 at @xmath277 . write ( [ eq : yoshida.2 ] ) as @xmath389}_{h}(t ) = \exp(at f_{ah } ) \exp(bt \d\sigma(f_{bh } ) ) \exp(atf_{ah}).\ ] ] application of the symmetric bch formula , together with the fact that @xmath82 acts by changing the signs on the @xmath90-components only , allows us to write the relation ( [ eq : yoshida.2 ] ) as @xmath391}_{h}(t ) & = & \exp((2a+b)t f + ( 2(at)(ah)^{2p } + ( bt ) ( bh)^{2p } ) k_{2p } \label{eq : yoshida.3 } \\ & & \qquad \mbox { } + ( 2at(ah)^{2p}-bt(bh)^{2p})p_{2p } + \oo{th^{2p+2 } } + \oo{t^{3}h^{2p } } + \cdots ) , \nonumber\end{aligned}\ ] ] where the @xmath392 comes from the @xmath393 term and the @xmath394 from the commutation of the @xmath274 and @xmath395 terms ( recall that no first order commutator appears in the symmetric bch formula ) . the numerical method is obtained letting @xmath277 . we require @xmath396 for consistency , and @xmath397 to annihilate the coefficient of @xmath398 , the lowest order @xmath90-term . the resulting method @xmath399}_{h}(t)$ ] retains the symmetry @xmath340 to order @xmath370 , as the first leading symmetry error is a @xmath400 term . this procedure allows us to gain two extra degrees in the retention of symmetry per iteration , provided that the underlying method is selfadjoint , compared with the thue morse sequence of @xcite that yields one extra degree in symmetry per iteration but does not require selfadjointness . as for the yoshida technique , the composition ( [ eq : yoshida.2 ] ) can be iterated @xmath136 times to obtain a time - symmetric method of order @xmath368 that retains symmetry to order @xmath401 . the disadvantage of ( [ eq : yoshida.2 ] ) , with respect to the classical yoshida approach is the fact that the order of the method is retained ( and does not increase by 2 units as the symmetry error ) . the main advantage is that all the steps are positive , in particular the second step , @xmath402 , whereas the @xmath372 is always negative for @xmath403 and typically larger than @xmath371 , requiring step - size restrictions for stiff methods . in the limit , when @xmath404 , @xmath405 , i.e. the step lengths become equal . thus , the proposed technique for improving symmetry is of particular interest in the context of stiff problems . this is illustrated by the following example and figure [ fig : stiff ] . [ ex:1 ] consider the pde @xmath406 , defined on the square @xmath407\times [ -1,1]$ ] , with a gaussian initial condition , @xmath408 , and periodic boundary conditions . the problem is semi - discretized on a uniform and isotropic mesh with spacing @xmath409 and is reduced to the set of odes @xmath410 , where @xmath411 and @xmath412 , @xmath413 being the circulant matrix of the standard second order divided differences , with stencil @xmath414 $ ] . for the time integration , we consider the splitting @xmath415 , where @xmath416 and @xmath417 and the second order self - adjoint method : @xmath418 we display the global error and the symmetry error for time - integration step sizes @xmath419 $ ] , where the parameter @xmath420 is chosen to be the largest step size for which the basic method @xmath230 in ( [ eq : pde ] ) is stable . the factor @xmath421 comes from the fact that both the yoshida and our symmetrization method ( [ eq : yoshida.2 ] ) require 3 sub - steps of the basic method . so , one step of the yoshida and our symmetrising composition can be expected to cost the same as the basic method ( [ eq : pde ] ) . as we can see in figure [ fig : stiff ] , the yoshida technique , with @xmath422 and @xmath423 , does the job of increasing the order of accuracy of the method from two to four , and so does the symmetry error . however , since @xmath424 , the @xmath372-step is negative , and , as a consequence , it is observed that the method fails to converge for the two largest values of the step size . conversely , our symmetrization method ( [ eq : yoshida.2 ] ) has @xmath425 and @xmath426 , with @xmath402 positive , and the method converges for all the time - integration steps . as expected , the order is not improved , but the symmetry order is improved by two units . the symmetry @xmath60 is applied by transposing the matrix representation @xmath427 of the solution at @xmath428 , before and after the intermediate @xmath402-step . otherwise , the two implementations are identical . ) ( right ) for the basic self - adjoint method ( [ eq : pde ] ) applied to a semidiscretization of @xmath406 . the order with the yoshida techique increases by two units , but the method does not converge for the two largest values of the step size . our symmetrization technique improves by two units the order of retention of symmetry ( but not global error ) . due to the positive step - sizes , it converges also for the two largest values of the step size . see text for details.,title="fig : " ] ) ( right ) for the basic self - adjoint method ( [ eq : pde ] ) applied to a semidiscretization of @xmath406 . the order with the yoshida techique increases by two units , but the method does not converge for the two largest values of the step size . our symmetrization technique improves by two units the order of retention of symmetry ( but not global error ) . due to the positive step - sizes , it converges also for the two largest values of the step size . see text for details.,title="fig : " ] in this paper we have shown that the algebraic structure of lie triple systems and the factorization properties of symmetric spaces can be used as a tool to : 1 ) understand and provide a unifying approach to the analysis of a number of different algorithms ; and 2 ) devise new algorithms with special symmetry / reversing symmetry properties in the context of the numerical solution of differential equations . in particular , we have seen that symmetries are more difficult to retain ( to date , we are not aware of methods that can retain a generic involutive symmetry in a finite number of steps ) , while the situation is simpler for reversing symmetries , which can be achieved in a finite number of steps using the scovel composition . so far , we have considered the most generic setting where the only allowed operations are non - commutative compositions of maps ( the map @xmath230 , its transformed , @xmath359 , and their inverses ) . if the underlying space is linear and so is the symmetry , i.e. @xmath429 , the map @xmath430 obviously satisfies the symmetry @xmath60 , as @xmath431 . because of the linearity and the vector - space property , we can use the same operation as in the tangent space , namely we identify @xmath60 and @xmath432 . this is in fact the most common symmetrization procedure for linear symmetries in linear spaces . for instance , in the context of the alternating - direction examples , a common way to resolve the symmetry issue is to first solve , say , the @xmath12 and the @xmath14 direction , solve the @xmath14 and the @xmath12 direction with the same initial condition , and then average the two results . in this paper we did not mention the use and the development of similar concepts in a more strict linear algebra setting . some recent works deal with the further understanding of scalar products and structured factorizations , and more general computation of matrix functions preserving group structures , see for instance @xcite and references therein . some of these topics are covered by other contributions in the present bit issue , which we strongly encourage the reader to read to get a more complete picture of the topic and its applications . * acknowledgements . * h.m .- k . , g.r.w.q . and a.z . wish to thank the norwegian research council and the australian research council for financial support . special thanks to yuri nikolayevsky who gave us important pointers to the literature on symmetric spaces . j. c. scovel . symplectic numerical integration of hamiltonian systems . in tudor ratiu , editor , _ the geometry of hamiltonian systems _ , volume 22 , pages 463496 . msri , springer - verlag , new york , 1991 . symplectic numerical integration of hamiltonian systems .

a natural step to extend most of the existing many - body single - particle self - consistent methods is to include the full momentum and energy dependence of the vertex corrections . historically , the self - consistent approach for vertex corrections was first considered by landau , abrikosov and khalatnikov in the context of the high energy behavior of quantum electrodynamics.@xcite the original goal was to develop a non - perturbative method which encodes the information in terms of a system of closed integral equations . the parquet equations , in principle , provide a framework for self - consistent determination of the self - energy and the vertex corrections . they were proposed for both boson - boson scattering and fermion - fermion scattering during the 1950s.@xcite methods similar to the parquet equations were first introduced in the context of many - body theory by de dominics and martin.@xcite one of the early practical applications was on the x - ray absorption and emission problem by roulet , gavoret , and nozires.@xcite since then , various problems have been studied by the parquet summation approach , most notably , the fermi liquid in a strong magnetic field,@xcite the disordered electron gas in a strong transverse magnetic field,@xcite the anderson impurity model,@xcite random potential problems,@xcite the hubbard model,@xcite helium-4,@xcite helium-3,@xcite local moment formation,@xcite the vortex liquid model,@xcite the matrix models,@xcite and nuclear structure calculations.@xcite while these applications of parquet formulation provide a lot of important insights , most of the calculations are based on various approximated forms of the parquet equations . it is obvious that going from a one particle to a two - particle self - consistent calculation represents a significant increase in the computational effort , as each two - particle vertex contains three independent momentum and frequency indices . from the point of view of practical calculation , the number of elements for each index is around a few thousands . therefore , the number of elements for the vertices are around tens of millions to a few billions . moreover , all the information is encoded in integral equations with complicated structure , in which simplification does not seem to be immediately possible . indeed , in the past , the most successful application using the full parquet equations was largely limited to the single anderson impurity model . @xcite with recent advances in computational infrastructure where peta - scale performance has become available , calculations for lattice models , such as the hubbard model are now feasible . for example , the solution of the parquet equations for a @xmath0 hubbard cluster with on - site coupling @xmath1 and temperature @xmath2 was recently obtained.@xcite however , limitations on computer performance and storage are apparently not the sole obstacles for obtaining the solution of the parquet equations . another major barrier is the stability of the solvers . the simple iteration method , which is widely adapted for the dynamical mean field method , often fails to provide a stable solution for the parquet equations . in most cases , a damping scheme has to be employed . even with the damping scheme , when the temperature is low or the coupling is large , finding a stable solution still seems to be rather difficult . @xcite given the large number of variables and the complexity of the parquet equations , instabilities in their solution may not be unexpected . methods based on the local gradient are not likely to be suitable as the hessian can not be readily calculated . most of the non - linear solvers only have local convergence properties . this may not pose a problem if we have a reasonable guess which is close enough to the true solution . unfortunately , it is not easy to obtain a good initial guess for the parquet equations . methods that in principle allow global " convergence , for example , the homotopy method or continuous loading method , have been proposed as possible ways to improve the calculations.@xcite while these tend to improve convergence , many steps are required for the solution to move along the homotopy path . moreover , practical experience seems to suggest that convergence may still not be achieved when the temperature is low or the coupling is strong . it is clear that a better solver is definitely required for the practical application of the parquet method within the context of the strongly correlated systems . one of the most prominent differences between the parquet formulation and most of the other approximation schemes such as rpa,@xcite self - consistent spin fluctuations approach,@xcite and fluctuation exchange approach@xcite is that the so - called crossing symmetry is obeyed by construction of the parquet equations . the crossing symmetry@xcite implies that a vertex in one channel can also produce the vertex in all other channels by pulling or crossing the vertex legs and multiplying by appropriate constants . it also implies the pauli exclusion principle is automatically satisfied . however , in the course of the iteration process , as long as the exact solution of the parquet equations is not obtained , the crossing symmetry is violated . the main point in the present paper is to highlight that the crossing symmetry is crucial for obtaining a stable solution . we devise a modified iteration scheme which can obtain a stable solution for the parquet equations at lower temperature and stronger coupling than that from the previous schemes.@xcite this is achieved primarily by restoring the crossing symmetry at each step of the iteration . it is important to notice that because of the large number of vertex functions , for production runs , massively parallel machines are absolutely necessary . since the vertex functions in different channels are mixed in the parquet formulation , an efficient scheme to transform the vertex functions storage in different nodes is critical to improving the overall efficiency of the calculations . some of the computational details have been explained in the previous publication.@xcite we have further improved the scheme which allows us to hide the communication latency across different nodes behind the local calculations within the nodes , which effectively further speeds up the calculations . the model used for testing the computational scheme for solving the parquet equations is the standard hubbard model at half - filling . the hamiltonian is @xmath3 where @xmath4 is the on - site repulsion and @xmath5 is the hopping matrix which establishes a unit of energy . the paper is organized as follows . in section ii , we reproduce the parquet equation which also serves to fix the notation . in section iii , we describe the iteration scheme for solving the parquet equations . in section iv , we discuss the violation of crossing symmetry . in section v , we present a modified iteration scheme which explicitly restores the crossing symmetry . we also discuss the limitation of the modified scheme and the possible directions for further developments . in section vi , we present the leading eigenvalues of the antiferromagnetic channel as a function of the temperature and coupling strength and find that the parameter region of stable solutions is greatly increased when the crossing symmetry is enforced . a brief summary is contained in section vii . in the appendix , we present a latency hiding scheme which allows substantial increase of the efficiency for solving the parquet equations . in order for the current paper to be reasonably self - contained , we provide in this section a brief description of the parquet equations , largely following ref . . the purpose is to highlight the structure of the equations and to define the notations that will be used in the subsequent sections . we do not intend to provide a derivation of the formulation . for this we refer the readers to the literature where ample discussions can be found.@xcite standard perturbative expansions attempt to describe all the scattering processes , at the lowest orders , as single- or two - particle feynman diagrams . in the single - particle formulation the self - energy describes the many - body processes that renormalize the motion of a particle in the interacting background of all the other particles . in the two - particle context , one is able to probe the interactions between particles using the so - called vertex functions , which are matrices describing the two particle scattering processes . for example , the reducible ( full ) two - particle vertex @xmath6 describes the amplitude of a particle - hole pair scattered from its initial state @xmath7 into the final state @xmath8 . here , @xmath9 represents a set of indices which combines the momentum @xmath10 , the matsubara frequency @xmath11 and , if needed , the spin @xmath12 and band index @xmath13 . since the total momentum and energy of the vertex are conserved , it is convenient to adapt the notation @xmath14 for the numerical implementation of the single band hubbard model . @xcite in general , depending on how particles or holes are involved in the scattering processes , one can define three different two - particle scattering channels . these are the particle - hole ( p - h ) horizontal channel , the p - h vertical channel and the particle - particle ( p - p ) channel . one can further discriminate the vertices according to their topology . starting from the reducible vertex @xmath15 introduced above , we may define the irreducible vertex @xmath16 corresponding to the subclass of diagrams in @xmath15 that can not be separated into two parts by cutting two horizontal green function lines . similarly , the fully irreducible vertex @xmath17 corresponds to the subclass of diagrams in @xmath16 that can not be split into two parts by cutting two green function lines in any channel . furthermore , since we are mostly interested in models that preserve the @xmath18 spin rotation symmetry , and since this is an exact symmetry for our two - dimensional calculations at non - zero temperature , it is convenient to preserve this symmetry . this is accomplished by decomposing the vertices in the so - called spin - diagonalized representation.@xcite in this representation , the spin degrees of freedom decompose the particle - hole channel into the density and the magnetic channels , and the particle - particle channel into the spin singlet and the spin triplet channels which we denote as @xmath19-channel , @xmath20-channel , @xmath21-channel , and @xmath22-channel respectively . they are defined as follows , @xmath23 and similarly for @xmath15 and @xmath17 . we reproduce the full set of equations for the parquet formulation in the spin diagonalized representation in the following.@xcite the schwinger - dyson equation is @xmath24 where @xmath25 is the single - particle green function , which itself can be calculated from the self - energy using the dyson equation , @xmath26 where @xmath27 is the bare green function . here , the indices @xmath28 , @xmath29 and @xmath30 combine momentum @xmath31 and matsubara frequency @xmath32 , i.e. @xmath33 . the reducible and the irreducible vertices in a given channel are related by the bethe - salpeter equation , @xmath34 @xmath35 where @xmath36 for the density and magnetic channels and @xmath37 for the spin singlet and spin triplet channels . the vertex ladders are defined as @xmath38 where @xmath39 is the product of two single - particle green functions . the parquet equations in the spin diagonalized representation are @xmath40 @xmath41 @xmath42 @xmath43 it is important to note at this point that if we substitute the irreducible vertices @xmath44 ( eqs . [ gamma_d],[gamma_m],[gamma_s ] , and [ gamma_t ] ) into the bethe - salpeter equation ( eqs . [ sd_bsph_eq ] and [ sd_bspp_eq ] ) the crossing symmetry in the full vertex @xmath15 is automatically satisfied regardless of the numerical values of the vertex ladders @xmath45 and @xmath46 , assuming the fully irreducible vertices , @xmath17 , are crossing symmetric . we write all the full vertices explicitly in the following using only the vertex ladders , @xmath45 , @xmath46 , and the fully irreducible vertices , @xmath17 . @xmath47 @xmath48 @xmath49 @xmath50 these relations allow us to restore the crossing symmetry for the full vertices without heavy computational overhead . the prominent technical problem at hand is whether or not we can solve this set of equations efficiently without resorting to any approximated scheme . an obvious difficulty is to handle the large number of variables . on going from the one - particle level calculation to two - particle level calculation , the number of variables which has to be monitored grows as the third power of the linear dimension of the system . if @xmath51 is the number of lattice sites times the number of discrete matsubara frequencies , i.e. , @xmath52 , the largest @xmath51 that can be handled is in the range @xmath53 i.e. , the number of variables can be over one billion . one can immediately see that practical calculations for reasonably large system sizes pose a serious problem , although not insurmountable with modern computational facilities where a large number of computer nodes are accessible . however , in addition to the large number of computations associated with solving the parquet equations , they also require a complex communication pattern between the different processes as we discussed in more detail in the appendix . moreover , a numerical instability is not unexpected , especially when the system in the proximity of a phase transition . the parquet formulation consists of two sets of equations . the first set , made of the parquet equations and the bethe - salpeter equation , determines the full vertex @xmath15 and the irreducible vertex @xmath44 given the one - particle self - energy @xmath54 and the fully irreducible vertex @xmath17 as the inputs . the second set of equations determine the one - particle quantities given the full vertex @xmath15 ; it includes the schwinger - dyson equation and the dyson equation . since the method is iteration based , the initial guess is crucial for obtaining a converged solution . in principle , the initial guess can be approximated , for example , by second order perturbation . however , in practice , this is not the optimal choice , especially when the self - energy from the second order perturbation is small . in this case the green function will quickly destabilize the calculation . this may relate to the fact that the damping from the imaginary part of the self - energy is quickly reduced . since we are supposing that we know the fully irreducible vertex @xmath17 , a practical scheme is to choose the irreducible , @xmath44 and full vertices , @xmath15 , equal to @xmath17 , and a large value ( a few times of the bandwidth ) for the imaginary part of the self - energy . fig . [ algorithm_a ] is an illustration of the flow diagram of the algorithm , where the fully irreducible vertices are the initial input for the calculation . the algorithm can be described as the following : \1 . set the initial conditions for the irreducible and full vertices , and the self - energy . update the green s functions and calculate the bare susceptibility , @xmath39 , which is given by the product of two green s functions . solve the parquet and the bethe - salpeter equations for the irreducible vertices , @xmath44 . simple iteration is used until the convergence criteria are met for the irreducible vertices . this completes the update of the vertices . the next step is to use the irreducible vertices obtained from the parquet equations to construct the full vertices . solve the bethe - salpeter equation to obtain the full vertices , @xmath15 , using the irreducible vertices from the previous step ; this is executed exactly by calling the lapack routines for the inverse of the matrices . @xcite with the full vertices obtained , we can update the self - energy . solve the dyson - schwinger equation to obtain the self - energy from the full vertices . simple iteration is used until certain convergence criteria are met for the self - energy . solve the dyson equation for the fully dressed green function from the self - energy . this completes the iteration loop , and the procedure is repeated from step 2 , until convergence is reached for both the self - energy and the irreducible vertices . in practice , step 2 which attempts to obtain the irreducible vertices needs to be iterated for a few times to get a reasonable convergence , even in the case where the coupling is weak and the temperature is high . on the other hand , step 4 which attempts to solve the self - energy from the updated full vertices is not iterated more than one time at each loop , so as to avoid instability ( we define instability here as the failure to obtain a converged solution from the iterative solver ) . attempting to solve the self - energy at the early stage of the iteration procedure where the full vertices are not well converged can generally lead to instability . although in the present paper , we only focus on the hubbard model , instabilities in the iteration process have also been observed in solving the parquet equations for nuclear structure calculations.@xcite a widely used method to avoid the instability in the iterative process is to introduce a damping factor in the updates of the variables . the updates are modified as follows . @xmath55 with this damping scheme , the solution for the half - filled hubbard model on a @xmath56 cluster has been successfully obtained for @xmath1 and temperature , @xmath2.@xcite however , in the strong coupling regime , obtaining a stable solution still seems to be difficult , even though a rather heavy damping is employed . we will demonstrate the instability problem of the simple iterative process by monitoring the leading eigenvalues @xmath57 defined as @xmath58 for @xmath59 and @xmath20 ; similarly @xmath60 for @xmath61 and @xmath22 . in principle , these leading eigenvalues signal a phase transition by going through @xmath62 , expressing the divergence of the susceptibilities in the corresponding channel . in fig . [ nosym ] , we plot the leading eigenvalues of the density , magnetic , spin singlet , and spin triplet channels as a function of the number of iteration steps calculated with this simple iteration method ( si ) . the calculation is done on a 2@xmath632 cluster with 32 frequencies and temperature @xmath64 ; the damping parameter is @xmath65 . a converged solution is obtained for @xmath1 ; however , for @xmath66 and @xmath67 the iterative solutions diverge . changing the damping or the initial self - energy does not help in obtaining a converged solution for the larger values of @xmath4 . these are illustrative examples which show the problem of using the simple iteration method for solving the parquet equations . for weak coupling and not too low temperature , converged solution can be obtained . beyond weak coupling the iteration becomes divergent . a widely used method to alleviate the divergence in the non - linear solver is the so - called continuous loading or homotopy method . the basis of the continuous loading method is to construct an auxiliary equation with a tuning parameter @xmath68 , so that its solution is trivial for @xmath69 but the solution of the original equation is recovered for @xmath70 . symbolically we can write down the set of equations of the parquet formulation as @xmath71 , where @xmath72 is a large vector . we define the auxiliary function as @xmath73 , where @xmath74 is a function with a trivial solution . in our study we choose @xmath75 as a vector containing all the elements of @xmath76 and @xmath77 , where @xmath78 and @xmath79 are the initial guesses for the irreducible vertices and the self - energy respectively . the iteration method is used to solve the function @xmath80 , instead of the @xmath72 . one can readily see that the solution of @xmath81 is trivial while the solution of the @xmath72 is recovered when @xmath70 . the idea is to find a converged solution for @xmath80 with a small enough @xmath68 where a converged solution can be readily obtained , and then gradually increase @xmath68 until it goes to @xmath62 . therefore , a series of @xmath68 values are needed , which we denote as @xmath82 . we plot the leading eigenvalues from the continuous loading method in fig . [ continue_nosym ] . the values of @xmath68 used are @xmath83 @xmath84 . @xmath85 and @xmath86 iterations are performed for @xmath87 and @xmath88 respectively . for @xmath1 , converged solution is obtained for @xmath70 . however , for @xmath66 and @xmath67 , the iterative solution diverges . the divergences appear before the homotopy parameter is pushed to @xmath70 . these examples illustrate the generic behavior when solving the parquet equations by the simple iteration method beyond weak coupling . they also illustrate that the continuous loading method may not be sufficiently robust to solve the problem . the damping factor @xmath89 used in these calculations is @xmath90 which we believe is a fairly small value , although it may still not be sufficient . a rule of thumb for choosing the damping parameter is that the damping parameter should be close to the value of the inverse of the residual between two consecutive iterations ; unfortunately , with the huge number of variables , this choice will result in a very small step and may not be a practical option.@xcite the exact solution of the parquet equations automatically satisfies the crossing symmetry . it is one of the most important differences between the parquet formulation and most other perturbative methods . however , within the iteration scheme presented in the last section , the crossing symmetry is not fulfilled unless the iteration converges to an exact solution . from the above section , we clearly find that the iteration method , even with the help of the continuous loading scheme , is not robust enough to obtain a converged solution beyond weak coupling . it is desirable to quantify the violation of crossing symmetry . the following six equalities are the consequence of the crossing symmetry ( see fig . [ crossing_symmetries ] for a diagrammatic representation of these crossing symmetry ( cs ) operations ) . @xcite @xmath91(p+p^{'}+q)_{-p^{'},-p } \equiv r_{1 } , \\ l_{2 } \equiv f_{s}(q)_{p , p^ { ' } } = \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\ ; \\ \nonumber [ -(3/2 ) f_{m}+(1/2)f_{d}](p+p^{'}+q)_{-p^{'},-p } \equiv r_{2 } , \\ l_{3 } \equiv f_{m}(q)_{p , p^ { ' } } = \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\ ; \\ \nonumber [ ( 1/2)f_{t}-(1/2)f_{t}](p+p^{'}+q)_{-p - q ,- p } \equiv r_{3 } , \;\\ l_{4 } \equiv f_{d}(q)_{p , p^ { ' } } = \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\ ; \\ \nonumber [ ( 3/2)f_{t}+(1/2)f_{t}](p+p^{'}+q)_{-p - q ,- p } \equiv r_{4 } , \;\\ l_{5 } \equiv f_{m}(q)_{p , p^ { ' } } = \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\ ; \\ \nonumber [ ( 1/2)f_{m}-(1/2)f_{d}](p^{'}-p)_{p , p+q } \equiv r_{5 } , \;\;\;\;\;\;\;\;\;\;\ ; \\ l_{6 } \equiv f_{m}(q)_{p , p^ { ' } } = \label{crossing_equalities6 } \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\ ; \\ \nonumber [ -(3/2)f_{m}+(1/2)f_{d}](p^{'}-p)_{p , p+q } \equiv r_{6}. \;\;\;\;\;\;\;\ ; \nonumber \end{aligned}\ ] ] in fig . [ nosym_crossing ] we plot the violation of crossing symmetry versus the number of iterations . it can be seen clearly that the crossing symmetry can not be perfectly restored , even for the case of @xmath92 , the measures of violation of crossing show oscillatory decreasing behavior , and the rate of decrease is quite slow even though the leading eigenvalues seem to be well converged . obviously , for the cases of @xmath66 and @xmath67 the crossing symmetry is severely violated and at the verge of the divergence there are sharp increases in the crossing symmetry violation . this may suggest that if the crossing symmetry can be restored , the divergence may be avoided beyond weak coupling . in fig . [ continue_nosym_crossing ] , we plot the crossing symmetry violation as a function of iteration steps with the continuous loading method and the same parameters as in fig . [ continue_nosym ] . it is important to note that the homotopy function @xmath93 does not respect the crossing symmetry except at @xmath70 . therefore , although the solution is converged , as long as @xmath94 , the crossing symmetry is violated . the data for @xmath95 , and @xmath67 are shown in the upper , the middle , and the lower panels respectively . the data for @xmath1 show a peak at the beginning of the iteration procedure when @xmath68 is increased and gradually converges to a finite value . for @xmath68 close to @xmath62 , the data show a similar oscillatory behavior as that from the simple iteration method . similar behaviors are also observed for @xmath66 and @xmath96 , however , the iterations fail to converge for @xmath68 close to @xmath62 ; the crossing symmetry is strongly violated . similar to that for the simple iteration method , at the verge of the divergence , there are sharp increases of the violation of crossing symmetry . cluster at temperature @xmath64 . the initial condition for the self - energy is set at @xmath97 , and that for the irreducible vertex is set at the bare hubbard coupling . the damping factor @xmath89 is set at @xmath90 . the solution is well converged for @xmath1 . however , divergence occurs for @xmath66 at the @xmath98th iteration step . for @xmath96 , divergence occurs at the @xmath99th iteration step . [ nosym],scaledwidth=26.0% ] cluster at temperature @xmath64 . the initial condition for the self - energy is set at @xmath97 , and that for the irreducible vertex is set at the bare hubbard coupling . the damping factor @xmath89 is set at @xmath90 . the solution is well converged for @xmath1 . however , divergence occurs for @xmath66 at the @xmath98th iteration step . for @xmath96 , divergence occurs at the @xmath99th iteration step . [ nosym],scaledwidth=26.0% ] cluster at temperature @xmath64 . the initial condition for the self - energy is set at @xmath97 , and that for the irreducible vertex is set at the bare hubbard coupling . the damping factor @xmath89 is set at @xmath90 . the solution is well converged for @xmath1 . however , divergence occurs for @xmath66 at the @xmath98th iteration step . for @xmath96 , divergence occurs at the @xmath99th iteration step . [ nosym],scaledwidth=26.0% ] and use them to define the auxiliary function as @xmath73 , where @xmath74 is a function with a trivial solution . we choose @xmath74 as a vector containing all the elements of @xmath76 and @xmath77 . the iteration method is used to solve the function @xmath80 , instead of the @xmath72 . the solution of the @xmath72 is recovered when @xmath70 . a series of @xmath68 values are needed , which we denote as @xmath82 . the function @xmath100 is solved by the simple iteration method with the initial conditions given by the converged solution of the function @xmath101 . for @xmath1 , we can push the value of @xmath68 to @xmath62 to obtain the converged solution for @xmath102 , and the solution of the parquet equations is recovered . however , the iteration procedure diverges for the cases of @xmath66 and @xmath96 ; they diverge at @xmath103 and @xmath104 respectively . these examples show that for the cases where simple iteration method diverges , the continuous loading method may not eliminate the divergence , even though the value of @xmath68 is pushed fairy close to @xmath62.[continue_nosym],scaledwidth=26.0% ] and use them to define the auxiliary function as @xmath73 , where @xmath74 is a function with a trivial solution . we choose @xmath74 as a vector containing all the elements of @xmath76 and @xmath77 . the iteration method is used to solve the function @xmath80 , instead of the @xmath72 . the solution of the @xmath72 is recovered when @xmath70 . a series of @xmath68 values are needed , which we denote as @xmath82 . the function @xmath100 is solved by the simple iteration method with the initial conditions given by the converged solution of the function @xmath101 . for @xmath1 , we can push the value of @xmath68 to @xmath62 to obtain the converged solution for @xmath102 , and the solution of the parquet equations is recovered . however , the iteration procedure diverges for the cases of @xmath66 and @xmath96 ; they diverge at @xmath103 and @xmath104 respectively . these examples show that for the cases where simple iteration method diverges , the continuous loading method may not eliminate the divergence , even though the value of @xmath68 is pushed fairy close to @xmath62.[continue_nosym],scaledwidth=26.0% ] and use them to define the auxiliary function as @xmath73 , where @xmath74 is a function with a trivial solution . we choose @xmath74 as a vector containing all the elements of @xmath76 and @xmath77 . the iteration method is used to solve the function @xmath80 , instead of the @xmath72 . the solution of the @xmath72 is recovered when @xmath70 . a series of @xmath68 values are needed , which we denote as @xmath82 . the function @xmath100 is solved by the simple iteration method with the initial conditions given by the converged solution of the function @xmath101 . for @xmath1 , we can push the value of @xmath68 to @xmath62 to obtain the converged solution for @xmath102 , and the solution of the parquet equations is recovered . however , the iteration procedure diverges for the cases of @xmath66 and @xmath96 ; they diverge at @xmath103 and @xmath104 respectively . these examples show that for the cases where simple iteration method diverges , the continuous loading method may not eliminate the divergence , even though the value of @xmath68 is pushed fairy close to @xmath62.[continue_nosym],scaledwidth=26.0% ] , versus the number of iterations for the simple iteration ( si ) method with the same parameters as fig . [ nosym ] . the six measures of crossing symmetry violation are defined as @xmath105 and @xmath106 are defined respectively as the left hand side and the right hand side of the eqs . [ crossing_equalities1 ] [ crossing_equalities6 ] . the data for @xmath1 shows an oscillatory but decreasing trend . this is expected for the case where the iteration provides a well converged solution . one should note that although the leading eigenvalues seem to be converged , the crossing symmetry is not perfectly constructed from the iteration . for @xmath66 and @xmath96 , the iteration fails to provide converged solutions , and the crossing symmetry is strongly violated . in particular , at the verge of the divergence , there is a sharp increase of the violation of crossing symmetry . [ nosym_crossing],scaledwidth=26.0% ] , versus the number of iterations for the simple iteration ( si ) method with the same parameters as fig . [ nosym ] . the six measures of crossing symmetry violation are defined as @xmath105 and @xmath106 are defined respectively as the left hand side and the right hand side of the eqs . [ crossing_equalities1 ] [ crossing_equalities6 ] . the data for @xmath1 shows an oscillatory but decreasing trend . this is expected for the case where the iteration provides a well converged solution . one should note that although the leading eigenvalues seem to be converged , the crossing symmetry is not perfectly constructed from the iteration . for @xmath66 and @xmath96 , the iteration fails to provide converged solutions , and the crossing symmetry is strongly violated . in particular , at the verge of the divergence , there is a sharp increase of the violation of crossing symmetry . [ nosym_crossing],scaledwidth=26.0% ] , versus the number of iterations for the simple iteration ( si ) method with the same parameters as fig . [ nosym ] . the six measures of crossing symmetry violation are defined as @xmath105 and @xmath106 are defined respectively as the left hand side and the right hand side of the eqs . [ crossing_equalities1 ] [ crossing_equalities6 ] . the data for @xmath1 shows an oscillatory but decreasing trend . this is expected for the case where the iteration provides a well converged solution . one should note that although the leading eigenvalues seem to be converged , the crossing symmetry is not perfectly constructed from the iteration . for @xmath66 and @xmath96 , the iteration fails to provide converged solutions , and the crossing symmetry is strongly violated . in particular , at the verge of the divergence , there is a sharp increase of the violation of crossing symmetry . [ nosym_crossing],scaledwidth=26.0% ] , versus the number of iterations for the continuous loading ( cl ) method with the same parameters and the same definition of @xmath107 as in fig . [ nosym_crossing ] . the homotopy function @xmath93 does not respect the crossing symmetry except at @xmath70 . therefore , although the solution may be converged for some values of @xmath68 , as long as @xmath94 , the violation of crossing symmetry is non - zero . for @xmath1 the crossing symmetry violations peak near the beginning of the iteration procedure where @xmath68 is small and gradually converge to a finite value when @xmath70 . similar behaviors are also observed for @xmath66 and @xmath96 ; however , since the iteration fails to converge for @xmath68 close to @xmath62 , the crossing symmetry is strongly violated . similar to that observed in the simple iteration method , at the verge of the divergence , there is a sharp increase of the violation of crossing symmetry . [ continue_nosym_crossing],scaledwidth=26.0% ] , versus the number of iterations for the continuous loading ( cl ) method with the same parameters and the same definition of @xmath107 as in fig . [ nosym_crossing ] . the homotopy function @xmath93 does not respect the crossing symmetry except at @xmath70 . therefore , although the solution may be converged for some values of @xmath68 , as long as @xmath94 , the violation of crossing symmetry is non - zero . for @xmath1 the crossing symmetry violations peak near the beginning of the iteration procedure where @xmath68 is small and gradually converge to a finite value when @xmath70 . similar behaviors are also observed for @xmath66 and @xmath96 ; however , since the iteration fails to converge for @xmath68 close to @xmath62 , the crossing symmetry is strongly violated . similar to that observed in the simple iteration method , at the verge of the divergence , there is a sharp increase of the violation of crossing symmetry . [ continue_nosym_crossing],scaledwidth=26.0% ] , versus the number of iterations for the continuous loading ( cl ) method with the same parameters and the same definition of @xmath107 as in fig . [ nosym_crossing ] . the homotopy function @xmath93 does not respect the crossing symmetry except at @xmath70 . therefore , although the solution may be converged for some values of @xmath68 , as long as @xmath94 , the violation of crossing symmetry is non - zero . for @xmath1 the crossing symmetry violations peak near the beginning of the iteration procedure where @xmath68 is small and gradually converge to a finite value when @xmath70 . similar behaviors are also observed for @xmath66 and @xmath96 ; however , since the iteration fails to converge for @xmath68 close to @xmath62 , the crossing symmetry is strongly violated . similar to that observed in the simple iteration method , at the verge of the divergence , there is a sharp increase of the violation of crossing symmetry . [ continue_nosym_crossing],scaledwidth=26.0% ] although it does not seem to be easy to analyze all the causes of the instability in the iteration , based on the discussion in the above section , our conjecture is that one of the possible reasons for the instability is that certain symmetries are violated in the course of the iteration process . a possible strategy to improve the iteration scheme is to impose those symmetries explicitly into the iteration process , so that at each step of the iteration these symmetries are not violated . the full vertices @xmath15 obtained by the solution of the bethe - salpeter equation can not guarantee the crossing symmetry , unless the exact solution is attained . therefore , so as to preserve the crossing symmetry , the simplest method is to use the full vertices obtained by solving the bethe - salpeter equation ( that is the full vertices obtained from the step 3 of the algorithm presented in the section iii ) , and feed them back into the parquet equation to reconstruct the crossing symmetric full vertices . [ algorithm_b ] illustrates the flow diagram for solving the parquet equations with explicit restoration of the crossing symmetry in the full vertices . here is the algorithm which explicitly preserves the crossing symmetry : . because of this explicit restoration of the crossing symmetry , in practice , the step 2 is only iterated for one time as the self - consistency is not required to generate the crossing symmetry.,scaledwidth=51.0% ] \1 . set the initial conditions for the irreducible and full vertices , and the self - energy . update the green s functions and calculate the bare susceptibility , @xmath39 . solve the parquet and the bethe - salpeter equations for the irreducible vertices , @xmath44 . simple iteration is used until the convergence criteria are met for the irreducible vertices . since we will restore the crossing symmetry of the full vertices in the step 3b , we find that it is not necessary to attain the self - consistency for step 2 . in practice , we iterate the parquet equations for one time only . the next step is to use the irreducible vertices obtained from the parquet equations to construct the full vertices . solve the bethe - salpeter equation to obtain the full vertices , @xmath15 , using the irreducible vertices from the previous step . this is executed exactly by calling the lapack routines for the inverse of the matrices . @xcite 3b . use the new irreducible vertices obtained in step 2 and the full vertices obtained in step 3a to form the vertex ladders . construct the full vertices from eqs . [ sd_bsph_eq ] and [ sd_bspp_eq ] using the vertex ladders . following these steps , the crossing symmetry is restored in the full vertex @xmath15 . with the full vertices obtained , we can update the self - energy . solve the dyson - schwinger equation to obtain the self - energy from the full vertices . simple iteration is used until the convergence criteria are met for the self - energy . solve the dyson equation for the fully dressed green function from the self - energy . this completes the iteration loop , and the procedure is repeated from step 2 until the criteria of convergence are met for both the self - energy and the irreducible vertices . the main difference between the current algorithm and the previous algorithm we present in section iii is in step 3 where the full vertices are constructed . in the previous algorithm the bethe - salpeter equation is solved many times to attain convergence . when the absolute convergence is attained , the crossing symmetry will be satisfied . in the current algorithm , we just explicitly solve the bethe - salpeter equation in step 3a to refresh the full vertices . once we obtain the full vertices , in step 3b , we construct the new vertex ladders and the new full vertices from the vertex ladders using the bethe - salpeter equation . by doing so , the crossing symmetry of the full vertices will be satisfied ; see eqs . [ f_d ] , [ f_m ] , [ f_s ] , and [ f_t ] . in fig . [ sym ] we show the leading eigenvalues using the same set of parameters used in fig . [ nosym ] . while the simple iteration scheme without crossing symmetry fails to converge for the case of @xmath108 and @xmath109 , it provides a converged solution when the crossing symmetry is explicitly restored . . the only difference is that the crossing symmetry in the full vertex , @xmath15 , is explicitly restored at each step of the iteration . this is easily achieved by constructing the full vertex directly from eqs . [ sd_bsph_eq ] and [ sd_bspp_eq ] . the simple iteration scheme without crossing symmetry fails for the case of @xmath108 and @xmath109 . with the crossing symmetry explicitly restored , converged solutions are obtained . [ sym],scaledwidth=26.0% ] . the only difference is that the crossing symmetry in the full vertex , @xmath15 , is explicitly restored at each step of the iteration . this is easily achieved by constructing the full vertex directly from eqs . [ sd_bsph_eq ] and [ sd_bspp_eq ] . the simple iteration scheme without crossing symmetry fails for the case of @xmath108 and @xmath109 . with the crossing symmetry explicitly restored , converged solutions are obtained . [ sym],scaledwidth=26.0% ] . the only difference is that the crossing symmetry in the full vertex , @xmath15 , is explicitly restored at each step of the iteration . this is easily achieved by constructing the full vertex directly from eqs . [ sd_bsph_eq ] and [ sd_bspp_eq ] . the simple iteration scheme without crossing symmetry fails for the case of @xmath108 and @xmath109 . with the crossing symmetry explicitly restored , converged solutions are obtained . [ sym],scaledwidth=26.0% ] . two symmetries are explicitly restored at each step of the iteration : the crossing symmetry ( cs ) in the full vertex , @xmath15 , and the time - reversal symmetry ( trs ) for the self - energy , @xmath54 , and both the irreducible vertex , @xmath44 , and the full vertex , @xmath15 , ( that is @xmath110 and similarly for @xmath44 ) . notice there is no substantial gain in the convergent rate compared to the case with only the crossing symmetry being restored . [ fullsym],scaledwidth=26.0% ] . two symmetries are explicitly restored at each step of the iteration : the crossing symmetry ( cs ) in the full vertex , @xmath15 , and the time - reversal symmetry ( trs ) for the self - energy , @xmath54 , and both the irreducible vertex , @xmath44 , and the full vertex , @xmath15 , ( that is @xmath110 and similarly for @xmath44 ) . notice there is no substantial gain in the convergent rate compared to the case with only the crossing symmetry being restored . [ fullsym],scaledwidth=26.0% ] . two symmetries are explicitly restored at each step of the iteration : the crossing symmetry ( cs ) in the full vertex , @xmath15 , and the time - reversal symmetry ( trs ) for the self - energy , @xmath54 , and both the irreducible vertex , @xmath44 , and the full vertex , @xmath15 , ( that is @xmath110 and similarly for @xmath44 ) . notice there is no substantial gain in the convergent rate compared to the case with only the crossing symmetry being restored . [ fullsym],scaledwidth=26.0% ] besides imposing the crossing symmetry on the full vertices , some of the internal symmetries can also be imposed on the irreducible vertices and the self - energy without much computational overhead . we illustrate the time - reversal symmetry in the self - energy @xmath111 and the vertices ( spatial reflection symmetry and parity invariance are assumed ) , @xmath112 as these symmetry operations do not mix vertices across different values of @xmath30 , and providing that the data is distributed with one or more @xmath30 at each node , the time reversal symmetry of the vertices can be imposed without invoking communications across different nodes . therefore , enforcing time - reversal symmetry will only cause a very minor computational overhead . other symmetries , such as the point group symmetry for the square lattice can be rather cumbersome . an expensive scheme involving heavy internode communication would be required to impose the complete set of point group symmetries . however , we may impose an important subset of the operations @xmath113 for which @xmath114 without expensive communications . in these cases , the vertices may be symmetrized by performing the sum @xmath115 where @xmath116 is the number of elements in this subset of operations . for general @xmath30 in the cluster brillouin zone there would be no @xmath89 such that @xmath117 apart from the identity . however , for the points of high symmetry , @xmath117 for all @xmath89 . generally , the instabilities first occur here , so imposing the point group symmetries at these @xmath30 values should have the greatest impact . in fig . [ fullsym ] we show the leading eigenvalues of various channels when both crossing and time - reversal symmetries are imposed for the same set of parameters being used from the data in fig . [ nosym ] and [ sym ] . spatial reflection symmetry , parity invariance and spin rotation invariance are assumed as appropriate for the two - dimensional hubbard model at non - zero temperature . we can see that there is only very marginal improvement for the convergence compared to the results without explicitly restoring the time - reversal symmetry ( see fig . [ sym ] ) . we also use the scheme described above to partially impose the point group symmetries . however , these symmetries resulted in no additional improvements and therefore no results are shown . with the improved scheme proposed in this paper , we are able to explore a wider range of temperature and coupling strength for the half - filled hubbard model . in fig . [ eigen_ref ] we show the leading eigenvalue for the most singular channel , the antiferromagnetic channel , @xmath118 , as a function of @xmath4 for a range of temperatures as low as @xmath119 . the data points enclosed in a black square correspond to the cases where the simple iteration without any symmetry restoration provides a converged solution . for all temperatures , @xmath118 increase sharply at weak coupling ( @xmath120 ) , and they tend to saturate at strong coupling ( @xmath121 ) . they are most sensitive to temperature at the intermediate coupling ( @xmath122 ) . we emphasize that convergency is not possible without the improved scheme , unless a large number of iterations are used to attain the crossing symmetry . we present improvements of numerical implementations for solving the parquet equations for the hubbard model . the main strategy is to enforce the symmetries in the iteration process . the most prominent advantage of the parquet formulation , compared to most of the other approaches , is that the crossing symmetry is exactly fulfilled . however , in general , it is true only if the exact solution is found . with the simple iteration method , the crossing symmetry is strongly violated prior an instability , suggesting that the instability is due to these symmetry violations . the continuous loading or homotopy method does not improve convergence significantly beyond the simple iteration method . we note that the solutions of the continuous loading function do not preserve crossing symmetry . this may partly explain why the continuous loading method does not provide significant improvement over simple iteration . we present a simple method to enforce the crossing symmetry at each step of the iteration which does not substantially increase the computational cost . the addition of these symmetry constraints can greatly improve the stability of the calculation , so that a wider range of parameters can be explored by the parquet formulation . along this line of thought , one can expect that the stability may be further improved if other symmetries are also imposed , the obvious ones being time - reversal and point group . however , these additional symmetries did not improve the stability significantly beyond that obtained with crossing symmetry alone . the parquet formulation still remains as one of the best approaches for calculating the two - particle vertex functions in a self - consistent manner . at present , solving the parquet equations for a large lattice size is still a very challenging task ; however , with the continuous advances of computational facilities , it should become more feasible in the foreseeable future . a promising direction , which allows immediate application of the parquet formulation , is to incorporate it as part of the multi - scale many - body approach.@xcite we would like to acknowledge very useful discussions with karen tomko , and we thank peter reis for his careful reading of the manuscript . this work was supported in part by the doe scidac grant de - fc02 - 06er25792 ( kmt , hf , syz , and mj ) and the u.s . national science foundation la - sigma grant eps-1003897 ( jr , jm , and mj ) . supercomputer support was provided by the nsf teragrid under grant number tg - dmr100007 . this research also used resources of the national center for computational sciences at oak ridge national laboratory , which is supported by the office of science of the u.s . department of energy under contract no . de - ac05 - 00or22725 . this appendix describes a highly effective implementation of the symmetry - enforcing variant of the parquet formulation described earlier in the paper . the communication bottleneck in this implementation is the expensive tensor rotations required to rotate the vertex ladders ( eqs . [ eq : phi_ladder ] and [ eq : psi_ladder ] ) between the forms used in the bethe - salpeter equation , eqs . [ sd_bsph_eq ] and [ sd_bspp_eq ] , to those used in the parquet equations , eqs . [ gamma_d ] [ gamma_t ] . if we distribute the equations between the processes executing on the compute nodes of a parallel machine using the transfer momenta @xmath30 , the tensor rotations are done with an expensive all - to - all communication among those processes , in which every node needs to communicate with all the other nodes . the mpi implementation of the all - to - all communication@xcite is a collective operation that is blocking , i.e. , each process has to wait until the message has been sent out . the key aspect of our implementation is the decomposition of the required communication so that non - blocking communication primitives can be effectively utilized . the non - blocking communication enables latency hiding by overlapping computations and communications . four different forms of tensor rotations are required : @xmath123 note that the indices in subscripts and those in parenthesis are equivalent , with the latter only distinguished by also labeling the nodes where data is distributed . the size of the tensors is @xmath124 , where @xmath51 is the number of momentum points times the number of discrete matsubara frequencies , i.e. , @xmath52 . all indices are in modulo arithmetic at each of the @xmath125 dimensions , where @xmath126 ( the cluster dimension ) , and the @xmath127 is for the matsubara frequency . because it takes many iterations ( up to a few hundred for low temperatures and strong coupling ) to obtain converged solutions , the total number of tensor rotations is significant and account for a large fraction of the computational time . we use the hybrid mpi / openmp model for the computations . the rank three tensors are decomposed and evenly distributed into @xmath128 virtual nodes . each virtual node consists of a few cores . the size of a virtual node ( i.e. , the number of cores ) is less than or equal to the size of a physical node . specifically , we slice the rank three tensors to a set of two - dimensional arrays based on the index in parenthesis , e.g. , @xmath30 and @xmath129 for the left and right sides of eq . [ rot1 ] , respectively . then , each two dimensional matrix is assigned to a virtual node . since we have @xmath130 layers of two dimensional slices , the total number of virtual nodes also becomes @xmath131 . in this scenario , every rotation requires data communications among all nodes . the following describes the data access patterns for our implementation of the tensor rotations . * step 1 : * this step involves no mpi communication and is done before any data is sent between nodes . the tensor elements are locally rearranged in order to collect specific elements to be grouped and sent to designated destination nodes . the index in parenthesis of the tensors on the right of eqs . [ rot1][rot4 ] represents the rank of a sending node in which a sliced two dimensional matrix resides . for rotations 1 and 2 , rank of sending node @xmath132 is @xmath133 for rotations 3 and 4 , @xmath132 is @xmath134 using eqs . [ rot12s ] and [ rot34s ] , and applying these to the corresponding rotations , the two dimensional matrix elements are grouped based on the rank of destination node @xmath30 from a given sending node @xmath132 . @xmath135 note that , here , @xmath132 is the node index ( the index of the sender ) and @xmath136 , so the @xmath28 and @xmath30 are the row and column indices of the matrix . we assume column - major order data access in mpi data communications which distribute columns of matrix @xmath137 to nodes of rank @xmath30 in the next step . * step 2 : * the columns of the two dimensional matrix @xmath137 are distributed among all nodes . at the sending nodes , each column of @xmath137 is sent to a destination node labeled by @xmath30 . the standard approach is to use _ mpi_alltoall_. however , as we show later , this task can be done using different combinations of point - to - point communications.@xcite in particular , non - blocking communication protocols can be applied to overlap communications and local computations . overall , this procedure is applied to all the tensor rotations and can be written as @xmath138 as shown in eq . [ step2 ] , the rank of destination nodes is determined by the column index @xmath30 of @xmath137 in sending nodes . the rank of sending nodes becomes column index @xmath132 of @xmath139 in the receiving nodes . the rank of sending nodes @xmath132 must be provided to receiving nodes in order to assign the correct column index to the received messages . * step 3 : * once messages have arrived at the destination nodes , the columns of the two dimensional matrix @xmath139 are rearranged to complete the tensor rotations . the column index of the rotated received matrix is related to the rank of the sending and receiving nodes by eqs . [ rot12s ] and [ rot34s ] . then , the rotations are finalized by using the following relations @xmath140 where @xmath30 is the index of a given receiving node and @xmath141 . this step is a local process , i.e. , no internode communication is necessary . while steps 1 and 3 are strictly local processes , step 2 is the only stage involving nonlocal mpi communications . the nature of the collective communications among all nodes in step 2 makes it suited to the use of _ mpi_alltoall_. in such a case , step 2 can start only after the completion of step 1 . because _ mpi_alltoall _ is a blocking communication , step 3 must wait to start until step 2 is finished . therefore , the total elapsed time to complete a tensor rotation is the sum of elapsed times of the three steps . when the problem size is large , the communication efficiency of _ mpi_alltoall _ is reduced significantly due to the increased network complexity associated with the bandwidth and latency among all participating nodes . our approach to handle these rotations more efficiently is to implement a latency hiding strategy by overlapping message communications ( step 2 ) and local computations ( steps 1 and 3 ) . to enable this , we have developed our own version of a routine that performs communications from all nodes to all nodes . at a basic level , the functionality of this routine is identical to that of the generic _ mpi_alltoall _ routine . however , our routine allows further data manipulations such that local computations are embedded between communications in the following way : on the sending node , the first column of @xmath137 is computed from the equations of step 1 . _ mpi_isend _ sends out the first column of @xmath142 while this column is being sent out , the next column of @xmath137 is prepared with step 1 . this procedure is repeated until all @xmath130 columns of @xmath137 , the group of the selected elements from @xmath143 , are sent out . this process overlaps steps 1 and 2 . latency hiding is also implemented in receiving nodes . we note that the sending nodes are also receiving nodes . they only differ by whether they are operating in the sending or the receiving _ mode_. on the receiving nodes , _ mpi_irecv _ is set to receive messages from arbitrary nodes by using _ mpi_any_source _ as a tag identifying the source of the message . for efficiency reasons , _ mpi_irecv _ is posted before _ mpi_isend _ of the sending process . mpi_test _ calls are used to check the completion of the arrival of the message . once message arrival is confirmed , the rank of the node that sent this message can be identified by inquiring using _ mpi_status_. this provides @xmath132 to assign to a corresponding column and to be used in step 3 . since the message arrival is column - by - column , the processing of each column of @xmath139 continues to step 3 while the next column is traveling through the network . this process is repeated until all columns are completed . this procedure completely overlaps steps 2 and 3 . depending on the size of problem , it is desirable to define a virtual node containing several cores ( assuming multicore hardware architecture ) based on the memory availability per node . among the cores , mpi communications are assigned to one core . the other cores are utilized by implementing openmp@xcite that parallelizes the local computational tasks in a node to all cores within the node . thus , openmp thread depth is set to match with the total number of cores per virtual node . specifically , we applied the _ do _ directive of openmp for iterations of index @xmath28 in the column selection processes of steps 1 and 3 . we test the efficiency of this latency hiding scheme using a non - blocking protocol against the standard _ mpi_alltoall_. all the experimental comparisons are conducted on the cray xt5 ( jaguar ) at the national center for computational sciences ( nccs ) at the oak ridge national laboratory . jaguar consists of 12 cores per node , with six cores per numa ( non - uniform memory access ) node , and two numas per node . first , we discuss hardware - driven constraints in implementing latency hiding . the non - blocking _ mpi_isend _ does not check for the arrivals of messages . with larger tensor size , the node usage and the size of individual columns becomes large . the _ mpi_isend _ from all participating nodes tries to dump a large column in each iteration . the next iteration starts regardless of message arrivals in the receiving nodes . as a consequence , a large amount of data rushes onto the network faster than the data can be absorbed by the receiving nodes . eventually , this causes memory overflow to the system buffer assigned to the message processing unit . to avoid this we have allocated more memory space to the system buffer . for simplicity , we assign one virtual node to a physical node . on the jaguar cray xt5 , this means one virtual node containing 12 cores . to utilize all cores in a node , the value of openmp thread depth is set to 12 . we gradually increase the problem size @xmath130 until jobs end with error indicating buffer overflow . then , we set a higher buffer size by controlling environmental variable _ mpich_unex_buffer_size_. for every incidence of error , we add 60 mb buffer size . the default value of _ mpich_unex_buffer_size _ is 60 mb on jaguar xt5 ( the total number of cores is less than 50,000 ) . the results are shown in the fig . [ minbuffer ] . up to @xmath144 , the 60 mb default buffer size is enough to handle the data traffic . increasing @xmath130 further forces us to use a larger buffer size . overall , the amount of added buffer size increases for larger problem sizes . we note that the results presented in fig . [ minbuffer ] are with the maximum number of cores per a virtual node . smaller core usage per node alleviates the buffer restriction . for example , hardware setup with a numa node per virtual node consumes less buffer memory due to the reduced total number of physical nodes participating in internode communication . we did not observe buffer memory overflow with the generic blocking _ mpi_alltoall _ routine . the performance of the latency hiding approach is evaluated in terms of wallclock time spent on a single tensor rotation and compared with the case of the standard _ mpi_alltoall _ applied for step 2 . for this , the elapsed time to complete the tensor rotation is averaged over nine independent runs . each run contains 40 repetitions of identical tensor rotations . at the end of each run , the elapsed time is also averaged for the 40 rotations . for all runs , we choose rotation 1 and the minimum buffer sizes shown in the fig . [ minbuffer ] are assumed . the comparison results are shown in fig . [ compare1 ] . except for @xmath145 and @xmath146 , latency hiding outperforms the case without latency hiding in significant amount . the performance differences are even higher for @xmath147 . for the _ mpi_alltoall _ case , there is a sudden speed - up at @xmath148 . we are exploring this behavior further . we believe that it is caused by changes in the data traffic controlled by the hardware . overall , latency hiding provides a higher speed - up for larger tensor sizes and core count . from the general trends , it can be expected that two - fold or more efficiency improvement for @xmath130 greater than @xmath149 can be obtained by implementing latency hiding with our non - blocking adaptation of the all - to - all routine for tensor rotations . e. anderson , z. bai , c. bischof , s. blackford , j. demmel , j. dongarra , j. du croz , a. greenbaum , s. hammarling , a. mckenney , and d. sorensen , _ lapack users guide , 3rd . edition _ , ( society for industrial and applied mathematics , 1999 ) .

despite decades of study , type ia supernovae ( snia ) continue to be an active topic for astrophysical research . the accepted physical mechanism for these events , that they are the result of thermonuclear explosions of degenerate material ( hoyle & fowler 1960 ) in white dwarf stars , is very well established but owing to the considerable complexity of this process many aspects of these spectacular events remain poorly understood . observationally , it is clear that all snia are not the same there is significant diversity in both brightness and decay timescale ( phillips 1993 ) . aside from its direct relevance to the study of snia themselves , understanding this diversity has important implications to other branches of astrophysics in particular to cosmology where inferences about the properties of distant snia play an important role in probing the rate of expansion of the universe . some correlations between supernova properties are already well established from observations of nearby snia ( e.g. phillips 1993 ) . however , it is becoming clear that the observed diversity is not adequately described by a single parameter ( benetti et al . 2004 , 2005 ) and that unraveling this diversity relies on a combination of careful observational study and sophisticated theoretical modelling . in recent years , there has been rapid development in the sophistication of numerical modelling of the hydrodynamics of snia explosions . in particular , while in earlier work supernovae explosions were modelled using first one - dimensional ( 1d ) computer codes ( e.g. nomoto , thielemann & yokoi 1984 ; hflich , wheeler & thielemann 1998 ) and later two - dimensions ( e.g. mller & arnett 1986 ; niemeyer , hillebrandt , woosley 1996 ) , the most up - to - date simulations are fully three - dimensional ( 3d , e.g. reinecke , hillebrandt & niemeyer 2002 ; gamezo et al . 2003 ; rpke 2005 ; rpke et al . these multi - dimensional models are crucial for the understanding of realistic flame propagation and hence nucleosynthesis in snia explosions and have clearly demonstrated that the earliest 1d models greatly underestimate the likely complexity of real snia . this naturally raises the question of whether and to what extent this multi - dimensional complexity may be responsible for the observed diversity of explosions . to connect hydrodynamical explosion models and observations of real snia light curves or spectra requires modelling of radiation transport in the supernova . the majority of the light escaping from a supernova explosion at around optical maximum originates from energy deposited in the ejecta by the absorption and compton scattering of @xmath0-rays emitted by radioactive isotopes . given that all snia explosion models are significantly optically thick at the earliest times ( @xmath1 day ) , the emission , transport and deposition of radiation need only be followed for times after most of the complex dynamics have ceased and the ejecta is in near - homologous expansion ; this greatly simplifies the radiation transport calculation and allows it to be decoupled from the hydrodynamical simulation . the level of sophistication of modern radiation transport computations for supernovae has gradually developed to keep pace with the development of the hydrodynamical explosion models : for earlier applications ( e.g. branch et al . 1981 , lucy 1987 , mazzali & lucy 1993 ) simple one - dimensional calculations were sufficient but there is now growing interest in fully three - dimensional simulations . in particular , lucy ( 2005 ) has recently described and tested a sophisticated monte carlo approach to 3d , time - dependent radiation transport . this method holds much promise for a wide range of applications and is already being adopted and developed in several contexts for example , kasen , thomas & nugent ( 2006 ) have developed a code based on that described by lucy ( 2005 ) which incorporates a treatment of polarisation and lifts the assumption of grey opacity used in the earlier code . independently , maeda , mazzali & nomoto ( 2006 ) have also adopted lucy s techniques and utilized them in multi - dimensional simulations of light curves for core collapse supernovae . in this investigation , monte carlo calculations using methods similar to those of lucy ( 2005 ) are used to investigate aspects of the influence of the 3d structure of snia on their observable properties . in section 2 , the computer code used for the calculations is described . although similar to the codes described by lucy ( 2005 ) and kasen et al . ( 2006 ) , there are key differences , specifically in the means by which the observable light curves are extracted from the monte carlo simulation . this code is then used to investigate the effect of departures from spherical symmetry for two physically distinct cases . first , in section 3 toy models are used to investigate the influence of large scale ( low - mode ) asymmetries on observable light curves . secondly , a representation of a modern 3d explosion model is used to investigate the implications of the complex , relatively small - scale inhomogeneity predicted ( section 4 ) . the emphasis here is on identifying and interpreting differential effects between 3d and 1d radiative transfer and thus several simplifying assumptions are made in the treatment of the micro - physics . conclusions are drawn in section 5 . the calculations presented in this paper have been performed using a monte carlo code which follows the propagation of radiative energy in 3d as a function of time . this code is closely based on that described by lucy ( 2005 ) and thus only a brief summary of the operation of the code is given here , with particular emphasis on the the departures from the approach described by lucy ( 2005 ) . the nomenclature used by lucy for the monte carlo quanta ( radioactive pellets , @xmath0-packets , @xmath2-packets ) is adopted throughout . to obtain light curves for a pre - specified supernova model , the code undertakes the following steps . a computation domain is assigned which is large enough to encompass the physical extent of the model . this is divided into a number of grid cells each of which is assigned an initial mass - density based on the input supernova model . pellets of radioactive material are placed in these cells , again in accordance with the chosen model . the radiative decays of these pellets by emission of @xmath0-rays and the subsequent propagation and thermalisation of the @xmath0-packets is followed via a monte carlo calculation . a grey - opacity treatment is adopted for the propagation of radiation ( @xmath2-packets ) in all other wave - bands . during the monte carlo calculation , estimators are determined for various properties of the radiation field in every grid cell ( see below ) . no information is retained regarding the variations of these properties within individual grid cells . after the monte carlo simulations are complete , the estimators are used to determine observer frame emissivities . a formal solution of the radiative transfer equation is then performed using these emissivities to determine the emergent radiation field . the code currently produces two classes of output : `` bolometric '' ( ultraviolet - optical - infrared , uvoir)-light curves which are obtained from the behaviour of the @xmath2-packets , and @xmath0-ray spectra and light curves which are derived from the @xmath0-packets . for the applications described in sections 3 and 4 , only the @xmath2-packet light curves are needed , however the analogous means by which @xmath0-ray properties are obtained are also described here for future reference . following lucy ( 2005 ) , a regular cartesian grid which expands with time is used . however , we allow the grid to expand continuously rather than in discrete jumps at the end of a each time - step . since the supernova ejecta is assumed to be in homologous expansion , this adds little complexity to computing the propagation of the monte carlo quanta and removes the need to check whether they skip across boundaries due to the modification of the grid at the start of each time - step . in the limit of small time - steps , this modification has no effect on the results of the calculation . the code described by lucy ( 2005 ) obtained light curves by directly counting the number of quanta that escaped the computational domain during each time - step . however , as discussed by lucy ( 2005 ) , that approach is not the most efficient with regard to minimizing monte carlo noise . significantly higher quality spectra and light curves can be obtained by using the paths of the monte carlo quanta to compute estimators for the emissivities in the model which can then be used , _ post hoc _ , to obtain the intensity via a formal solution of the radiative transfer equation . methods using such an approach have previously been used and shown to be highly successful ( see e.g. lucy 1999 ) . for spherical models such as the test case used by lucy ( 2005 ) the method of counting packets can always be used reliably , its inefficiency being countered by the use of a large number of quanta . for complex 3d structures , however , this approach becomes unsatisfactory since the angular dependence of the escaping radiation field can only be addressed by angular binning of the quanta . this has further negative impact on the signal - to - noise of the computed spectra and rapidly becomes prohibitive if more than a few angle - bins are to be considered . in contrast , the use of emissivity estimators and a formal solution of the radiative transfer equation allow the spectra and light curves to be computed correctly for individual lines - of - sight to the supernova without introducing extra monte carlo noise . thus such methods are strongly favoured for the study of any models which depart from spherical symmetry . the implementation of these methods for the calculation of uvoir light - curves and @xmath0-ray spectra are described in the next two sub - sections . the light - curve for uvoir radiation as viewed by a distant observer in direction @xmath3 is given by : @xmath4 here , the integral is performed over the plane perpendicular to @xmath3 and @xmath5 is the emergent intensity of a ray which is destined to reach the observer at time @xmath6 . the intensity depends on @xmath7 , the impact - vector of the ray given by @xmath8 where @xmath2 is any position on the ray trajectory . @xmath9 is the luminosity the observer would imply were they to assume that the emission was isotropic . the true ( unobservable ) luminosity of the supernovae , @xmath10 , is given by @xmath11 this @xmath10 is the quantity obtained by directly summing all emerging packets in a monte carlo simulation . to evaluate equation 1 for a particular viewing direction ( @xmath3 ) at a particular time ( @xmath6 ) , a large sample of rays ( typically @xmath12 ) are chosen such that they cover the complete range of @xmath13 required by the size of the supernova ejecta . these rays are launched simultaneously from a plane perpendicular to ( @xmath3 ) behind the supernovae having initially zero intensity . the time of launch is chosen such that the ray with @xmath14 crosses the coordinate origin at time @xmath6 . for each ray , the emergent intensity is determined by numerically solving the radiative transfer equation along its trajectory : @xmath15 where @xmath16 is the element of pathlength along the ray while @xmath17 and @xmath18 are the observer frame , direction- , time- and position - dependent emissivity and opacity , respectively . since the trajectory is a light ray , the position and the time @xmath19 are related by @xmath20 . the emergent intensity , @xmath21 , is the value of @xmath22 at the point where the ray trajectory finally leaves the supernova ejecta . in all the calculations presented here , the uvoir opacity per unit density in a grid cell is assumed to be constant and isotropic in the co - moving frame . thus the observer frame opacity , @xmath23 can be readily calculated . in contrast , the emissivity @xmath24 is not known _ a priori _ and is obtained from a monte carlo simulation . in the current version of the code , there are two distinct uvoir emissivity source terms : one due to thermalisation of @xmath0-ray packets ( @xmath25 ) and one due to scattering of uvoir photon packets ( @xmath26 ) . following lucy ( 2005 ) , estimators for these emissivities in the co - moving frame for a particular grid cell during a particular time step are obtained from the monte carlo energy packet trajectories using : @xmath27 and @xmath28 in equation 4 , the summation runs over all the trajectories of @xmath0-ray packets within the cell ( which has volume @xmath29 ) during the time - step ( which has duration @xmath30 ) ; @xmath16 is the trajectory length and @xmath31 is the packet energy determined in the observer frame . the packet is travelling in direction @xmath32 and the velocity of the ejecta at the mid - point of the trajectory is @xmath33 . the co - moving frame @xmath0-ray thermalisation opacity ( @xmath34 ) is frequency - dependent and includes contributions from both compton scattering and photoelectric absorption . note that the compton term accounts only for the rate at which @xmath0-rays transfer energy to compton electrons ( which are assumed to thermalise in situ ) . in equation 5 , the summation is over uvoir packet trajectories and @xmath35 is obtained from the adopted uvoir opacity coefficient . it has been assumed here that the emissivity is isotropic in the co - moving frame and that terms @xmath36 and smaller can be neglected . note that although the co - moving frame emissivity ( @xmath37 ) is isotropic , the observer frame emissivity ( @xmath17 ) is not , owing to the doppler terms in the transformation between frames . the scheme described thus far is valid provided that the grid cells are individually optically thin . in practice , this condition is violated at early times when the ejecta is compact and dense . at such times , it becomes unacceptable to assign a uniform emissivity to each grid cell . this can be overcome , however , by weighting each contribution to the emissivities with a factor which accounts for the probability of energy absorbed and re - emitted during the related trajectory escaping to infinity . the weighting factor used here , which must be applied individually to each contributing pathlength included in the sums in equations 4 and 5 is @xmath38 where @xmath39 and @xmath40 is the total @xmath2-packet optical depth from the mid point of the trajectory @xmath16 to the edge of the supernova in the direction @xmath3 . this form of the weighting factor is valid provided that _ either _ the total optical depth across a grid cell is small _ or _ that the contributing pathlengths ( @xmath16 ) are all small compared to the typical length scale on which the physical properties ( e.g. mass density ) of the model vary . therefore , in calculations which make use of these weighted estimators , the pathlengths that energy packets can travel in a single step are not permitted to exceed a predetermined maximum step size , @xmath41 . for each time step in the calculation , @xmath41 is set to _ either _ one - tenth the width of a grid cell _ or _ the distance corresponding to @xmath42 in the densest grid cell ; the larger of these two values is chosen . furthermore , when the weighted emissivities are used , the opacity term in equation 3 must be neglected since the probability of the emitted energy escaping has already been addressed . this weighting overcomes the problem of having optically thick cells at early times but it has two significant drawbacks . first , the required computations of @xmath43 are time demanding and lead to a substantial increase in code execution time . secondly , since the weighting factor is angle dependent , separate sets of estimators are required for each viewing angle that is to be investigated ( in contrast , the unweighted estimators are independent of @xmath3 ) . therefore , this method should only be used for computing the light - curve at early times when the density in the grid cells is high and be replaced with the unweighted scheme once the opacity of individual cells has dropped sufficiently . to test the implementation of the method described above , the code has been applied to the test model used by lucy ( 2005 ) . this is a spherical model , based on that used by pinto & eastman ( 2000a ) , having total mass @xmath44 , @xmath45ni mass @xmath46 and a maximum velocity of @xmath47 km s@xmath48 . the distribution of ni is centrally peaked . following lucy ( 2005 ) , the grey absorption cross - section of 0.1 @xmath49 g@xmath48 is adopted for uvoir radiation , and the photoelectric absorption coefficient is taken from ambwani & sutherland ( 1988 ) , adopting a mean value of @xmath50 . a 100@xmath51 cartesian grid is adopted for this test and @xmath52 monte carlo packets were used in the calculations . the light curve obtained by lucy ( 2005 ) for this model is shown in figure 1 ( solid line ) . this agrees very well with the light curve computed here by directly counting the number of monte carlo packets escaping this model ( diamond symbols in figure 1 ) . it also matches well with the light - curve computed for a particular viewing angle using the weighted estimators described in section 2.4 ( the particular viewing direction was chosen randomly here ) . however , when the estimators are not weighted ( i.e. if equations 4 and 5 are used directly ) , the computed light curve is significantly overestimated at early times ( @xmath53 days , in this case ) . as discussed in section 2.4 , this is because the grid cells of the model are optically thick at early times ; the maximum optical depth across one grid cell is plotted , as a function of time , in figure 2 . this shows that significant errors arise if the unweighted estimators are used when this optical depth is @xmath54 . the @xmath0-ray spectrum is obtained following the same principles as that used for the uvoir light curve described above . the important difference being that the treatment of @xmath0-rays is fully frequency dependent . to deal with the frequency dependence , a grid of frequency points is used to divide the spectrum into small frequency intervals . one point in this grid is set to the rest frequency of each of the radioactive @xmath0-ray emission lines in the range of interest and the remaining points are spaced logarithmically between . to compute the spectrum , the same scheme of solving the radiative transfer equation along a set of rays trajectories through the model is used . here , however , the radiative transfer along each trajectory is computed multiple times , once for each frequency point in the frequency grid , thereby determining the emergent radiation field as a function of both frequency and time . as for the grey - computations described above , the opacity term in the radiative transfer equation is known ( the sum of compton and photoelectric terms ) . there are again two emissivity terms which need to be considered . the first , direct emission of @xmath0-rays by radioactive decay can also be expressed analytically for every grid cell in terms of the half - lives of the radioactive isotopes and their initial concentrations in the cell . the second emissivity term is due to compton down - scattering . the treatment of this term requires that both the angular- and frequency - dependence of the compton process be considered . it is determined via a set monte carlo estimators ( one per frequency interval per grid cell per time step ) ; the estimator for the frequency interval @xmath55 in the grid cell @xmath56 for timestep @xmath57 is given ( in the observer frame ) by @xmath58 where the sum runs over all @xmath0-ray trajectories which lie in cell @xmath56 ( having volume @xmath59 ) and have frequency appropriate for scattering into the frequency interval @xmath55 ( of width @xmath60 ) during the timestep @xmath57 ( which has duration @xmath61 ) . @xmath62 is the number density of target electrons in the cell , @xmath31 is the observer - frame energy of the @xmath0-ray packet and @xmath63 is the compton differential cross - section for scattering into the direction of the line - of - sight ( @xmath3 ) in the observer frame . this cross - section depends upon the angle between the trajectory and @xmath3 and is determined by applying the klein - nishina formula for the cross - section in the co - moving frame . in this section , two toy models of elliptical supernovae are used to investigate possible observational consequences of large scale asphericity in supernova explosions . such an investigation is motivated by observational evidence for global asphericity in snia obtained via polarimetry ( see e.g. howell et al . 2001 ; wang et al . the origin of this asphericity is not well known : mostly likely it is determined by the details of the explosion process itself but may have its roots in the properties of a rapidly rotating progenitor . here , however , the objective is not to gain insight to the physical origin of such a geometry but rather to study how it might effect both the amplitude and shape of observed light curves in comparison with spherical explosions . earlier calculations of radiation transport in elliptical supernovae have been discussed by hflich ( 1991 ) . in that work , a multi - dimensional monte carlo code was also used . however , the treatment of energy packet generation and emission was simplified via the use of a parameterised photosphere in contrast to the full treatment of @xmath0-transport and deposition employed here . a simple elliptical model has been constructed , closely related to the spherical model used as a test case in section 2 . the adopted model has the same total mass and same @xmath45ni mass as the spherical model . it is also assumed to be in homologous expansion and to have uniform density . however , the maximum velocity ( and hence spatial extent ) is taken to be smaller in the @xmath64-direction than in the @xmath65- and @xmath66-directions ( symmetry is still assumed under rotation about the @xmath64-axis ) . such a model is intended as a simple description for cases in which either the explosion mechanism or the properties of the progenitor lead to a large - scale ( low angular mode ) departure from sphericity . two particular realisations of the model will be considered here . for both the maximum velocity in the @xmath65- and @xmath66-directions was kept at 10@xmath67 cm s@xmath48 . the models differ in the chosen maximum velocity in the @xmath64-direction : for one , this velocity was fixed at @xmath68 cm s@xmath48 , thereby giving an axis ratio of 2:1 ; while for the other @xmath69 cm s@xmath48 was adopted to give an axis ratio of 5:4 . the axis ratio of 2:1 may be regarded as extreme it is comparable to the axis ratio that might be present in a very rapidly rotating progenitor ( see e.g. yoon & langer 2005 ) but there is little evidence to suggest that this would be preserved during an explosion . as an extreme case , however , this model is useful for providing a clear indication of the sense in which asphericity can affect the light curve . the second ratio adopted ( 5:4 ) is comparable to the @xmath70 per cent asphericity found by howell et al . ( 2001 ) for sn1999by and is therefore more likely to be indicative of the propertied of real snia explosions . as discussed in section 1 , the interest here is in probing the effect of a 3d treatment of the radiative transfer compared to spherically symmetric 1d calculations and so the simplifying assumption of a constant , grey - uvoir opacity is retained . for all the calculations discussed in the section , this grey - uvoir absorption cross - section remains fixed at @xmath71 @xmath49g@xmath48 . light - curves have been computed for observers viewing the ellipsoidal supernovae from infinity along both the major- and minor - axes . these were computed using the weighted - estimators method described in section 2.4 . the two light curves for the model with axis ratio 2:1 are plotted in figure 3 , along with the angle - averaged light curve for the same model ( i.e. the light curve obtained from the arithmetic mean of the light curves seen by a large number of observers from random viewing angles ) . when viewed down the short - axis , the light curve is considerably brighter than as observed down the long - axis . around maximum light , the difference in brightness is approximately a factor of 3.5 . at later times , as the supernova becomes less optically thick , the difference becomes smaller approximately a factor of 2.5 at around 50 days . the light curve peaks slightly earlier when viewed down the long axis ( at @xmath72 days compared to @xmath73 days if viewed down the short axis ) . also , the light curve decays more slowly if viewed down the long axis ; this is characterised by the bolometric @xmath74-values is the change in magnitude between maximum light and 15 days after maximum light . ] of 0.55 ( long - axis ) and 0.64 ( short - axis ) . qualitatively similar , but quantitatively smaller differences are seen in the light curves computed for different viewing angles using the model with the axis ratio 5:4 . for this model , representative light curves are show in figure 4 . the scale of the angular variation is determined by the interplay of several effects . consider viewing an opaque ellipsoid with axis ratio 2:1 and uniform surface brightness as characterised by uniform surface temperature . one would expect to find the flux to be twice as large when viewing along the short axis compared to the long axis , simply due to the increase in apparent surface area . the computed ratio exceeds this for two closely related reasons . first , the choice of uniform density in the models means that the column density to any particular 2:1 ellipsoidal surface within the model is always less when viewing down the short axis than the long . thus the optical depth to the surface is smallest along the short axis making the light curve brighter when viewed from this direction . secondly , contours of constant radiation density ( or equivalently emissivity ) do not exactly follow the 2:1 ellipsoidal geometry of the model ; in the outer regions there can be significant departures , always in the sense that the radiation energy density is highest at the points of intersection by the short axis . this is illustrated in figure 5 where the variation of the @xmath2-packet emissivity ( @xmath75 ) is shown along the ellipsoidal axes . a higher radiation energy density on the short axis makes the light curve brighter when viewed down that axis and so enhances the variation in brightness with viewing angle . the origin of this effect also lies in the lower column densities along the short axis . at late times , a quasi - static description of the radiation field becomes valid since packets are not trapped in the model for a significant number of timesteps . under such circumstances , the lower opacity along the short axis means that photons preferentially diffuse in that direction , making the energy density on an 2:1 ellipsoidal surface peak at the points of intersection with the short axis . at earlier times , this effect is enhanced by the time - dependent nature of the calculations ; fewer packets manage to reach outer ellipsoidal shells than would be predicted in a quasi - static description because they have had insufficient time to diffuse far enough . the difference in diffusion time means that this affects the energy density along the long axis more significantly than along the short axis and thus acts to enhance the effect expected from the quasi - static case . as time passes , the shapes of the contours of radiation energy density evolve slightly and the outward decline becomes somewhat less steep ( this can be seen by comparing the @xmath76 days and @xmath77 days results in figure 5 ) . however , throughout the time range considered here , a departure from 2:1 ellipticity remains . since both the effects described above are the result of angular variations in the optical depth to 2:1 ellipsoidal surfaces , both persist while the ejecta remains optically thick . they slowly decrease in strength during the decay phase as expansion causes the optical depths to drop ; at very late times the entire ejecta will become optically thin to uvoir radiation such that the light curve will become independent of viewing angle . for the model adopted here , however , this nebular phase will not begin until several hundred days later than the times considered here . the viewing - angle dependency of the light curves computed from these simple models may have some interesting ramifications for understanding the observed properties of snia light curves . as pointed out by wang et al . ( 2003 ) in the context of sn2001el , directional dependence of the luminosity as predicted for elliptical models of supernovae would lead to dispersion in the observed peak magnitudes ( based on the earlier work by hflich 1991 and their implied asphericity of @xmath78 per cent for sn2001el , they speculate that this dispersion would be around 0.1 mag ) . the results obtained here support this argument and indicate that if the degree of asphericity were larger in some cases ( e.g. sn 1999by ; howell et al . 2001 ) the spread in the peak magnitude could be greater , @xmath79 mag . furthermore , the full light curves computed here allow this dispersion relative to the known relationship between light curve shape and peak luminosity to be examined . this trend , the so - called `` phillips relation '' following phillips ( 1993 ) , expresses the negative correlation between peak brightness and the @xmath74-parameter measured from observed light curves of snia . for both the models considered here , the variation of @xmath74 with viewing angle is in the opposite sense to the standard relationship . this is illustrated in figure 6 where the six light curves shown in figures 3 and 4 are represented as points in the @xmath74-@xmath80 plane . the gradient of the standard phillips relation ( describing the mean observed relationship between @xmath74 and @xmath80 ) in the b - band is plotted for comparison in the figure . this effect would lead to a detectable scatter about the phillips relation and thus may have a major role to play in understanding the diversity of supernova observations : the results plotted in figure 6 would suggest that if snia explosions were moderately elliptical ( such that the model with axis ration of 5:4 were approximately applicable ) , viewing angle effects could explain a scatter of several tenths of a magnitude about the mean relationship . significant caution must be applied in interpreting this result since the grey - treatment adopted here does not allow band - limited light curves to be studied for direct comparison with observations quantitative differences may occur if the frequency - dependence of the opacity were taken into account . furthermore , the models used here have predicted an angular variation of the radiation energy density and thus , by implication , the temperature of the ejecta . such a variation further contradicts the used of a uniform opacity and highlights the need for the consideration of more detailed micro - physics . also , the particularly simple model chosen ( uniform density with centrally concentrated @xmath45ni and time - independent mean opacity ) produces significantly smaller absolute values of @xmath74 than are typically observed thus further work using more realistic models of aspherical supernovae are needed . modern snia explosion models predict complex , three - dimensional sub - structure ( e.g. reinecke et al . 2002 ; gamezo et al . 2003 ; rpke 2005 ; rpke et al . 2006 ) within the explosion . in contrast , most models used to compute light curves for comparison with observation have a smooth , one - dimensional density-/composition - profile . in this section , the effect of the predicted inhomogeneity on model light curves will be investigated . in principle , there are two classes of effect which are of interest here : first , inhomogeneity can change the viewing - angle averaged light curve compared to that of a spherically symmetric model . secondly , inhomogeneity could also lead to angular dependence of the light curve the degree of angular dependence may be reasonably expected to depend on a combination of the dynamic range of the inhomogeneity in density and the physical length scale of the variations . the length scale is relevant since if only small scale inhomogeneity is present it will tend to be averaged out when integrating over solid angle and so not introduce significant angular variation of the light curve . given that the code used here adopts several simplifying assumption most importantly , perhaps , that of a grey absorption coefficient for uvoir radiation the emphasis here is on understanding and assessing the differential effect of introducing 3d structure by comparing with an equivalent 1d model using a fixed set of well - understood approximations . such a calculation is an important first step in understanding the role of 3d structure and is a useful starting point for further work where the micro - physics is improved ( see section 5 for further discussion ) . the model used for this test calculation is based on the 3d explosion model computed by rpke ( 2005 ) . for this model , rpke ( 2005 ) followed the hydrodynamics of an exploding white dwarf star ( with total mass 1.4 @xmath81 ) for 10 seconds on a schematic grid of 512@xmath82 cartesian cells . only one spatial octant was simulated and symmetry under reflection was assumed to describe the remaining octants ( thus numerically only 256@xmath82 grid cells were used ) . the distribution of mass - density and mass - fraction of iron - group elements in each grid cell at the end of their simulation was made available for this work . light curves have previously been simulated from 1d representations of this model by blinnikov et al . ( 2006 ) . in order to make the calculations here tractable in terms of both computer memory required and photon statistics in each grid cell the model adopted here uses only 170@xmath51 grid cells . it was obtained from the 512@xmath51 grid by first removing the outermost cell from both ends of the grid in each of the three cartesian directions resulting in a 510@xmath51 grid . the mass in the cells removed in this process was negligible . the resolution was then reduced by a factor of 3 by subdividing the 510@xmath51 grid into 3@xmath51 blocks and replacing each block with a single cell whose density was equal to the mean density of the original 27 cells . given the grey treatment of uvoir radiation currently adopted in the code , it is not necessary to specify the detailed composition of the material in each grid cell . however , it is necessary to specify the initial distribution of @xmath45ni which provides the source of radiative energy . the hydrodynamics code used by rpke ( 2005 ) provides as estimate of the fractional mass of iron - group elements in each grid cell but does not give a reliable estimate of the breakdown of this material into specific isotopes and elements . the @xmath45ni mass - fractions used here were obtained by adopting a constant ratio for the mass of @xmath45ni to the total mass of iron - group elements in all grid cells . this ratio was fixed to yield a total @xmath45ni - mass of 0.28 @xmath81 , as derived by kozma et al . ( 2005 ) for this model ; note that the nucleosynthesis calculations described by kozma et al . ( 2005 ) do not produce compositional information in sufficient detail to reconstruct the full 3d - distribution of @xmath45ni - mass owing to the modest ( 27@xmath51 ) number of tracer particles they used in comparison to the number of grid cells in the 3d model . in order that the differential effect of the 3d structure could be assessed , a 1d comparison model was made by averaging the 3d model over spherical shells , taking care to conserve the radial distributions of total mass and @xmath45ni - mass . in order to illustrate the degree of inhomogeneity in the 3d model , figure 7 shows the velocity distribution of density in the spherically averaged model with points indicating the actual densities of individual grid cells in the 3d model . this shows that across most of the velocity range of the model , there is a spread in density of at least a factor of three between different cells with similar velocities . this is comparable to the dynamic range of density for a given velocity in the original hydrodynamical model ; thus one may be confident that the model used here contains a fairly reliable representation of the inhomogeneity implied by the hydrodynamics . we continue to adopt a grey uvoir - absorption cross - section . however , to correctly evaluate the influence of 3d structure on the light curve , it is necessary to consider that compositional inhomogeneity would cause the cross - section per gram to be a function of position . it goes beyond the scope of this paper to undertake full calculations of the composition dependence of the opacity ; instead , a simple one - parameter description of the opacity is adopted , following mazzali & podsiadolwski ( 2006 ) . they consider the opacity to be determined by the mass fraction of iron - group elements on the assumption that the opacity per gram is a factor of ten higher for the iron - group than for lighter elements . adopting the same assumption , the uvoir - absorption cross - section per gram used in this section is given by @xmath83\ ] ] where @xmath84 is the mass fraction of iron - group elements , which varies from cell to cell , and the normalisation factor @xmath85 is chosen such that the mean value of @xmath86 in the ejecta is fixed to @xmath87 @xmath49 g@xmath48 . although crude , this parameterisation captures the essential physics that the heavy elements dominate the mean opacity and only requires the compositional information which is directly available from the explosion model ( namely , the total iron - group mass ) . note that , for ease of comparison with the constant-@xmath86 calculations used in the earlier sections of this paper , the time - dependence imposed on @xmath86 by mazzali & podsiadolwski ( 2006 ) is not included here . in view of the simplifications used in terms of both numerical resolution and particularly the very simple treatment of opacity the results obtained here should not be regarded as a definitive prediction of the radiation properties of the rpke et al . ( 2005 ) explosion model . rather , the emphasis is on using a model which has a reliable representation of the degree of inhomogeneity in a real explosion model to understand the role played by the complex structure in the radiation transport . the light curve obtained with the 3d model described above is shown as the solid histogram in figure 8 ; this is the light curve averaged over viewing angles . at peak , this light curve is dimmer than those plotted in figures 1 , 3 and 4 ; a consequence of the lower @xmath45ni - mass in the present model . the light curve peaks earlier ( at @xmath88 days ) as a result of differences in the distribution of @xmath45ni with velocity : in the model used here , the @xmath45ni - distribution is less centrally concentrated than that adopted in sections 2.5 and 3 . to establish the influence of the inhomogeneity on the light curve , the 1d model obtained by spherically averaging the 3d model ( see above ) was also used to compute a light curve . this light curve is shown by the dashed curve in figure 8 . from the figure , it is apparent that the inhomogeneity causes the light curve to be a little brighter at early times and very slightly fainter during the decay phase . this occurs because in the complex 3d medium a fraction of the @xmath2-packets find trajectories that preferentially pass through lower density cells and therefore encounter lower opacities than is possible in a spherically symmetric model with the same total mass . this allows some @xmath2-packets to escape more quickly making the light curve brighter during the pre - maximum rise phase . it also means that slightly fewer packets remain trapped in the ejecta which leads to the marginal dimming in the post - maximum phase . the scale of the effect in this model is rather modest : at most it amounts to a difference of @xmath89 mag and this occurs only at very early times , @xmath78 days before maximum light . at around maximum light , the effect is limited to a change in magnitude of around @xmath90 mag ( 5 per cent in luminosity ) . in accordance with `` arnett s rule '' ( arnett 1982 ; arnett , branch & wheeler 1985 ) which states that the emitted bolometric luminosity is roughly equal to the instantaneous rate of generation of radioactive luminosity at maximum light the increase in peak luminosity resulting from the 3d structure is accompanied by a faster rise to maximum ; the peak luminosity occurs approximately one day earlier in the 3d calculations than the 1d case . the 3d model was also used to look for viewing - angle dependency of the light curve . light curves were extracted for a set of eight randomly chosen viewing angles and compared to the angle - averaged light curve . no significant ( @xmath91 per cent ) departures from the average light curve were found ; this can be understood since large scale departures from sphericity are not present in this model , partially as a result of the assumed reflection symmetries ( see above ) . it is interesting to note that the results from the model considered here show that inhomogeneity would act in the sense of pushing model results closer to observation ; for example , a recent comparison of light curves computed from 1d models with observations ( blinnikov et al . 2006 ) does suggest that the 1d models produce light curves which rise too slowly and underestimate peak luminosities ( see their figures 8 13 ) . direct comparisons with the calculations of blinnikov et al . ( 2006 ) are not possible here owing to the considerably more complex treatment of opacity that they adopt , but it may reasonably be speculated that their rise times would also be shorter if multi - dimensional effects were incorporated . the scale of the effect here is rather too small to directly have major implications for the confrontation of models and observations . however , the effect is not completely negligible and it is plausible that the calculations presented here may underestimate the true scale of this phenomenon ; in particular , both the reduction in resolution from that used in the explosion model and the grossly simplified one - parameter treatment of the opacity may both suppress 3d density and composition effects that would be present in a complete treatment . furthermore , only one explosion model has been considered here and this model is constructed from a simulation describing only one octant ( see above ) . in the future , it will be necessary to examine a range of models including fully 3d models where the greater complexity may enhance the effects . we have described and tested a new 3d , time - dependent monte carlo code for modelling radiation transport in snia . the code adopts the methods presented by lucy ( 2005 ) and also incorporates a scheme which uses monte carlo radiation field estimators to allow observables to be extracted for specific viewing angles ; this approach helps to suppress monte carlo noise and eliminates the need to obtain intensities by angular binning of emergent monte carlo packets . this code has been used to investigate two classes of three dimensional effects in snia models . first , two elliptical snia toy models were used to investigate how large scale asphericity might alter observable light curves . as expected from previous simplified treatments ( e.g. hflich 1991 ) , it was found that light curves were brighter when viewed down the minor - axis than the major - axis . the brightness enhancement is largest at early times and becomes smaller during the decline phase as the supernova ejecta becomes less optically thick . for a model with axis ratio comparable to that suggested by polarisation data for real snia ( howell et al . 2001 ; wang et al . 2003 ) , the differences in peak brightness and light curve shapes between viewing angles is detectable and in principle could play a role in the observed scatter about the mean relationship between light curve brightness and width . secondly , a model with structure based on the results of recent three - dimensional hydrodynamical simulations was used to study the effect of inhomogeneities in both density and composition on light curves . it was found that 3d structure could lead to light curves which are both brighter at early times and which peaked sooner after the explosion that is found from 1d models . for the particular model considered , the effect was rather modest ( @xmath92 5 per cent around maximum light ) but study of other explosion models is required to quantify the possible diversity in this effect . considerable further work is required to fully understand the role of 3d effects on radiation transport in snia . the use of the grey approximation in the treatment of uvoir radiation is currently the greatest limiting factor to the practical applicability of these results . in calculations involving a uniform grey absorption coefficient , the line - of - sight opacity depends solely on the total column density . under realistic conditions for snia ejecta , the opacity is dominated by spectral lines ( see e.g. pinto & eastman 2000b ) and is thus a strong function of frequency , velocity gradient , composition and the ionization / excitation state of the plasma . indeed , when strong lines dominate , the opacity has little direct sensitivity to the column density and is instead mostly a function of the density of spectral lines in frequency space . furthermore , photon escape from highly opaque material is facilitated by frequency redistribution to those regions of the spectrum where there are relatively few spectral lines ( see e.g. pinto & eastman 2000b for a detailed discussion ) ; in the grey calculations this means of escape is not available and all the energy packets must burrow through the imposed opacity . given this , it is likely that the grey approximation overestimates the role of geometry in determining photon propagation . however , further calculations are required to determine and understand this in detail . in the context of the inhomogeneous model discussion in section 4 , a more realistic treatment of the role played by compositional inhomogeneity would be of particular interest ; only a crude attempt at describing the variation of opacity with composition has been used here ( equation 8) and a sophisticated treatment involving the very real differences between the atomic properties of different elements may lead to interesting effects . fortunately , as discussed by lucy ( 2005 ) , the monte carlo method is well suited to incorporating realistic physics in a complex geometry which will allow the interplay of more detailed micro - physics and 3d structure to be studied in the near future ; this would involve a more sophisticated treatment of the opacity , perhaps similar to the methods employed by blinnikov et al . ( 2006 ) or kasen et al . from such a calculations it would be possible to extract not only a bolometric light curve , but also light curves for specific photometric bands , an important step in the progression towards confrontation of theory and observation . light curve calculations also need to be performed for a wider range of explosion models than the exploratory set discussed here . this is necessary in order to quantify how the various effects might differ in scale depending on various properties of the explosion and so help to understand what role they might play in establishing both the average properties and diversity of sn observations . eventually , it would also be valuable to undertake a similar assessment of the role of 3d effects in full nlte modelling of uvoir - spectra ; much of the greatest diagnostic power lies in spectral modelling including both intensity and polarimetry and thus it is important that commonly used modelling assumptions , such as that of spherical symmetry , are studied in the context of modern explosion models . i am very grateful to f. rpke for making available the data used for the trial 3d model described in section 4 and for advice relating to hydrodynamical models and their interpretation ; to w. hillebrandt for useful comments and suggestions relating to various aspects of this work ; to l. lucy for making available the numerical results of his test light curve computations ( used for figure 1 ) ; and to d. sauer and r. kotak for helpful discussions . i also thank the referee , p. pinto , for his careful reading of this manuscript and constructive comments and suggestions . arnett w. d. , 1982 , apj , 253 , 785 + arnett w. d. , branch d. , wheeler j. c. , 1985 , nature , 314 , + 337 + ambwani k. , sutherland p. , 1988 , apj , 325 , 820 + benetti s. , et al . , 2004 , mnras , 348 , 261 + benetti s. , et al . , 2005 , apj , 623 , 1011 + blinnikov s. i. , rpke f. k. , sorokina e. i. , gieseler m. , + reinecke m. , travaglio c. , hillebrandt w. , stritzinger + m. , 2006 , a&a , 453 , 229 + branch d. , falk s. w. , uomoto a. k. , wills b. j. , mccall + m. l. , rybski p. , 1981 , apj , 244 , 780 + branch d , 1992 , apj , 392 , 35 + gamezo v. n. , khokhlov a. m. , oran e. s. , chtchelkanova + a. y. , rosenberg r. o. , 2003 , science , 299 , 77 + hflich p. , 1991 , a&a , 246 , 481 + hflich p. , wheeler j. c. , thielemann f. k. , 1998 , apj , + 495 , 617 + howell d. a. , hflich p. , wang l. , wheeler j. c. , 2001 , + apj , 556 , 302 + hoyle f. , fowler w. a. , 1960 , apj , 132 , 565 + kasen d. , thomas r. c. , nugent p. , 2006 , apj , 651 , 366 + kozma c. , fransson c. , hillebrandt w. , travaglio c. , + sollerman j. , reinecke m. , rpke f. k. , spyromilio j. , + 2005 , a&a , 437 , 983 + lucy l. b. , 1999 , a&a , 345 , 211 + lucy l. b. , 1987 , in eso workshop on the sn 1987a , ed . + i. j. danziger , 417 , european southern observatory , + garching + lucy l. b. , 2005 , a&a , 429 , 19 + maeda k. , mazzali p. a. , nomoto k. , 2006 , apj , 645 , 1331 + mazzali p. a. , lucy l. b. , 1993 , a&a , 279 , 447 + mazzali p. a. , podsiadlowski ph . , 2006 , mnras , 369 , l19 + niemeyer j. c. , hillebrandt w. , woosley s. e. , 1996 , apj , + 471 , 903 + nomoto k. , thielemann f .- k . , yokoi k. , 1984 , apj , 286 , + 644 + mller e. , arnett w. d. , 1986 , apj , 307 , 619 + phillips m. m. , 1993 , apj , 413 , l105 + pinto p. a. , eastman r. g. , 2000a , apj , 530 , 744 + pinto p. a. , eastman r. g. , 2000b , apj , 530 , 757 + reinecke m. , hillebrandt w. , niemeyer j. c. , 2002 , a&a , + 391 , 1167 + rpke f. k. , 2005 , a&a , 432 , 969 + rpke f. k. , gieseler m. , reinecke m. , travaglio c. , + hillebrandt w. , 2006 , a&a , 453 , 203 + wang l. , et al . 2003 , apj , 591 , 1110 + yoon s .- c . , langer n. , 2005 , a&a , 435 , 967 + [ lastpage ]

core collapse supernovae are dominated by neutrinos , therfore many supernova properties may depend on the nature of neutrino - nucleon interactions . this provides an opportunity to study characteristic symmetries and features of the standard model weak interactions . supernovae may provide macroscopic manifestations of charge conjugation @xcite or parity violation @xcite . in the laboratory at high energies , antineutrino - nucleon cross sections are systematically smaller than neutrino - nucleon cross sections . this is related to charge conjugation and parity violation in the standard model . nevertheless , most core collapse supernova simulations still use the _ same _ lowest order cross section @xmath6 for both neutrinos and antineutrinos , @xmath7 . \label{1}\ ] ] here @xmath8 is the ( incoming ) neutrino energy , @xmath9 the scattering angle and the vector @xmath10 and axial vector @xmath11 coupling constants are listed in table i. in this paper , we discuss free space corrections to eq . ( [ 1 ] ) from nucleon recoil , weak magnetism , strange quarks and single nucleon form factors . these corrections may be important for supernova simulations because they are present at all densities . in contrast , the density dependent nucleon correlations discussed by @xcite and @xcite are important only at very high densities , well inside the neutrino sphere . electron antineutrinos were detected from supernova 1987a via their capture on protons . weak magnetism reduces this cross section and impacts the deduced neutron star binding energy and antineutrino temperature . weak magnetism also changes the rate of antineutrino capture in the neutrino driven wind above a protoneutron star . this may change the electron fraction @xmath12 ( ratio of electrons or protons to baryons ) and nucleosynthesis yields in the wind @xcite . however , to confirm this change in @xmath12 , one should perform supernova simulations including weak magnetism because this may change the neutrino luminosities and emitted spectra which will also impact @xmath12 . in this paper , we present simple formulae so that these corrections can be incorporated in simulations . we also give estimates of the change in emitted electron antineutrino spectrum and the differences expected between the spectrum of mu or tau antineutrinos and neutrinos . the reduction in opacities from recoil and weak magnetism should increase luminosities of both neutrinos and antineutrinos as more energy is transported to the neutrino sphere . this could increase heating behind the shock and more then compensate for the smaller antineutrino absorption cross section . weak magnetism should also reduce the cooling from positron capture on neutrons further increasing the net heating behind the shock . these changes in heating could make a simulation more likely to explode . finally , weak magnetism allows mu and tau antineutrinos to escape the star faster than mu and tau neutrinos . this should lead to a large muon number ( number of mu neutrinos minus antineutrinos ) or tau number of up to @xmath13 for the hot protoneutron star @xcite . this is accompanied by a nonzero chemical potential for mu and tau neutrinos . accurate treatments of neutrino transport are now feasible based on the boltzmann equation @xcite . therefore , it is important to systematically improve neutrino opacities . in section ii , we catalog a number of possible opacity corrections . in section iii , we present exact neutrino - nucleon cross sections for both charged and neutral currents . these are accurate to all orders in @xmath14 where @xmath8 is the neutrino energy and @xmath15 the nucleon mass . we also present corrections from single nucleon form factors and strange quark contributions in the nucleon . in section iv we discuss recoil and weak magnetism corrections to mu and tau neutrino , energy and lepton number fluxes and spectra . section v discusses corrections for electron antineutrinos . we also discuss possible changes in electron fraction @xmath12 and heating rates . finally , we conclude in section vi that these weak magnetism and recoil corrections should be incorporated in future supernova simulations . recently , accurate numerical algorithms have been developed to solve the boltzmann equation for neutrino transport @xcite . the high accuracy of these simulations now warrants systematic improvements in the neutrino opacities . therefore , we discuss a number of opacity corrections . the corrections listed in table ii can be divided into two main groups . numbers 1 through 5 are classified as model independent because they can be calculated exactly and are independent of the model used to describe dense matter . in contrast , corrections 6 through 13 are model dependent . not only do these corrections depend on the model , but it is important that these corrections be consistent with the equation of state . corrections 1 , phase space , and 2 , matrix element are perhaps the simplest . these are the only ones that are present in the limit of very low densities . they represent more accurate calculations of free space neutrino - nucleon scattering . by phase space we mean corrections from the @xmath16 value of the reaction such as the neutron - proton mass difference or because the outgoing neutrino momentum is slightly different from the incoming momentum . in this paper we focus on number 2 , corrections to the matrix element . recoil corrections , 2a , arise because the nucleon is not infinity heavy and recoils slightly from the neutrino . weak magnetism , 2b , arise from the parity violating interference between the weak magnetic moment of a nucleon and its axial current . weak magnetism is important because it has opposite sign for neutrinos and antineutrinos . thus weak magnetism increases the opacity for neutrinos and decreases the opacity for antineutrinos . the first effects of the dense medium are 3 , pauli blocking of some outgoing nucleon or electron states , and 4 , the fermi or thermal motion of the initial nucleon or electron . these can be calculated exactly . the final model independent correction in table ii is 5 , coulomb corrections . if heavy nuclei are present , the opacity may be dominated by coherent elastic scattering from nuclei . however , these nuclei have strong coulomb interactions and form a strongly correlated classical liquid . the correlations between ions can greatly reduce the opacity for low energy neutrinos that have wavelengths comparable to the distance between ions . this correction is model independent because the coulomb interaction is known . for example , ref . @xcite calculated the exact static structure factor of the ions with a simple monte carlo simulation . in principle , coulomb correlations are also present for neutrino electron scattering @xcite . corrections 613 depend on the model used for the strong interactions between nucleons . the mean field approximation , 6 , is perhaps the simplest way to treat the interactions . nonrelativistic mean field models incorporate mean field effects in the density of states with an effective mass . relativistic mean field models have both lorentz scalar and vector mean fields and the scalar field changes the effective dirac mass . it is important that the opacities be consistent with the model used for the equation of state ( eos ) . this can be achieved by using linear response theory @xcite to calculate the response of the medium to a neutrino probe . the random phase approximation ( rpa ) or sum of ring diagrams provides the linear response of a mean field theory ground state . therefore , if one uses a mean field eos and calculates the opacities in rpa , where the rpa effective interaction is the same as that used for the eos , then the opacities and eos will be consistent . for example , the relativistic rpa opacities of ref . @xcite are consistent with their mean field eos . we note that the nonrelativistic calculations of burrows and sawyer @xcite include corrections 1 , 37 however they do not contain any matrix element correction , 2 . likewise the relativistic rpa calculations of @xcite contain corrections 1 , 37 and 2a recoil but do not contain most of 2b weak magnetism . in the present paper we discuss weak magnetism corrections . in later work we will incorporate these into full nonrelativistic and relativistic rpa calculations . there are many @xmath17 correlations beyond a simple rpa approximation , 8 , for example those responsible for @xmath18 @xcite . in addition there are meson exchange current corrections , 9 . the stony brook group has done some work calculating corrections from hyperons or other strange hadron components , 10 , @xcite . more work could be done calculating opacities for meson condensates or other exotic phases of dense matter , 11 . if superfluid or superconducting phases are present there will be pairing corrections to the opacity , 12 . these can be important for the late time cooling of a neutron star @xcite or perhaps if there is a color superconducting phase @xcite . finally there could be important corrections to opacities in nonuniform matter , 13 . for example , in the low density pasta phases present in a neutron star inner crust or for a nonuniform meson condensate@xcite . all of the above assumed magnetic fields do not play important roles . if magnetic fields are important , they could substantially complicate the opacity , 14 , see for example @xcite . magnetic fields might be important in jets or to break the spatial symmetry to explain neutron star kicks @xcite . to conclude this section , now that accurate neutrino transport is possible , it is important to systematically improve the neutrino opacities . one should include weak magnetism and other free space corrections along with the many density dependent effects . furthermore , these free space corrections can be included exactly and without any model dependence . in this section we discuss recoil , weak magnetism , nucleon form factor , and strange quark corrections to the zeroth order cross section , eq . . then we present differential , transport and total cross sections that are exact to all orders in the neutrino energy over the nucleon mass @xmath14 . perhaps the simplest correction comes from the phase space of the scattered lepton because its energy @xmath19 is less than the incident energy @xmath8 . if we use full relativistic kinematics for the nucleon and we assume it is originally at rest , then , @xmath20 \equiv \eta k , \label{2}\ ] ] @xmath21^{-1 } , \label{3}\ ] ] with @xmath22 . we expect corrections to depend on the small parameter , @xmath23 to the first order in @xmath24 , @xmath25 . \label{5}\ ] ] if one uses the phase space with @xmath19 but still uses the matrix element for @xmath26 then eq . ( [ 1 ] ) becomes , @xmath27 , \label{6}\ ] ] @xmath28 @xmath29 . \label{8}\\\ ] ] we note that burrows and sawyer s work @xcite in the limit of low densities reduces to @xmath30 , see appendix i. if one evaluates the matrix element to first order in e and includes the phase space corrections one has , @xmath31 . \label{9}\ ] ] we note that the full recoil correction in eq . ( [ 9 ] ) is 50% larger than that for @xmath30 in eq . ( [ 8 ] ) . equation ( 9 ) still neglects the parity violating interference between axial and vector currents . this interference is dominated by weak magnetism . the full weak vector current @xmath32 has dirac @xmath33 and pauli or tensor @xmath34 contributions , @xmath35 where @xmath36 is the momentum transfered to the nucleon . the weak current @xmath32 is related by cvc ( conserved vector current ) to the electromagnetic current . therefore the large anomalous magnetic moments of the proton and neutron give rise to a large @xmath37 . the couplings @xmath10 and @xmath37 are collected in table i. the cross section to the first order in @xmath24 is , @xmath38 , \label{11}\ ] ] with plus sign for @xmath39 and minus sign for @xmath40 . weak magnetism increases @xmath39 and decreases @xmath40 , while recoil decreases both @xmath39 and @xmath40 cross sections . therefore weak magnetism and recoil corrections approximately cancel for @xmath39 but add for @xmath40 . note , burrows et al . @xcite , see also for example @xcite , include weak magnetism for the charged currents but neglect it for neutral currents . equation ( 11 ) is a good approximation for @xmath39 . however it can become negative for large @xmath40 energies . to cure this problem we include corrections to all orders in @xmath24 . the exact cross section can be calculated , see also @xcite , @xmath41 + c_a^2 \left [ 1+\eta^2 + \eta ( 1-x ) \right ] \pm 2c_a ( c_v+f_2 ) ( 1-\eta^2)\nonumber \\ \left . + \eta^2 { e^2\over 2 } ( 1-x ) \left[f^2_2 ( 3-x ) + 4c_v f_2 ( 1-x ) \right ] \right\ } . \label{12}\end{aligned}\ ] ] in general the @xmath42 and @xmath43 terms are small because they only enter at order @xmath44 . however , their inclusion is important to ensure a positive cross section at high energies . note , schinde considers the @xmath45 term but neglected all @xmath37 contributions @xcite . the finite size of the nucleon can be included in eq . ( [ 12 ] ) by using appropriate form factors , @xmath46 that are functions of the four momentum transfer squared , @xmath47 for example , for @xmath48 , @xmath49 with @xmath50 1.26 and the axial mass @xmath51 1 gev . form factors only enter at order @xmath44 . they reduce the cross section by about 10% at @xmath52 , and even less at lower energies . however , for completeness we include simple parameterizations of the form factors in appendix ii . because the corrections are so small , we will ignore the form factors in the rest of this paper . the opacity for @xmath53 from charge current reactions depends on the total cross section for @xmath54 , @xmath55 the zeroth order result is , @xmath56 while the correction factor from just phase space is , @xmath57 @xmath58+c_a^2 \left [ { 2e-1 \over e(1 + 2e ) } + { 1\over 2e^2 } ln ( 1 + 2e)\right]\right\ } / ( c_v^2 + 3c_a^2 ) , \label{19}\ ] ] @xmath59 the exact result is , @xmath60 @xmath61 , \label{22}\end{aligned}\ ] ] @xmath62 for @xmath3 using table i this is , @xmath63 while for @xmath53 we have , @xmath64 we note the main effect of weak magnetism and recoil is to reduce the @xmath65 opacity by the large amount in eq . ( [ 24 ] ) . for a typical neutrino energy @xmath66 20 mev this is a 15% reduction . the full correction factor @xmath67 , is plotted in fig . [ fig1 ] along with the lowest order forms in eqs . ( [ 24],[25 ] ) . we see that the lowest order is fine for neutrinos but fails for antineutrinos above about 50 mev . for mu and tau neutrinos the scattering opacity depends on the transport cross section , @xmath68 note , strictly speaking this is true only in the limit @xmath69 nevertheless eq . ( [ 26 ] ) provides a good estimate when @xmath70 . the zeroth order transport cross section is , @xmath71 the transport cross section including only phase space corrections is , @xmath72 @xmath73 + c^2_a \left [ { 2e + \left(2e^2 -e-1\right ) ln \left(1 + 2e\right ) \over e^3 \left(1 + 2e\right ) } \right ] \right\ } \nonumber \\ / \left [ { 2\over 3 } \left ( c_v^2 + 5c_a^2 \right)\right ] , \label{29}\end{aligned}\ ] ] @xmath74 finally the exact transport cross section is @xmath75 @xmath76 \right . \nonumber \\ + c_a^2 \left [ { 1+e\over 2e^3 } ln ( 1 + 2e ) - { 10e^3 + 27e^2 + 18e + 3 \over 3e^2 ( 1 + 2e)^3}\right ] \nonumber \\ \pm \left(c_v + f_2\right ) c_a \left [ { 1\over e^2 } ln \left ( 1 + 2e\right ) - { 2 + 10e+ { 28\over 3}e^2 \over e ( 1 + 2e)^3 } \right ] \nonumber \\ + c_v f_2 \left[{1\over e^2}ln ( 1 + 2e ) -{2\over 3 } \left ( { 3 + 15e + 22e^2\over e ( 1 + 2e)^3}\right)\right]\nonumber \\ \left . + f^2_2 \left[{1\over 4e^2 } ln ( 1 + 2e ) + { 8e^3 -22e^2 - 15e-3 \over 6e ( 1 + 2e)^3 } \right ] \right\ } / \nonumber \\ \left [ { 2\over 3 } \left(c^2_v + 5c_a^2 \right ) \right ] , \label{32}\end{aligned}\ ] ] @xmath77 the full correction , eq . ( [ 32 ] ) , is plotted in fig . ( [ fig2 ] ) . corrections for @xmath78 are very close to those for @xmath79 elastic scattering . we also plot the lowest order result , eq . ( [ 33 ] ) . this fails for antineutrinos at high energies . neutrino nucleon elastic scattering is sensitive to possible strange quark contributions in the nucleon . strange quarks do not contribute to @xmath80 because the nucleon has no net strangeness . however both @xmath81 and @xmath82 can be modified . parity violating electron scattering constrains @xmath37 @xcite . in this section we concentrate on @xmath11 because @xmath37 only contributes at order @xmath14 . strange quark contributions are assumed to be isoscalar while the dominant contribution to @xmath11 is isovector . therefore , strange quarks are expected to increase the cross section for @xmath79 scattering and decrease that for @xmath78 . we write , @xmath83 with @xmath84 and the plus sign for @xmath79 and the minus sign for @xmath78 scattering . the strange quark contribution @xmath85 is expected to be negative or zero . the best limit comes from a brookhaven @xmath79 scattering experiment @xcite . we assume , @xmath86 a better measurement of @xmath85 from a laboratory experiment would be very useful . we make a simple estimate of strange quark effects starting from the zeroth order transport cross section in eq . ( [ 27 ] ) . we consider the neutral current opacity for a mixture of neutrons and protons of electron fraction @xmath12 . for simplicity we set @xmath87 for @xmath79 scattering . the ratio of the opacity with @xmath85 to that with @xmath88 is , @xmath89 + 10 g_a g_a^s(1 - 2y_e)+1-y_e\over 5g_a^2 + 1-y_e}.\ ] ] for @xmath90 , @xmath91 represents a 21% reduction in the opacity at @xmath92 or a 15% reduction at @xmath93 . thus strange quarks could reduce opacities by 10 to 20% . note , strange quarks do not contributed to charged current interactions . in this section , we use the weak magnetism and recoil corrections for transport cross sections , eq . ( [ 32 ] ) to discuss muon and tau neutrino properties in supernovae . first , we examine muon and tau lepton number currents . because neutrinos now have shorter mean free paths than antineutrinos , there will be net @xmath94 and @xmath95 densities and nonzero @xmath94 and @xmath95 chemical potentials . we then calculate the increase in the @xmath94 and @xmath95 energy flux because of weak magnetism and recoil . this should increase the neutrino luminosities . finally we examine corrections to the emitted spectra of @xmath94 , @xmath95 and @xmath96 @xmath97 . we show that the @xmath98 are expected to be hotter than @xmath99 because of weak magnetism . for simplicity , we neglect charged muons . these could play some role at very high densities and or temperatures . without muons , our results are identical for muon and tau neutrinos . therefore , we write @xmath5 in this section where @xmath100 could be @xmath101 or @xmath2 . we start by considering high densities , well inside the neutrino sphere , where a simple diffusion approximation is valid . earlier work @xcite calculated the lepton number current to first order in @xmath102 . we extend this work to all orders . the transport mean free path for a @xmath5 of energy @xmath8 is , @xmath103 with the zeroth order mean free path @xmath104 evaluated at an arbitrary reference energy @xmath105 , @xmath106^{-1 } , \label{35}\ ] ] and @xmath107 is the neutron number density . note , for simplicity we assume pure neutron matter , @xmath108 . because the correction factor @xmath109 , eq . ( [ 32 ] ) , is so similar for @xmath110 and @xmath111 scattering our results are not expected to change much for @xmath112 . the lepton number current for @xmath113 in a simple diffusion approximation is , @xmath114}. \label{36}\ ] ] here @xmath115 is the neutrino chemical potential . for @xmath4 we have , @xmath116 this is now longer then @xmath117 because of the different weak magnetism correction factors @xmath118 . the @xmath4 lepton number current has the same form as eq . ( [ 36 ] ) with @xmath115 replaced by @xmath119 , @xmath120 } . \label{38}\ ] ] the total lepton number current is , @xmath121 } } \right . \left . - \int^\infty_0 { dk \over r^t_{\bar\nu } } { 1 \over 1 + { \rm exp } \left [ ( k+\mu_{\nu_{x } } ) /t \right ] } \right ] . \label{39}\end{aligned}\ ] ] if @xmath122 the lepton number of the star ( number of @xmath5 - number of @xmath4 ) will rapidly change . this buildup in lepton number gives rise to a nonzero chemical potential @xmath115 . reference @xcite argued that the system rapidly reaches steady state equilibrium where @xmath123 . we numerically solve for the chemical potential @xmath115 so that , @xmath124 this is shown in fig . [ fig3 ] as a function of @xmath125 . this chemical potential insures @xmath126 . if we expand eq . ( [ 40 ] ) to first order in @xmath127 and use @xmath128 and @xmath129 expanded to the first order in @xmath14 one reproduces the lowest order chemical pot @xmath130 of ref . @xcite , @xmath131 with @xmath132 for @xmath111 scattering . the lowest order @xmath130 is seen in fig . [ fig3 ] to be an excellent approximation to the exact result even at @xmath133 mev . note , this nonzero @xmath115 can lead to muon number ( number of @xmath94 minus number of @xmath96 ) or tau number for the protoneutron star as large as @xmath13 @xcite . we calculate the energy flux carried by @xmath5 by multiplying the integrand in eq . ( [ 36 ] ) by @xmath8 , @xmath134 we use eq . ( [ 41 ] ) for @xmath135 with @xmath136 and add the @xmath4 energy flux . we assume the temperature @xmath125 has only a radial dependence so that , @xmath137 . \label{43 } \end{aligned}\ ] ] in the absence of weak magnetism and recoil corrections , @xmath138 and the energy flux becomes , @xmath139 in fig . [ fig4 ] we plot the ratio of the full energy current with weak magnetism and recoil to the zeroth order result @xmath140 . _ we see that weak magnetism and recoil substantially enhances the energy flux . _ this should raise the neutrino luminosity of the protoneutron star . at a temperature of 30 mev the enhancement is over 50% . many supernova simulations combine @xmath5 and @xmath4 into one effective species . in this approximation @xmath141 . therefore , we evaluate @xmath142 in eq . ( [ 43 ] ) while setting @xmath143 . this would correspond to using a ( geometric ) average for the correction factor , @xmath144 , \label{45}\ ] ] for both @xmath5 and @xmath4 . we see that @xmath145 is slightly larger then @xmath140 . this is because the chemical potential somewhat reduces the number of @xmath146 and hence their energy current . even so , @xmath147 is a much better approximation then @xmath148 . the above lepton number and energy fluxes are based on a diffusion approximation . this is valid inside the neutrino sphere where @xmath149 is much less than the size of the system . we now discuss weak magnetism and recoil corrections to the emitted @xmath5 and @xmath4 spectra . we will use a simple monte carlo model of raffelt @xcite to estimate the emitted spectra . we emphasize that this simple model should be checked against full simulations that include weak magnetism and recoil . nevertheless , we expect the model to provide a first orientation . raffelt @xcite discusses the formation of @xmath5 spectra . at high densities , reactions such as @xmath150 , @xmath151 , or @xmath152 keep @xmath5 in chemical and thermal equilibrium with the matter . we define the energy sphere as the approximate location where these reactions , which all have small cross sections , become too slow to maintain thermal equilibrium . next the @xmath5 propagate through a scattering atmosphere where neutrinos diffuse because of @xmath153 which has a much larger cross section . however the small energy transfer in nucleon scattering is assumed to be too small to maintain thermal equilibrium . finally , the neutrinos escape from the neutrino sphere . note , the average spectrum for neutrinos and antineutrinos may be modified somewhat by nucleon - nucleon bremstrahlung in the scattering atmosphere @xcite . however , we expect weak magnetism in nucleon scattering to provide an estimate for the _ difference _ between the spectrum of neutrinos and antineutrinos . this should be explicitly checked in future work that includes weak magnetism for nucleon - nucleon bremstrahlung . we model the energy sphere as a black body with temperature @xmath154 12 mev . this high temperature is chosen so that the final temperature of the emitted @xmath5 agrees with detailed simulations . alternatively , @xmath155 is the hot matter temperature well inside the neutrino sphere at densities high enough ( @xmath156 so that the @xmath5 are thermalized . we assume the same @xmath155 for both @xmath5 and @xmath4 because they are produced in pairs . we next assume the @xmath5 propagate through a scattering atmosphere of optical depth , @xmath157 here @xmath158 is the thermally averaged optical depth in the absences of recoil and weak magnetism corrections . raffelt assumes a boltzmann distribution for the initial thermalized neutrinos for which @xmath159 . note , the optical depth is the thickness of the scattering atmosphere in units of the energy dependent mean free path . a value of @xmath160 30 corresponds to one typical simulation @xcite . raffelt calculates the survival probability @xmath161 for a @xmath5 of energy @xmath8 to make it through the atmosphere and escape . otherwise it is assumed to be scattered back to the energy sphere and absorbed . a fit to ref . @xcite monte carlo results is , @xmath162^{-1 } \left [ 1- { 0.033 \over 1 + 1.5 ( \ell_\tau + .17)^2 + 0.5 ( \ell_\tau + .32)^6 } \right ] , \label{47}\ ] ] with @xmath163 . therefore the final emitted spectrum is , @xmath164 for antineutrinos the optical depth is , @xmath165 and the @xmath4 spectrum is , @xmath166 the normalization @xmath167 is chosen so that @xmath168 . this corresponds to steady state equilibrium and no net change in the lepton number . note , @xmath167 is related to the chemical potential @xmath115 . the @xmath5 and the @xmath4 spectra are shown in fig . 5 . both of these curves differ from a boltzmann distribution . nevertheless the average energy can be characterized by a spectral temperature , @xmath169 table ii collects t values . we find that t for @xmath4 is about 7% larger than that for @xmath5 almost independent of the choice of @xmath158 . thus weak magnetism insures that the @xmath4 spectrum is hotter than that for @xmath5 . at high energies @xmath170 is small . therefore @xmath171 becomes significantly larger than @xmath172 . this is shown in fig . 5 as an increasing ratio of @xmath173 to @xmath172 . thus the high energy tail in the spectrum is expected to be antineutrino rich . the difference between the @xmath5 and the @xmath4 spectra in fig . [ fig5 ] is a macroscopic manifestation of charge conjugation , c , violation in the standard model . if the weak interactions had conserved c then @xmath174 . in principle , this difference is directly observable . however , it may be very difficult to distinguish a detected @xmath5 from a @xmath4 . instead , a superposition of the @xmath5 and the @xmath4 spectra may be more easily observable . weak magnetism , by separating the @xmath5 and the @xmath4 spectra and by increasing the high energy tail of the @xmath4 may lead to an observable broadening of the combined @xmath5 and @xmath4 spectrum . the reduction in @xmath3 charged current interactions from weak magnetism and recoil can change the emitted @xmath3 spectrum , decrease the electron number current and increase the @xmath175 energy flux . these changes , in turn , impact the @xmath12 in the neutrino driven wind which is a possible site for @xmath176process nucleosynthesis @xcite . finally these changes could be important for analyzing the detected @xmath177 events from sn1987a or a future galactic supernova . the increased @xmath3 mean free path from recoil and weak magnetism will increase the @xmath177 number current or decrease the total electron lepton number current @xmath178 . this can increase the @xmath53 density in the star . for simplicity , consider a situation where @xmath179 is near zero in a simulation without recoil or weak magnetism . then in steady state equilibrium , we expect recoil and weak magnetism to lead to a @xmath179 of order the @xmath115 from section iii eq . . this chemical potential will lead to an increase in the electron fraction because , in @xmath180 equilibrium , @xmath181 . this will lead to a small change in the electron fraction of order , @xmath182 . \label{52}\ ] ] here @xmath183 is the electron chemical potential and @xmath184 the electron fraction without weak magnetism and @xmath12 is the new electron fraction . using eq . ( [ 41 ] ) for @xmath179 at t = 10 mev , @xmath185 = 15 mev one has , @xmath186 an increase of 10% . note , eqs . ( [ 52],[53 ] ) slightly overestimate the change in @xmath12 because the charged current opacity for @xmath3 increases with @xmath12 . nevertheless , weak magnetism could lead to a modest increase in @xmath12 . in principle , weak magnetism will increase the lepton number diffusion time scale . however this time scale is very sensitive to temperature . we also expect weak magnetism to increase the @xmath3 energy flux in a similar way to fig . 4 . with weak magnetism , the protoneutron star should cool faster and the lower temperatures will decrease the lepton number diffusion time . these contrary effects should be investigated in a full simulation . we consider the following simple model @xcite to estimate the change in the @xmath3 spectrum with weak magnetism . let the @xmath3 neutrino sphere be at a temperature @xmath187 without weak magnetism . for simplicity we assume a boltzmann spectrum @xmath188 . now we include weak magnetism and recoil described by @xmath189 . this reduction in the opacity will shift the neutrino sphere inwards to higher densities and temperatures . we define a new energy dependent temperature @xmath190 to include this shift , @xmath191^u t_0 . \label{54}\ ] ] here @xmath192 is the ratio of temperature to density gradients , @xmath193 , \label{55}\ ] ] for the protoneutron star at the neutrino sphere . the extra -1 in the denominator follows because the optical depth involves a path integral from the neutrino sphere to infinity @xcite . realistic values of @xmath192 could be @xmath194 0.25 to 0.35 @xcite . the energy dependent temperature @xmath190 defines the emitted spectrum , @xmath195 if we expand @xmath196 , with @xmath197 and expand @xmath198 to first order in @xmath14 , @xmath199 e^{-k / t_0 } . \label{57}\ ] ] we define a spectral temperature @xmath200 as , @xmath201 , \label{58}\ ] ] for @xmath192 = 0.3 and @xmath202 5 mev one has , @xmath203 thus weak magnetism and recoil can increase the emitted @xmath3 average energy by of order 10% . it is important to check our simple model with full simulations . figure 6 shows the full spectrum , eq . ( [ 57 ] ) . we see that weak magnetism and recoil shift the strength to higher energies and increase the high energy tail significantly . this is shown by a large ratio of the spectrum with weak magnetism to @xmath204 at high energies . we now discuss some possible implications of weak magnetism on neutrino heating , electron fraction and neutrino cooling in the low density region outside the gain radius . the gain radius is the point were cooling from neutrino emission balances heating from neutrino absorption and scattering . the net amount of heating outside the gain radius could be important for the success of the explosion . the absorption of @xmath3 outside the gain radius will be reduced by the smaller cross section from weak magnetism . however the cross sections grows with the square of the energy so the new absorption rate may be proportional to , @xmath205 the first factor could increase the rate by about 20% while the second factor @xmath206 0.8 could decrease the rate by a similar amount . therefore , the net rate of @xmath3 absorption might be little changed . thus the electron fraction , which depends on the absorption rate , should not change greatly . however , the @xmath3s absorbed have a higher energy so the net heating should increase by about 10% because of the higher average energy in eq . ( [ 59 ] ) . this assumes the luminosity of the @xmath3 changed only because of the change in @xmath200 . if the luminosities of @xmath53 and @xmath3 increased further because of the increased energy fluxes in fig . 4 then the total heating could be even larger . finally we discuss cooling from positron capture outside the gain radius , @xmath207 the rate for eq . ( [ 61 ] ) will be reduced by weak magnetism , @xmath208 at @xmath125 = 2 mev the factor in parenthesis reduces the cooling rate by 9% . thus weak magnetism can change the @xmath3 luminosity to increase the heating while , at the same time , reducing the cross section to decrease the cooling . these results should be checked with full simulations . in this paper we examine recoil and weak magnetism corrections to @xmath39-nucleon interactions . these are important because they are present at all densities , even at the relatively low densities near the neutrino sphere . furthermore , the corrections are model independent . we calculate simple , exact correction factors eqs . ( [ 12],[22],[32 ] ) to include recoil and weak magnetism in supernova simulations . perhaps the most important effect of weak magnetism and recoil is to increase energy fluxes of both @xmath4 and @xmath3 antineutrinos . we calculate , in a diffusion approximation , an increase of order 15% in the total energy flux for temperatures near 10 mev . this should raise the neutrino luminosity . weak magnetism and recoil will also change the emitted spectrum of @xmath0 and @xmath3 . we estimate that @xmath4 will be emitted about 7% hotter than @xmath5 because @xmath4 have longer mean free paths . likewise weak magnetism may increase the @xmath3 temperature by of order 10% . this increase in temperature coupled with the increase in neutrino luminosity should overcome the reduced absorption cross section and increase the heating in the low density region outside of the neutrino sphere . this , in turn , could be important for the success of an explosion . we find large corrections . however , supernova simulations are very complicated with many degrees of feedback . therefore , it is important to check our results with a full simulation that includes boltzmann neutrino transport and weak magnetism corrections . we thank john beacom , hans - thomas janka and georg raffelt for extensive discussions of the physics . we thank the institute for nuclear theory in seattle for its hospitality when this work was started and acknowledge financial support from doe grant de - fg02 - 87er40365 . llll + reaction & @xmath10 & @xmath11 & @xmath37 + @xmath209 & @xmath210 0.035 & @xmath211 0.630 & @xmath212 1.019 + @xmath213 & @xmath214 & @xmath215 -0.630 & @xmath216 -.963 + @xmath217 & 1 & @xmath50 1.260 & @xmath218 3.706 + [ table1 ] here @xmath50 1.260 , @xmath219 0.2325 , @xmath220 1.793 and @xmath221 1.913 . lll + & & correction + 1 . & & phase space + 2 . & & matrix element + & a. & recoil + & b. & weak magnetism + & c. & form factors + & d. & strange quarks + 3 . & & pauli blocking + 4 . & & fermi / thermal motion of initial nucleons + 5 . & & coulomb interactions + 6 . & & mean field effects + 7 . & & nn correlations in rpa + 8 . & & nn correlations beyond rpa + 9 . & & meson exchange currents + 10 . & & other components such as hyperons + 11 . & & other phases such as meson condensates or quark matter + 12 . & & corrections from superfluid/ superconductor pairing + 13 . & & nonuniform matter + 14 . & & magnetic field effects + [ table2 ] [ cols= " < , < , < , < " , ] c.j . horowitz and gang li , phys . b443 * ( 1998 ) 58 . c.j . horowitz and gang li , phys . d * 61 * ( 2000 ) 063002 . a. vilenkin , 1979 , unpublished ; ap j * 451 * ( 1995 ) 700 . c. j. horowitz and j. piekarewicz , nuc a640 * ( 1998 ) 281 . phil arras , dong lai , phys rev . d * 60 * ( 1999 ) 043001 . sanjay reddy , madappa prakash , james m. lattimer and jose a. pons , phys . c * 59 * , ( 1999 ) 2888 . a. burrows and r.f sawyer , phys . c * 58 * ( 1998 ) 554 ; phys . c * 59 * ( 1999 ) 510 . horowitz and gang li , phys . lett . * 82 * ( 1999 ) 5198 . horowitz , astro - ph/0108113 . see for example , m. rampp and h .- janka , ap jl * 539 * ( 2000 ) 33 , or s.w . bruenn , k.r . denisco , and a. mezzacappa , astro - ph/0101400 . c. j. horowitz , phys . d * 55 * ( 1997 ) 4577 . c. j. horowitz , phys . * 66 * ( 1991 ) 272 . a. l. fetter and j. d. walecka , `` quantum theory of many particle systems '' , mcgraw - hill , ny 1971 , p 171 . s. reddy , g. f. bertsch , and m. prakash , phys . 475b * ( 2000 ) 1 . steen hannestad and georg raffelt , apj . * 507 * ( 1998 ) 339 . e. g. flowers , m. ruderman , and p. g. sutherland , apj . * 205 * ( 1976 ) 541 ; a. d. kaminker , p. haensel and d. g. yakovlev , ast . and astrophys . * 373 * ( 2001 ) l17 . gregory w. carter and sanjay reddy , phys . d * 62 * ( 2000 ) 103002 . sanjay reddy , madappa prakash and james m. lattimer , phys . d * 58 * ( 1998 ) 013009 . stephen j. hardy , markus h. thoma , phys . d * 63 * ( 2001 ) 025014 . adam burrows , timothy young , philip pinto , ron eastman and todd a. tompson , apj . * 539 * ( 2000 ) 865 . p. vogel , and j. f. beacom , phys . d * 60 * ( 1999 ) 053003 . ahrens et al . , phys . d * 35 * ( 1987 ) 785 . schinde , ap j suppl . * 74*(1990 ) 249 . b. mueller et al . , phys . lett . * 78 * ( 1997 ) ; k. a. aniol et al . , * 82 * ( 1999 ) 1096 . raffelt , astro - ph/0105250 . m.j . musolf and t.w . donnelly , nucl . a546 * ( 1992 ) 509 . in this appendix we discuss the low density limit of burrows and sawyer s calculations @xcite and show that it reduces to our phase space result , @xmath222 , in eq . ( [ 6 ] ) . burrows and sawyer start by considering a pure vector interaction . ( 2 ) for the differential rate of neutrino scattering @xmath223 is , @xmath224\lambda^{00}s(q,\omega ) . \label{a1.1}\ ] ] here the momentum transfered to the neutrons is @xmath225 and the energy transfered is @xmath226 . the neutrino distribution is @xmath227 and @xmath228 is the neutrino tensor , eq . ( 3 ) of ref . the dynamic structure function @xmath229 for the nuetrons is , @xmath230 2\pi \delta(\omega+\epsilon_p - \epsilon_{p+q } ) , \label{a1.2}\ ] ] where @xmath231 is the neutron distribution function and @xmath232 the energy of a neutron of momentum @xmath233 . one can recover our eq . ( [ 6 ] ) by , ( a ) neglecting the final state neutrino pauli blocking , ( b ) integrating eq . ( [ a1.1 ] ) over @xmath234 , ( c ) assuming that the density is low enough so that @xmath235 in eq . ( [ a1.2 ] ) , @xmath236 @xmath237 here @xmath107 is the neutron density . finally ( d ) we assume , in the limit of low density , that the @xmath234 depdendence of @xmath229 is sharply peaked near , @xmath238 , with @xmath239 given by our eq . ( [ 2 ] ) . adding a simillar expression for the axial interactions gives , @xmath240 with @xmath222 given by our eq . ( [ 6 ] ) . in this appendix we collect simple parameterizations of the single nucleon form factors for use in eq . ( [ 13 ] ) . we write the weak form factors in terms of the electromagnetic dirac , @xmath241 , and pauli , @xmath242 , form factors for @xmath243 or @xmath244 . here , see eq . ( [ 14 ] ) , @xmath254 @xmath255 g / \left ( 1 + \tau \right ) , \label{a2.11}\ ] ] @xmath256 @xmath257 @xmath258 with @xmath259 @xmath260 and finally , @xmath261 these parameterizations are from ref .

majorana fermions ( mfs ) are their own anti - particles . they have been the subject of theories for more than seven decades without an experimental signature . however recently it has been suggested that these exotic particles can be discovered in condensed matters systems , in particular in topological superconductors ( tscs ) @xcite . tscs are superconducting systems that admit exceptional boundary states owing to nontrivial topology associated with bulk spectra inside the quasi - particle excitation gaps . these exceptional boundary states are coherent superpositions of electrons ( particles ) and holes ( anti - particles ) which , upon proper compositions , represent realizations of majorana states . the presence of majorana states in tscs is robust because of their deep roots in the global properties of bulk states . this leads to renewed efforts to reveal mfs in laboratories . two examples of tscs are chiral p - wave superconductors in one dimension ( 1d ) and two dimensions ( 2d ) . in 1d , a chiral p - wave superconductor may accommodate a pair of majorana bound states ( mbss ) separately at its two ends @xcite . the energy of this pair of mbss at large separation is exponentially close to the chemical potential by definition the zero energy of the superconductor , and each mbs hence provides a container for a mf to stay . recently , systems equivalent to 1d p - wave superconductors have been proposed based on proximity effects in 2d topological insulators @xcite , or even more close to experimental reality , in 1d semiconducting quantum wires @xcite . to detect mbss in these systems @xcite , and possibly to manipulate the mfs therein , is of great interest in current research @xcite . 2d chiral p - wave superconductors @xcite , and equivalent systems based on the proximity effect in topological insulators @xcite or semiconductors @xcite , are hosts for chiral majorana modes ( @xmath0mms ) , which are gapless , charge - neutral edge excitations . @xmath0mms can serve as coherent transmission channels for mfs , and hence are ideal ingredients for building mf interferometers @xcite ( see , e.g. , fig . [ fig : setup ] ) . unlike conventional ( electronic ) interferometers , these mf interferometers do not permit an arbitrary magnetic flux insertion due to the presence of the underlying superconductors . instead , magnetic flux in the mf interferometers is trapped in superconducting vortex cores and remains quantized in integer multiples of a flux quantum . in addition , each vortex here allows for one mbs @xcite , and the parity of the number of vortices gives rise to a @xmath1 -type of mf interferometry @xcite . mm ) coupled to one or several normal metal leads . the chiral majorana mode lives at the edge of the underlying topological superconductor ( tsc ) , which has a full pairing gap and is grounded . the interior of the topological superconductor may accommodate a number of vortices , each permitting a magnetic flux quantum to penetrate through and containing a majorana bound state . majorana fermions traveling in the chiral majorana mode pick up a phase that encodes ( the parity of ) the number of vortices , then scatter into normal leads . to generate charge current and noise majorana fermions interfere pairwise.,title="fig:",scaledwidth=32.0% ] + both examples mentioned above belong to the same universality class , labeled @xmath2 in the altland - zirnbauer classification @xcite . this means that in these systems time - reversal symmetry is broken but particle - hole symmetry ( phs ) as far as the bogoliubov - de gennes theory is concerned is present . effectively these systems are spinless , and the phs operator can be simply expressed as @xmath3 with @xmath4 the complex conjugate operator and @xmath5 the pauli matrix reversing the electron and hole components . we will limit ourselves to this universality class in this paper , and use the inherent phs as the only constraint to the scattering theory that will be developed here . scattering theory in mesoscopic physics @xcite has been proven a very useful tool in dealing with electronic coherent transport phenomena . recent developments have also shown powerful application of scattering theory to the topological classification of condensed matter @xcite , extending an approach known from adiabatic pumping processes @xcite . in this paper we employ scattering theory to investigate interferometers built on @xmath0mms . we pursue especially a general understanding of how mfs being exotic ( quasi- ) particles which are charge - neutral can be involved in charge transport . for such a purpose we adopt consistently the majorana basis in our treatment , where the scattering theory of ( charge ) current and noise takes interestingly an off - diagonal form . namely , it is the pairwise interference between transmitted mfs , instead of the self - interference of individual particles ( as for electrons or holes ) , that plays the central role in observables like current and noise . in the following , we will first analyze the scattering at a junction between a @xmath0 mm and a normal metal lead , which turns out to be also pairwise in the properly - chosen majorana basis . in other words , the scattering at the junction can be significantly simplified in terms of both its physical picture and its mathematical parametrization . next we will apply the simplified scattering picture to compute current and noise in several examples of chiral mf interferometers , including prototypes like fabry - perot @xcite , mach - zehnder @xcite and hanbury brown - twiss two - particle interferometers@xcite . the emphasis of these calculations is placed on demonstrating the way majorana scattering amplitudes enter the expressions for current and noise , and interpreting the results in a consistent manner . most of the examples used in this paper have been discussed in the literature , therefore we will not dwell on detailed discussions about the implications of specific results . nevertheless our approach shows its elegance in dealing with chiral mf interferometers and in understanding them . to address the question how mfs can be involved in charge transport , it is necessary to analyze how they are contacted with a normal ( metal ) lead which is indispensable for measurements . in the chiral mf interferometry , where the propagating pathways for mfs can be devised to serve various particular interests , junctions between @xmath0mms and normal leads become a universal , as well as crucial , ingredient and deserve a dedicated investigation . in this section we will focus on such a contact problem , leaving the propagation of chiral mfs to be handled in the next section with specific examples of interferometers . to this end we consider a @xmath0 mm tunneling - coupled to a normal lead which contains @xmath6 transmission modes around the fermi energy and is connected to an electron reservoir with chemical potential @xmath7 at its far end .. ] the junction is described by the scattering theory adapted to deal with interfaces between normal and superconducting systems @xcite . namely , the @xmath6 transmission modes in the normal lead are artificially doubled into their electron and hole copies , which are then treated independently ( but constrained by the phs ; see below ) with different occupation functions @xmath8 where @xmath9 with @xmath10 the boltzmann constant and @xmath11 the temperature , @xmath12 with @xmath13 the chemical potential in the superconductor . throughout this paper we assume @xmath14 is much smaller than the superconducting gap , @xmath13 is fixed by grounding the superconductor , and energy @xmath15 is always measured relative to @xmath13 . in order to remove the artificial doubling from actual physical effects , it suffices to count only contributions by states with @xmath16 . specifically , the scattering matrix , @xmath17 , for the junction at energy @xmath15 is defined by [ eq : sm_eh ] ^(+)(e ) + _ e^(+)(e ) + _ h^(+)(e ) + = s(e ) ^(-)(e ) + _ e^(-)(e ) + _ h^(-)(e ) + where @xmath18 is the annihilation operator for chiral mfs , @xmath19 is the @xmath6-vector of electron / hole annihilation operators in the normal lead and the subscripts @xmath20 always stand for the outgoing / incoming states . the operators here are each normalized by ( the square root of ) the velocity of the corresponding mode , such that @xmath17 is a ( @xmath21)-dimensional unitary matrix as a result of quasi - particle probability current conservation . we emphasize that @xmath17 by itself does not respect charge conservation because of the implicit presence of the tsc upon which the @xmath0 mm has to live . often it turns out to be more convenient to work with the scattering matrix in an alternative basis the basis is called the majorana basis and hence the scattering matrix is denoted by @xmath22 which is defined as follows [ eq : sm_mf ] ^(+)(e ) + ^(+)(e ) + = s_m(e ) ^(-)(e ) + ^(-)(e ) + where @xmath23 with @xmath24 the @xmath6-dimensional identity matrix acting within the space of electron or hole modes . evidently @xmath22 is also unitary , and is related to @xmath17 by a change of basis in the normal lead : @xmath25 the inherent phs in the current problem imposes that , by definition , @xmath26 both of which apply to outgoing and incoming states separately . the latter of the above equations implies @xmath27 which allows us to regard the scattering modes represented by @xmath28-operators as artificial majorana modes . the scattering matrices constrained accordingly by the phs satisfy : in the electron - hole basis , @xmath29 where @xmath30 with the pauli matrix @xmath5 interchanges the flavors of electron and hole ; in the majorana basis , @xmath31 it is clear that at @xmath32 , @xmath22 becomes a real orthogonal matrix . indeed , in the rest of this paper , we consider the low - energy limit where the energy dependence of the scattering matrices in eqs . and can be dropped ( accordingly we will also drop the energy dependence of the operators for the scattering modes ) . as a result @xmath22 is always taken to be a real orthogonal matrix with determinant @xmath33 ( i.e. @xmath34 ) . this significantly simplifies the theory that will be developed in the following . in this section our goal is to understand the contact between a @xmath0 mm and a normal lead by decomposing the scattering matrix at the junction into its physical part and its ( physically ) irrelevant part . the physical part is what is relevant to physical observables in the scattering process , and can be taken as the canonical form of the scattering matrix ; the irrelevant part by definition does not make physical contributions . to achieve this goal we will work exclusively with scattering matrices in the majorana basis , @xmath22 , in the current section . the advantage of working in the majorana basis is that @xmath22 , being a real orthogonal matrix in the low - energy limit , is both a faithful and a convenient representation of the scattering processes which involve mfs . such a representation naturally includes all possible scattering events at the junction without additional constraints as eq . ( i.e. the set of @xmath22 in the low - energy limit is precisely @xmath35 , while the corresponding set of @xmath17 is only a subset of @xmath36 ) . this greatly facilitates the decomposition that we will perform in this section . the disadvantage of working in the majorana basis is that the artificial " majorana modes in the normal lead , defined in eq . , are not associated directly with the distribution of electrons / holes in the reservoir . therefore we need to return to the electron - hole basis in the next section to compute physical observables such as average current and noise . nevertheless we will see that the decomposition of @xmath22 pays off not only in simplifying the computations , but also in understanding physically the results . the idea to decompose a scattering matrix into its physical and irrelevant parts is to notice that the physics of a single scattering event is unchanged under @xmath37-transformations of the ( electron ) basis for the @xmath6 degenerate modes ( separately for outgoing and incoming ones ) in the normal lead . decompositions following this idea are conventionally referred to as polar decompositions of a scattering matrix when the physical part contains only the transmission eigenvalues @xcite . in the current problem the same idea gets slightly more complicated by the demand of the phs , which insists that the valid transformations are of the form @xmath38 with @xmath39 . in the majorana basis , if eq . is also imposed on the transformed operators ( with tildes ) , this means [ eq : tf_mf ] ^(+ ) = ^(+ ) , ^(- ) = ^(- ) , where @xmath40 accordingly the decomposition of @xmath22 is given by @xmath41 when @xmath42 and @xmath43 are properly chosen , @xmath44 contains a minimal number of parameters - 2\,\mbox{dim}[u(n ) ] = n$ ] , which turns out to be the actual number of parameters in @xmath44 . ] that are relevant to physical observables , and hence is the canonical form of @xmath22 . as a consequence of the phs , the decompositions here do not in general bring @xmath44 into a diagonal form as conventional polar decompositions do ( the diagonal form is even impossible in the electron - hole basis because the @xmath0 mm can not be coupled only to a single electron / hole mode at all ) . instead the decompositions of @xmath22 turn out to take the form of euler decompositions : mathematically this means @xmath44 appears to be a combination of uncoupled ( 2d-)planar rotations ; physically this means the scattering of many majorana modes is essentially pairwise . in the rest of this section , we will derive such an euler decomposition first in the simplest case when @xmath45 , and then in the arbitrary @xmath6-channel case . ) normal lead and a @xmath0 mm , in the electron - hole basis of the normal lead ( a ) , and in the properly - chosen majorana basis ( b ) . in the latter basis , only one majorana mode in the normal lead is coupled to the @xmath0 mm , the other is simply reflected with amplitude @xmath33.,title="fig:",scaledwidth=40.0% ] + when @xmath45 , @xmath22 is of dimension @xmath46 . namely , @xmath22 is equivalent to a rotation in three dimensions . such a rotation can be parametrized by three parameters , known as the euler angles , as follows [ eq : euler1 ] s_m = ( cc 1 & 0 + 0 & r ( ) ) ( cc r ( ) & 0 + 0 & 1 ) ( cc 1 & 0 + 0 & r ( ) ) , where [ eq : r0 ] r ( ) ( cc & - + & ) (= , , ) is a planar rotation by an angle @xmath47 . the above decomposition is exactly what is dictated by eq . . to see this , simply notice that [ eq : r0u0 ] u_1^r ( ) u_1 = ( cc e^-i & + & e^i ) , thus we can identify @xmath48 and @xmath49 . namely , if we redefine the electron and hole operators by trivial gauge transformations ( in the normal lead ) @xmath50 and redefine @xmath51 according to eq . , then the majorana scattering matrix contains one single parameter and takes the form [ eq : sm_1 ] _ m = r ( ) & 0 + 0 & 1 . is the canonical form of the majorana scattering matrix when @xmath45 , which says that , effectively , the @xmath0 mm @xmath18 is coupled to one ( artificial ) majorana mode @xmath52 in the normal lead , leaving the other ( artificial ) majorana mode @xmath53 to be completely reflected with amplitude @xmath33 ( see fig . [ fig : dec1 ] ) . ) normal lead and a @xmath0 mm , in the electron - hole basis for the normal lead ( a ) , and in the properly - chosen majorana basis ( b ) . in the latter basis , only one majorana mode in the normal lead is coupled to the @xmath0 mm , the others are reflected in pairs and one is reflected with amplitude @xmath33.,title="fig:",scaledwidth=45.0% ] + when @xmath54 , there is no decomposition like eq . readily available . nevertheless one can show that ( see appendix [ app : dec ] ) , the @xmath55-dimensional majorana scattering matrix can always be decomposed as eq . with [ eq : sm_n ] _ m = 1 , where @xmath56 couples the modes @xmath18 and @xmath52 , @xmath57 with @xmath58 couples the modes @xmath59 and @xmath60 , and the last diagonal entry @xmath33 signifies a complete reflection of mode @xmath61 . note that by definition ( see eq . ) , @xmath62 and @xmath63 are partners derived from the same normal mode @xmath64 , that is , @xmath65 the concept of such pairs of majorana modes turns out to be fundamental in relating majorana scattering matrices to charge transport , as will be revealed in the next section . the above canonical form of majorana scattering matrices should be no surprise after we have understood the @xmath45 case : the @xmath0 mm @xmath18 effectively couples to a single majorana mode @xmath52 determined by the specifics of the junction in the normal lead , which leaves its partner @xmath66 determined solely by @xmath52 to be coupled to the rest of the majorana modes in the normal lead ; @xmath66 then picks " another ( single ) majorana mode @xmath53 to couple to , leaving in turn @xmath67 behind ; this process goes on until the last one @xmath61 is left completely alone ( see fig . [ fig : decn ] ) . interestingly , this re - pairing picture is a good analogy to the re - pairing of mfs that happens at the topological phase transition in the kitaev model for 1d p - wave superconductors @xcite here , the re - pairing of majorana modes happens when a normal lead makes contact to a 2d p - wave superconductor with a chiral majorana edge mode . in this section we compute the average current and the zero - frequency noise power in various prototypical chiral mf interferometers . to this end we introduce for convenience two matrices : @xmath68 , which is a diagonal matrix containing the fermi distribution functions for all incoming modes , and @xmath69 , which is another diagonal matrix containing the weights for charge current carried by electron and hole modes in lead @xmath70 ( @xmath33 for electron modes and @xmath71 for hole modes ; @xmath72 for modes that do not belong to lead @xmath70 ) . explicitly , we define @xmath73 where @xmath74 is the total number of contacts , @xmath75 is the fermi distribution function for electrons / holes in contact @xmath70 , @xmath76 is the number of transmission modes ( without double counting ) in lead @xmath70 , @xmath77 is the empty matrix corresponding to the modes in lead @xmath78 . we will use the following formulas from the scattering theory adapted for normal - superconducting hybrid systems @xcite : the time - averaged current at normal contact @xmath79 , @xmath80 , is given by @xmath81\;;\end{aligned}\ ] ] the average current flowing through the grounded superconducting contact is given , owing to current conservation , by ( omitting the minus sign ) @xmath82 the zero - frequency noise power , defined as @xmath83 with the current fluctuation @xmath84 , is given by @xmath85\;.\end{aligned}\ ] ] in the above formulas we have used a shorthand notation @xmath86 defined as follows @xmath87 note that @xmath17 here is the full scattering matrix in the electron - hole basis , which includes @xmath74 contacts but does not explicitly include the @xmath0 mm ( see , e.g. , eq . in sec . [ ssec : fpi1 ] ) . the full majorana scattering matrix @xmath22 is related to @xmath17 by @xmath88 before we proceed presenting the results , we confirm that the transformations in eqs . and indeed lead to no change in physical observables . that is , we verify that @xmath89 and @xmath90 are unchanged upon replacing @xmath17 by @xmath91 , given by @xmath92 the proof is straightforward after noticing that @xmath93 = 0\quad \mbox{and } \quad \forall \nu : [ v_{\oplus } , ( \sigma_z)_{\nu } ] = 0,\end{aligned}\ ] ] where @xmath94 $ ] stands for the commutator . therefore we will adopt only the canonical form @xmath44 in the actual calculations , and also suppress the tildes from now on . we start with the simplest setup : a single normal lead with a single mode is probing a closed @xmath0 mm , or , a chiral majorana loop ; the interior of the tsc enclosed by the chiral majorana loop may accommodate @xmath95 vortices , each allowing for one mbs at zero energy . this setup , as well as the setup that will be discussed in the next subsection ( sec . [ ssec : fpi1 ] ) , has been discussed by law , lee and ng @xcite . here we reformulate the problem in a consistent majorana language , with extensions that are intriguing in their own right . we assume that the mbss are all deep inside the tsc , such that the coupling between the mbss and the @xmath0 mm or the normal lead , which exponentially decays with respect to their spatial separation , is negligible . with this assumption the vortices are taken into account only as an additional phase @xmath96 that the chiral mf picks up after moving in a full circle , and the acquired total phase hence is @xmath97 , where the single @xmath98 comes from the berry phase , and @xmath99 , with @xmath100 the circumference of the loop , @xmath101 the group velocity of the @xmath0 mm , is the level spacing for the chiral majorana loop . ) normal lead probing a chiral majorana loop : the left panel case ( a ) can be extended into the right panel one ( b ) , to model , e.g. , the presence of an additional point contact or scattering at two coupled mbss ( see text for details).,title="fig:",scaledwidth=45.0% ] + following our discussion in section [ ssec : scdec ] , the effective scattering process for the current example in the majorana language is pictured in fig . [ fig : l11]a , and the majorana scattering matrix is given by @xmath102 where @xmath103 , and [ eq : r1 ] r_1= with @xmath104 the ( local ) reflection amplitude of @xmath105 at the junction connecting to the still unclosed @xmath18 ( namely , @xmath106 in eq . ) . note that the above majorana scattering matrix is implicitly energy - dependent because of the energy - dependence of @xmath107 ; it is unitary , but not necessarily real except at @xmath32 . inserting @xmath22 into eqs . and , together with , we find the average current and the zero - frequency noise power ( auto - correlator ) to be @xmath108\delta{f}(e ) , \label{eq : il11}\\ p & = \frac{e^2}{h}\int_{e\geq 0 } de\ , \bigl\{2\bigl[1-\re(r_1r_2^*)\bigr]\theta(e ) \nonumber\\ & \qquad\qquad\qquad\quad + \im(r_1r_2^*)^2\delta{f}(e)^2\bigr\ } , \label{eq : pl11}\end{aligned}\ ] ] where @xmath109+f_h(e)\bigl[1-f_h(e)\bigr ] \nonumber\\ & = -k_b t \frac{\partial}{\partial e}\bigl[f_e(e)+f_h(e)\bigr ] . \label{eq : theta}\end{aligned}\ ] ] note that @xmath110 is positive ( negative ) definite if @xmath111 ( @xmath112 ) , zero at equilibrium , while @xmath113 is positive definite at finite temperature . as a convention , we shall intentionally keep the formal complex conjugate of real variables ( e.g. @xmath114 ) for a reason that will be clear in a moment . let us now examine the expressions for the current eq . and noise eq . closely . the net current in eq . is evidently attributed to the andreev process : an incoming electron or hole , weighted effectively by the difference of the fermi distribution functions @xmath115 , is reflected with probability " @xmath116 $ ] , and the missing charges are absorbed by the superconducting condensate as cooper pairs . here , negative @xmath117 means that a particle with an opposite charge with respect to the incoming particle is reflected with probability @xmath118 . the noise in eq . has two contributions : the first term in the curly brackets gives the thermal noise which persists at equilibrium at finite temperature ; the second term gives the shot noise which vanishes either at equilibrium , where @xmath119 , or upon total normal or andreev reflections , where @xmath120 . the remarkable expression for the reflection probability " , @xmath117 , highlights one essential feature of mf transport , namely , charge is always carried " by the interference between ( a pair of ) majorana modes , even though an individual majorana mode is charge - neutral . this can be heuristically understood as follows . take one ( quasi - particle ) scattering state given by , in the majorana basis , @xmath121 $ ] ; in the electron - hole basis , the same state reads @xmath122 with @xmath123 and @xmath124 ; we immediately see that the current density carried by this state is @xmath125 , which is precisely the interference term between the two majorana modes . when @xmath126 and @xmath127 are identified with the reflection amplitudes in @xmath22 , @xmath117 becomes the reflection probability " that appears in eqs . and . in order to stress the central importance of interference between majorana modes in this work , we will on most occasions keep formal expressions similar to @xmath117 without simplifying expressions written with formally complex conjugate but in reality real variables . at zero - temperature and low bias ( @xmath128 ) , @xmath129 , @xmath130 , and eqs . and can be reduced to @xcite : @xmath131 the fano factor , defined as @xmath132 , is give by @xmath133 and takes the value @xmath134 if @xmath95 is even , or @xmath72 if @xmath95 is odd . physically the two situations here are realized exactly when the chiral mf self - interferes maximally destructive or constructive upon reflection , hence a stationary state is disallowed or allowed at zero energy . @xmath135 therefore corresponds to the reflection of a majorana mode that is completely off - resonance ( @xmath136 ) or on - resonance ( @xmath137 ) with such a stationary state @xcite . as a consequence , the scattering process is a total normal reflection ( @xmath138 ) or a total andreev reflection ( @xmath139 ) in either case , the shot noise vanishes . in units of @xmath140 , as a function of @xmath141 for the setup in fig . [ fig : l11]a . the resonance peak for @xmath107 equal to an even multiple of @xmath98 and the ( anti - resonance ) valley for @xmath107 equal to an odd multiple of @xmath98 is a robust feature regardless of the coupling strength , parametrized by @xmath104 , between the normal lead and the @xmath0 mm . this plot is essentially a reproduction of the results of law et al . @xcite . the purpose of this reproduction is to facilitate comparisons with our later results.,title="fig:",scaledwidth=40.0% ] + for more general cases beyond the low - bias limit , where @xmath107 is allowed to vary with energy continuously between an even and an odd multiple of @xmath98 , we plot the zero - temperature differential conductance @xmath142 , as a function of @xmath141 and in units of @xmath140 , in fig . [ fig : gl11 ] with different @xmath104 . it is clearly seen that a quantized resonance peak is established at @xmath143 ( @xmath144 ) and completely suppressed at @xmath145 ( @xmath144 ) @xcite ; decreasing @xmath104 broadens the resonance peak but does not remove such a general feature . the present simple setup can be slightly extended by including another chiral majorana loop , labeled by @xmath146 , which couples only to the original loop @xmath18 with tunneling amplitude @xmath147 ( see fig . [ fig : l11]b ) . this extended setup can be used to model , for example , the effect of an additional point contact formed by confining the path of the original @xmath0 mm , or scattering upon two coupled mbss . the computation for the extended setup requires a minimal effort , which amounts to replacing @xmath107 in eq . with @xmath148 , defined as @xmath149 where @xmath150 with @xmath151 the magnetic flux inside @xmath146 , and @xmath152 is the local reflection amplitude between the two chiral majorana loops . the expressions for the average current and the noise power are unchanged from eqs . and . in units of @xmath140 , as a function of @xmath153 with @xmath154 for the setup in fig . [ fig : l11]b . we set @xmath155 in the upper panel and set @xmath156 in the lower panel ; for both panels @xmath157 is assumed and @xmath104 is fixed to be @xmath158 . in the upper panel the resonance peak splits into two with their distance proportional to the energy splitting of the two coupled chiral majorana loops ; in the lower panel the resonance peak persists whenever @xmath107 is an integer multiple of @xmath98 as long as the coupling between the two chiral majorana loops is finite.,title="fig:",scaledwidth=40.0% ] + for simplicity we consider @xmath159 and @xmath160 . when the coupling between the two loops are turned off , we know from previous discussions that a resonance - led total andreev reflection occurs at @xmath32 ( more generally , at @xmath161 with @xmath162 an integer ; we will focus on the energy range around zero ) . upon turning on the coupling even with infinitesimal @xmath147 the resonance at zero energy is immediately switched off because @xmath148 has a @xmath98 shift from @xmath107 and hence @xmath126 changes sign . indeed , since the energy spectrum of the coupled loops is given by @xmath163 ( which is also equivalent to the resonance condition @xmath164 ) , we find that the resonance shifts to @xmath165 when @xmath166 . this is in close analogy to a system in which two mbss ( e.g. at the two ends of a chiral p - wave superconducting wire ) are coupled and lifted from zero energy @xcite . it deserves to be mentioned that if the current case is modified by setting one of the phases @xmath167 or @xmath168 to be @xmath72 and the other @xmath98 , then the resonance occurs at zero energy regardless of the coupling strength between the loops as long as it is finite . differential conductances for both situations , @xmath169 or @xmath98 , are plotted in fig . [ fig : gl11_2 ] . : the upper panel case ( a ) can be extended to the lower panel case ( b ) to model the presence of an additional point contact or two coupled mbss ( see text for details).,title="fig:",scaledwidth=40.0% ] + our second example deals with the fabry - perot interferometer composed of two single - mode normal leads ( labeled by @xmath100 and @xmath170 ) probing a chiral majorana loop simultaneously . we assume that the two normal leads are mediated only by the @xmath0 mm and other tunneling processes are negligible . [ fig : l21]a illustrates the effective scattering in the majorana language , and the corresponding majorana scattering matrix is given by @xmath171 where @xmath172 , and @xmath173 with the subscript @xmath174 meaning that the reflection / transmission amplitude is local at the junction ( corresponding to eq . , @xmath175 , @xmath176 , for two leads separately ) . note that we have chosen a gauge such that @xmath107 belongs entirely to the lower arm . ) and in units of @xmath140 , as a function of @xmath141 for the fabry - perot interferometer in fig . [ fig : l21]a , where @xmath177 is fixed to be @xmath158 and the two normal contacts are equally biased . in general , resonance peaks ( anti - resonance valleys ) of the differential conductances at @xmath143 ( @xmath178 ) persist as a robust feature . asymmetric coupling at the two junctions ( @xmath179 ) results in unbalanced resonance peaks in @xmath180 and @xmath181 , while the resonance peak in @xmath182 remains quantized at @xmath183 . particularly when one of the junctions is shut off ( e.g. @xmath184 ) , the result returns to what is shown in fig . [ fig : gl11 ] , where the resonant - andreev - reflection is completely local . this plot is again essentially a reproduction of the results in ref . @xcite , presented here for the purpose of comparison with later results.,scaledwidth=40.0% ] the average current and the zero - frequency noise power are given by @xmath185\delta{f}_{l } , \label{eq : il21}\\ & p_{ll}= \frac{e^2}{h}\int_{e\geq 0 } de\ , \bigl\{2\bigl[1-\re(r_{l1}r_{l2}^*)\bigr ] \theta_{l } \nonumber\\ & \qquad\qquad\qquad\qquad + \im(r_{l1}r_{l2}^*)^2\delta{f}_l^2 \nonumber\\ & \quad + \frac{1}{2}t_{lr}\bigl[\theta_{r } - \theta_{l } + \frac{1}{2}\sum\limits_{a , b = e , h}({f}_{la}-{f}_{rb})^2\bigr]\bigr\ } , \label{eq : pal21}\\ & p_{lr}= p_{rl } \nonumber\\ & = \frac{e^2}{h } \int_{e\ge0 } de\ , ( -\frac{1}{2})\re(t_{lr}r_{l2}^*t_{rl}r_{r2}^*)\delta{f}_{l}\delta{f}_{r } , \label{eq : pcl21}\end{aligned}\ ] ] where @xmath186 . @xmath187 and @xmath188 can be obtained by interchanging subscripts @xmath100 and @xmath170 in the expressions for @xmath189 and @xmath190 . in the presence of multiple normal contacts , it is particularly interesting to look at the average current @xmath191 flowing through the grounded superconducting contact . assuming the two normal contacts here are equally biased ( @xmath192 ) , @xmath191 is simply given by @xmath193 \delta{f } , \label{eq : isl21}\end{aligned}\ ] ] where we have used the explicit expressions of the reflection amplitudes . an immediate observation based on the above result is that the resonance peak in the differential conductance @xmath182 at @xmath194 remains quantized at @xmath183 regardless of the coupling of the two normal leads to the @xmath0 mm ( see fig . [ fig : gl21 ] ) . formally the expression for @xmath189 ( or @xmath187 ) is the same as eq . in the single lead case , meaning that one contact does not contribute to the average current at the other contact regardless of the electron / hole occupation in either contact . naively this can be seen as a result of the charge neutrality of the @xmath0 mm that bridges two contacts . a more precise interpretation , however , is that the majorana scattering states sourced from different contacts ( e.g. @xmath195 transmitted from contact @xmath170 and @xmath196 ) are not phase coherent when only single - particle scattering is considered , therefore can not lead to finite average current according to our discussion in sec . [ ssec : l11 ] . this statement will be clear after we further examine the cross - correlation between the two contacts . the charge - current cross - correlator @xmath197 ( or @xmath198 ) , given by eq . , is manifestly non - vanishing at low temperature and low bias if , and only if , both contacts are biased ( @xmath199 ) and exchange of majorana states is allowed ( @xmath200 ) . this is despite the fact that the mediating @xmath0 mm carries no charge current by itself . indeed , the non - vanishing cross - correlation here is entirely due to those scattering events that involve two - particle exchange and permit coherence between scattering states from different sources , and hence is entirely the exchange noise " . such an exchange contribution is irrelevant for the average current but present in noise . meanwhile , the absence of other types of noise ( e.g. thermal noise , partition noise ) in the cross - correlator again highlights the essential role that the interference between majorana modes plays in charge transport . in contrast to the cross - correlation , the auto - correlation for contact @xmath100 , given by eq . , contains both the equilibrium thermal contribution ( the first term in the kernel ) and nonequilibrium contributions ( the rest two terms ) . contact @xmath170 manifests itself in @xmath190 only away from equilibrium ( except for modifying @xmath177 to @xmath201 ) , which is clearly different from a conventional normal - metal - superconductor - normal - metal device@xcite . of the integral for the cross - correlator @xmath197 in eq . , excluding the factor due to fermi distribution functions , as a function of @xmath202 . the coupling between the normal leads and the @xmath0 mm is assumed to be symmetric . situations with various coupling strength are plotted . between resonance ( @xmath143 ) and anti - resonance ( @xmath178 ) points ( cf . fig . [ fig : gl21 ] ) , @xmath203 changes sign.,title="fig:",scaledwidth=45.0% ] + at zero - temperature and low bias ( @xmath204 , equal for both normal contacts ) , the average current and the zero - frequency noise power can be reduced to @xcite : @xmath205 the fano factors , for individual normal leads defined as @xmath206 , and for the total noise power ( or equivalently , for the superconducting contact ) defined as @xmath207 , where @xmath208 , are given by @xmath209 we see that the even / odd @xmath95 cases represent rather distinct processes . for the conductance and fano factors of both contacts @xmath100 and @xmath170 are equal . this indicates that the statistics of the electron transport is in fact identical at these leads and the process consists of a random sequence of crossed andreev reflection processes @xcite . in contrast for @xmath95 odd , electrons are drawn at different rates from both contacts but such as to generate a perfect stream of cooper pairs entering the superconductor noiselessly ( @xmath210 ) . meanwhile , the cross - correlators change from being positive for even @xmath95 to being negative for odd @xmath95 , indicating increasing importance of local andreev reflections . for a more general picture of the scattering processes beyond the low - bias limit , we plot in fig . [ fig : pl21 ] the cross - correlation noise power spectrum ( the kernel of the integral in eq . excluding the factor due to fermi distribution functions ) @xmath211 as a function of @xmath107 . by using the explicit expressions of the scattering amplitudes , one finds clearly that @xmath203 changes sign between resonance ( @xmath212 ) and anti - resonance ( @xmath213 ) points ( cf . [ fig : gl21 ] ) . the interpretation for this sign change is in fact the same as our preceding discussion for the low - bias cases . in units of @xmath140 , as a function of @xmath153 with @xmath154 for the extended fabry - perot interferometer in fig . [ fig : l21]b . we assume @xmath155 , @xmath157 and @xmath214 . situations with various @xmath147 between the two loops are plotted . when @xmath215 , this setup is simply two copies of the single - lead setup in fig . [ fig : l11 ] ; when @xmath216 , this setup reduces to the original fabry - perot interferometer in fig . [ fig : l21]a , with @xmath217 playing the role of @xmath107 in the original setup . for the case of @xmath218 and @xmath157 , @xmath191 does not depend on @xmath147 at symmetric coupling ( @xmath219 ) , and its expression reduces to eq . ( see also fig . [ fig : gl21]c ) with @xmath107 substituted by @xmath220 . the plot shows that quite generally , @xmath182 peaks at resonance ( upon eigenstates of the coupled loops ) points , and is suppressed at anti - resonance points.,scaledwidth=45.0% ] as a function of @xmath202 for the extended fabry - perot interferometer in fig . [ fig : l21]b . in the upper panel we assume @xmath155 , and in the lower panel @xmath156 ; for both panels we take @xmath157 and @xmath214 . situations with various transmission amplitudes @xmath147 between the two loops are plotted . clearly between resonance and anti - resonance points ( cf . [ fig : gl21_2 ] ) , @xmath203 changes sign.,scaledwidth=45.0% ] as in the single lead case discussed in sec . [ ssec : l11 ] , the present setup can be extended in the same way by including another chiral majorana loop which couples to the original one ( see fig . [ fig : l21]b ) . for the extended setup , the following substitutions for the scattering amplitudes need to be made @xmath221 , \label{eq : rl1_2}\\ & r_{r1 } = z\bigl[r_{r0}+r_{l0}e^{i(\varphi+\varphi^\prime)}-r(e^{i\varphi^\prime}+r_{l0}r_{r0}e^{i\varphi})\bigr ] , \label{eq : rr1_2}\\ & t_{lr } = zt_{l0}t_{r0}t\,e^{i(\varphi+\varphi^\prime ) } , \label{eq : tlr_2}\\ & t_{rl } = -zt_{l0}t_{r0}t , \label{eq : trl_2}\\ & z\equiv \bigl[t^2+(r - r_{l0}e^{i\varphi})(r - r_{r0}e^{i\varphi^\prime})\bigr]^{-1}. \label{eq : z_2}\end{aligned}\ ] ] note that the above amplitudes reduce to eq . at @xmath222 , and to eqs . - at @xmath223 , @xmath216 . we plot the differential conductance @xmath182 and the cross - correlation noise power spectrum @xmath203 for the present case in fig . [ fig : gl21_2 ] and [ fig : pl21_2 ] , respectively . similar to our previous discussions , the coupling between the two loops ( assumed again to be of the same circumferences ) leads to split resonance peaks if @xmath155 , or a halved period if @xmath224 , for the differential conductance @xmath182 ; the sign of @xmath203 shows consistently changes between resonance and anti - resonance points upon eigenstates of the coupled loops . ) normal lead probing a chiral majorana loop . only the ( single ) majorana mode coupled to the chiral majorana loop can feel " the phase picked up in the loop.,title="fig:",scaledwidth=30.0% ] + our previous two examples deal with single - mode ( @xmath45 ) leads , which are simple but nevertheless reveal most of the interesting physics . the real strength of the decompositions discussed in sec . [ sec : contact ] , however , becomes obvious only in dealing with multi - mode ( @xmath54 ) leads . in this section we start again with a single - lead setup similar to that in sec . [ ssec : l11 ] but with @xmath54 ( see fig . [ fig : l1n ] ) . according to sec . [ ssec : mcdec ] , the majorana scattering matrix for this setup can be written as @xmath225 where @xmath226 , @xmath126 is given by eq . , and @xmath227 and @xmath228 with @xmath229 in accordance with eq . . note that the basis for this scattering matrix is reordered from the original one defined by eq . , to be @xmath230 , where one should keep in mind that @xmath231 and @xmath232 are the partners associated with the @xmath233-th original mode as given by eqs . and . in terms of the elements of this scattering matrix the average current and the zero - frequency noise power ( auto - correlator ) are : @xmath234\delta{f } , \label{eq : il1n}\\ p & = \frac{e^2}{h}\int_{e\geq 0 } de\ , \bigl\{\sum\limits_{i=1}^{n}2\bigl[1-\re(r_ir_{i+1}^*)\bigr]\theta \nonumber\\ & \qquad + \sum\limits_{i=1}^{n}\frac{1}{2}\bigl[1-\re(r_ir_{i+1}^*)^2+\im(r_ir_{i+1}^*)^2\bigr]\delta{f}^2 \nonumber\\ & \qquad + \sum\limits_{i=2}^{n}\re(t_ir_{i-1}^*t_ir_{i+1}^*)\delta{f}^2\bigr\}. \label{eq : pl1n}\end{aligned}\ ] ] clearly the total average current is a sum of contributions from all @xmath6 modes individually : @xmath235 and @xmath236 are reflection amplitudes for @xmath231 and @xmath232 and their interference term @xmath237 registers the outgoing current contributed by mode @xmath233 ; scattering between different modes ( i.e. @xmath238 ) does not appear directly in the average current similar to the case of scattering between contacts in a fabry - perot interferometer . the noise power , by contrast , contains not only the auto - correlation of each individual mode ( the first two terms in eq . ; eq . ) , but also the cross - correlation between different modes ( the last term in eq . eq . ) . in general , @xmath239 are parameters determined specifically by the details of the junction between the normal lead and the tsc , and do not necessarily lead to quantized conductance regardless of the value of @xmath107 . this is certainly an important consequence of the presence of multiple modes in the normal lead . an interesting exception @xcite , however , is the zero - temperature low - bias conductance when @xmath240 and @xmath95 is odd , where one finds @xmath241 , thus @xmath242 and @xmath243 . wimmer et al . @xcite have proposed this exceptional case to be a robust transport signature for detecting a mbs , with which a zero - energy eigenstate of the chiral majorana loop can be identified . multi - mode normal leads coupled to a single chiral majorana loop . in each lead , only one majorana mode is coupled to the chiral majorana loop , and hence is involved in transmission between contacts ( any direct tunneling between contacts is ignored ) ; the other majorana modes ( omitted in this illustration ) are locally , pairwise reflected.,title="fig:",scaledwidth=40.0% ] + as the final example for the fabry - perot - type interferometers making use of a single @xmath0 mm , we discuss a generic setup with @xmath74 normal contacts each admitting @xmath244 ( @xmath245 ) transmission modes in the lead coupled to the @xmath0 mm ( see fig.[fig : lkn ] ) . the majorana scattering matrix for this setup is given by @xmath246 where @xmath247 and @xmath248 and @xmath249 with @xmath58 take the values according to eq . for contact @xmath79 . the subscript @xmath174 again means that the reflection / transmission amplitude is local at the junction and contains no @xmath107-dependence . the basis here for each contact is ordered the same way as in . the average current and the zero - frequency noise powers are given by @xmath250\delta{f}_{\alpha } , \label{eq : ilkn}\\ & p_{\alpha\alpha}= p_{\alpha\alpha}^{(0 ) } + \frac{e^2}{h}\int_{e\geq 0 } de\,\cdot \nonumber\\ & \quad\sum\limits_{\beta\ne\alpha}\frac{1}{2}t_{\alpha\beta}\bigl[\theta_{\beta } - \theta_{\alpha } - \delta{f}_{\alpha}^2 + \frac{1}{2}\sum\limits_{a , b = e , h}({f}_{{\alpha}a}-{f}_{{\beta}b})^2\bigr ] , \label{eq : palkn}\\ & p_{\alpha\beta } = \frac{e^2}{h } \int_{e\ge0 } de\ , ( -\frac{1}{2})\re(t_{\alpha\beta}r^*_{2(\alpha)}t_{\beta\alpha}r^*_{2(\beta)})\delta{f}_{\alpha}\delta{f}_{\beta } , \nonumber\\ & \hspace{60mm}(\alpha\ne\beta ) \label{eq : pclkn}\end{aligned}\ ] ] where @xmath251 takes the form of the single - lead auto - correlator . a remarkable fact in these results is that in the presence of many modes , the @xmath107-dependence of the average current or the auto - correlators , carried only by @xmath252 and @xmath253 , can be overwhelmed by the large @xmath107-independent contributions from @xmath248 and @xmath249 ( @xmath58 ) at imperfect interfaces between normal leads and the tsc ; the cross - correlators , however , preserve the @xmath107-dependent terms as their sole contributions , hence will optimally manifest the information embedded in the transmission of @xmath0 mm . without performing further quantitative analysis for this example , we simply point out two recurrent messages in the above general results : the average current at one contact is _ only _ contributed by the interference between pairs of majorana modes in the corresponding lead , and the current cross - correlation between two contacts is _ purely _ due to coherent exchange of two mfs sourced from the two contacts . these two messages together recapitulate the crucial role that interference between majorana modes plays in charge transport . mms is simply an identity matrix . in both setups the outgoing leads are connected to electron reservoirs that are grounded.,title="fig:",scaledwidth=48.0% ] + so far we have been focusing on the junctions between normal leads and a single @xmath0 mm , and the fabry - perot - type interferometers which comprise only such junctions . alternatively another type of junctions , composed of two @xmath0mms with opposite chirality and connected to a chiral normal channel , can be used to build interferometers of the mach - zehnder type @xcite or the hanbury brown - twiss type @xcite ( see fig . [ fig : mz_hbt ] ) . the junctions between two @xmath0mms and a chiral normal channel are reflectionless and can be described by two - times - two scattering matrices @xcite , where the two components at the normal side correspond to electron and hole modes or two majorana modes depending on the basis . assuming again the scattering matrix at the junction is energy - independent at low enough energy , the majorana scattering matrix belongs to @xmath254 owing to the phs . it follows that by employing the same strategy as presented in sec . [ ssec : gen ] , the canonical form of the scattering matrix , in the properly chosen majorana basis , is simply an identity matrix . in other words , the junction effectively splits or joins , without mixing , two @xmath0mms . this certainly simplifies significantly our discussions which we present below in terms of specific examples . the mach - zehnder interferometer illustrated in fig . [ fig : mz_hbt]a has been investigated by fu and kane @xcite , and akhmerov , nilsson and beenakker @xcite . here we simply reformulate the problem in the majorana language we have been speaking so far , to echo some general features of majorana transport that have been discussed in the previous part of this paper . the majorana scattering matrix for the mach - zehnder interferometer(see fig . [ fig : mz_hbt]b ) , defined as @xmath255 is effectively given by @xmath256 where we can choose a specific gauge such that @xmath257 with @xmath258 , and @xmath259 . the average current and zero - frequency noise power can be easily obtained : @xmath260\;\delta{f}_{l } , \label{eq : is_mz}\\ & p_{ll } = \frac{e^2}{h}\int_{e\geq 0 } de \ ; \theta_l , \label{eq : pll_mz } \\ & p_{rr } = \frac{e^2}{h}\int_{e\geq 0 } de \ ; \bigl[\theta_l + \im(t_1t_2^*)^2\,\delta{f}_l^2\bigr ] , \label{eq : prr_mz } \\ & p_{lr } = p_{rl } = -\frac{e^2}{h}\int_{e\geq 0 } de \ ; \re(t_1t_2^*)\;\theta_l\ , , \label{eq : pc_mz}\end{aligned}\ ] ] where @xmath261 and @xmath262 when the explicit expressions of @xmath263 and @xmath264 are taken . clearly the outgoing current in eq . appears to be the interference between two @xmath0mms the transmission probability " @xmath265 has a physically identical interpretation as the reflection probability " @xmath117 appearing in eq . . an obvious difference , however , between the present setup and a fabry - perot - type interferometer is that contact @xmath100 here indeed provides coherent sources for the mf pairs instead of single ones that are drained by contact @xmath170 . as a direct consequence , the cross - correlator in eq . involves only the thermal fluctuation in contact @xmath100 , while all the other contributions ( e.g. partition or exchange noise ) are absent due to the one - way nature of the present setup . the hanbury brown - twiss interferometer shown in fig . [ fig : mz_hbt]c has recently been investigated by strbi , belzig , choi , and bruder @xcite . here we reformulate the problem , again in the majorana language . the majorana scattering matrix for the hanbury brown - twiss interferometer(see fig . [ fig : mz_hbt]d ) , defined as @xmath266 is effectively given by @xmath267 in this case we can choose a specific gauge such that @xmath268 , and @xmath269 . the average current and zero - frequency noise power are given by @xmath270\,,\quad ( \beta = u , d ) \label{eq : pbb_hbt } \\ & p_{ud } = p_{du } \nonumber\\ & = \frac{e^2}{h}\int_{e\geq 0 } de \ ; ( -\frac{1}{2})\re(t_{ul}t^*_{ur}t_{dr}t^*_{dl})\delta{f}_{l}\delta{f}_{r } \ , , \label{eq : pud_hbt}\end{aligned}\ ] ] and all other cross - correlators are zero . taking the explicit expressions , we have @xmath271 . as pointed out by strbi _ @xcite , a remarkable fact in the present setup is that although the average current at the two drains @xmath272 and @xmath2 vanishes identically , the cross - correlation between the two contacts is in general non - zero . this fact , in the majorana language , can be interpreted as follows : the mf pair in either drain are necessarily from different sources , and are therefore phase incoherent for single - particle scattering which the average current measures the incoherence between mf pairs implies vanishing average current ; two - particle scattering allows coherence between mf pairs in both drains , and is included in the measurement of noise power the coherent injection of two electrons / holes from the two sources leads to non - vanishing cross - correlation between current in the two drains . the scenario here bears close analogy to the @xmath45 fabry - perot interferometer discussed in sec . [ ssec : fpi1 ] , where the average current contributed by a foreign contact is always zero but the cross - correlation between two contacts is in general non - vanishing . the underlying physics in both scenarios , as it has been for essentially all our examples , boils down to the central importance of interference in the charge transport with majorana modes . in this work we emphasized a scattering approach using a global majorana basis . both the normal state leads and the system which intrinsically carries chiral majorana states are described in the same basis . as a consequence , from a purely technical particle scattering point of view , majorana scattering problems are rather similar to scattering problems in normal mesoscopic structures . however , majorana modes are neutral and carry neither charge nor ( electric ) current . in contrast to normal scattering problems the charge and current operators are not diagonal in the majorana scattering states . therefore , unlike in normal systems , current is not determined by transmission probabilities of majorana fermions . we have shown that charge and current appear due to interference of pairs of majorana modes . for averages of single particle quantities , like charge and current , to be non - vanishing , interfering majorana modes must necessarily have originated in the same contact . in contrast , in two particle quantities , like the current noise , two - particle exchange permits interference of two majorana particles even if they originate from different leads . the physics of neutral excitations is certainly an important field of future research . it can be expected that the theory of majorana particles represents an instructive example of this development . we thank c. w. j. beenakker , p. w. brouwer , a. r. akhmerov , m. wimmer , i. martin and a. f. morpurgo for discussions . m. b. thanks the kitp of beijing and santa barbara for hospitality where part of this work was carried out . the work is supported by the swiss national science foundation , the european marie - curie training network nanoctm and the swiss national center of competence in research on quantum science and technology . we prove in this appendix that for an arbitrary matrix @xmath273 , one can find a decomposition [ eq : appdeco ] _ n= , such that @xmath274 and @xmath275 satisfy @xmath276 where @xmath277 ( @xmath278 ) stands for the @xmath162-th row of @xmath42 ( @xmath43 ) , and the @xmath279@xmath280@xmath279 matrix @xmath281 is defined as @xmath282 before we start the proof , we point out that @xmath42 ( @xmath43 ) satisfying eq . ( eq . ) is nothing but @xmath42 ( @xmath43 ) given in eq . written in a differently ordered basis ( the purpose of such reordering is to make the central matrix at the right - hand - side of eq . taking a compact block - diagonal form ) . this can be shown by using explicitly @xmath283 hence we have , from eq . , @xmath284 where @xmath285 and @xmath286 are @xmath6-dimensional unitary matrices ( cf . eqs . and ) . it is then straightforward to see that by rearranging , for @xmath287 , the @xmath233-th row and column to the ( @xmath288)-th and the ( @xmath289)-th row and column to the ( @xmath290)-th , the rearranged @xmath42 and @xmath43 satisfy eqs . and . where both @xmath292 and @xmath293 are normalized @xmath279-dimensional row vectors and @xmath294 is a @xmath279@xmath280@xmath279 matrix . we assume @xmath295 ( namely , the upper - left corner element of the matrix @xmath291 is not @xmath296 ) , such that both @xmath292 and @xmath293 are well - defined up to an irrelevant sign . then we construct two matrices @xmath297 by setting @xmath298 and choosing @xmath299 and @xmath300 with @xmath301 so that the orthogonality and eqs . and are all satisfied ( which can be easily shown to be always possible ) . we find @xmath302 where @xmath303 . it remains to show that @xmath42 and @xmath43 defined above satisfy eqs . and . to this end we observe @xmath310 , hence eqs . and are equivalent to @xmath311 = 0,\quad [ \mathcal{w } , \lambda ] = 0,\end{aligned}\ ] ] where @xmath94 $ ] stands for the commutator . we also keep in mind that , by construction , all factor matrices at the right - hand - sides of eqs . and satisfy the above commutation relations . this immediately verifies the validity of our construction , and thus completes the proof .

this paper is a forward modeling investigation of the plasma dynamics of polar plumes . these so - called `` polar rays '' or polar plumes ( e.g. , van de hulst 1950 ; saito 1958 ; harvey 1965 ; koutchmy , 1977 ; deforest et al . 1997 ) trace open magnetic field lines to different heights depending on observation type . in extreme ultraviolet ( euv ) emission line spectroheliograms , plumes appear as short spikes near the polar limb ( bohlin et al . 1975a ; ahmad & withbroe 1977 ; widing & feldman 1992 ; walker et al . 1993 ) with unchanged shape and sudden brightness changes over typical lifetimes ranging from several hours to days ( see deforest et al . 1997 ; del zanna et al . euv observations can be used to characterize the bases of polar plumes and also their plasma characteristics ( densities , temperatures , flows , abundances , etc . ) and also other processes such as heating , wave propagation , etc . ( e.g. bohlin et al . 1975b ; walker et al . 1988 , 1993 ) . soft x - rays show mainly the hot plasma at the foot - points of the plumes ( e.g. , ahmad & webb , 1978 ) that are also seen as weak radio sources ( gopalswamy et al . , 1992 ) . deforest et al . ( 1997 , 2001a ) have shown that most white light polar rays can be associated directly with the euv polar plumes and their foot - points on the solar surface . above the solar limb , polar plumes appear brighter than the surrounding media , suggesting that they are denser than the background corona or inter - plume regions ( e.g. van de hulst 1950 ; saito 1958 & 1965 ; etc ) . the coronal - background electron density is about @xmath2 @xmath3 and falls sharply with height ( see different density models in raouafi & solanki 2006 ) . in comparison , the density of polar plumes ranges from @xmath4 @xmath3 ( del zanna et al . 1997 ; young et al . 1999 ) . young et al . ( 1999 ) reported electron densities corresponding to the strong brightenings at plume foot - points of the order of @xmath5 @xmath3 and found no decrease with height up to 70 mm . they also inferred temperatures of the order of 2 mk . in a recent publication , wilhelm ( 2006 ) using off - limb spectroscopic measurements from the spectrometer sumer ( wilhelm et al . 1995 ) on board soho ( domingo et al . 1995 ) found that plumes are approximately 5 - 6 times denser than inter - plume regions up to @xmath6 . the plume - interplume density ratio changes little in this height range . note that these are off - limb measurements and they do not take into account the plume area projected on the solar disk . the contribution of polar plumes to the fast solar wind arising from the polar coronal holes has been a subject of debate and controversy . in general , plumes are thought to harbor smaller outflow speeds than inter - plume regions ( e.g. , habbal 1992 ; wang 1994 ; hassler et al . 1999 ; etc ) . thus , polar plumes observed above the solar limb seem to be nearly in hydrostatic equilibrium and show small or no line shifts , suggesting that the bulk of the acceleration of the solar wind takes place at larger heights ( see del zanna et al . 1997 ; kohl et al . 1997 ; wilhelm et al . 2000 ; teriaca et al . 2003 ) . doppler dimming techniques applied to soho / uvcs ( kohl et al . 1995 ) observations of the lines have shown outflows in inter - plume dark lanes ranging from 100 to 150 km s@xmath7 at @xmath8 from sun center and smaller speeds in bright plumes ( 0 - 65 km s@xmath7 ; see giordano et al . however , gabriel et al . ( 2003 ) used the doppler dimming technique applied to soho / sumer data and found the intensity ratio of the doublet to be smaller in plumes than in inter - plumes . they concluded that outflow speeds of the solar wind are greater in plumes than in inter - plumes . unfortunately , they did not analyze line profiles , in particular the line widths , on which the doppler dimming effect depends . euv coronal emissions in spectral lines that are formed within narrow temperature ranges are excellent temperature diagnostics for different coronal structures . soho / eit ( delaboudinire et al . 1995 ) images ( 195 that forms at @xmath9 mk and /@xmath10 171 at @xmath11 mk ) reveal that plumes are cooler than the background corona ( deforest et al . 1997 ; young et al . 1999 ) . spectral analysis have also shown that uv lines are narrower in plumes than in inter - plume regions ( e.g. 1032 line widths are 10 - 15% lower ; hassler et al . 1997 ; etc ) . this suggests that the plume material is relatively cooler than or has less turbulence than dark lane plasma ( inter - plume regions ) . wilhelm et al . ( 1998 ) used intensity ratios of temperature sensitive lines for accurate measurements of temperatures in the low corona . they found evidence for an increasing temperature with height in the background corona , but no rise of temperature in plumes ( see also young et al . 1999 ) . the geometry of plumes is an important key in understanding their contribution to the fast solar wind . the high - speed solar wind at even moderate solar latitudes is thought to originate within coronal holes at high latitudes and to be affected by superradial expansion of the coronal hole magnetic field ( zirker 1977 ) . from white light observations plumes are thought to also expand superradially with altitude ( ahmad & withbroe 1977 ; munro & jackson 1977 ; fisher & guhathakurta 1995 ) . deforest et al . ( 1997 ) found that plumes expand highly superradially ( with a half - cone angle of 45 ) in their lowest 20 - 30 mm , and more slowly above . deforest et al . ( 2001b ) also found that below @xmath12 plumes expand superradially at the same rate as the background coronal hole . however , woo & habbal ( 1999 ) and woo et al . ( 1999 ) have claimed to observe radial , rather than superradial , expansion of plumes in the coronal holes . speculation on interaction between plume and inter - plume material has been reported , in particular at high altitudes in the corona . saito and tanaka ( 1957a , b ) were the first to associate plumes with magnetic flux concentrations ( faculae ) in the polar regions . polar plumes arise from footpoints 2 - 4 mm ( @xmath13 ) wide and expand rapidly with height ( see deforest et al . a simulation has shown that a typical plume width at the base of the corona is 20 to 28 mm ( klaus wilhelm , private communication ) . several authors ( harvey 1965 ; newkirk & harvey 1968 ; lindblom 1990 ; allen 1994 ; etc . ) have suggested that polar plumes arise mainly from unipolar magnetic structures coinciding with the supergranular network boundaries . however , deforest et al . ( 1997 ) have shown that not all of the flux concentrations give rise to polar plumes . other authors suggest that unipolar magnetic flux concentrations reconnect with ephemeral emergent flux having opposite polarity ( saito 1965 ; wang 1994 , 1998 ; wang & sheeley 1995 ; etc ) . this is supported by the high temperature observed at the plume footpoints , producing high pressure plasma from the dipole that can escape along open field lines . deforest et al . ( 2001b ) found that polar plumes are of episodic nature : `` polar plumes are both transient and persistent : they are recurring structures that brighten for only 1 day but reappear at approximately the same location for up to 2 - 3 weeks '' . they speculate that this behavior is the result of multiple magnetic reconnections of the unipolar flux concentrations at the network boundaries encountering multiple ephemeral dipoles driven by supergranular motion . we consider model plumes having different outflow speeds and turbulence and calculate the line profile shape of the plume plasma . we use the observed profile shapes of the lines as primary criteria to select the appropriate height variation of those quantities . we then refine them to reproduce the other measured quantities , i.e. line widths , total intensities and intensity ratios . in the following sections , we study the effect of polar plume locations along the line of sight on the line profiles and intensities of the doublet and ly@xmath0 . the comparison with the observed quantities deduced from soho / uvcs spectra allows us to constrain the dynamic properties of the plasma inside plumes . in section 2 , we present a summary of plume spectroscopic observations by soho / uvcs and highlight the spectral properties of 1032 . a polar plume model is described in section 3 and the results from this model are presented and discussed in sections 4 - 7 . conclusions and prospects are to be found in section 8 . a preliminary version of some of the results has been reported by raouafi et al . the ultraviolet coronagraph spectrometer ( uvcs ; kohl et al . 1995 ) has provided important spectral information on coronal lines in polar holes . the main spectral lines recorded by uvcs are ly@xmath0 , the doublet at 103 nm and the doublet centered at around 61 nm ( in the order of decreasing emission ) . spectral analysis of uvcs data has shown that ly@xmath0 has , to a good approximation , a gaussian shape at all heights accessible to this instrument . kohl et al . ( 1999 ) pointed out small deviations from gaussians at the profile peaks , allowing for two line components resulting from the contributions of different plasma media with different properties along the los . giordano et al . ( 1997 ) have also shown that ly@xmath0 profiles are narrower when ray paths intersect high density structures ( plumes ) than otherwise ( inter - plumes ) . the change with height of the line profiles is more complex than for ly@xmath0 . while at low altitudes ( say below @xmath14 ) the observed profiles are relatively gaussian shaped , with small deviations in the wings and peaks , more pronounced deviations of the observed profiles from a single gaussian are clearly seen in the observations at higher altitudes up to about @xmath15 . at this height , the profiles are well represented by two gaussian components : a narrow and a broad one . the latter appears to be associated with regions of low density ( inter - plume regions ) while the narrow component shows a clear association with the denser regions along the los ( higher emission regions : e.g. , polar plumes ) . the spectral behavior of the two components as a function of height above the limb is quite different . the width of the broad component increases dramatically with height , while that of the narrow one remains almost constant ( see kohl et al . 1997 ; 1999 ) . lines have a narrow component at @xmath16 , accounting for only a small fraction of the observed spectral radiance . for higher altitudes , the observations are not good enough to distinguish whether or not a narrow coronal component exists . kohl et al . ( 1999 ) have shown also that the @xmath17 values for the narrow components of and are smaller than those for 1216 at the same heights , and the narrow value is smaller than that of . at the solar activity minimum the polar coronal holes are well developed and bordering streamers appear not to reach the uvcs observation altitudes , making it very likely that narrow components are contributions of fine structures inside the polar coronal hole . polar plumes are the most prominent features at the heights at which uvcs observes , which makes them the main source of the narrow component observed in the spectral lines . figs . [ september19 - 231997 ] and [ june031996 ] display spectral profiles of 1032 recorded directly above the pole on 1997 september 19 - 23 and 1996 june 03 , respectively , using the @xmath18 long uvcs slit . eit images and kitt peak vacuum telescope magnetograms clearly show the presence of polar plumes against the solar disk and magnetic concentrations with footpoints at @xmath19 ( fig . [ september19 - 231997 ] ) and @xmath20 ( fig . [ june031996 ] ) away from the north pole . the profiles in the first figure are formed of two distinct components ( narrow and broad ) . two gaussians are needed to give satisfactory fits to the profiles at @xmath21 and partly also around @xmath22 , although with a small contribution of the narrow component . however , the profiles in the second figure are very close to a gaussian in shape , although they show a small degree of asymmetry . thus , the spectral profiles depend relatively strongly on the distance of the plume footpoints to the pole and also on to the brightness of the plume . in other words , they depend on where the plume crosses the los at a given altitude . this difference in the profile shapes is a valuable tool to study plasma dynamic in these coronal fine structures . the characteristics of the plumes spectral contribution ( narrow component ) to lines as observed by uvcs are summarized as follows : * dominates the profiles below @xmath23 and decreases above ; * not seen in the profiles beyond @xmath24 . only the broad component remains above this altitude . note that the noise level in the profiles increases with height , which might screen small contributions from fine coronal structures along the los ; * mostly located at the center of the broad component which is close to the rest wavelength of the spectral lines . thus , no significant doppler shifts are observed for the narrow component ; * the width of narrow component hardly changes as function of height in the range where this latter is observed ( see kohl et al . in order to study the influence of polar plumes on the los - integrated profiles of and , we consider a number of plumes aligned along the los , with foot points placed at different angular locations in the polar coronal hole ( see fig . [ ppmes_spat_dist ] ) . the importance of the angular locations of the plume footpoints is enhanced because of the superradial expansion of the magnetic field . the footpoint location determines the los outflow speed of the plume plasma at the location where the plume crosses the los . based on the work of deforest et al . ( 2001b ) and others we assume that plumes expand superradially at the same rate as the surrounding background corona ( inter - plumes ) according to the global magnetic model by banaszkiewicz et al . the electron density is considered to be greater in plumes than in inter - plumes , following wilhelm et al . ( 1998 ) and wilhelm ( 2006 ) . the density contrast is largest just above the solar surface and decreases gently with height , such that it disappears above @xmath25 ( see set in fig . [ ippp_densities ] ) . the inter - plume electron density stratification is obtained from soho data ( sumer , uvcs and lasco ( brueckner et al . 1995 ) ; see doyle et al . 1999a , b : solid line in fig . [ ippp_densities ] ) . the density profile of plumes ( dashed line in fig . [ ippp_densities ] ) is obtained by considering a higher density at the footpoints at the solar surface ( @xmath26 ; see del zanna et al . 1997 ; young et al . 1999 ) and a lower temperature inside plumes ( @xmath27 ) than in inter - plumes ( @xmath28 and @xmath29 , respectively , for the model by doyle et al . 1999a , b ) . the variation of both densities as a function of height is given by eq . ( 1 ) in doyle et al . ( 1999a , b ) . we assume that the considered coronal species have simple maxwellian velocity distributions inside both plume and inter - plume regions , but with different velocity turbulence values , @xmath30 , that are functions only of height in both structures . the plume outflow speed , @xmath31 , is assumed to be proportional to the solar wind speed in inter - plume regions , @xmath32 , and the proportionality coefficient is assumed to depend only on the radial distance to sun center : @xmath33 , where @xmath34 is obtained through the mass - flux conservation equation . because of the lack of conclusive observational results concerning @xmath35 and @xmath36 we consider height profiles for each . two cases are considered for the plume velocity turbulence : @xmath37 constant ( @xmath38 for the lines following observational values from kohl et al . 1997 , 1999 ) and a height - dependent turbulent velocity @xmath39 . the value of @xmath40 is approximately the observed width of the narrow component as measured at @xmath41 ( see fig . 3 of kohl et al . , 1999 ) . in the latter case a low value of the turbulence is prescribed at low altitudes , which increases to reach the inter - plume value above @xmath42 . three cases are considered for @xmath35 : constant , linear and non - linear . the data presented in figs . [ september19 - 231997 ] and [ june031996 ] correspond to plumes based between @xmath43 from the pole . in order to compare our simulations with these observed profiles , we limit ourselves in this section to the very left plume in fig . [ ppmes_spat_dist ] that has its foot - point centered at @xmath44 from the pole . this plume lies between the examples shown in figs . [ september19 - 231997 ] and [ june031996 ] . in the present section , we consider the four test cases ( sections 4.1 - 4.4 ) presented in table 1 . note that we have computed the lines also for the remaining combinations of @xmath45 and @xmath31 , but nothing new was to be learned from them . & @xmath46 & @xmath47 + 2 . & @xmath46 & @xmath48 + 3 . & @xmath49 & @xmath48 + 4 . & @xmath49 & @xmath50 + for the inter - plume regions , the plasma properties are similar to those found by raouafi & solanki ( 2004 & 2006 ) . the theoretical formalism used here for the line profile computations is discussed by raouafi & solanki ( 2004 ) and references therein . in the present section , we limit ourselves to discussing the profiles of the 1032 line . 1037 is also computed and is found to behave similarly to the 1032 line . ly@xmath0 profiles are also computed , but this line is less sensitive to the behavior of @xmath51 and @xmath52 and its shape does not change as much as that of 1032 . thus , in contrast to the lines , the ly@xmath0 profile is close to a gaussian for most plume parameters and at most heights . in a first step , we consider a plasma outflow speed inside plumes that is equal to a constant fraction of that in the inter - plume regions . plume speeds that are smaller and greater than inter - plume values are considered ( @xmath53 & @xmath54 ) . @xmath45 is height independent and equal to 50 km s@xmath7 . [ fig_o6_pp_pap_asc_vppcvip ] displays the profiles of the los - integrated 1032 line at @xmath8 and @xmath55 for different @xmath56 values . for small @xmath56 , the plume contribution ( narrow component ) dominates the profiles at low altitudes , showing some resemblance with the observed profiles ( see solid lines in figs . [ fig_o6_pp_pap_asc_vppcvip]a - b ) . however , at high altitudes the narrow component is still clearly present with a strength well above the typical noise level present in the observed profiles . for greater values of @xmath56 ( figs . [ fig_o6_pp_pap_asc_vppcvip]c - d ) , the plume contribution is doppler shifted away from the line peak already at low altitudes ( due to the location of the considered plume relatively far from the pole ) . this produces strongly asymmetric profiles , unlike the observed ones . also , the plume contribution is still significant at high altitudes . the present case ( @xmath57 ) does not reproduce the overall observed profiles . consequently , it does not reflect the plasma conditions and mechanisms and taking place in polar plumes . nonetheless , the computations suggest that outflow speeds in plumes are likely to be much smaller ( @xmath58 ) than inter - plume values at low altitudes . this avoids large contributions from plumes far from the central wavelength of the observed line irrespective of their footpoint location . plume outflow speeds close to or greater than than the inter - plume values at low altitudes will be excluded in the rest of the analysis . in this section , we keep a height - independent velocity turbulence inside plumes ( @xmath59 km s@xmath7 ) . however , the plume - interplume outflow speed ratio is now allowed to vary linearly with height , such that @xmath60 where @xmath61 and @xmath62 are constants chosen such that @xmath63 is zero or very close to zero at the solar surface and increases toward unity at high altitudes , roughly above @xmath64 . a possible mechanism for such an acceleration of the plume plasma could be interaction with the surrounding background medium having higher outflow speeds . we consider the three height profiles of @xmath65 that are given in fig . [ asconsvpplvip]a . los - integrated profiles computed along rays reaching @xmath15 and @xmath55 from disk center are plotted in figs . [ asconsvpplvip]b - c . the obtained profiles are to some extent similar to those in fig . [ fig_o6_pp_pap_asc_vppcvip ] for @xmath66 . although the profiles at low altitudes now all have the narrow component nearly centered at the broad component , which is close to the observations , those obtained at high altitudes have the same significant narrow contributions shifted to the wings of the profile . such a signal is not observed in the corona . the plume outflow speed is that of case 2 , while the velocity turbulence of the plume plasma is assigned a height - dependent function . a variation in the velocity turbulence appears natural if cool plume and hot inter - plume material interacts . due to the lack of observational results concerning the variation of both parameters , we assume , for simplicity , similar profiles for the turbulence inside plumes and inter - plumes @xmath67 . thus , @xmath36 has the following functional form ( see raouafi & solanki 2004 , 2006 ) @xmath68 where @xmath69 , @xmath70 , @xmath71 and @xmath72 are adjustable parameters whose values have been selected in order to get as close as possible to the observations . since there is observational evidence for cooler plume plasma than in inter - plumes , we assume that @xmath36 is smaller than @xmath67 at low altitudes and converges gently towards the large values of @xmath73 ( see fig . [ vpplvvip_alpsppnl]a ) . this guarantees the presence of a narrow component at low heights and progressively broader plume profiles as we move further above the solar limb . figs . [ vpplvvip_alpsppnl]b - h displays the 1032 line profiles obtained for lines - of - sight ranging from @xmath74 to @xmath75 above the pole for the plume outflow speed profiles shown in fig . [ asconsvpplvip]a . at altitudes below @xmath76 the contribution of polar plumes to the line profile is narrower than that of inter - plumes and is approximately centered at the rest wavelength of the line . this results in approximately gaussian - shaped line profiles that are slightly narrower than those obtained without polar plumes ( see fig . [ vpplvvip_alpsppnl]b ) . for @xmath77 a profile that can be well represented by two gaussians and that resembles the observed profiles is obtained ( figs . [ vpplvvip_alpsppnl]c - f ) . the broad component is due to inter - plume material , the narrow component to plumes . due to the low plume outflow speeds at these altitudes , the doppler shift of the narrow component is low . above @xmath78 , the increase in the plume velocity turbulence results in increasingly broader plume contributions , so that the line profile ( see figs . [ vpplvvip_alpsppnl]g - h ) takes on the approximate shape of a single gaussian for the dashed velocity profile in fig . [ asconsvpplvip]a ( dotted lines in the same figure ) . for the other velocity models , the plume component is narrower , in particular , for the dotted plume velocity model , so that one can still distinguish two line components at high altitudes . hence , high plume outflow speeds at @xmath79 that are closer to inter - plume values fit the observations better than smaller ones . the present case is in better agreement with the observation than cases 1 and 2 , although the agreement is not perfect , both at very low and very high altitudes . e.g. at low altitudes the plume contribution does not dominate the los - integrated profiles . probably the plume outflow speed at these altitudes is still high enough to dim the radiative plume component ( shifted from resonance due to the small velocity turbulence ) . improvement can be obtained by considering a non - linear height - dependent @xmath65 allowing for lower and higher plume speeds at low and high altitudes , respectively . this is the aim of case 4 . the velocity turbulence in plumes @xmath36 is again described by eq.([nl_alpspp ] ) and is illustrated by the triple - dot - dashed curve in fig . [ vpplvvip_alpsppnl]a . the plume outflow speed is now considered to be related in a non - linear way to the inter - plume speed : @xmath80 where @xmath35 is illustrated by the solid line in fig . [ vppnlvvip_alpsppnl]a . the plume - inter - plume solar wind speed ratio is less than 10% below @xmath81 from sun center , and increases abruptly to more than 75% by about @xmath82 , then converges gently toward unity . this profile allows more contribution from the plume radiative component at @xmath83 and favors broader plume profiles at @xmath84 . the los - integrated 1032 line profiles are displayed in figs . [ vppnlvvip_alpsppnl]b - h , together with the plume ( dotted profile ) and inter - plume ( long - dashed line ) contributions . the plume contribution is now , as expected , more significant at low altitudes , in particular below @xmath85 and dominates the profile below @xmath86 ( see fig . [ vppnlvvip_alpsppnl]b ) . the change in the plume contribution is due to combined effects of the electron density decrease but mostly to the variation of the outflow speed and turbulence of the plasma inside plumes . the calculated profiles are close to a gaussian shape at @xmath87 and display increasingly a bi - gaussian shape above this altitude up to @xmath88 ( figs . [ vppnlvvip_alpsppnl]c - d ) . above @xmath89 , the obtained profiles can not be represented by one gaussian and the two components ( narrow and broad ) can be distinguished even without fitting . the width of the plume component increases relatively rapidly above @xmath81 and so does its doppler shift . this sudden change in the profile of the plume contribution allows the two ( broad and narrow ) components to merge completely above @xmath90 . the resemblance of the computed profiles to the observed ones is now uniformly good ( compare profiles in figs . [ september19 - 231997 ] & [ vppnlvvip_alpsppnl ] ) . note that greater gas densities in plumes above @xmath91 than the one considered here allow for a greater contribution from plumes that results in broader los - integrated profiles at high altitudes , that are even closer to the observed ones . any remaining small differences between computed and observed profiles could also reflect the footpoint location of the modeled plume , or the possible presence of multiple plumes along the los in the observations . case 4 best reproduces the observed profile shapes . we expect this model to be the closest to the plasma conditions in the polar plumes . in other words , the plume material remains cooler and much slower than the inter - plume plasma up to @xmath81 from sun center , where a rapid increase in both outflow speed and effective temperature takes place , leading to plume dynamic properties converging towards those of inter - plume region within the next solar radius . the rapid change in plume plasma properties is likely to be due to interaction between the cool and slow plume and hot and fast inter - plume materials , although some authors speculated that this interaction takes place much higher in the solar corona ( see introduction ) . in the next sections , we will consider this case only for an extensive study of the profiles and total intensities of the different spectral lines considered here . the contribution of polar plumes located very close to or on the pole will also be discussed . fig . [ vppnlvvip_alpsppnl_wtirat ] displays the height dependence above the pole of the 1032 line width ( fig . [ vppnlvvip_alpsppnl_wtirat]a ) and total intensity ( fig . [ vppnlvvip_alpsppnl_wtirat]b ) as well as of the intensity ratio of the doublet ( figs . [ vppnlvvip_alpsppnl_wtirat]c - d ) from slightly above the solar limb to @xmath92 from sun center . the profiles underlying this figure were computed according to case 4 ( section 4.4 ) . they were fit with two gaussians below @xmath93 and with one gaussian above . the widths of the broad and narrow components and of the single gaussian fit are represented by different symbols . also plotted are values obtained from uvcs spectra together with their corresponding error bars according to kohl et al . ( 1997 ) and cranmer et al . the narrow component widths hardly change with height up to @xmath93 and fit the measured values relatively well . the widths of the broad component , however , change rapidly with height , leading to a non - gaussian shape of the composite profiles . higher up the obtained line widths are comparable to those obtained by raouafi & solanki ( 2006 ; fig . 8) although they lie somewhat close to the measurements , in particular , around @xmath93 . total intensities are also comparable to those computed by raouafi & solanki ( 2006 ) although the intensities below @xmath93 fit the observed values for the 1032 line better . the change in total intensities can be more clearly seen in the intensity ratio of the doublet . above @xmath94 , the ratios obtained from the composite profiles ( full circles ) are improved compared to the case without plumes ( open circles ) , in particular , at low altitudes . all the computed ratios now lie inside the error bars of uvcs measurements . note that only simple maxwellian velocity distribution are used and no anisotropy in the kinetic temperature of the ions is considered . below @xmath94 , the intensity ratios with plume contribution taken into account are smaller than those without plume contribution ( see bottom panel of fig . [ vppnlvvip_alpsppnl_wtirat ] ) . in our computations , this is due to the large electron density numbers in plumes compared to inter - plume densities and is not a solar wind effect as reported by gabriel et al . ( 2003 & 2005 ) . in addition , we showed previously that plume wind speeds faster or just close to inter - plume speeds give line profiles different from those observed in the polar coronal holes . here we demonstrate that the fast solar wind originates mainly from inter - plume regions and that the plume plasma remains slower and cooler up to approximately @xmath95 from sun center ( see fig . fig . [ lyalp_prwti]a displays the velocity turbulence used for the computation of ly@xmath0 ( solid line for inter - plume and dotted line for plume ) . we assume the same @xmath96 variation of the plume velocity turbulence as in the case of the ions : small values at low altitudes and a rapid increase toward inter - plume values above @xmath81 . the plume - inter - plume ratio of the outflow speed of the hydrogen atoms is assumed to be the same as for the ions in the previous section ( see fig . [ vppnlvvip_alpsppnl]a ) . note that the hydrogen outflow speed in inter - plume regions is lower than that of ions ( see raouafi & solanki 2006 ) . the computed line profiles of ly@xmath0 are shown in fig . [ lyalp_prwti]b - j ( solid - black lines ) together with the plume ( dot - dashed lines ) and inter - plume ( dotted lines ) contributions for different rays crossing the polar solar axis at different heights . the red curves are single gaussian fits to the computed profiles . the high quality of the fits at all heights demonstrates how close the computed profiles are to a gaussian shape . however , small deviations from the single gaussian shape are noticeable , in particular , at low altitudes in the wings and peaks of the profiles . kohl et al . ( 1999 ) observed similar deviations and interpreted them as due to the contribution of polar plumes . note that slightly changing ( @xmath97 ) the model parameters ( i.e. the outflow speed and turbulence of the plasma inside plumes ) does not change much the obtained profiles , which remain gaussian shaped to a good approximation . the widths of the computed line profiles are displayed in fig . [ lyalp_prwti]k as a function of the projected heliocentric distance of the corresponding rays ( full circles ) . the widths obtained without plumes ( open circles ) are also displayed for comparison , together with the best fit to uvcs observations ( solid curve ) and the error bars of the observed values ( vertical lines ; see cranmer et al . line widths obtained in both cases ( with and without polar plumes ) are within the errors bars of the measured values . however , the widths obtained with plumes taken into account fit the observed values better . by optimizing the number and position of plumes and their properties , an even better match to the observations may be obtained . [ lyalp_prwti]l shows the total intensity of ly@xmath0 . the best fit to the observed intensities along with the error bars are also displayed for comparison . at low altitudes , intensities obtained with plumes taken into account are very similar to those without plumes and are slightly greater than the measured ones . however , at higher altitudes intensities obtained with plumes provide an improved fit to the observations . generally , the presence of plumes along the los slightly increases the total intensity due to the density enhancement in plumes . fig . [ footpoints_fig]a - d shows the los - integrated profiles of the 1032 line with contributions from each of the four right - hand plumes in fig . [ ppmes_spat_dist ] taken into account separately . the contributions from the two polar plumes based at colatitude @xmath98 and @xmath99 ( the narrow component of the line profile ) dominate the profiles up to @xmath78 and remain clearly visible up to @xmath100 ( dotted and solid lines ) which is not the case for the observed profiles . the reason for the continuing visibility of the truly polar plumes up to large altitudes is that the rapid acceleration of the solar wind in the plume hardly affects the plume profile , since it is almost perpendicular to the los . the contribution from the plume at @xmath101 is important at low altitudes but relatively small at high altitudes ( dashed lines ) . this suggests that polar plumes preferentially originate more than @xmath102 away from the pole . the contribution from the plume with footpoint at @xmath103 ( dot - dashed lines ) is very similar to that shown in fig . [ june031996 ] due to plumes at @xmath20 from the solar pole . various authors have studied locations of polar plume foot - points within polar regions . they traced plumes arising near the edges of coronal holes , relatively far away from the solar pole ( eclipse observations ; see saito 1965 ) . in contrast , we did not find any report on plumes very close to the polar axis . this case may be very rare for unknown physical reasons ( for example , bipolar flux may not emerge near the pole ) . polar coronal plumes can be easily identified off - limb in white light ( lasco , eclipses ) and euv ( eit , trace ) images . on the solar disk , for instance by using eit images , it is also possible to trace them back to the solar surface , in particular , using images recorded in the wavelength bands containing the hot lines 171 and 195 that are regularly recorded by eit . x - ray images also allow the determination of the locations of plume foot - points , that correspond very often to bright points reflecting the high temperatures there . once the plume foot - points are identified , it is relatively easy to link them to the underlying magnetic structures using magnetograms . data recorded daily over several consecutive days or better still weeks help to survey the latitudes of polar plume foot - points . for this purpose , periods of time when the solar axis is tilted toward the observer are more useful since the observation of a major area of the polar coronal hole including the pole is possible . these times correspond to march and september of each year . eit images recorded at solar minimum seem to support this hypothesis . an average of all daily unsigned solis ( synoptic optical long - term investigations of the sun ) high sensitivity magnetograms recorded in september 2005 ( see bottom panel of fig . [ footpoints_fig ] ) show magnetic flux concentrations preferentially between @xmath104 and @xmath105 latitude , although we are not at solar minimum yet . another reason why plumes are more likely to have their footpoints some degrees away from the pole has got to do with the relative area available for plumes between 70 and @xmath106 is 3 times larger than at latitudes above @xmath106 . however , preliminary analysis of solis magnetograms tends not to support the area argument . this subject will be addressed in detail in a subsequent publication . the aim of the present study is to obtain information on the height evolution of the plasma dynamics inside polar plumes . we compared profile shapes of the doublet ( 1032 and 1037 ) and the ly@xmath0 spectral lines observed by soho / uvcs at different altitudes in polar coronal holes to synthetic profiles that include the contribution of polar plumes placed at different locations along the los . observational constraints on these fine coronal structures ( densities , temperatures , geometry , ... ) have been taken into account following different reports in the literature . since the outflow speed and kinetic temperature evolution with height are not conclusively known , we consider different height profiles for the outflow speed and plasma turbulence inside plumes and study how these choices influence the line profiles of the doublet , which are more sensitive to the detailed dynamic structure along the los than ly@xmath0 . by comparing the synthetic profiles to the observed ones in the polar coronal holes , we determined plausible height profiles for the outflow speed and plasma turbulence in polar plumes with respect to the surrounding background corona . we find that models with constant ( height - independent ) velocity turbulence in the plumes show narrow components with a significant contribution to the los - integrated profile at all heights , in particular above @xmath107 where there is no observational indication of the presence of such a narrow component in the observed profiles . this is independent of whether the plume outflow speed equals a constant or a height - dependent fraction of the inter - plume outflow speed . synthetic profiles with shapes very similar to the observed ones are obtained when the plasma turbulence in plumes is height - dependent with similar variation to that in inter - plume regions and converges towards the latter at greater heights . at the same time , the outflow speed is small near the solar surface , but approaches the inter - plume value at greater heights . the narrow component is found to contribute significantly to the resultant profiles at altitudes below @xmath108 and is increasingly dimmed above that altitude . this model suggests that the plume plasma remains much slower and cooler ( in the sense that @xmath109 ) than inter - plume material up to @xmath110 . above @xmath110 these parameters experience a rapid rise towards inter - plume values reached by @xmath111 . the rapid acceleration and increase in the effective temperature of the plasma inside plumes is probably the result of interaction with the faster and hotter inter - plume material . the entrainment or mixing of material between plumes and inter - plumes is hindered by the magnetic field . the smaller the magnetic energy density relative to the thermal or the kinetic energy density , the more efficiently entrainment takes place . now consider the following numbers : from @xmath112 to @xmath113 the density @xmath114 decreases by a factor of approximately @xmath115 , while the area of a plume increases by roughly @xmath116 . consequently the kinetic energy density , @xmath117 , remains roughly constant over this height range , while the magnetic energy density , @xmath118 drops by roughly a factor of @xmath115 . this may explain why the plum properties remain distinct from those of the inter - plume gas up to about @xmath113 and then start to merge . as far as the actual interaction is concerned , it may be enhanced by , e.g. , the kelvin - helmholtz instability which can act at the boundary between two streams with sufficiently different speeds . in addition electron collisions may not be sufficient at heights @xmath119 to keep the different species in equilibrium . consequently , plume species are heated and accelerated differently and rapidly as the density drops sharply with height . this may not be the case below this altitude where the density is high enough to keep all species in equilibrium contrasting inter - plumes regions where the ( electron ) density is smaller allowing for the plasma equilibrium to brake down at lower altitudes and thus the hotter and faster plasma compared to plume regions . we find that observed profile shapes , line widths and total intensities of the doublet and of ly@xmath0 are better reproduced when the influence of appropriate plumes is included . in particular , the intensity ratios of the doublet below @xmath74 are smaller when the plume contribution is taken into account . this is due to the density enhancement in the plume region . this result is in disagreement with the interpretation preferred by gabriel et al . ( 2003 & 2005 ) that such a reduction in the ratio is due to the doppler dimming effect , which would imply a higher outflow speed in plumes than in inter - plume regions . in our analysis , such high outflow speeds in the plumes produce synthetic profiles whose narrow component is doppler shifted towards a profile wing , in contrast to the observations . it is noteworthy that including polar plumes increases the agreement of the computed profiles with the observed ones and with critical parameters like line width or line ratio even if no anisotropy in the temperature ( or turbulence velocity ) is included in the computations . note that the properties of the plumes were constrained basically using the line profile shape only and the quantitative comparison with the line parameters and ratio was only carried out later . this strengthens the thesis put forward by raouafi & solanki ( 2004 & 2006 ) that there is no compelling need for large anisotropies in temperature in coronal holes based on published uvcs data . note that we have not tried to optimally reproduce the observed quantities by carrying out an exhaustive search in parameter space . thus the obtained results can possibly be improved further by applying a @xmath120 minimization procedure to fit the line profiles . plumes with footpoints within @xmath19 of the pole display significant narrow components even at altitudes above @xmath78 , contrary to the observations . eit images during the solar minimum together with high sensitivity magnetograms from solis show evidence supporting the hypothesis that plumes originate away from the pole . this subject will be studied in detail in a subsequent paper . the authors would like to thank craig deforest for constructive comments and critics that greatly improved the manuscript and john raymond for helpful discussions . the nso is operated by the association of universities for research in astronomy , inc . under cooperative agreement with the national science foundation . ner work is supported by nso and the nasa grant nnh05aa12i . cranmer , s. r. , kohl , j. l. , noci , g. , et al . 1999 , , 511 , 481 deforest , c. e. , hoeksema , j. t. , gurman , j. b. , thompson , b. j. , plunkett , s. p. , howard , r. , harrison , r. c. , hassler , d. m. 1997 , , 175 , 393 fisher , r. a. , & guhathakurta , m. 1995 , , 447 , l139 gabriel , a. h. , bely - dubau , f. , & lemaire , p. 2003 , , 589 , 623 gabriel , a. h. , abbo , l. , bely - dubau , f. , llebaria , a. , & antonucci , e. 2005 , , 635 , l185 harvey , j. w. 1965 , , 141 , 832 hassler , d. m. , wilhelm , k. , lemaire , p. , & schuehle , u. 1997 , , 175 , 375 hassler , d. m. , dammasch , i. e. , lemaire , p. , , brekke , p. , curdt , w. , mason , h. e. , vial , j .- c . , & wilhelm , k. 1999 , science , 283 , 810 kohl , j. l. , esser , r. , gardner , l. d. , et al . 1995 , , 162 , 313 kohl , j. l. , noci , g. , antonucci , e. , et al . 1997 , , 175 , 613 kohl , j. l. , esser , r. , cranmer , s. r. , et al . 1999 , , 510 , l59 koutchmy , s. 1977 , solar phys . , 51 , 399 . saito , k. , & tanaka , y. 1957a , , 9 , 106 saito , k. , & tanaka , y. 1957b , , 9 , 210 saito , k. 1958 , , 10 , 49 saito , k. 1965 , , 17 , 1 teriaca , l. , poletto , g. , romoli , m. , & biesecker , d. a. 2003 , , 588 , 566 walker , a. b. c. , deforest , c. e. , hoover , r. b. , & barbee , t. d. w. : 1993 , , 148 , 239 wang , y .- m . 1994 , , 435 , l153 wang , y .- m . , & sheeley , n. r. , jr . 1995 , , 452 , 457 wang , y .- m . 1998 , , 501 , l145 widing , k. g. , & feldman , u. 1992 , , 392 , 715 wilhelm , k. , marsch , e. , dwivedi , b. n. , hassler , d. m. , lemaire , p. , gabriel , a. h. , huber , m. c. e. 1998 , , 500 , 1023 wilhelm , k. , dammasch , i. e. , marsch , e. , & hassler , d. m. 2000 , , 353 , 749 wilhelm , k. 2006 , , 455 , 697 woo , r. , & habbal , s. r. 1999 , , 510 , l69 woo , r. , habbal , s. r. , howard , r. a. , & korendyke , c. m. 1999 , , 513 , 961 young , p. r. , klimchuk , j. a. , & mason , h. e. 1999 , , 350 , 286 zirker , j. b. , ed . 1977 , coronal holes and high speed wind streams ( boulder : univ . colorado press )

this paper is concerned with the classification of legendrian rational unknots in lens spaces . the lens space in question may be equipped with a tight or an overtwisted contact structure , but in the latter case we require that the knot complement be tight . legendrian knots in overtwisted contact @xmath2-manifolds with tight complement are called _ non - loose _ or _ exceptional_. the classification of legendrian unknots in @xmath3 is due to eliashberg and fraser @xcite . in the case of the tight standard contact structure @xmath4 on @xmath3 , their classification is up to isotopy ; in the case of exceptional unknots in an overtwisted contact structure , up to coarse equivalence . recall that two legendrian knots @xmath5 , @xmath6 , in contact @xmath2-manifolds are called _ coarsely equivalent _ if there is a contactomorphism @xmath7 carrying @xmath8 to @xmath9 . here the contact structures are understood to be ( co-)oriented , and the contactomorphism is supposed to preserve the ( co-)orientation . the classification of legendrian knots in overtwisted contact manifolds up to isotopy is complicated by the fact there are contactomorphisms topologically but not contact isotopic to the identity , cf . the discussion in ( * ? ? ? * section 4.3 ) . in the present paper we extend the classification result of eliashberg and fraser to rational unknots in lens spaces . our focus will lie on the exceptional case , and we too are content with the classification up to coarse equivalence . the classification in the tight case is essentially due to baker and etnyre @xcite , although they give an explicit description only for @xmath0 with @xmath10 odd . we obtain a complete classification ( both in the tight and the exceptional case ) for the lens spaces @xmath0 with @xmath10 any integer . as an illustration of the general case we also discuss the classification for @xmath1 . in particular , we determine the range of the classical invariants realisable by such rational unknots , and the @xmath2-dimensional homotopy invariant of the contact structures containing exceptional knots . this classification is achieved as follows . the number of distinct legendrian rational unknots with tight complements is determined via the classification of tight contact structures on solid tori ; this strategy has previously been employed by etnyre @xcite . we then describe the expected number of legendrian rational unknots explicitly in a contact surgery diagram of the lens space . for the @xmath2-sphere , such a description is due to plamenevskaya @xcite . in all cases the knots are distinguished by the rational analogues of the classical legendrian knot invariants . here is an outline of the paper . in section [ section : unknots ] we recall the topological classification of rational unknots in lens spaces . in section [ section : classical ] we describe a result of lisca et al . @xcite about the computation of the classical legendrian knot invariants of a legendrian knot presented in a contact surgery diagram . we extend their result from integral to rational homology spheres and from nullhomologous to rationally nullhomologous knots . the invariants in this case are the rational legendrian knot invariants @xcite . in section [ section : plane ] we recall how to compute the @xmath2-dimensional homotopy invariant of a contact structure from a surgery diagram . this will be used in some cases to show that the contact structure defined by a certain surgery diagram is overtwisted . sections [ section:3-sphere ] to [ section : l52 ] contain the classification for @xmath3 , @xmath11 , @xmath0 and @xmath1 , respectively . the classification for @xmath3 and @xmath11 is of course subsumed by that for @xmath0 . nonetheless , the separate description of those two simple cases allows us to include some additional details and to make the whole classification scheme more transparent . many of the necessary computations are relegated to section [ section : compute ] . we understand that blent tosun has been working on the classification problem discussed in this paper using a parallel approach , but staying closer to the argument in @xcite rather than relying on surgery diagrams for the existence part of the classification . this may in fact be advantageous for dealing with the general @xmath12 . the lens space @xmath12 with @xmath13 and @xmath14 coprime to @xmath10 is defined as the quotient space of @xmath15 under the @xmath16-action generated by @xmath17 this gives @xmath12 a canonical orientation , and our contact structures are assumed to be positive for that orientation . alternatively , @xmath12 with the described orientation can be obtained by surgery along a single @xmath18-framed unknot in @xmath3 . a _ rational unknot _ @xmath19 in some @xmath2-manifold @xmath20 is a knot with a rational seifert disc , i.e. some cable of @xmath19 on the boundary @xmath21 of a tubular neighbourhood of @xmath19 is supposed to bound a @xmath22-disc @xmath23 embedded in @xmath24 , cf . as discussed in that paper , the union @xmath25 equals the complement of an open ball in a lens space , so for the study of rational unknots one may restrict attention to the case of @xmath20 being a lens space . moreover , a rational unknot we always mean a rational unknot that is not an honest unknot , i.e. the homological order of the knot is supposed to be greater than @xmath26 . ] in @xmath12 is then necessarily the spine of one of the heegaard tori . recall that the genus 1 heegaard splitting of a lens space is unique up to isotopy @xcite . hence , up to isotopy there are at most four oriented rational unknots in @xmath12 , namely @xmath27 , @xmath28 , where @xmath29\co 0\leq\theta\leq 2\pi / p\}\subset l(p , q),\ ] ] and likewise for @xmath30 . for @xmath31 this reduces in fact to one possibility , and for @xmath32 , @xmath33 the knot @xmath34 is isotopic to @xmath35 , both being fibres in an @xmath36-bundle over @xmath37 . in homology one has @xmath38=q[k_1]$ ] , so for @xmath39 there are indeed four rational unknots up to isotopy , see ( * ? ? ? * lemma 5.2 ) . in the surgery picture , @xmath40 can be represented as in figure [ figure : standard - unknot ] . the knot @xmath30 would be the spine of the solid torus glued in to perform the surgery . by exchanging the role of the two heegaard tori , which induces the orientation - preserving diffeomorphism @xmath41 for @xmath42 mod @xmath10 , one gets a similar picture for @xmath30 , with the surgery coefficient replaced by @xmath43 . @xmath40 [ r ] at 0 18 @xmath44 [ l ] at 70 20 .,title="fig : " ] by expanding the rational surgery into integral surgeries along a link of unknots , one can give representations of @xmath40 and @xmath30 in a single surgery diagram . for instance , in the case @xmath45 , @xmath46 we can take @xmath47 . since @xmath48 and @xmath49 we can represent @xmath40 and @xmath30 as in figure [ figure : k1k2 ] . a slam dunk , cf . * figure 5.30 ) , will then produce figure [ figure : standard - unknot ] or the analogous picture for @xmath30 , respectively . @xmath40 [ r ] at 0 18 @xmath30 [ l ] at 175 18 @xmath50 [ tl ] at 38 12 @xmath51 [ tl ] at 63 12 @xmath50 [ tr ] at 109 12 @xmath51 [ tr ] at 135 12 .,title="fig : " ] since we are interested in a classification up to coarse equivalence , we need to take the action of the diffeomorphism group into account . in this topological setting , coarse equivalence of two knots is supposed to mean that there is an orientation - preserving diffeomorphism of the ambient manifold sending one knot to the other . up to coarse equivalence there is exactly one oriented rational unknot in @xmath12 for @xmath52 mod @xmath10 , and exactly two for @xmath53 mod @xmath10 . the orientation - preserving diffeomorphism of @xmath12 induced by @xmath54 sends @xmath55 to @xmath56 , @xmath28 . thus , topologically we can ignore the orientation of @xmath55 . for @xmath52 mod @xmath10 , the orientation - preserving diffeomorphism of @xmath12 induced by @xmath57 exchanges @xmath40 and @xmath30 . for @xmath53 mod @xmath10 , there are only two orientation - preserving diffeomorphisms of @xmath12 up to isotopy : the identity and the one described above , see @xcite . so @xmath40 and @xmath30 can not be coarsely equivalent . however , as pointed out in @xcite for the tight case , the two different orientations of @xmath40 may correspond to non - equivalent legendrian realisations . similar considerations apply in the exceptional case . this gives some information about the contactomorphism group of the corresponding contact structure . in this section we first recall from @xcite how to define the classical invariants for _ rationally _ nullhomologous legendrian knots . we then show how to compute these invariants for knots presented in a surgery diagram of the ambient manifold , provided this manifold is a rational homology sphere . this constitutes a mild extension of a result due to lisca et al . * lemma 6.6 ) . let @xmath58 be a rationally nullhomologous legendrian knot in a contact @xmath2-manifold , i.e. @xmath59 is of some order @xmath60 in @xmath61 . then there is a rational seifert surface @xmath62 for @xmath63 , i.e. a surface that is embedded , except along its boundary , which is an @xmath64-fold covering of @xmath59 . note that the boundary of @xmath65 need not be connected . ( one can replace @xmath59 by a suitable embedded curve or collection of curves on the boundary of a tubular neighbourhood of @xmath59 , representing the class @xmath63 in the tubular neighbourhood , and then find an embedded surface with that curve ( or those curves ) as its boundary . ) let @xmath66 be a push - off of @xmath59 in the direction of the contact framing , i.e. the framing determined by @xmath67 . then the _ rational thurston bennequin invariant _ of @xmath59 is @xmath68 i.e. the rational linking number of @xmath59 and @xmath66 . any two rational seifert surfaces for @xmath59 differ by a class in @xmath69 . since @xmath66 is rationally nullhomologous , its intersection number with such a class is zero , so @xmath70 is well defined . now assume that @xmath59 is oriented . the contact structure @xmath71 is a trivial plane bundle when restricted to the rational seifert surface @xmath65 , and the _ rational rotation number _ @xmath72 is defined by writing @xmath73 for the number of full turns of the positive tangent vector to @xmath59 , as @xmath59 is traversed @xmath64 times , relative to the trivialisation of @xmath74 . in general , this number will depend on the relative homology class represented by @xmath65 . if the euler class @xmath75 is a torsion class , then @xmath72 is well defined . as shown in @xcite , any ( closed , connected ) contact @xmath2-manifold @xmath76 has a contact @xmath77-surgery presentation @xmath78 , i.e. there is a legendrian link @xmath79 in @xmath3 with its standard tight contact structure such that contact @xmath77-surgery along the components of @xmath80 produces @xmath76 . now let @xmath81 be a contact @xmath77-surgery presentation of a rational homology @xmath2-sphere @xmath76 . then @xmath82 will be a finite abelian group , so the euler class @xmath75 is a torsion class . furthermore , let @xmath83 be a legendrian knot that becomes rationally nullhomologous in @xmath76 . denote the link components of @xmath79 by @xmath84 , and set @xmath85 , depending on whether @xmath86 belongs to @xmath87 or @xmath88 . so @xmath89 is the integral surgery coefficient of the link component @xmath86 . write @xmath20 for the linking matrix of @xmath79 , i.e.@xmath90 define an extended matrix by @xmath91 in other words , @xmath92 is the linking matrix of @xmath93 , with the convention that @xmath94 is set to @xmath95 . as a final piece of notation , we write @xmath96 for the rotation number of @xmath86 , @xmath97 , and @xmath98 for the thurston bennequin invariant of @xmath99 , all regarded as knots in @xmath100 . we can now formulate a lemma that tells us how to compute the classical invariants of @xmath99 when it is regarded as a legendrian knot in the surgered manifold @xmath76 . for the case that @xmath76 is an integral homology sphere , this lemma is due to lisca et al . * lemma 6.6 ) . ( the condition that @xmath101 be an integral homology sphere is not contained in the statement of their lemma , but in the paragraphs preceding it , and it is used implicitly in their argument . ) [ lem : loss ] the rational invariants of @xmath102 are given by @xmath103 and @xmath104 these formulae are exactly the same as those in @xcite , with @xmath105 and @xmath106 replaced by their rational counterparts . however , since it is not entirely clear where the order of @xmath59 in @xmath61 might or might not be relevant for the argument , we deem it worth to include a proof . before we turn to that proof , we want to discuss the behaviour of the relative euler class of a contact structure under a contact @xmath77-surgery along a legendrian knot @xmath86 . a neighbourhood of @xmath86 can be identified with a neighbourhood of @xmath107 , equipped with the contact structure @xmath108 . we think of @xmath109 as the solid torus we cut out during the surgery . the boundary @xmath110 of this solid torus is a convex surface with two dividing curves @xmath111 . these curves lie in the class of the longitude @xmath112 giving the contact framing . write @xmath113 for the meridian of @xmath114 and @xmath115 for the standard longitude @xmath116 . then @xmath117 . thus , in terms of the standard meridian and longitude @xmath118 , the slope of the dividing curves is @xmath119 . the convex torus @xmath110 has a linear legendrian ruling given by the @xmath120-curves , which represent the class @xmath121 . write @xmath122 for meridian and longitude , respectively , of a solid torus we glue in to perform the surgery . contact @xmath123-surgery can be described by the gluing maps @xmath124 contact @xmath125-surgery , by the maps @xmath126 note that in both cases @xmath127 gets glued to @xmath112 . so the slope of the dividing curves on the boundary of the solid torus we want to glue in is again @xmath119 , now with respect to @xmath122 . the legendrian ruling of the convex torus @xmath114 in the original model can be changed from the class @xmath128 to either @xmath113 or @xmath129 by an isotopic deformation of the @xmath22-torus through convex tori of slope @xmath119 and linear legendrian ruling , staying inside any arbitrarily small neighbourhood of the original torus . after this modification , we see that contact @xmath77-surgery simply corresponds to regluing a standard solid torus with slope @xmath119 and linear legendrian ruling in the class @xmath130 . now suppose that we perform contact @xmath77-surgery along a legendrian knot @xmath59 in a contact @xmath2-manifold @xmath131 . the relative euler class @xmath132 is poincar dual to the class in @xmath133 represented by the zero set of a generic section of @xmath134 that coincides with the tangent direction along @xmath59 . by what we just discussed , we may assume that this section coincides with the legendrian ruling on the boundary of a standard tubular neighbourhood of @xmath59 , and this section will extend without zeros over the solid torus we glue in when performing a surgery along @xmath59 . translated into our situation at hand , this implies the following statement . [ lem : euler ] under the natural map @xmath135 induced by inclusion , the poincar dual of the relative euler class @xmath136 maps to the poincar dual of @xmath137 . write @xmath138 for the meridian of @xmath86 , and @xmath139 for the longitude determined by the surface framing . then @xmath140 where @xmath141 denotes a copy of the integers generated by the class @xmath113 . in @xmath142 , the longitude @xmath139 represents the class @xmath143 surgery with coefficient @xmath89 along @xmath86 means that we glue a new meridional disc along @xmath144 , @xmath145 . it follows that @xmath146 on the other hand , from the mayer vietoris sequence of the triple @xmath147 and the assumption that @xmath101 be a rational homology sphere ( as well as some obvious identifications under excision isomorphisms ) we have the short exact sequence @xmath148 with @xmath61 a finite abelian group . we have @xmath149 . the class @xmath150 maps to @xmath95 in @xmath151 ; the class @xmath130 , to @xmath26 . so the sequence reduces to @xmath152 ( alternatively , this follows from @xmath99 being rationally nullhomologous in @xmath101 . ) hence the poincar dual of the relative euler class @xmath137 over the rationals , @xmath153 is some rational multiple of @xmath150 . beware that over the integers @xmath150 is not , in general , a primitive element in @xmath154 . for instance , for @xmath155 and @xmath99 representing the generator of @xmath156 , the class @xmath150 is twice the generator of @xmath157 . the definition of the rotation number of a legendrian knot can be interpreted in terms of relative euler classes . this translates into @xmath158 for the rational rotation number @xmath159 of @xmath102 we argue similarly . if the order of @xmath99 in @xmath61 is @xmath64 , and @xmath65 is a rational seifert surface for @xmath99 in @xmath101 , then @xmath160\rangle= \operatorname{\mathrm{pd}}(e(\xi , l_0))\bullet\sigma.\ ] ] for this intersection product , only the free part of @xmath161 is relevant , and since the intersection product @xmath162 $ ] equals @xmath64 , we conclude that @xmath163 hence , by lemma [ lem : euler ] , @xmath164 the relations in the presentation of @xmath154 can be written formally as @xmath165 the surgery description of @xmath101 defines a @xmath166-dimensional handlebody @xmath167 with boundary @xmath168 . both @xmath169 and @xmath170 are isomorphic to @xmath171 . the relevant part of the homology exact sequence of the pair @xmath172 is @xmath173 since @xmath101 is a rational homology sphere , we have @xmath174 , so the matrix @xmath20 is invertible over @xmath175 . therefore , in @xmath176 we have @xmath177 this proves the first formula in the lemma . we now turn to the thurston bennequin invariant . the contact framing of @xmath99 ( both in @xmath100 and in @xmath76 ) is given by @xmath178 . on the other hand , from the exact sequence ( [ eqn : sequence ] ) we see that there is a unique @xmath179 such that @xmath180 is nullhomologous in @xmath181 , where as before @xmath64 denotes the order of @xmath99 in @xmath101 . then the rational thurston bennequin invariant of @xmath99 in @xmath76 can be computed as @xmath182 where the intersection product should be interpreted as a product on the boundary of a tubular neighbourhood of @xmath99 . from ( [ eqn : lambda ] ) we have @xmath183 and the fact that this is nullhomologous in @xmath181 means that it can be expressed as a linear combination of the relations in @xmath154 , which yields @xmath184 with ( [ eqn : tb ] ) we get @xmath185 a surgery presentation @xmath78 of a given contact @xmath2-manifold @xmath76 determines a @xmath166-dimensional @xmath22-handlebody @xmath167 with boundary @xmath168 . following @xcite and @xcite we are now going to explain how to determine the homotopical data of @xmath71 as a tangent @xmath22-plane field from such a presentation . given a choice of spin structure @xmath186 on @xmath101 , there is an invariant @xmath187 of the homotopy type of @xmath71 over the @xmath22-skeleton of @xmath101 . when the first chern class @xmath188 is a torsion class , the homotopy obstruction over the @xmath2-skeleton can be described by a rational number @xmath189 . in ( * theorem 4.12 ) it is shown how to compute @xmath190 from a surgery diagram containing only contact @xmath123-surgeries . a spin structure @xmath186 can be specified in terms of a characteristic sublink in the surgery diagram . blent tosun has informed us of a way to compute the @xmath22-dimensional homotopy invariant @xmath190 from any contact surgery diagram , including those with @xmath26-handles and contact @xmath125-surgeries . this would allow one to give a complete homotopy classification of the contact structures on @xmath12 we describe in terms of surgery diagrams in the following sections . the @xmath2-dimensional invariant can be computed as follows . write @xmath191 for the signature of @xmath167 , and @xmath192 for its euler characteristic . let @xmath193 be the surface obtained by gluing a seifert surface of @xmath86 with the core disc of the handle corresponding to @xmath194 . the homology class of @xmath195 in @xmath169 is completely determined by @xmath196 . generalising a result of gompf , the following was shown in @xcite . [ prop : d3 ] suppose that @xmath188 is torsion , and @xmath197 for each @xmath198 . then @xmath199 where @xmath200 denotes the number of components of @xmath87 , and @xmath201 is the cohomology class determined by @xmath202 for each @xmath203 . see @xcite of an extensive discussion of this formula , in particular concerning the computation of the term @xmath204 . the standard ( and unique tight ) contact structure @xmath4 on @xmath3 has @xmath205 . on @xmath0 there are , by @xcite and @xcite , exactly @xmath206 tight contact structures up to isotopy . they can be obtained by contact @xmath123-surgery on @xmath3 along a single unknot with invariants @xmath207 and @xmath208 obtained from the standard legendrian unknot ( @xmath209 ) by any mix of @xmath210 positive or negative stabilisations . the @xmath211-invariant of the corresponding contact structure on @xmath0 is given by @xmath212 topologically trivial legendrian knots in arbitrary tight contact @xmath2-manifolds were shown by eliashberg and fraser @xcite to be classified up to legendrian isotopy by the classical invariants @xmath105 and @xmath106 , and they determined the range of these invariants . so part ( a ) of the following theorem is a weaker formulation of their result , which we include for completeness and comparison with the case of exceptional knots . the exceptional unknots in the @xmath2-sphere @xmath3 have also been classified , up to coarse equivalence , by eliashberg and fraser ( * ? ? ? * theorem 4.7 ) . an alternative proof of their result was given by etnyre and vogel , see @xcite . we are going to give yet another proof of this classification , which contains _ in nuce _ all the key ideas required to extend the result to lens spaces . our argument for determining an upper bound on the number of exceptional knots is parallel to that of @xcite . the proof is then completed by finding as many explicit realisations of exceptional unknots as this bound allows . in the case of @xmath3 , these explicit realisations are due to plamenevskaya @xcite . [ thm : elfr ] ( a ) let @xmath213 be a legendrian unknot . then @xmath214 with @xmath215 a negative integer , and @xmath216 lies in the range @xmath217 any such pair of invariants @xmath218 is realised , and it determines @xmath59 up to coarse equivalence , i.e. for each @xmath219 we have @xmath220 distinct legendrian unknots . \(b ) let @xmath221 be an exceptional unknot in an overtwisted contact structure @xmath71 on @xmath3 . then @xmath71 is the contact structure determined up to isotopy by @xmath222 , and @xmath223 these invariants determine @xmath59 up to coarse equivalence , and any pair of invariants in this set is realised . examples of legendrian unknots in @xmath100 that realise the invariants stated in the theorem are given by arbitrary stabilisations of a standard legendrian unknot with @xmath224 and @xmath225 . examples of exceptional unknots with the stated invariants have been described by plamenevskaya @xcite in terms of the front projection of the knot in a contact surgery diagram , see figure [ figure : olga - lens ] . @xmath226 at 0 167 @xmath227 at 0 55 @xmath228 at 88 167 @xmath59 [ bl ] at 66 132 @xmath229 [ l ] at 66 148 @xmath229 [ l ] at 66 107 @xmath59 [ bl ] at 64 39 @xmath229 [ l ] at 64 14 @xmath59 [ bl ] at 146 39 @xmath229 [ l ] at 146 14 @xmath119 [ l ] at 146 58 @xmath119 [ l ] at 146 76 @xmath119 [ l ] at 146 132 @xmath119 [ l ] at 146 150 -sphere.,title="fig : " ] each of these diagrams gives a copy of @xmath3 , as can be seen by simple kirby moves , cf . a straightforward computation with the formula from proposition [ prop : d3 ] shows that each diagram gives a contact structure on @xmath3 with @xmath230 . so this contact structure is overtwisted ( and determined up to isotopy by this value of @xmath211 thanks to eliashberg s classification of overtwisted contact structures @xcite , cf . contact @xmath123-surgery along @xmath59 cancels the @xmath125-surgery along the parallel knot , see ( * ? ? ? * section 3 ) or ( * ? ? ? * prop . this leaves us with a diagram containing only contact @xmath123-surgeries , or one with a single @xmath125-surgery along the standard legendrian unknot . the latter produces the tight contact structure on @xmath231 , see ( * ? ? ? * lemma 4.3 ) , the former a stein fillable and hence tight contact structure . this shows that in all examples the knot @xmath59 is exceptional . we claim that the knot @xmath59 in figure [ figure : olga - lens](a ) has @xmath232 , the one in ( b ) has @xmath233 ( depending on a choice of orientation of @xmath59 ) , and the one in ( c ) ( with @xmath234 unknots along which we perform @xmath123-surgery ) has @xmath235 . plamenevskaya determines @xmath236 by keeping track of the contact framing through kirby moves ; no comment is made about @xmath216 . in fact , the claimed values of @xmath236 and @xmath216 follow easily from lemma [ lem : loss ] . see section [ section : compute ] for some details of these computations . so we are left with showing that these invariants ( in the tight and exceptional case , respectively ) determine @xmath59 up to coarse equivalence , and that no other values of the classical invariants can be realised . given a legendrian unknot @xmath59 in @xmath100 or an exceptional unknot @xmath59 in @xmath3 , decompose the @xmath2-sphere along a torus as @xmath237 , with @xmath238 a standard neighbourhood of @xmath59 . more precisely , with @xmath239 denoting meridian and longitude of the solid tori @xmath240 , we assume that the gluing is described by the identifications @xmath241 , @xmath242 , and @xmath243 is a convex torus with two dividing curves of slope @xmath244 , where @xmath245 . both in the tight and the exceptional case , the contact structure on @xmath246 is tight , and the boundary @xmath247 is convex with two dividing curves of slope @xmath215 . moreover , up to coarse equivalence @xmath59 is determined by the contact structure on @xmath246 . according to giroux @xcite and honda @xcite , the number of tight contact structures on a solid torus @xmath248 inducing a fixed characteristic foliation on @xmath249 divided by two curves of slope @xmath250 is given by @xmath251 where the @xmath252 are the terms in the continued fraction expansion @xmath253;\ ] ] for slope @xmath119 there is a unique structure . for @xmath254 the continued fraction expansion is given by @xmath255 and @xmath256 , i.e. we have @xmath220 distinct tight structures , which corresponds to the @xmath220 realisations of a legendrian unknot with @xmath257 in @xmath100 described above . for @xmath258 , the contact structure on @xmath246 would have to be overtwisted , so this case does not occur in @xmath100 or when @xmath59 is exceptional . finally , for @xmath259 we first have to modify @xmath246 by a dehn twist such that the dividing curves have a slope @xmath260 , in order to apply the classification result cited above . a dehn twist of @xmath246 that replaces @xmath261 by @xmath262 changes the slope to @xmath263 , since @xmath264 for @xmath265 we have to take @xmath266 , which gives @xmath267 . for this slope there is exactly one tight contact structure on @xmath246 . for @xmath268 , we have to take @xmath269 , resulting in a slope @xmath270 . in this case there are exactly two tight contact structures on @xmath246 , for inductively one sees that the continued fraction expansion of @xmath271 is @xmath272 $ ] . thus , for @xmath273 the number of tight contact structures on @xmath246 equals the number of examples in figure [ figure : olga - lens ] . it follows that these examples constitute a complete list of exceptional unknots . the following is the analogue of theorem [ thm : elfr ] for the real projective space @xmath274 . [ thm : rp3 ] ( a ) let @xmath275 be a legendrian rational unknot in the unique tight contact structure on @xmath11 . then @xmath276 with @xmath215 a negative integer , and @xmath72 lies in the range @xmath277 any such pair of invariants @xmath278 is realised , and it determines @xmath59 up to coarse equivalence , i.e. for each @xmath219 we have @xmath220 distinct legendrian rational unknots . \(b ) up to coarse equivalence , the exceptional rational unknots @xmath59 in an overtwisted @xmath279 are in one - to - one correspondence with the following set of values of the classical invariants : @xmath280 the overtwisted contact structure containing the knots of the first series has @xmath211-invariant equal to @xmath281 , the one containing the second series has @xmath282 . in other words , there is exactly one exceptional rational unknot with @xmath283 , there are three with @xmath284 , and there are four each for @xmath285 with @xmath268 . legendrian rational unknots in @xmath286 that realise the stated values of the invariants are given as follows . represent @xmath287 by @xmath123-surgery along a single standard legendrian unknot in @xmath3 with @xmath224 and @xmath225 , and let @xmath99 be a push - off of the surgery curve with @xmath288 positive and @xmath289 negative stabilisations , @xmath290 . observe that by lemma [ lem : loss ] the push - off without any stabilisations has @xmath291 . examples of exceptional rational unknots with the stated invariants are shown in figures [ figure : rp3-tb3 ] and [ figure : rp3-tb5 ] . with some simple kirby moves one sees that in all cases @xmath59 is an isotopic copy of the standard rational unknot @xmath292 . we illustrate this in section [ section : compute ] for the example in figure [ figure : rp3-tb5](a ) ; there we also explain how @xmath70 can be computed from such kirby moves instead of lemma [ lem : loss ] . @xmath226 at 0 97 @xmath227 at 103 97 @xmath228 at 203 97 @xmath59 [ bl ] at 76 56 @xmath229 [ l ] at 76 44 @xmath229 [ l ] at 76 33 @xmath229 [ l ] at 76 24 @xmath59 [ bl ] at 177 56 @xmath229 [ l ] at 177 21 @xmath59 [ bl ] at 286 74 @xmath229 [ l ] at 286 19 -space i.,title="fig : " ] @xmath226 at 0 185 @xmath227 at 145 185 @xmath293 [ r ] at 0 96 @xmath293 [ l ] at 221 96 @xmath59 [ bl ] at 69 39 @xmath229 [ l ] at 69 14 @xmath119 [ l ] at 65 56 @xmath119 [ l ] at 65 74 @xmath119 [ l ] at 65 136 @xmath119 [ l ] at 65 167 @xmath59 [ br ] at 145 31 @xmath229 [ r ] at 145 23 @xmath119 [ r ] at 155 56 @xmath119 [ r ] at 155 74 @xmath119 [ r ] at 155 136 @xmath119 [ r ] at 155 167 -space ii.,title="fig : " ] the invariants of these exceptional examples are listed in table [ table : rp3 ] . a sample computation of these invariants is given in section [ section : compute ] . since the @xmath211-invariant differs from @xmath294 , all contact structures given by these surgery diagrams are overtwisted . .invariants of the exceptional rational unknots in @xmath11 . [ cols="^,^,^,^,^",options="header " , ] in this section we collect some hints for the computation of the invariants in various of the examples described above . we consider the example in figure [ figure : olga - lens](c ) . here the linking matrix @xmath295 is the @xmath296-matrix @xmath297 where we have numbered the surgery curves @xmath298 from bottom to top in the figure . by successive subtraction of the @xmath299 from the @xmath300 row , @xmath301 , we obtain @xmath302 hence @xmath303 . the first row of @xmath304 is @xmath305 this information suffices to compute the rotation number of @xmath59 ( with clockwise orientation ) : @xmath306 by successive subtraction of the @xmath299 from the @xmath300 row , starting from @xmath307 , in the @xmath308-matrix @xmath309 we obtain @xmath310 by expanding the determinant of this matrix along the last row one shows inductively that @xmath311 hence @xmath312 we consider the example in figure [ figure : rp3-tb5](a ) . the kirby moves that transform the surgery link into a single unknot with topological framing @xmath51 , and @xmath59 into the rational unknot of figure [ figure : standard - unknot ] , are shown in figure [ figure : rp3-kirby ] . these moves are analogous to those in plamenevskaya s example ( * ? ? ? * figure 4 ) ; we say more about them further down , where we use them to compute @xmath70 without appealing to lemma [ lem : loss ] . \(i ) at 0 192 @xmath51 at 9 169 @xmath59 [ bl ] at 40 187 @xmath119 [ b ] at 26 155 @xmath51 [ tl ] at 50 160 @xmath51 [ tl ] at 73 160 @xmath51 [ t ] at 115 160 @xmath50 [ t ] at 136 157 ( ii ) at 0 131 @xmath59 [ bl ] at 40 128 @xmath229 [ r ] at 0 113 @xmath229 [ r ] at 0 104 @xmath229 [ b ] at 26 95 @xmath51 [ tl ] at 50 100 @xmath51 [ tl ] at 73 100 @xmath51 [ t ] at 115 100 @xmath50 [ t ] at 136 97 ( iii ) at 0 75 @xmath59 [ bl ] at 28 73 @xmath229 [ r ] at 1 55 @xmath229 [ r ] at 1 45 @xmath229 [ t ] at 26 34 @xmath51 [ tl ] at 50 41 @xmath51 [ tl ] at 73 41 @xmath51 [ t ] at 115 41 @xmath50 [ t ] at 136 38 ( iv ) at 0 15 @xmath59 [ r ] at 60 16 @xmath51 [ l ] at 101 16 ( a).,title="fig : " ] the linking matrix @xmath295 is the @xmath308-matrix @xmath313 where , as before , we have numbered the surgery curves @xmath84 from bottom to top . observe that this @xmath314 equals the @xmath315 from the previous example , with a single change in the very last entry of the matrix . hence , arguing as before , one obtains @xmath316 . by expanding the determinant of @xmath309 , transformed as in the previous example , along the last row , and using the result from the previous example , one obtains @xmath320 hence @xmath321 alternatively , @xmath70 can be computed by keeping track of the framing of @xmath59 during the kirby moves in figure [ figure : rp3-kirby ] . in @xmath3 we have @xmath322 , so initially the contact framing of @xmath59 is given by @xmath323 , where @xmath130 is ( and remains throughout the following moves ) the longitude corresponding to the surface framing in @xmath3 . this framing curve can be thought of as a parallel copy of @xmath59 , also going through the @xmath324-box . to get from ( i ) to ( ii ) , we make two positive blow - ups , i.e. we add two @xmath125-framed unknots to the picture ( corresponding to taking the connected sum with two copies of @xmath325 ) , and then slide @xmath59 and the @xmath123-framed knot over them to undo the @xmath324-linking . this adds two positive twists to the framing of @xmath59 , so the framing is now @xmath130 . to get from ( ii ) to ( iii ) , we slide @xmath59 over the parallel @xmath125-framed unknot . this adds @xmath229 to the framing of @xmath59 . to get from ( iii ) to ( iv ) , we first blow down the two @xmath125-framed unknots not linked with @xmath59 . this has no effect on @xmath59 , but changes the third @xmath125-framed unknot into a @xmath123-framed one . now we blow down the chain of unknots , starting with the @xmath123-framed one . at each step , the adjacent @xmath324-framed unknot gets framing @xmath119 , and the framing of @xmath59 increases by @xmath26 . since we have to blow down a total of @xmath326 @xmath123-framed unknots to obtain ( iv ) , the framing of @xmath59 finally becomes @xmath327 . now recall equation ( [ eqn : tb ] ) from the proof of lemma [ lem : loss ] and the argument preceding it . the unique @xmath179 such that @xmath328 is nullhomologous in the surgered manifold given by figure [ figure : rp3-kirby](iv ) is @xmath329 . hence @xmath330 giving us the same result for @xmath70 as before . here is how to compute @xmath189 for this example . the surgery diagram is equivalent to @xmath326 unlinked @xmath123-framed unknots and a further unlinked @xmath324-framed unknot . so the signature of the @xmath166-dimensional filling @xmath167 is @xmath331 , its euler characteristic is @xmath332 . in order to compute @xmath204 , we follow the algorithm described in @xcite . the poincar dual @xmath333 in terms of the obvious generators of @xmath334 , the meridional discs to the surgery curves is given by the vector @xmath335 of rotation numbers . the homomorphism @xmath336 induced by inclusion is described , again in terms of the obvious bases , by the linking matrix @xmath20 , so the class @xmath337 that maps to @xmath338 can be thought of as a row vector with @xmath339 . so this vector @xmath340 is given by the sum of the vectors in ( [ eqn : row1 ] ) and ( [ eqn : row2 ] ) . then @xmath341 then with equation ( [ eqn : d3 ] ) , observing that @xmath342 in this example , we obtain @xmath343 @xmath59 [ r ] at 2 137 @xmath119 at 42 137 @xmath95 [ tl ] at 68 128 @xmath95 [ tl ] at 79 119 @xmath59 [ r ] at 0 78 @xmath229 [ b ] at 61 86 @xmath344 [ t ] at 63 56 @xmath229 [ r ] at 51 77 @xmath229 [ r ] at 73 77 @xmath59 [ r ] at 9 18 @xmath345 [ l ] at 77 20 ( a).,title="fig : " ] the linking matrix is the @xmath346-matrix @xmath20 with zeros on the diagonal and all other entries equal to @xmath26 . it is a simple exercise to show that @xmath347 . correspondingly , @xmath348 . it follows that @xmath349 since @xmath350 , we have @xmath351 . next we consider the example in figure [ figure : lp1](b ) . here the topological kirby diagram shows directly that @xmath59 is the rational unknot in @xmath0 . the linking matrix @xmath20 is the @xmath352-matrix @xmath353 , the matrix @xmath92 is @xmath354 hence @xmath355 the rotation numbers @xmath96 , @xmath356 , equal @xmath357 ( with @xmath59 oriented clockwise ) . hence @xmath358 so for @xmath359 in the range @xmath360 we get @xmath344 different legendrian realisations . finally , we come to the example in figure [ figure : lp1](c ) . here the computations are minor modifications of those for the example we discussed in the case of @xmath11 . the kirby moves for showing that @xmath59 is the rational unknot are as in figure [ figure : rp3-kirby ] . the linking matrix @xmath20 differs from the one in that previous case by the substitution of @xmath361 for @xmath50 . thus , one finds @xmath362 and @xmath363 , which yields @xmath364 the first row of @xmath304 is now @xmath365 the last row is @xmath366 with the rotation number of the surgery curve at the top of figure [ figure : lp1](c ) being @xmath367 , one computes with lemma [ lem : loss ] that @xmath72 , with either orientation of @xmath59 , takes for @xmath368 the values claimed in theorem [ thm : lp1 ] . we close with some comments about the computation of the @xmath211-invariant . for the example in figure [ figure : lp1](a ) , the only term in formula ( [ eqn : d3 ] ) that is not entirely obvious is the signature . topologically , the surgery diagram consists of @xmath344 @xmath95-framed unknots with a common @xmath123-linking . by making a @xmath125-blow up and sliding the corresponding @xmath125-framed unknot over this link , we obtain @xmath344 unlinked @xmath125-framed unknots , all of which are linked once with the extra @xmath125-framed unknot . by sliding the @xmath344 unknots off the extra one , we obtain an unlink consisting of a single @xmath353-framed unknot and @xmath344 @xmath125-framed unknots . this describes a filling of signature @xmath10 . since we had to add a @xmath125-framed unknot to arrive at this picture , we have @xmath369 . for the example in figure [ figure : lp1](c ) , one sees @xmath370 by an argument similar to that for ( a ) . since the surgery diagram corresponds to adding @xmath215 @xmath22-handles , we have @xmath371 . the vector of rotation numbers is given by @xmath372 , i.e. we need to solve the equation @xmath373 over @xmath175 . this is achieved by @xmath374 then one computes as in section [ subsection : rp3 ] . a major part of the work on this project was done during an inspiring `` research in pairs '' stay at the mathematisches forschungsinstitut oberwolfach in september 2012 . we thank the forschungsinstitut for its support , and its efficient and friendly staff for creating an exceptional research environment . 10 , rational linking and contact geometry , _ perspectives in analysis , geometry , and topology _ , progr . math . * 296 * ( birkhuser , basel , 2012 ) , 1937 . , grid diagrams and legendrian lens space links , _ j. symplectic geom . _ * 7 * ( 2009 ) , 415448 . , diffotopies des espaces lenticulaires , _ topology _ * 22 * ( 1983 ) , 305314 . , a legendrian surgery presentation of contact @xmath2-manifolds , _ math . proc . cambridge philos . _ * 136 * ( 2004 ) , 583598 . , surgery diagrams for contact @xmath2-manifolds , _ turkish j. math . _ * 28 * ( 2004 ) , 4174 . , classification of overtwisted contact structures on @xmath2-manifolds , _ invent . _ * 98 * ( 1989 ) , 623637 . , topologically trivial legendrian knots , _ j. symplectic geom . _ * 7 * ( 2009 ) , 77127 . , on knots in overtwisted contact structures , _ quantum topol . _ , to appear . , _ an introduction to contact topology _ , cambridge stud adv . math . * 109 * ( cambridge university press , cambridge , 2008 ) . , structures de contact en dimension trois et bifurcations des feuilletages de surfaces , _ invent . math . _ * 141 * ( 2000 ) , 615689 . , handlebody construction of stein surfaces , _ ann . of math . ( 2 ) _ * 148 * ( 1998 ) , 619693 . , _ @xmath166-manifolds and kirby calculus _ , grad . stud . math . * 20 * ( american mathematical society , providence , ri , 1999 ) . , on the classification of tight contact structures i , _ geom . _ * 4 * ( 2000 ) , 309368 . , heegaard floer invariants of legendrian knots in contact three - manifolds , _ j. eur ( jems ) _ * 11 * ( 2009 ) , 13071363 . , isometries of elliptic @xmath2-manifolds , _ j. london math . soc . ( 2 ) _ * 65 * ( 2002 ) , 167182 . , generalised thurston bennequin invariants for real algebraic surface singularities , _ manuscripta math . _ * 117 * ( 2005 ) , 273298 . , on legendrian surgeries between lens spaces , _ j. symplectic geom . _ * 10 * ( 2012 ) , 165181 .

the study of quantum chromodynamics at high density is almost as old as the theory itself @xcite . in recent years the field has attracted wide interest in the wake of a revival of the idea of color superconductivity ( csc ) @xcite . the stimulus for this revival was the observation @xcite that the instanton - induced quark quark interaction can be much stronger than that induced by simple one - gluon exchange , and can thus give a transition temperature on the order of 100 mev . subsequent work @xcite showed that the perturbative color - magnetic interaction also gives rise to a strong pairing interaction . these and other dynamical considerations @xcite underlie a picture of the ground state of high - density qcd in which the @xmath3 gauge symmetry is spontaneously broken by a bcs - like condensate . the details of the breaking , which include both the higgs ( or meissner ) effect and the rearrangement of global symmetries and goldstone bosons , depend on quark masses , chemical potentials , and temperature . prominent in the list of possibilities are those of color - flavor locking in three - flavor qcd @xcite and crystalline superconductivity with broken translation invariance when there are two flavors with different densities @xcite . for a review see @xcite . as noted , csc at high density is so far a prediction of weak - coupling analysis . one expects the coupling to become weak only at high densities , and in fact it turns out that reliable calculations demand extremely high densities @xcite . the use of weak - coupling methods to make predictions for moderate densities is thus not an application of qcd , but of a model based on it . it is imperative to confirm these predictions by non - perturbative methods . standard lattice monte carlo methods , unfortunately , fall afoul of well - known technical problems when the chemical potential is made nonzero , although we do note remarkable progress made recently in the small-@xmath4 regime @xcite . in this paper we initiate a study of high - density quark matter based on lattice qcd in the strong - coupling limit . we work in the hamiltonian formalism , which is more amenable than the euclidean formalism to strong - coupling perturbation theory and to qualitative study of the ensuing effective theory @xcite . the fermion kinetic hamiltonian is a perturbation that mixes the zero - flux states that are the ground - state sector of the electric term in the gauge hamiltonian . in second order , it moves color - singlet fermion pairs around the lattice ; the effective hamiltonian for these pairs is a generalized antiferromagnet , with spin operators constructed of fermion bilinears . we depart from studies of the vacuum by allowing a background baryon density , which is perforce static in second order in perturbation theory . our aim at this stage is to discover the ground state of the theory with this background . in third order ( when @xmath5 ) the baryons become dynamical ; we display the effective hamiltonian but make no attempt to treat it . the symmetry group of the effective antiferromagnet is the same as the global symmetry group of the original gauge theory . this depends on the formulation chosen for the lattice fermions . following @xcite , we begin with naive , nearest - neighbor fermions , which suffer from species doubling @xcite and possess a global @xmath0 symmetry group that contains the ordinary chiral symmetries [ as well as the axial @xmath6 as subgroups . we subsequently break the too - large symmetry group with next - nearest - neighbor ( ) couplings along the axes in the fermion hopping hamiltonian . a glance at the menu of fermion formulations reveals the reasons for our choice . wilson fermions @xcite have no chiral symmetry and make comparison of results to continuum csc difficult if not impossible . staggered fermions @xcite likewise possess only a reduced axial symmetry while suffering a reduced doubling problem . the overlap action @xcite is non - local in time and hence possesses no hamiltonian ; attempts @xcite to construct an overlap hamiltonian directly have not borne fruit . finally , domain - wall fermions @xcite have been shown @xcite to lose chiral symmetry and regain doubling when the coupling is strong . as we discuss below , while the theory still exhibits doubling in the free fermion spectrum , we are not interested in the perturbative fermion propagator but in the spectrum of the confining theory . we take it as a positive sign that the unbroken symmetry is now @xmath7 . this symmetry is what we want for the continuum theory , except for the axial @xmath8 . the latter can still be broken by hand . our emphasis on the global symmetries is a consequence of the fact that the gauge field is not present in the ground - state sector and does not reappear in strong - coupling perturbation theory . in other words , confinement is a _ kinematic _ feature of the theory , leaving no possibility of seeing the higgs - meissner effect directly . this is but an instance of confinement - higgs duality , typical of gauge theories with matter fields in the fundamental representation @xcite . our aim is thus to identify the pattern of spontaneous breaking of global symmetries . for various values of @xmath9 and @xmath10 , this can be compared to weak - coupling results @xcite . this paper is largely an exposition of formalism , along with partial results . we study the nearest - neighbor antiferromagnetic hamiltonian , both with and without a uniform baryonic background density . we transform the quantum hamiltonian into a path integral for a non - linear @xmath11 model , where the manifold of the @xmath11 field depends on the baryon background . we then investigate the limit of large @xmath9 and show that the global @xmath0 symmetry is indeed spontaneously broken . adding in the couplings is a problem of vacuum alignment @xcite . we do this in the vacuum sector and recover the result @xcite that the @xmath1 chiral symmetry is broken to the vector @xmath12 . the analysis for the finite - density theory is more involved and we defer it to a future publication . other groups have recently studied the strong - coupling effective hamiltonian for naive and wilson fermions at non - zero chemical potential @xcite . we differ from their approaches in eschewing mean field theory in favor of the exact transformation to the @xmath11 model , which is amenable to semiclassical treatment . as noted above , we base our program on fermions ; we also work at fixed baryon density . let us walk through the paper . we review in sec . [ sec : heff ] the derivation of the effective hamiltonian of lattice gauge theory in strong - coupling perturbation theory @xcite . the second - order hamiltonian [ @xmath13 is an antiferromagnet with @xmath0 spins ; the global symmetry group is @xmath0 for the nearest - neighbor theory , broken to @xmath1 by terms . the baryon number at each site determines the representation of @xmath0 carried by the spin at that site . in second order , baryon number is static ; it becomes mobile in the next order , where ( for @xmath5 ) the new term in the effective hamiltonian is a baryon hopping term . the baryon operators responsible for the hopping are composite operators that do not obey canonical anti - commutation relations . if this were not the case , then the effective hamiltonian in third order would strongly resemble that of the @xmath14@xmath15 model @xcite , @xmath16 here @xmath17 is an annihilation operator for an electron at site @xmath18 with spin @xmath19 , and the number operators @xmath20 and spin operators @xmath21 are constructed from it . the added term @xmath22 is a more complicated hopping and interaction term . the @xmath14@xmath15 model describes a doped antiferromagnet ; it arises as the strong - binding limit of hubbard model , a popular model for itinerant magnetism and possibly for high-@xmath23 superconductivity . the model is not particularly tractable and , absent new theoretical developments , does not offer much hope for progress in our finite - density problem . it is nonetheless worth pondering the fact that a model connected , however tentatively , with superconductivity appears in a study of high - density nuclear matter . in the remainder of this paper , we work only to @xmath24 , where the baryons are fixed in position . motivated by the similarity of our hamiltonian to the heisenberg antiferromagnet , we apply condensed matter methods developed for that problem . indeed , condensed matter physicists have generalized the @xmath25 , spin-@xmath26 heisenberg model to @xmath27 in many representations @xcite , which corresponds to adding flavor and color degrees of freedom to the electrons . plane . ] these are exactly the generalizations needed for our effective hamiltonian . with @xmath9 colors and @xmath28 ( single - component ) flavors , a site of the lattice can be constrained to contain a color - singlet combination of @xmath29 particles . the flavor indices of the spin then make up a representation of @xmath27 whose young diagram has @xmath9 columns and @xmath30 rows ( see fig . [ fig : young1 ] ) . we set @xmath31 and the correspondence is complete ( until we include terms in the hamiltonian ) . in sec . [ sec : nls ] we derive a @xmath11 model representation for the partition function of the antiferromagnet . following read and sachdev @xcite , we employ spin coherent states @xcite to define the @xmath11 field . @xmath28 and @xmath30 determine the target space of the @xmath11 model to be the symmetric space @xmath32 $ ] ; the number of colors @xmath9 becomes an overall coefficient of the action . multiplies the quantum hamiltonian , and hence serves only to set the energy scale . ] as for the quantum hamiltonian , the nearest - neighbor theory is symmetric under @xmath33 while the terms break the symmetry to @xmath1 ( while leaving the manifold unchanged ) . the @xmath9 multiplying the action invites a large-@xmath9 analysis , and in sec . [ sec : zerob ] we study the vacuum sector , meaning zero baryon number , thereby . we return to an exercise proposed and solved by smit @xcite , in generalizing the vacuum sector to allow baryon number @xmath34 on alternating sites ; this means specifying conjugate representations of @xmath33 on alternating sites , with respectively @xmath30 and @xmath35 rows . as shown by read and sachdev @xcite , in this situation one can carry out an alternating @xmath33 rotation to convert the antiferromagnet into a ferromagnet with _ identical _ spins on alternating sites , and the classical ( @xmath36 ) analysis gives a homogeneous ground state . the result is , as one might expect , that @xmath33 is broken to @xmath37 in the classical vacuum ; the ground state energy is independent of @xmath30 . the @xmath2 corrections to the energy do depend on @xmath30 , however , and they select the self - conjugate @xmath38 configuration ( _ i.e. _ , @xmath39 everywhere ) as the lowest - energy vacuum . thus the true ground state breaks @xmath40 . calculation includes the effect of the time - derivative terms in the action that were dropped in the leading order , and these terms _ do _ remember the difference between the ferromagnet and the antiferromagnet . ] when we add terms to the action as a perturbation , we find that the ground state breaks @xmath41 to the vector @xmath12 , as expected . we turn to non - zero baryon density in sec . [ sec : nonzerob ] . we study homogeneous states , in which all sites carry the same representation of @xmath33 , with @xmath42 . the classical vacuum of the @xmath11 model is more elusive than for the vacuum sector , since now there are identical manifolds on adjacent sites but the coupling is antiferromagnetic . we begin by studying the two - site problem , and we learn that when one of the classical spins is fixed then when the energy is minimized the other spin is still free to wander a submanifold of the original symmetric space . if we replicate this to the infinite lattice then we have a situation where the even spins , say , are fixed in direction while each odd spin wanders the submanifold , independent of the other odd spins . this means a ground state whose degeneracy is exponential in the volume , similar to some frustrated models or the antiferromagnetic potts model @xcite . the cure to this disease comes from the @xmath43 fluctuations , which couple the odd spins to each other and make them align . in the end we find that the @xmath33 symmetry is broken by the vacuum to @xmath44 . perturbing this ground state with the terms is technically difficult , and we do not attempt it here despite its obvious physical interest . we close with a brief summary and discussion . the @xmath43 calculation in the @xmath45 case is relegated to an appendix , as are other ( but not all ) technical details . for an @xmath46 gauge theory with @xmath10 flavors of fermions , we write the lattice hamiltonian @xmath47 here @xmath48 is the electric term , a sum over links @xmath49 of the @xmath46 casimir operator on each link , @xmath50 next is the magnetic term , a sum over plaquettes , @xmath51 finally we have the fermion hamiltonian , @xmath52 the fermion field @xmath53 carries color @xmath54 and flavor @xmath55 at site @xmath56 . the function @xmath57 is a kernel that defines the lattice fermion derivative . it can yield a naive , nearest - neighbor action if @xmath58 ; a long - range slac derivative @xcite if @xmath59 ; or anything in between , such as a action obtained by truncating the slac kernel to its @xmath60 terms . for @xmath61 the ground state of @xmath62 is determined by @xmath48 alone to be any state with zero electric field , whatever its fermion content , @xmath63 and are degenerate with respect to all the fermionic degrees of freedom . we consider perturbation theory in @xmath64 . both @xmath65 and @xmath66 are sums of operators that are strictly bounded , independent of @xmath67 except for the explicit coefficient in @xmath65 . we can dismiss first - order perturbations by noting that @xmath65 and @xmath66 are multilinear in link operators @xmath68 and @xmath69 , which are raising / lowering operators for the electric field ; thus there are no non - zero matrix elements within the zero - field sector . we proceed to higher orders , and seek an effective hamiltonian that acts in the zero - field sector @xcite . define @xmath70 to be the projector onto the subspace of all the @xmath71 states . then perturbation theory in @xmath72 gives an effective hamiltonian , @xmath73 here @xmath74 projects onto the subspace orthogonal to the @xmath71 states ; the operator @xmath75 supplies energy denominators , so that @xmath76 the intermediate states @xmath77 contain flux excitations . in second and third order the patterns of flux can only be strings of length @xmath18 in the fundamental representation of the color group . thus the energy denominators are @xmath78 where @xmath79 is the quadratic casimir of the fundamental representation of @xmath46 . the perturbations @xmath65 and @xmath66 are explicitly of @xmath24 and @xmath80 , respectively ; each energy denominator gives a factor of @xmath81 . thus to @xmath82 we can forget about @xmath65 . our result to this order is @xmath83 since @xmath66 has no non - zero matrix elements within the @xmath71 sector , we have dispensed with @xmath84 in . the first term in arises for any value of @xmath9 and is @xmath24 ; the case @xmath85 must be treated carefully , but all cases @xmath86 are generic . the second term is special to @xmath5 and is @xmath82 . we calculate explicitly the first term in @xmath87 . each term in @xmath66 creates a string of flux of length @xmath18 , which must be destroyed by the conjugate term . thus @xmath88\sum_{\bn\mu } \left(\psi_\bn^{\dag\alpha f}\alpha_\mu\psi_{\bn+j\muhat}^{\beta f}\right ) \langle0|\left(\prod u\right)_{\alpha\beta } \left(\prod u^\dag\right)_{\gamma\delta}|0\rangle \left(\psi_{\bn+j\muhat}^{\dag\gamma g}\alpha_\mu\psi_\bn^{\delta g}\right ) , \label{h2eff1}\ ] ] where we define @xmath89 ^ 2 } { \frac12g^2c_f|j|}>0.\ ] ] the matrix element of the gauge fields yields @xmath90 , independent of @xmath18 . as they appear in , each @xmath91 is next to a @xmath92 on a different site . this invites a fierz transformation on the product of fermion fields , which we write generally as @xmath93 here @xmath94 are combined site , flavor , and color indices , and we have assumed that @xmath95 and @xmath96 are always different while @xmath18 and @xmath95 might be equal [ as in ] . the matrices @xmath97 are the 16 dirac matrices , normalized to @xmath98 , and we have defined @xmath99 this sign factor is @xmath100 according to whether @xmath97 commutes or anticommutes with @xmath101 ; it will be a constant companion in our calculations . as they appear in @xmath102 , the indices @xmath103 are the same site and color but different flavors , and likewise @xmath104 . leaving the flavor indices explicit , we obtain @xmath105 each fermion bilinear in parentheses is a color singlet located at a given site . the second term contains the baryon density is zero in the half - filled state , and thus @xmath106 . ] @xmath107 , and the sum @xmath108 is the total baryon number @xmath109 . we now combine the dirac indices with the flavor indices and write @xmath110 we have defined new matrices @xmath111 as direct products of the @xmath112 dirac matrices and the @xmath12 flavor generators , @xmath113 and we have normalized them conventionally according to @xmath114 the @xmath111 generate a @xmath33 algebra , with @xmath115 . an alternating flip @xmath116\psi_\bn\ ] ] ( spin diagonalization @xcite ) removes the @xmath101 matrices from the odd-@xmath18 terms in @xmath66 , and hence removes the sign factors @xmath117 from the odd-@xmath18 terms in @xmath102 . we have finally @xmath118 the odd-@xmath18 terms are of the form @xmath119 which can be written in any basis for the @xmath0 algebra . the even-@xmath18 terms , however , contain @xmath120 which is defined only in the original basis ( [ dirprod ] ) . in the zero - field sector in which we work , gauss law constrains the fermion state at each site to be a color singlet . the drained state @xmath121 , with @xmath122 for all @xmath123 , is the unique state with @xmath124 . the other color singlet states may be generated by repeated application of the baryon creation operator , @xmath125 with @xmath9 operators @xmath92 . ( here and henceforth , the indices @xmath126 combine the flavor and dirac indices . ) as noted above , at each site @xmath56 the operators @xmath127 generate a @xmath33 algebra , with @xmath128 . the drained state is obviously a singlet under this algebra . the creation operator @xmath129 is in the symmetric representation of @xmath33 with one row and @xmath9 columns ( see fig . [ fig : baryon ] ) . repeated application of @xmath129 to the drained state gives the state @xmath130 if there are @xmath30 operators @xmath131 , then the state @xmath132 lies in the representation with @xmath9 columns and @xmath30 rows ( see fig . [ fig : young1 ] ) . its baryon number is @xmath133 . the second - order effective hamiltonian @xmath102 preserves @xmath134 , the baryon number on each site . thus any distribution of @xmath134 defines a sector within which @xmath102 is to be diagonalized . in other words , baryons constitute a fixed background in which to study `` mesonic '' dynamics . the baryon number at each site fixes the representation of @xmath33 at that site , which is the space of states in which the charges @xmath135 act . the @xmath136 terms in are of the form @xmath137 , and they commute with the generators @xmath138 of a global @xmath33 symmetry group . this symmetry is in fact familiar from the lattice hamiltonian of naive , nearest - neighbor fermions : spin diagonalization of @xmath10 naive dirac fermions transforms the hamiltonian into that of @xmath139 staggered fermions . in the weak coupling limit , there are in fact @xmath140 fermion flavors the doubling problem . this doubling is partially reflected in the accidental @xmath141 symmetry , which is intact in the @xmath142 limit and is respected by the effective hamiltonian . retaining terms in the fermion hamiltonian ( and thus in @xmath143 ) that involve odd separations @xmath18 does not break this symmetry . the nielsen - ninomiya theorem @xcite guarantees that any fermion hamiltonian of finite range will possess the full doubling problem . this is a statement , however , about weak coupling only , since the dispersion relation of free fermions is irrelevant if the coupling is strong and the fermions are confined . it is interesting that the accidental @xmath0 symmetry nonetheless survives into strong coupling as a vestige of doubling . the terms in with _ even _ @xmath18 , on the other hand , break the @xmath33 symmetry , as do even-@xmath18 terms in the original fermion hamiltonian . it is easy to see via spin diagonalization , which leaves the even-@xmath18 terms unchanged , that the only generators left unbroken are the @xmath144 corresponding to @xmath145 which form the @xmath146 chiral algebra . this of course makes no difference to the nielsen - ninomiya theorem , which will enforce 8-fold doubling in the perturbative propagator even without the @xmath0 symmetry . if we are interested in the realization of the global symmetries of the continuum theory , though , we can study this lattice theory which has the same symmetry . the simplest theory one may study is thus one containing nearest - neighbor and terms . we shall proceed to discard terms with longer range ; we shall begin with the nearest - neighbor theory , with its accidental doubling symmetry , and later break this symmetry to @xmath146 with the terms . two essential differences will always remain between this lattice theory and the continuum theory . one is the presence of the axial @xmath8 symmetry on the lattice . this symmetry is exact , broken by no anomaly , and may make the drawing of conclusions for the continuum theory less than straightforward unless it is broken by hand . the other difference is the fact that the effective hamiltonian for baryons ( see below ) is also a short - ranged hopping hamiltonian . if the baryons were almost free , we would say that they are surely doubled like the original quarks . the fact that the simplicity of the hopping terms is only apparent , and that the baryons are still coupled strongly to mesonic excitations , offers the possibility that doubling may not return . the third - order term in @xmath87 , which only exists in the case of @xmath5 , is calculated via @xmath147 for a single link , we have @xmath148 and the same can be proven for a chain of links , @xmath149 thus [ @xmath126 are here ( temporarily ) flavor indices ] , @xmath150 the kernel is @xmath151 again , spin diagonalization simplifies the odd-@xmath18 terms , but not the even-@xmath18 terms . the result is @xmath152 with @xmath153 where @xmath154 is the usual staggered - fermion sign factor , and @xmath155_{ii ' } b_{\bn+j\muhat}^{i'}\zeta_\mu(\bn)+\text{\em h.c.},\ ] ] where @xmath156 . the baryon operators are @xmath157 where we have written @xmath158 to represent the compound index @xmath159 , taking values in the symmetric three - index representation of @xmath33 ( @xmath126 once more combine flavor and dirac indices ) . the odd-@xmath18 part of @xmath160 , like that of @xmath143 , is symmetric under the @xmath33 doubling symmetry . the even-@xmath18 part breaks @xmath33 to @xmath146 . @xmath160 is a baryon hopping term . as mentioned in the introduction , however , its simplicity is deceptive . the baryon operators @xmath161 are composite and hence do not obey canonical anticommutation relations , _ i.e. , _ @xmath162 the separation of @xmath160 into a canonical kinetic energy and an interaction term is a challenge for the future . because of the complexity of the third - order effective hamiltonian , we restrict ourselves henceforth to the second - order theory , in which baryons are a fixed background . the theory defined by @xmath143 is a generalized spin model , with spins chosen to be in representations of @xmath33 according to the baryon distribution . we review @xcite in this section how to convert the spin model into a @xmath11 model . the @xmath11 field at each site will move in a manifold determined by the baryon number at that site . we employ a generalization of spin coherent states @xcite to derive a path integral for the spin model of @xmath143 . we recall that a given site carries generators @xmath135 of @xmath33 in a representation with @xmath9 columns and @xmath30 rows , with @xmath133 . first we choose a basis for the lie algebra of @xmath33 . this consists of the generators @xmath163 , with @xmath164 , whose matrix elements in the fundamental rep are @xmath165 the corresponding charges are @xmath166 where we have subtracted a constant for convenience . the cartan subalgebra consists of the operators @xmath167 we build the coherent states from the state of highest weight . the highest - weight state @xmath168 in the representation is an eigenstate of the cartan generators , @xmath169 -(n_c/2)|\psi_0\rangle&{\rm for}\ i = m+1,\ldots , n . \label{cartanh}\ ] ] in this state , the generators take the simple form @xmath170 with @xmath171 the state @xmath168 is invariant ( up to a phase ) under the subgroup of @xmath33 that commutes with @xmath172 ; this is @xmath37 . the most general rotation of @xmath168 is carried out with the generators @xmath173 that are _ not _ in the corresponding subalgebra . we choose parameters @xmath174 , with @xmath175 $ ] and @xmath176 $ ] , and write @xmath177 the only generators @xmath178 that appear in are those that lower an @xmath179 that starts from @xmath180 in while raising another @xmath179 that starts from @xmath181 . any other generator would annihilate @xmath168 and thus give no effect in the exponential . the coherent states are normalized , @xmath182 and over - complete , @xmath183 in the integral is over the coset space @xmath32 $ ] ( see below ) . matrix elements of the generators are given by @xmath184 where the matrix @xmath185 is given by a unitary rotation from @xmath172 , @xmath186 the matrix @xmath187 is built out of the @xmath188 matrix @xmath174 , @xmath189 . \label{ua}\ ] ] @xmath11 is both hermitian and unitary . the manifold of matrices @xmath11 is the coset space @xmath32 $ ] , a sub - manifold of @xmath33 . this is because for any matrix @xmath187 , one can generate an orbit @xmath190 by multiplying with a matrix @xmath191 where @xmath192 and @xmath193 . all matrices in the orbit will give the same matrix @xmath11 when inserted into , and thus in integrating over the configuration space of @xmath11 one must choose only a single representative of each orbit . this set of representatives , the coset space @xmath194 $ ] , is the quotient space of the non - invariant subgroup @xmath37 . the measure over the coset space must be invariant under unitary rotations , @xmath195 where @xmath196 represents the rotation @xmath72 in hilbert space . equation ( [ sq ] ) shows that this is a rotation @xmath197 and by , this means that a measure in @xmath68 must be invariant under @xmath198 . this fixes the measure uniquely to be the haar measure in @xmath33 , and thus one can integrate over the coset space by integrating with respect to @xmath68 over @xmath33 and using . a representation whose young diagram has @xmath35 rows is the conjugate to the representation with @xmath30 rows . its coherent state space can be constructed to look the same , with only a sign difference . to do this we start with the _ lowest_-weight state , which satisfies [ cf . ] @xmath199 ( n_c/2)|\psi_0\rangle&{\rm for}\ i = m+1,\ldots , n . \end{array}\right.\ ] ] this introduces a minus sign into . the subsequent steps are identical , with only the replacement of by @xmath200 here , too , @xmath11 is given in terms of @xmath172 and @xmath68 by . the partition function @xmath201 can be written as a path integral by inserting the completeness relation ( [ complete ] ) at every slice of imaginary time . this gives @xmath202 where the action is @xmath203.\ ] ] the hamiltonian @xmath204 is a transcription of the quantum hamiltonian to the classical @xmath11 matrices . starting with the quantum operator @xmath205 , we have @xmath206 expressed in these variables , the quantum hamiltonian is term from indeed disappears . we have dropped an additive constant that is independent of @xmath109 . ] @xmath207 where @xmath208 . we transcribe this according to to obtain the classical hamiltonian , @xmath209 where @xmath210 recall that each @xmath211 is an @xmath212 matrix ranging over the coset space appropriate to site @xmath56 . the time - derivative term in @xmath213 is a berry phase @xcite . it can be expressed in terms of the matrix @xmath68 that determines @xmath11 via . the result is is of the form given in , and in that case @xmath68 can not be integrated over all of @xmath33 . ] @xmath214 . \label{action_sigma}\ ] ] @xmath215 will vary from site to site if @xmath30 does . if one takes the route of for a representation with @xmath35 rows , then the kinetic term for that site acquires a minus sign ( see below ) . the number of colors has largely dropped out of the problem , since @xmath11 is just an @xmath212 unitary matrix field . the explicit factors of @xmath9 in eqs . ( [ hsigma ] ) and ( [ action_sigma ] ) invite a semiclassical approximation in the large-@xmath9 limit . this of course neglects the @xmath9-dependence of the couplings @xmath216 , but the common scale of the couplings only serves to set an energy scale . the ground state will be independent of this scale , although correlations and the temperature scale will reflect it . we take the point of view that after all @xmath5 , and we are interested in properties of the effective theory for this value only . the @xmath217 limit , for fixed couplings @xmath218 , will be a device for investigating the properties of a generalized effective theory . the simplest way to set the baryon density to zero is just to choose @xmath219 on each site , meaning @xmath220 . it turns out to be just as easy to consider a slightly generalized case @xcite , in which @xmath221 is chosen to alternate , @xmath222 , on even and odd sublattices . this means to choose a representation with @xmath223 rows on even sites and @xmath224 rows on odd sites , which gives a pair of conjugate representations of @xmath33 . in view of , we can substitute @xmath225 on the odd sites and thus have identical manifolds on all sites . the hamiltonian is then @xmath226 @xmath30 is a new parameter in the theory , and we can ask what value of @xmath30 gives the lowest energy for the ground state . we will see that @xmath38 is indeed preferred , but only in the next - to - leading order in @xmath2 . in the large-@xmath9 limit , we seek the classical saddle point of @xmath213 . we assume the saddle is at a configuration @xmath227 that is independent of time , and so we drop the time derivative . we begin with the nearest - neighbor hamiltonian , ) is preserved by the alternating signs in the time - derivative term . ] @xmath228 again , the matrices @xmath211 are hermitian and unitary , and the expansion coefficients @xmath229 are real and satisfy @xmath230 . the minimum of @xmath62 is clearly at a constant field , @xmath231 , which can be diagonalized to @xmath232 by a @xmath33 rotation . this is a nel state in the original variables . the @xmath33 symmetry is broken to @xmath37 and there are @xmath233 goldstone bosons . the classical energy density ( per link ) is @xmath234 , independent of @xmath30 . thus at leading order in @xmath2 , the optimal value of @xmath30 is undetermined , and any alternating background of baryon number is equally good . since it turns out that the @xmath2 corrections select @xmath38 , let us consider the effect of the term in @xmath62 for this case only . the perturbation is @xmath235 it breaks the @xmath33 symmetry to @xmath236 . assuming that @xmath237 , we again seek the minimum energy configuration in the form of a constant field ; we minimize @xmath238 among the @xmath33-equivalent @xmath239 states that minimize the nearest - neighbor action . it is not hard to show ( see appendix [ app : nnn ] ) that @xmath240 is minimized for @xmath241 . this is a condensate that is symmetric under the vector generators @xmath242 but not under the axial generators @xmath243 , and thus it breaks the chiral symmetry to the vector subgroup , @xmath244 . returning to the nearest - neighbor theory , we consider fluctuations around the @xmath232 minimum of @xmath213 . first we rescale @xmath245 in order to put the kinetic and potential terms on an equal footing , giving @xmath246 , \label{srescaled}\ ] ] with @xmath247 . recalling , we can write @xmath248 where @xmath249 is anti - hermitian and anticommutes with @xmath172 . it is more convenient to work with the hermitian matrix @xmath250 in terms of which we expand @xmath251 if we further expand @xmath252 in the basis of generators of @xmath33 , @xmath253 we find that the @xmath254 corresponding to generators of @xmath37 vanish ; this is the subgroup under which the vacuum is symmetric . the field @xmath252 thus contains @xmath233 real degrees of freedom , corresponding to the goldstone bosons . we expand @xmath255 and @xmath211 in powers of @xmath252 ; using , we obtain to second order @xmath256.\ ] ] the coefficient matrix is @xmath257 ) , \label{cetaeta}\ ] ] and the classical energy is @xmath258 . @xmath259 is antisymmetric and purely imaginary ; we show in appendix [ app : cd ] that @xmath259 has eigenvalues @xmath100 , each with degeneracy @xmath260 . we change basis so as to diagonalize @xmath259 , and write the index @xmath261 as the compound @xmath262 with @xmath263 and the @xmath264 corresponding to the eigenvalue of @xmath259 . since the original @xmath254 are real , we have @xmath265 thus we eliminate @xmath266 and write @xmath267 . \label{sln}\ ] ] the alternating sign in is what makes the theory antiferromagnetic . it forces us to differentiate between even and odd sites , and we transform to momentum space as follows ( dropping the @xmath268 superscript ) : @xmath269 l^\alpha_{2,\bk}e^{i\bk\cdot(\bn-\hat z)}&\quad \text{$\bn$ odd . } \end{array}\right.\ ] ] the even sites comprise an _ fcc _ lattice with lattice constant 2 , and the momenta @xmath270 take values in its brillouin zone . we obtain @xmath271 here @xmath272 -jd\gamma(\bk)&jd+\partial_\tau \end{array}}\right)\ ] ] and @xmath273 . the gaussian path integral over the @xmath96 field now gives the free energy , @xmath274,\ ] ] where @xmath275 and @xmath276 . for the ground state energy , we take @xmath277 and obtain ( restoring all constants ) @xmath278 . \label{eq : e0}\ ] ] this is exactly the result of smit @xcite . the @xmath43 corrections lift the degeneracy of the ground states with different values of @xmath30 . the integral in is positive and its coefficient contains the number of goldstone bosons . thus the state of lowest energy is that with @xmath38 , and the symmetry breaking scheme is @xmath40 . further breaking by the terms was discussed above . the zero - density theories considered in the preceding section were defined by selecting representations with @xmath30 and @xmath35 rows on alternating sites . for any @xmath30 , this led to a @xmath11 model with identical degrees of freedom on all sites after redefinition of the spins on the odd sublattice and ferromagnetic couplings . factor in the kinetic energy retained information about the antiferromagnetic nature of the quantum problem ; it did not affect the classical analysis . ] we eventually settled on @xmath38 as the background that gives the ground state of lowest energy . introducing non - zero baryon density means changing @xmath30 on some sites of the lattice . since in general there will be representations on different sites that are not mutually conjugate , different sites will carry @xmath11 variables that do not live in the same submanifold of @xmath33 . we here limit ourselves to the simpler case of uniform @xmath30 , where adjacent sites carry identical spins but the coupling is _ antiferromagnetic_. in order to learn how to work with such a theory , we begin by studying the two - site problem . the results of this study will lead directly to an _ ansatz _ for the ground state of a lattice with a fixed density of baryons . consider , therefore , two sites with quantum spins @xmath279 and @xmath280 that carry representations of @xmath33 with @xmath281 and @xmath282 rows , and @xmath9 columns ( see fig . [ fig : twospins ] ) . the quantum hamiltonian is @xmath283 an antiferromagnetic coupling . the corresponding classical @xmath11 model has the interaction hamiltonian @xmath284 where @xmath285 the two @xmath172 matrices reflect the different values of @xmath286 according to @xmath287 the @xmath217 limit is the classical limit , in which we seek values of @xmath288 that minimize @xmath204 . a global @xmath33 rotation , _ viz . _ , @xmath289 can be used to diagonalize @xmath290 so that @xmath291 . now we have to minimize @xmath292 . the case of conjugate representations , @xmath293 , is easy : @xmath294 is a unitary rotation of @xmath295 , which ( in this case ) can be rotated into @xmath296 . this is the unique antiferromagnetic ground state . @xmath290 and @xmath294 can be copied to the odd and even sublattices of an infinite lattice to give the classical nel state considered in the preceding section . the case @xmath297 is more complex . we consider @xmath42 for definiteness ; the other case is similarly handled . we write explicitly [ from eqs . ( [ qulu])([ua ] ) ] @xmath298 -\frac{\displaystyle\sin\left(2\sqrt{a^\dag a}\right ) } { \displaystyle\sqrt{a a^\dag}}\,a^\dag & -\cos\left(2\sqrt{a^\dag a}\right ) \end{array}\right ) . \label{eq : general_q}\ ] ] @xmath299 is a square matrix of dimension @xmath35 and @xmath300 is a square matrix of dimension @xmath30 . since @xmath294 is a rotation of @xmath172 , @xmath301 and hence the energy is @xmath302.\end{aligned}\ ] ] @xmath303 is minimized when all the eigenvalues of @xmath304 are equal to @xmath305 . this means that the @xmath35 column vectors @xmath306 form an orthogonal set in @xmath30 dimensions , with @xmath307 since @xmath308 by assumption , such a set of vectors can always be found . since @xmath309 , we have @xmath310 and so the off - diagonal blocks of vanish . the lower - right block of @xmath294 is the unit matrix @xmath311 . we know that @xmath312 since @xmath313 and thus the upper - left block must have eigenvalues @xmath100 . equating traces of @xmath294 and @xmath172 , we find that the upper - left @xmath314 block must take the form @xmath315 where @xmath316 and @xmath317 . @xmath318 represents the coset @xmath319 $ ] . we conclude that the classical ground state of this @xmath45 two - site problem is degenerate , even beyond breaking the overall @xmath33 symmetry . the solutions can be written as @xmath320 \sigma_2&=&\left ( \begin{array}{cc } \sigma^{(m ) } & 0 \\ 0 & \bm1_{n - m } \end{array } \right ) \label{eq : q2}\end{aligned}\ ] ] ( to which a global @xmath33 rotation can be applied ) . a particular instance of @xmath318 is @xmath321 , given by . the symmetry of the vacuum is the set of rotations that leaves both @xmath290 and @xmath294 invariant , namely , @xmath322 . the classical solution of the two - site problem will guide us in approaching the problem of an infinite lattice below . we expect that spontaneous symmetry breaking will give a vacuum of the same character , with continuous degeneracy . there are , however , two kinds of degeneracy in the two - site problem : that which results from breaking the global @xmath33 to @xmath37 , and that which comes of breaking the @xmath323 subgroup to @xmath324 the latter degeneracy is connected with freedom in choosing the orientation of @xmath294 _ relative to _ @xmath290 . it is instructive to see how quantum mechanical fluctuations lift the classical degeneracies . the quantum two - site problem is easy to solve . we rewrite the hamiltonian ( [ 2site_qh ] ) as @xmath325 . \label{2site_qh2}\ ] ] @xmath326 and @xmath327 are constants , the quadratic casimir operator in the @xmath30-row , @xmath9-column representation of @xmath33 . the first term in is minimized by coupling @xmath279 and @xmath280 to the representation that minimizes the casimir , which is the representation with @xmath328 rows and @xmath9 columns ( see fig . [ fig : twosite ] ) . the ground state has discrete degeneracy equal to the dimension of this representation . the exact quantum solution naturally shows no sign of spontaneous symmetry breaking and hence it is not of much relevance to the infinite volume problem . more interesting is the problem where the state of @xmath279 is _ fixed _ and @xmath280 is allowed to vary . this breaks by hand the global @xmath33 while allowing quantum fluctuations to lift any remaining degeneracy in the relative orientation of the two spins , so it can be regarded as quantization in the presence of spontaneous symmetry breaking . in effect , this is mean field theory . we replace the hamiltonian ( [ 2site_qh ] ) by @xmath329 to minimize the energy we maximize the mean field by choosing the state of @xmath279 to be the highest - weight state . this state diagonalizes the generators @xmath179 of the cartan subalgebra while other generators of @xmath33 have expectation value zero . thus @xmath330 the operators @xmath179 all commute , and their eigenvalues make up the weight diagram of the representation . traceless diagonal generators of @xmath27 . these can be obtained by isolating the @xmath8 member of the set @xmath179 and taking linear combinations of the rest . ] @xmath331 is a dot product of the weight vectors of the two spins . the energy is minimized by choosing for @xmath280 a state that lies opposite the highest - weight state in the weight diagram . as shown in the example of fig . [ fig : twositeweights ] , this still leaves a degeneracy , albeit a discrete one . we stress that this degeneracy comes from freedom in the relative orientation of @xmath279 and @xmath280 ; it remains after quantum fluctuations lift the continuous degeneracy of the classical system . in the @xmath217 limit , the discrete degeneracy becomes infinite and presumably it is well described by the continuous degeneracy of the classical problem . at @xmath36 we seek the saddle point of the action , which we assume to be a time - independent configuration . the classical hamiltonian of the @xmath11 model is @xmath332 seeking an antiferromagnetic ground state , we set @xmath232 on the sublattice of even sites . the odd sites are then governed by @xmath333 this is just the two - site problem studied above , replicated over the lattice . as we saw above , the ground state configuration is degenerate with respect to the configuration at each odd site , @xmath334 a uniform choice for the odd sites , @xmath335 for instance , breaks the @xmath33 symmetry to @xmath322 ; a non - uniform choice can break the symmetry all the way to @xmath336 . the entropy of this classical ground state is evidently proportional to the volume . as noted for the two - site problem , the continuous degeneracy of the ground state is an artifact of the classical , @xmath217 limit . quantum fluctuations will spread each odd spin s wave function over the @xmath319 $ ] manifold . a mean - field _ ansatz _ for the even spins , as noted in the discussion of the two - spin problem , will still leave a discrete degeneracy for the odd spins ; the symmetry breaking scheme will depend on how the odd spins are allocated to the available states . furthermore , one can contemplate making a non - uniform _ ansatz _ for the _ even _ spins as well , reducing the phase space available for odd spins with unequal neighbors but adding entropy on the even sublattice . the situation is reminiscent of that in the antiferromagnetic potts model @xcite , the phase structure of which is not yet understood . this potts - like discrete degeneracy , however , is an artifact of the mean - field approach that , like the classical _ ansatz _ , assumes a fixed state for the spins on the even sites . an essential difference between our @xmath11 model and the potts model is that our degrees of freedom are continuous and will fluctuate as soon as they are allowed to do so . a given odd spin will _ not _ be surrounded by a uniform fixed background ; the neighboring even spins will be influenced by _ all _ their odd neighbors , and will induce an interaction among the odd spins that makes them rotate together . this should reduce the entropy of the ground state to zero . the systematic way to see this effect is to carry out a @xmath2 expansion around the classical _ ansatz _ , which we do in appendix [ app : corrections ] . the result is a ferromagnetic interaction among the @xmath211 on the odd sites . thus the ground state turns out to be the two - site solution , replicated _ uniformly _ over the lattice : @xmath337 \sigma_\bn&=&\left(\begin{array}{cc } \sigma^{(m)}&0\\ 0&\bm1_{n - m } \end{array}\right),&\ \text{$\bn$ odd } , \end{array}\ ] ] where @xmath338 is a _ global _ degree of freedom . the symmetry group of the vacuum is @xmath322 . let us summarize the results presented in this paper . in the vacuum sector , we have rederived smit s result for the lowest - energy configuration of alternating @xmath339 sites . the result is indeed @xmath219 ; the @xmath0 symmetry of the nearest - neighbor theory is spontaneously broken to @xmath340 . we extended this result to the theory and found that its @xmath1 chiral symmetry is broken to the vector @xmath12 flavor subgroup . adding net baryon number to the system , we examined the case of uniform baryon density , @xmath341 . here our study was limited to the nearest - neighbor theory , and we found a nel - like ground state that breaks @xmath33 to @xmath342 . the number of goldstone bosons @xmath343 thus depends on the baryon density @xmath221 as @xmath344 directions for future work begin with adding interactions to the @xmath345 theory and extracting from it a prediction for the breaking of the continuum - like chiral symmetry . ( the axial @xmath8 symmetry can be broken by hand . ) another direction is to gain greater freedom in fixing the baryon density . a constant value of @xmath346 means a baryon density that is close to the maximum allowed on the lattice ; the density can be lowered by setting @xmath347 only on a sparse sublattice , along the lines shown in @xcite . an ultimate goal , as for the hubbard model , is the incorporation of the third - order term in the effective hamiltonian in order to have a theory with dynamical baryons . perhaps an instructive half - measure would be to study the second - order theory in the presence of a disordered baryon background . the strong - coupling effective theory can be regarded as a qcd - like model , possessing gauge invariance and the correct degrees of freedom . in that case the lattice spacing is merely a parameter , an overall scale . more insight can be gained by considering the strong - coupling theory to represent qcd at large distances , derived by some renormalization - group transformation from a weak - coupling short - distance hamiltonian . on the one hand , one would expect any such effective hamiltonian to contain many terms of great complexity ; on the other hand , a simple lattice theory such as ours might offer a qualitative approximation to the real theory ( as long as one accepts the loss of lorentz invariance ) . we can estimate the lattice spacing to be some scale at which the running qcd coupling is large , certainly greater than the radius of a proton . the limitation that the lattice puts on the density then becomes a physical issue . taking the lattice spacing @xmath348 to be on the order of 1 fm , the highest baryon density allowed by the lattice is @xmath349 @xmath350 . for low values of @xmath10 this may not be enough to see finite - density phase transitions , in particular a transition to color superconductivity . perhaps a way out is to consider an unphysically large number of flavors . this work was begun while b.s . was on sabbatical leave at the center for theoretical physics at mit . he thanks the members of the center for their hospitality . thanks the organizers of the 19th jerusalem winter school in theoretical physics for their hospitality . we thank assa auerbach , richard brower , krishna rajagopal , subir sachdev , and uwe - jens wiese for helpful discussions . this work was supported by the u.s . department of energy under grant no . df - fc02 - 94er40818 and by the israel science foundation under grant no . 222/02 - 1 . the @xmath351 signs are defined only when the @xmath352 are written in the basis @xmath353 . we choose a chiral basis for the gamma matrices , so that @xmath354 , @xmath355 , and @xmath356 . the energy ( [ epsprime ] ) is a sum of squares , @xmath357 with the constraint @xmath358 . the coefficients @xmath359 take on the values @xmath360 . the minimum of @xmath240 occurs when all @xmath352 are zero except those corresponding to @xmath361 , namely , those for @xmath362 and @xmath363 ; the energy is independent of these @xmath352 . thus the set of solutions can be written in the form @xmath364 recalling that @xmath365 , we have @xmath366 , so @xmath367 . a chiral rotation @xmath368 , with @xmath369 turns @xmath370 into @xmath371 , which is invariant only under vector transformations generated by @xmath372 . to analyze the matrices @xmath259 and @xmath373 , given by eqs . ( [ cetaeta ] ) and ( [ dmatrix ] ) , we begin with the @xmath33 generators @xmath111 that lie outside the subalgebra @xmath374 that commutes with @xmath375 we choose for them a basis @xmath376 and @xmath377 with @xmath378 and @xmath379 , given by @xmath380 since the coset space @xmath381 is a symmetric space , the commutator @xmath382 $ ] lies in @xmath62 ; in order for @xmath383)$ ] to be nonzero , the commutator must have a nonzero component in the cartan subalgebra of @xmath62 . this is only possible if @xmath384 and @xmath385 . thus in this basis @xmath259 takes the form @xmath386 diagonalizing @xmath259 gives eigenvalues @xmath100 . the generators corresponding to the basis that diagonalizes @xmath259 are @xmath387 as for @xmath388 : the only anticommutators among the @xmath389 that do not vanish [ note the bounds on @xmath390 are between @xmath391 and @xmath392 , _ viz . , _ @xmath393 noting that @xmath394 , we find that @xmath373 takes the block - diagonal form @xmath395 \\ & = & \delta_{qq ' } \sum_{\bmm(\bn ) } ( \bm1_{m}-\sigma^{(m)}_\bmm)_{pp'}.\end{aligned}\ ] ] we summarize this by writing @xmath396 where @xmath397 is an @xmath314 matrix given by @xmath398 it is easy to prove that the eigenvalues of @xmath303 range from @xmath399 to @xmath400 . in particular , @xmath388 is positive . we build on the @xmath36 vacua described in sec . [ sec : nonzerob ] by allowing fluctuations around them . we let the @xmath211 on the even sites fluctuate around @xmath172 ; we let the @xmath211 on odd sites roll freely around the @xmath319 $ ] manifold covered by , and also execute small oscillations off the manifold into the @xmath32 $ ] coset space . our goal is an effective action for the classical part ( [ submanifold ] ) of the odd spins . to reach this , we integrate out the even spins ; the off - manifold fluctuations of the odd spins must be included for consistency in the @xmath2 expansion . the counterpart of the action ( [ srescaled ] ) for our problem has an antiferromagnetic spin - spin interaction , with no @xmath401 factors . we separate it into odd , even , and coupled terms , @xmath402 here @xmath403 , where @xmath404 are the nearest neighbors of the even site @xmath56 . we expand the field on the even sites around @xmath232 in the manner of , @xmath405 while for the odd sites we write ( see appendix [ app : oddspins ] ) @xmath406 with @xmath407 and @xmath408 @xmath252 describes the fluctuations of the even spins around their classical value @xmath172 . @xmath255 rotates the odd spins within the manifold of their classical values , while @xmath409 describes their fluctuations outside that manifold . we further define @xmath410 the classical field on the odd sites . we leave @xmath411 alone and expand @xmath412 and @xmath413 around the classical values of the fields , @xmath414 here @xmath415 is the rotated fluctuation field on the odd sites , and the hermitian matrices @xmath416 and @xmath417 are sums over the odd neighbors of the even site @xmath56 , @xmath418 since both @xmath416 and @xmath172 are block diagonal , the first trace in each integral in is zero . now we organize the partition function as follows : @xmath419 z_{\text{even } } , \label{eq : z}\ ] ] with @xmath420.\end{aligned}\ ] ] @xmath421 is a product of decoupled single - site integrals . again we expand in the group algebra , @xmath422 where the sum is over the @xmath233 generators of @xmath33 that are not in @xmath37 . @xmath417 can be expanded similarly and we obtain the following form for the integral over the even fields : @xmath423,\ ] ] where @xmath424 the matrix @xmath259 is the same as in , @xmath425),\ ] ] while the new matrix @xmath388 varies with the site @xmath56 according to the average @xmath416 of its neighboring spins , @xmath426 we study the two matrices in appendix [ app : cd ] . diagonalizing @xmath259 as before , we arrive at @xmath427 \right\},\ ] ] where @xmath428 is the @xmath314 matrix @xmath429 the quantities @xmath430 ( and @xmath431 ) for each @xmath432 are complex @xmath30-component vectors ; they are rotations of the @xmath233 real components @xmath254 ( and @xmath433 ) into the basis that diagonalizes @xmath259 . the matrix @xmath397 is given by ; it carries the dependence on @xmath416 . performing the gaussian integration we get @xmath434 . \label{eq : zeven}\ ] ] finally we separate the integral ( [ eq : z ] ) over the odd spins into an integral over the classical field @xmath435 and an integral over the fluctuations around it . we obtain @xmath436 \right\}. \label{eq : z1}\end{aligned}\ ] ] here @xmath437 stands for an odd site , @xmath56 for an even one . equation ( [ eq : z1 ] ) gives an effective action for the classical odd spins @xmath438 . these enter the exponents through @xmath439 [ via eqs . ( [ eq : e ] ) and ( [ eq : m ] ) ] and through @xmath440 [ via eqs . ( [ tildel ] ) and ( [ eq : lo ] ) ] . the action in the first exponent is minimized when each matrix @xmath441 has the largest number of zero eigenvalues , each of which makes @xmath442 approach @xmath443 . it is easy to check that @xmath397 has @xmath328 zero eigenvalues ( the maximal number ) when the @xmath438 on _ all _ the odd sites @xmath404 align , _ \label{eq : minimum}\ ] ] moreover , when holds , all the @xmath445 s align parallel to each other and @xmath446 is maximized ; also the eigenvalues of @xmath447 are maximized ( to @xmath448 ) . thus the action in the second exponent also has a minimum at this point in configuration space . these effects add up to an effective _ ferromagnetic _ interaction among the @xmath449 nearest neighbors @xmath437 of each even site @xmath56 . this effective interaction will align the classical spins on the odd sublattice to the same direction in their submanifold , @xmath319 $ ] . the divergences in the effective action have their origin in the fact that the semiclassical corrections are calculated as gaussian integrals in the even fluctuation fields @xmath252 , and the coefficient matrix @xmath450 acquires zero eigenvalues . the correct range of integration over @xmath252 is of course not infinite , but rather the volume of the @xmath32 $ ] manifold . this will regulate the divergences , but leave the effective action for the odd spins attractive . in the classical analysis , the fields on the odd sites take values in the sub - manifold @xmath319 $ ] of the manifold @xmath32 $ ] . we denote these values @xmath451 , @xmath452 here @xmath453 with @xmath454 , and @xmath455 @xmath318 contains @xmath456 independent degrees of freedom . any @xmath457 $ ] can be written as @xmath458 -\frac{\displaystyle\sin\left(2\sqrt{a^{\dag}a}\right ) } { \displaystyle\sqrt{a^{\dag}a}}\,a^{\dag } & -\cos\left(2\sqrt{a^{\dag}a}\right ) \end{array } \right ) \label{eq : sig_gen}\ ] ] [ cf . ] , which coincides with if @xmath459 recall that @xmath348 is an @xmath188 matrix , so the zero block has dimensions @xmath460 . we allow motion out of the sub - manifold by allowing @xmath348 to vary further , @xmath461 the @xmath462 degrees of freedom in @xmath463 complement the @xmath456 degrees of freedom inherent in @xmath464 to give @xmath233 , the dimensionality of the entire @xmath194 $ ] coset space . writing @xmath11 with the generalized @xmath348 we have @xmath465 0 & \cos\left(2\sqrt{\bar{a}\bar{a}^{\dag}}\right ) & -\bar{a}\,\frac{\displaystyle\sin\left(2\sqrt{\bar{a}^{\dag}\bar{a}}\right ) } { \displaystyle\sqrt{\bar{a}^{\dag}\bar{a } } } \\[2pt ] 0 & -\frac{\displaystyle\sin\left(2\sqrt{\bar{a}^{\dag}\bar{a}}\right ) } { \displaystyle\sqrt{\bar{a}^{\dag}\bar{a}}}\,\bar a^\dag & -\cos\left(2\sqrt{\bar{a}^{\dag}\bar{a}}\right ) \end{array } \right ) u^{\dag},\ ] ] with @xmath466 we can also write this as @xmath467}(\bar{a},\bar{a}^{\dag } ) \end{array } \right ) u^{\dag}.\ ] ] @xmath468}$ ] is a matrix in the manifold @xmath469 $ ] . indeed for @xmath470 , we have @xmath471 } = \left ( \begin{array}{cc } -\bm1_{n - m } & 0 \\ 0 & \bm1_{n - m } \end{array } \right ) \equiv -\bar{\lambda}.\ ] ] since @xmath469 $ ] is a self - conjugate manifold , its structure near @xmath468}=-\bar\lambda$ ] is the same as its structure near @xmath468}=\bar\lambda$ ] , which corresponds to @xmath472 . expanding @xmath468}$ ] around @xmath473 gives . j. c. collins and m. j. perry , phys . lett . * 34 * , 1353 ( 1975 ) . b. c. barrois , nucl . b * 129 * , 390 ( 1977 ) . d. bailin and a. love , nucl . b * 205 * , 119 ( 1982 ) ; phys . b * 107 * , 377 ( 1981 ) ; phys . rept . * 107 * , 325 ( 1984 ) . m. alford , k. rajagopal , and f. wilczek , phys . b * 422 * , 247 ( 1998 ) . r. rapp , t. schfer , e. v. shuryak , and m. velkovsky , phys . * 81 * , 53 ( 1998 ) . d. t. son , phys . rev . d * 59 * , 094019 ( 1999 ) . t. schfer and f. wilczek , phys . b * 450 * , 325 ( 1999 ) [ arxiv : hep - ph/9810509 ] . m. alford , k. rajagopal and f. wilczek , nucl . b * 537 * , 443 ( 1999 ) . m. g. alford , j. a. bowers , and k. rajagopal , phys . d * 63 * , 074016 ( 2001 ) [ arxiv : hep - ph/0008208 ] ; j. a. bowers and k. rajagopal , phys . d * 66 * , 065002 ( 2002 ) [ arxiv : hep - ph/0204079 ] . k. rajagopal and f. wilczek , arxiv : hep - ph/0011333 . k. rajagopal and e. shuster , phys . d * 62 * , 085007 ( 2000 ) [ arxiv : hep - ph/0004074 ] . z. fodor and s. d. katz , jhep * 0203 * , 014 ( 2002 ) [ arxiv : hep - lat/0106002 ] ; z. fodor , s. d. katz , and k. k. szabo , arxiv : hep - lat/0208078 . de forcrand and o. philipsen , nucl . b * 642 * , 290 ( 2002 ) [ arxiv : hep - lat/0205016 ] . b. bringoltz and b. svetitsky , arxiv : hep - lat/0209005 . t. banks , l. susskind , and j. kogut , phys . d * 13 * , 1043 ( 1976 ) . j. smit , nucl . b * 175 * , 307 ( 1980 ) . r. c. brower and b. svetitsky , phys . d * 61 * , 114511 ( 2000 ) [ arxiv : hep - lat/9912019 ] . h. b. nielsen and m. ninomiya , nucl . b * 185 * , 20 ( 1981 ) [ erratum-_ibid . _ * 195 * , 541 ( 1982 ) ] ; _ ibid . _ * 193 * , 173 ( 1981 ) . k. g. wilson , in _ new phenomena in subnuclear physics _ , erice 1975 , edited by a. zichichi ( plenum , new york , 1977 ) , part a p. 69 . l. susskind , phys . d * 16 * , 3031 ( 1977 ) . r. narayanan and h. neuberger , phys . b * 302 * , 62 ( 1993 ) [ arxiv : hep - lat/9212019 ] ; nucl . phys . * b412 * , 574 ( 1994 ) [ arxiv : hep - lat/9307006 ] ; * b443 * , 305 ( 1995 ) [ arxiv : hep - th/9411108 ] ; h. neuberger , phys . rev . d * 57 * , 5417 ( 1998 ) . m. creutz , i. horvath , and h. neuberger , nucl . suppl . * 106 * , 760 ( 2002 ) [ arxiv : hep - lat/0110009 ] . d. b. kaplan , phys . b * 288 * , 342 ( 1992 ) [ arxiv : hep - lat/9206013 ] . y. shamir , nucl . b406 * , 90 ( 1993 ) . s. d. drell , m. weinstein , and s. yankielowicz , phys . d * 14 * , 1627 ( 1976 ) . s. myint and c. rebbi , nucl . b * 421 * , 241 ( 1994 ) [ arxiv : hep - lat/9401009 ] . g. t hooft , in _ recent developments in gauge theories _ , cargse 1979 , ed . g. t hooft _ et al . _ ( plenum , new york , 1980 ) ; e. h. fradkin and s. h. shenker , phys . d * 19 * , 3682 ( 1979 ) ; l. f. abbott and e. farhi , phys . b * 101 * , 69 ( 1981 ) ; nucl . b * 189 * , 547 ( 1981 ) . t. schfer , nucl . b * 575 * , 269 ( 2000 ) . a. le yaouanc , l. oliver , o. pene , j. c. raynal , m. jarfi , and o. lazrak , phys . d * 37 * , 3691 ( 1988 ) ; * 37 * , 3702 ( 1988 ) ; * 38 * , 3256 ( 1988 ) . e. b. gregory , s. h. guo , h. kroger , and x. q. luo , phys . d * 62 * , 054508 ( 2000 ) [ arxiv : hep - lat/9912054 ] . y. umino , phys . d * 66 * , 074501 ( 2002 ) [ arxiv : hep - ph/0101144 ] . m. e. peskin , nucl . b * 175 * , 197 ( 1980 ) . a. auerbach , _ interacting electrons and quantum magnetism _ ( springer - verlag , 1994 ) . s. randjbar - daemi , a. salam , and j. strathdee , phys . b * 48 * , 3190 ( 1993 ) [ arxiv : hep - th/9210145 ] . t. kato , prog . phys . * 4 * , 154 ( 1949 ) ; a. messiah , _ quantum mechanics _ ( north - holland , amsterdam , 1961 ) , ch . 16 . b. schlittgen and u .- j . wiese , nucl . suppl . * 83 * , 718 ( 2000 ) [ arxiv : hep - lat/0005020 ] ; phys . d * 63 * , 085007 ( 2001 ) [ arxiv : hep - lat/0012014 ] . a. n. berker and l. p. kadanoff , j. phys . a * 13 * , l259 ( 1980 ) ; j. r. banavar , g. s. grest , and d. jasnow , phys . rev . b * 25 * , 4639 ( 1982 ) .